Book of Preprints

Advanced Grid-Connected
PV Systems
September 2011

INES CEA/LITEN
50 avenue La Léman – BP 377-73375 LE BOURGET DU LAC
Tél. :+33 4 79 44 45 46 - Fax : +33 4 79 68 80 49
E-mail: G-GRE-L2S.fr – jens.merten@cea.fr
www.liten.cea.fr 1/124
2/124
 INES
National Institute of Solar Energy

Situation of INES:

3/124
 FOREWORD

This book of preprints gathers some publications realized by the personnel of INES dealing
with advanced grid connected photovoltaic systems.

The first section of this book deals with heart for this kind of systems, which is the power
converter. Some publications deal with the inverters for the grid connected system.
However, novel concepts for DC grids and their converters are investigated, especially in the
view of optimizing their design in terms of the Global Energy Requirement Criterion, which
is determined by the primary energy consumption of the devices over their life cycle.
The second section deals with system monitoring and diagnostic tools, which allow the in-
situ detection of module failures without interruption of the system operation.

The third section is the largest one, reflecting the emphasis of activities on Energy
Management Systems, which minimize the power fluctuations on the grid. The prediction
of the PV system output is a key factor here. Energy Management is used for the Solar
Mobility, which is the recharge of electrical vehicles from grid connected PV-systems, which
is demonstrated at INES. Energy Management is also a key element for modern buildings
equipped with PV which are seeking to minimize overall energy consumption. For the
modeling of energy management algorithms of any PV application, a powerful modeling
platform has been created at INES.

The last section is dealing with the integration of storage systems, which allow provide a
controlled and predictable power injection into the grid. Again, demonstrators with are
operating at INES.

By the way, INES is the French National Institute of Solar Energy which has started
operations at the end of 2005. Actually, INES counts on 280 researchers from several French
institutions (Commissariat à l’énergie atomique et énergies alternatives (CEA), Centre
national de recherche scientifique (CNRS), Université de Savoie, and others). Their mission is
to carry out research activities and to develop innovative solutions to increase the use of solar
energy. The research activities cover the complete PV value chain ranging from the
purification and crystallization of the Silicon material over crystalline and organic cells and
modules, over complete systems and concentrator systems, up to the integration of PV into
buildings. There are also groups dealing with solar thermal systems. One section of INES
deals with the training of installers.

I hope that the reader finds some useful information from this collection of articles and will be
happy to provide further information if required.

Jens Merten
Head of Lab for solar systems (L2S) at INES

4/124
 TABLE OF CONTENTS

I - PV POWERPLANTS AND THEIR COMPONENTS

1.1 POWER CONVERTERS

Page
A TRANSFFORMELESS INVERTER AND RCD: WHAT’S THE PROBLEM?
T. Tran-Quo - H.Colin – C.Duvauchelle – B.Gaiddon – C. Kieny – C.Le Thi Minh S.Bacha - S. Assanou
– G.Moine – Y.Tanguy 11

PERFORMANCE EVALUATION OF PHOTOVOLTAIC GRID CONNECTED INVERTERS

B H. Colin – M. Vervaart – Y. Lausenaz – Y. Delesse 17

C DESIGN OPTIMIZATION METHODOLOGY FOR POWER CONVERTERS BASED

ON GLOBAL ENERGY REQUIREMENT CRITERIA. APPLICATION TO A DC-DC FLYBACK
STRUCTURE 21
C.Jaouen - B.Multon - F.Barruel

D WIRING DESIGN BASED ON GLOBAL ENERGY REQUIREMENT CRITERIA: A FIRST STEP

TOWARDS OPTIMIZATION OF DE DISTRIBUTION VOLTAGE
31
C.Jaouen - B.Multon - F.Barruel

1.2 PV SYSTEM MONITORING AND DIAGNOSTICS

A IN-SITU DIAGNOSTIC SYSTEM FOR PV MODULES AND PV SYSTEMS

G. Martin – F.Barruel 41

B EFFECTS OF SHADOW ON A GRID CONNECTED PV SYSTEM

N. Chaintreuil – F. Barruel – X.Le Pivert – H.Buttin – J.Merten 45

C PERFORMANCE ANALYSIS OF THE FIRST MEDIUM SIZE PV-PLANT IN BURKINA FASO ON

THE CITY GRID OF OUAGADOUGOU
51
M. Vervaart – J. Merten

D RELIABILITY OF LARGE-SCALE GRID CONNECTED PHOTOVOLTAIC SYSTEM

G.Zini – C.Mangeant – J.Merten 57

II – MANAGING THE FLUCTUATING OUTPUT

2.1 GUARANTIED PV: COUPLING WITH STORAGE SYSTEMS

A MANAGEMENT OF PHOTOVOLTAIC POWER GENERATION WITH A VANADIUM ReDox
FLOW ENERGY STORAGE
69
F.Bourry - P.Besson - X. Le Pivert

B PRATICAL IMPLEMENTATION OF GRID CONNECTED ENERGY STORAGE SYSTEMS

WITH LITHIUM-ION BATTERY: THE GROW-DERS PROJECT
73
H.Colin - X. Le Pivert - J.Merten - A.Barona - JF. Cousseau - P. De Boer-Meulman - G. Bloemhof - J.
Bozelie - H. Dietschman - G. Kourtis - M. Okasinki - E. Raaijen
C INNOVATIVE ENERGY STORAGE SYSTEMS FOR PV GRID-CONNECTED APPLICATIONS
H.Colin - J.Merten - A.Graillot – X. Valve – G. Sarre – A. Fedzin –
79
P. Gaillard - JP Smaha

5/124
 Page
2.2 ENERGY MANAGEMENT

A ENERGY MANAGEMENT STRATEGIES FOR OPTIMAL CHARGING OF ELECTRICAL

VEHICLES WITH PHOTOVOLTAIC PRODUCTION
87
H.Guillou, V.D. Cung, D.L. Ha, M. Jacomino

B HOW MUCH PV ENERGY WILL I PRODUCE TOMORROW? A FORECASTING TOOL WHICH

FITS THE FUTURE CONDITIONS ON THE FRENCH ELECTRICITY MARKET
93
S. Lespinat - F.Cugnet – X. Le Pivert

C DESIGN OF A MANAGEMENT AND SIMULATION’S TOOL FOR SOLAR CAR PARK

T.Vu – F.Barruel – J.Merten 97

D ENERGY MANAGEMENT SYSTEM FOR A PHOTOVOLTAIC GRID-CONNECTED BUILDING

D.L.Ha - S.Ploix - F.Wurtz - P.Perichon - J.Merten 101

E ADVANCED SIMULATION PLATFORM M2C FOR THE DEVELOPMENT OF HOME ENERGY

MANAGEMENT SYSTEM
107
D.L. Ha – B.Cinquin-Lapierre – P. Besson – F. Bourry

F OPTIMAL POWER MANAGEMENT FOR GRID CONNECTED PV SYSTEMS WITH BATTERIES

Y.Riffonneau – S.Bacha – Member IEEE – F.Barruel – S.Ploix 113

6/124
 I.
PV POWERPLANTS AND THEIR
COMPONENTS

1.1 POWER CONVERTERS

1.2 PV SYSTEM MONITORING AND

DIAGNOSTICS

7/124
8/124
1.1 POWER CONVERTERS

9/124
10/124
 TRANSFORMERLESS INVERTERS AND RCD: WHAT'S THE PROBLEM?

T. Tran-Quoc1, H. Colin2, C. Duvauchelle3, B. Gaiddon4,C. Kieny1, C. LE Thi Minh1, S. Bacha5, S. Aissanou5,

G. Moine6, Y. Tanguy6
1
- IDEA – G2elab, BP. 46, 38402 Saint Martin d’Hères, France
2
- INES/CEA, BP 332, 50 avenue du lac Léman, F-73377 Le Bourget-du-lac, France
3
- EDF R&D - 1, avenue du Général de Gaulle – 92141 Clamart Cedex - France
4
- HESPUL, 114 boulevard du 11 novembre 1918, F-69100 Villeurbanne, France
5
- G2elab, BP. 46, 38402 Saint Martin d’Hères, France
6
- TRANSENERGIE SA, 3D, allée Claude Debussy, 69130 Ecully, France

ABSTRACT: Considering the structure of PV systems, a stray capacitance can appear between the PV arrays and the
ground. When transformerless inverters are used, this capacitance can cause leakage currents to the ground.
According to the French standards, a Residual-Current Device (RCD) has to be installed at the AC side of the PV
installation, for the protection of individuals. Yet, when the value of the leakage currents reaches a threshold (30 mA in
homes in France), the RCD may switch off and unintended disconnections of the PV installation occur with
accompanying production losses.
Based on simulations and experimental tests, this project aims at giving information to PV system designers and inverter
manufacturers about the best suitable type of RCD to use for several PV system configurations.
This issue is relevant for countries with TT grounded networks like France where many operations of PV systems with
unexplained disconnections of RCD have been reported.

1 INTRODUCTION - Experimental: tests with different types of inverters

and RCDs to quantify the leakage currents and study the
When transformerless PV inverters are used, the stray disconnection actions.
capacitance between the PV arrays and the ground can
cause leakage currents to the ground (Fig. 1):
- These leakage currents flow from the connection of
PV panel

the PV structure to the ground (necessary for safety
reasons and to prevent lightning damages)
- The stray capacitance is formed from the module
electrically active layers and the surrounding metallic
structures [1], thus the capacitance magnitude will
depend on parameters such as the module surfaces,
the distance between the electric charges and metallic
structures, and the nature of the insulation material
- The capacitive current is created from this stray
capacitance and the alternating voltages of polarities
- Transformerless inverters do not isolate the DC from
the AC side, and allow the current to circulate via the
ground connections and through the inverter
According to the French standards, a Residual-
Current Device (RCD) has to be installed at the AC side
of the PV installation, for the protection of individuals.
Yet, when the value of the leakage currents reaches a
threshold (30 mA in homes in France), the RCD
disconnects the PV installation.
TT grounding systems and RCD are topics that have
not been explored in detailed so far. The purpose of this
study, which is part of a research project funded by the
French Agency for Environment and Energy
Management (ADEME), is to fully characterise
capacitive discharge currents that occur with
transformerless inverters, in order to determine which
type of RCD should be used to design safe and efficient
PV systems.
This paper presents modelling that have been
undertaken so far as well as details on experimental tests
that have been done with the intention to have a better
understanding on that issue, in particular:
- Theoretical: simulations investigating the influence
of different parameters on the leakage current
(topologies of inverters, types of PV modules,
resistance to the ground, ...)
Grid
Inverter
Ground
Figure 1: Schematic of leakage current according to the
grounding system in France
2 WHY TO USE RESIDUAL CURRENT DEVICES
For a PV system, it is necessary to take two different
measures of protection:
- Basic protection: protection against direct contacts,
- Protection against faults: protection in case of
isolation fault between a live conductor and the
ground (indirect contact).
In France, NF C 15-100 (or IEC 60479-1) standard
requires the installation of protective devices that will
cover those risks. The use of these devices will naturally
lead the neutral grounding scheme. For a TT grounded
networks (such as in France), this device is the basic unit
of the protection of persons, its use is mandatory to
ensure safety throughout the electrical installation.
The RCD is a protective device that monitors the
residual current resulting from the vector sum of currents
within conductors. In normal condition, the sum of the
currents of all conductors (phase + neutral + ground) is
zero [8].
The RCD is defined by the IEC 60755 international
standard which provides different types of protections, as
well as disconnection threshold or sensitivities:
- Class AC (sinusoidal alternating current)
- Class A (sinusoidal alternating current or pulsed DC

11/124
 component)
- Class B: this RCD is a device designed to protect
networks with rectified alternation and filtered by
capacitive load
- Class HI (High Immunity): RCD used for electronics
or for sensitive devices.
3 APPROACH
This project aims at giving concrete answers to PV
system designers and inverter manufacturers about the
best suitable type of RCD to use for several PV system
configurations. This issue is relevant for countries with
TT network like France where many operations of PV
systems with unexplained disconnections of RCD have
been reported. This causes a loss of production, loss of
financial benefit for the owner of the installation, loss of
confidence in the PV technology and most of all a safety
problem since system owners may be tempted to suppress
the RCD or increase the disconnection threshold to a
higher value, generating a potentially dangerous
situation.
This problem of leakage currents has already been
investigated in the literature [1-7]. Experimental tests
done within previous research projects such as [9]
concluded that the leakage current magnitude mainly
depends on the inverter typology but without any
recommendation concerning the type of RCD that should
be used to make PV systems perform well and safe.
First, simulations have been performed with the help
of EMTP-RV software in order to identify the factors that
influence the leakage currents. Different types of
inverters and modules have been designed in different
configurations. The influence of stray capacitance value,
capacitance distribution on both poles (DC+ and DC-)
and the influence of the resistance of the neutral
(impedance to ground) are also studied. Experimental
tests have been then conducted on real components.
Several types of inverters combined with different types
of RCD (AC, A, B…) will be tested in order to measure
the maximal admissible capacitance before the trigger
action of the RCD and then to determine the best suitable
type of RCD for each configuration.
Finally, a solution that allows to differentiate the
capacitance leakage current from the fault current is
proposed. This solution can be used to avoid the
undesirable disconnection of PV by RCD. Suggestions
and solutions for reducing the current magnitude are also
investigated.

T1: Full-bridge inverter, no DC/DC converter, T2: Full-bridge inverter, DC/DC boost converter,
without transformer without transformer
PV array DC/AC Inverter PV array DC/DC Converter DC/AC Inverter
+

+
+

?s

T3: Half-bridge inverter, DC/DC boost converter, T4: Full-bridge inverter, DC/DC forward converter,
without transformer with HF transformer
PV array DC/DC Converter DC/AC Inverter PV array DC/DC Converter DC/AC Inverter
+ +
D1

D3
+

Tr0_1

+
+

1 2
+
+

Tr0_2
1
+
+

1 2
?i
+

1
D2

D4

T5: Full-bridge inverter, DC/DC full-bridge T6: Full-bridge inverter, no DC/DC converter, with
converter, with HF transformer LF transformer
PV array DC/DC Converter DC/AC Inverter PV array DC/AC Inverter
R2
+
D1

D3

LF Transf.
Tr0_1
+
+

+
1 2 1 2

2
2
D2

D4

Figure 2: Different inverter models developed with EMTP-RV for studying the leakage current

12/124
 LV2 LV3 LV4 LV5
c
p1 p2 p1 p2 b p1 p2
N1 N2 a N1 N2 N1 N2
PI b PI PI
c

P
+

+
P
P

N
N
N

N
LV6
a
p1
N1 N2
p2
PV-3kW
PI L6b

PV
N
+

PVa
PVb
PVc
LV14

N
LV1

N
20 kV Network LV7 LV9
1 2 c c
+ p1 p2 p1 p2 p1 p2
N1 N2 N1 N2 N1 N2
20/0.42 PI PI PI

P
+

+
1

N
LV8 LV10 LV11
a
p1 p2 p1 p2 p1 p2 b p1 p2
N1 N2 N1 N2 N1 N2 N1 N2
PI PI PI PI

P
+

+
P

N
N
LV12 LV13
b a
p1 p2 p1 p2
N1 N2 N1 N2
PI PI
P

P
+

+
N

Figure 3: LV network with a single phase PV simulated with EMTP-RV

4.2 Simulations
4 MODELLING AND SIMULATION
The simulations show that for a 3 kWp PV system
4.1 Modelling using a transformerless inverter, the leakage current
could reach 39 mA (Table I and Fig. 4), according to the
In this part, several models of inverter were hypothesis defined.
developed in EMTP-RV in order to study the influence
of inverter typology on leakage current (see Fig. 2): Table I: Leakage current magnitude according to inverter
- Full-bridge inverter, no DC/DC converter, typology (Module capacitance: 1µF and R_earth = 30 Ω)
without transformer (Typology T.1)
- Full-bridge inverter, DC/DC boost converter, Topologies Leakage currents
without transformer (Typology T.2) (mA)
- Half-bridge inverter, DC/DC boost converter, T1 : Full-bridge inverter – no 38
without transformer (Typology T.3) DC/DC converter- transformerless
- Full-bridge inverter, DC/DC forward converter, T2 : Full-bridge inverter – DC/DC 39
with HF transformer (Typology T.4) boost converter - transformerless
- Full-bridge inverter, DC/DC full-bridge T3 : Half bridge inverter – 9
converter, with HF transformer (Typology T.5) DC/DC boost converter -
- Full-bridge inverter, no DC/DC converter, with transformerless
LF transformer (Typology T.6). T4 : Full-bridge inverter – DC/DC 0
Forward converter – HF
Then, a model of a real LV network with a PV transformer
system has been modelised (Fig. 3). This LV network is T5 : Full-bridge inverter – DC/DC 0
supplied by a 20/0.4 kV transformer of 160 kVA with 14 full-bridge converter –HF
buses, 10 loads and 1 PV system (3 kW). A three phase transformer
load is connected at bus 3 and single phase loads are T6 : Full-bridge inverter – no 0
connected to other buses. DC/DC converter – LF
transformer
The influence of stray capacitance values, capacitance
distribution on both poles (DC+ and DC-) and the
influence of the resistance of the neutral network
(impedance to ground) on the leakage current are studied.

13/124
 0.08 i_RCD 300
VDC_Neg
i_RCD_rms VDC_Pos
0.06 200

Voltage (V)
0.04 100
C u rren t (A )

0.02 0

-100
0
-200
-0.02
-300
-0.04
-400 900 910 920 930 940 950 960 970 980 990 1000
-0.06 t (ms)

-0.08 Figure 6: Voltage on DC+ and DC- of a transformerless

900 910 920 930 940 950 960
t (ms)
Figure 4: 39 mA leakage current obtained by simulation
for T2 typology
Results obtained from simulations indicate that the
leakage current strongly depends on the inverter
typology:
- for inverter with transformer, leakage current is
negligible (Topologies T4 to T6);
- for transformerless half bridge inverter, leakage
current is very small (9mA, T3) - So we note that
leakage currents are not necessarily linked with all
transformerless inverters, indeed a proper choice in
the inverter typology will strongly reduce this
current. This comes from the reduced voltage
variation amplitude of the polarities.
- for transformerless inverter, with a PV array
capacitance of 0,8 µF , the leakage current can cause
the trigger actions of the RCD (Table II).
970 980 990 1000
inverter
The leakage current does not depend on the
distribution of capacitance on both positive and negative
polarities (Table III). This current depends on the grid
grounding system resistance: in damp weather conditions
this resistance reaches very low values, and this can lead
to the disconnection of the RCD. In our simulations, this
variation is relatively small (10% variation when the
earth resistance varies from 1 Ω to 30 Ω). But sometimes,
this variation is significant and can lead to a
disconnection of the PV system. Simulations have been
carried out with a constant 0.7µF capacitance and with
several earth resistance values. While a 10 Ω earth
resistance doesn’t generate any critical leakage current,
an earth resistance value lower to 1 Ω generates a leakage
current upper to the limit of 30 mA (see Table IV).
Table II: Influence of PV module capacitance

Fig. 5 shows the result of the harmonic analysis C (µF) i_leakage (mA)
(FFT) for a wave form of the leakage current shown in
Fig. 4. It shows that the fundamental component (50Hz) 1 39
is very important while other harmonics are very small. 0.8 31
Fig. 6 shows the voltage on both poles of DC part of a
transformerless inverter (DC+ and DC- ; output of PV 0.6 24
arrays for T2 typology – see Fig. 2). We can notice that
we have an important fundamental component while this Table III: Influence of repartition of PV capacitance on
current is direct. The other harmonics are due to the RLC 2 polarities
circuit in the system. As the leakage currents have just
little DC components, standard AC RCD type should C (µF) i_leakage (mA)
correctly protect transformerless-based PV systems. VDC+ (0.5) & VDC- (0.5) 39
VDC+ (1) & VDC- (0) 39
VDC+ (0) & VDC- (1) 39

Table IV: Influence of the neutral earth resistance

(0.7µF)

R_earth (Ω) i_leakage(mA)

30 27
10 28
1 30

4.3 Conclusion of the simulations

The simulations show that in general cases, there is

no problem of leakage current; this is why the probability
Figure 5: Harmonics of leakage current of disconnection of RCD is very small.
In case where the leakage current is important
(> 30 mA), following solutions can be carried-out:

14/124
- Using inverters with transformer or transformerless
half bridge inverter
- Using PV mono or polycrystalline modules (these
modules with small tray capacitance)
- Using an advanced control mode inverter.
The results presented in Figure 8 show in this system
configuration and the available test conditions that:
- The leakage current value is rather proportional to the
inverter power (1 to 2 mA/kVA),
- Extrapolated value for nominal power lead to a
maximum of 9,4 mA for inverter #2:

5 EXPERIMENTAL TESTS Leakage currents versus power

7
Based on the theoretical results obtained so far,
several experimental tests have been conducted with real 6
system components (inverters and RCD) in order to have
a concrete evaluation of leakage currents and see how 5

RM S leakage current (mA)

RCD operate in a PV system.
4

Such experimental tests took place in a laboratory 3

facility equipped with different types of transformerless
Inverter 1
inverters available on the market and most common RCD 2 Inverter 2

types : AC, A, B, … (Fig. 7). Three types of Inverter 3

Linéaire (Inverter 2)
1
transformerless inverters are used for these experimental Linéaire (Inverter 1)
Linéaire (Inverter 3)
tests: 0
- Inverter # 1: the same as T1 in Fig. 2 0 500 1000 1500 2000 2500 3000 3500 4000
Inverter power (W)
- Inverter # 2: the same as T1 in Fig. 2 with H5
topology Figure 8: Leakage current versus inverter power
- Inverter # 3: the same as T2 in Fig. 2.
5.2 Behaviour of RCD
Tests are undertaken to give empirical answers on 3
topics : The objective of this experimental test is to analyse
- The RMS leakage current value for different the behaviour of most RCD types available in case of
transformerless inverters with regard to the power leakage current.
delivered, For this test, 4 different RCD types are connected in
- The behaviour of several types of RCD in case of string to see which RCD first trips for different levels of
capacitive leakage current, leakage currents (see Fig. 9).
- The impact of using a high immunity RCD on human The value of the leakage current is set by increasing
safety in case of direct contact with a live conductor. manually the capacitance value (see Fig. 7).

Figure 9: Several RCD types under test

First results show that:

Figure 7: Inverter under test with capacitance between
the positive and negative poles and the ground to
increase the leakage current
5.1 RMS leakage current
The objective of this experimental test is to quantify
the RMS leakage current value of three transformerless
inverters available on the market.
All inverters were supplied by a 5 kWp PV array of
polycrystalline type. Their nominal power is ranging
from 3,8 to 4,6 kVA.
- The order of disconnections of the RCD may vary
(according to the conditions of testing),
- RCD’s behaviour strongly depends on the inverter:
for instance, inverter #3 is very sensitive and
disconnects before any action of a RCD; for the two
others, the disconnection sequence is different from
each other.
5.3 Leakage current and human safety
The third test deals with human safety. The inverters
are tested with the most immunised RCD (i.e. the one
disconnecting for the highest value of current). A virtual

15/124
direct contact is simulated with the help of a resistance
placed in series with the variable capacitors (set to the
highest value before tripping), either on DC+ or DC-
pole, to check if the operation and tripping of the RCD
complies with human safety requirements.
Here as well, results depend on the inverter type: with
inverter #3, which detects the default, RCD trips in case
of direct contact and would thus perform its duty; with
inverters #1 and #2, RCD’s behaviour differs depending
on whether the resistance is connected on DC+ or DC-:
in a particular situation no action is observed, which
would endanger human safety.
6 CONCLUSIONS
This study aimed at giving some hints to PV system
designers and inverter manufacturers about the best
suitable type of RCD to use for several PV system
configurations.
This issue is relevant for countries with TT grounded
networks like France where many operations of PV
systems with unexplained disconnections of RCD have
been reported.
The simulations made within this project showed that
in general cases, there is no problem of leakage current;
this is why the probability of disconnection of RCD is
very small.
In case where the leakage current is important
(> 30 mA), following solutions can be carried out:
- Using inverters with transformer or transformerless
half bridge inverter
- Using PV mono or polycrystalline modules (these
modules have small stray capacitance)
- Using an advanced control mode inverter.
The first experiments clearly showed the complexity
of the study as results depend on inverter types and
testing conditions. They also emphasized the relevance of
this topic regarding production sustainability and safety.
In the continuation of the project, further
experimental tests will be carried out in order to complete
the first results.
7 REFERENCES
[1] J.C Hernandez, P.G. Vidal, A. Medina,
“Characterization of the insulation and leakage
currents of PV generators: Relevance for human
safety”, Renewable Energy, ELSEVIER, No. 35, 2010
[2] Lin Ma, Fen Tang, Fei Zhou, Xinmin Jin and Yibin
Tong, “Leakage current analysis of a single-phase
transformer-less PV inverter connected to the grid”,
IEEE Conference, 2007
[3] Roberto González, Jesús López, Pablo Sanchis, and
Luis Marroyo, “Transformerless Inverter for Single-
Phase Photovoltaic Systems”, IEEE Transaction on
Power Electronics, Vol. 22, No. 2, March 2007
[4] Roberto González, Eugenio Gubía, Jesús López, and
Luis Marroyo, “Transformerless Single-Phase Multilevel-
Based Photovoltaic Inverter”, IEEE Transaction on
industrial Electronics, Vol. 55, No. 7, July 2008
[5] Kerekes, T.; Teodorescu, R.; Borup, U
“Transformerless Photovoltaic Inverters Connected to the
Grid”, APEC 2007 conference, February 2007, Anaheim,
California USA
[6] Oscar Lopez, Francisco D. Freijedo, Alejandro G.
Yepes, Pablo Fernandez-Comesana, Jano Malvar, Remus
Teodorescu, and Jesus Doval-Gandoy, “Eliminating
ground current in a transformerless photovoltaic
application”, IEEE Transaction on power conversion,
Vol. 25, march 2010
[7] “Courants de décharge capacitifs - Remarques
concernant la conception des onduleurs sans
transformateur”, Information technique, SMA
[8] Schneider Electric, “La protection différentielle dans
les installations électriques basse tension”, Inter
Sections, magazine de Schneider Electric de
l’enseignement technologique et professionnel
[9] DISPOWER project, “Identification of general safety
problems, definition of test procedures and design-
measures for protection”, 2006
16/124
 PERFORMANCE EVALUATION OF PHOTOVOLTAIC GRID CONNECTED INVERTERS

H. Colin, M. Vervaart, Y. Lausenaz, Y. Delesse

CEA – DRT/LITEN/DSEN/GRETHE/LIS
CE Cadarache GRETHE/LIS bât. 351 13108 St-Paul-lez-Durance France
Tel : 33 (0)4 42 25 75 34 Fax: 33 (0)4 42 25 73 65 E-mail: herve.colin@cea.fr, mark.vervaart@cea.fr,
yvan.lausenaz@cea.fr, yves.delesse@cea.fr

ABSTRACT: In the framework of the European project NNE5-1999—00430 HIP-HIP (House Integrated
Photovoltaic – High-Tech in Public), dealing with the implementation of a large programme of installations of grid
connected PV systems in six European countries for a total of 3 MWp, and of national contract, the ADEME (French
Agency for Environment and Energy Management) has entrusted GENEC with making an evaluation of the
performance of typical grid inverters used in the project or commonly installed by professionals: validation of the
main characteristics of six inverters by determining their efficiencies (inverter efficiency), the quality of the signal
(power factor, harmonic distortion), their self-consumption, the protections (ac and dc disconnection detection) and
the precision of the displayed parameters (when applicable).The technical characteristics of these equipments are
validated by specific tests and compared to the manufacturers’ data. Some of the inverters are available in many
samples, thus differences of behaviour have been investigated as well.
Keywords: Evaluation, Grid-Connected, Inverter

1 INTRODUCTION The tests have permitted to cover a wide range of

devices from the small inverters for AC-modules (about
The development of photovoltaic grid-connected 100 Wp) and devices of intermediate power (1 and 1.5
systems has been very important over the last years in kWp) up to high-powered inverters (2.5 kWp). The
developed countries such as Japan, the US and Western impact of temperature on the inverter efficiency has also
Europe. been studied.
In Europe, Germany, mainly, but also the
Netherlands, Italy and Spain have set large programmes 3 INVERTERS TESTED
of solar roofs. In France, due to the high cost of systems
and the rather low price of photovoltaic electricity sale, The various inverters tested are presented in the
this field is only in its infancy. following Table 1:
In order to make this sector more dynamic, in
addition to more attractive selling prices and national Table 1: Inverters tested
subsidies (ADEME, local organizations), France has Pdc (W) Vdc (V) Vac (V)
actively taken part to the European project NNE5-1999— Inverter 1 2600 125-500 198-251
00430 HIP-HIP (House Integrated Photovoltaic – High- Inverter 2 1650 120-300 230 (+10/-15%)
Tech in Public), dealing with the implementation of a Inverter 3 90-140 24-40 207-244
large programme of installations of grid connected PV Inverter 4 1500 139-400 198-251
systems in six European countries for a total of 3 MWp , Inverter 5 1000 80-190 230 (+10/-15%)
and keeps on promoting photovoltaic grid-connected Inverter 6 2200 125-400 230 (+10/-15%)
systems.
In this framework, the laboratory has been entrusted 4 TEST CONFIGURATION
with making an evaluation of the performance of typical
grid inverters commonly installed by professionals [1]. 4.1 Testing bench
The inverters are installed in the test lab and are
2 SCOPE AND OBJECTIVES supplied with energy by photovoltaic modules placed
outside. The electricity is delivered to the grid via an
2.1 Objectives isolation transformer.
The purpose of the study is to do some tests on
typical inverters that may be installed in grid-connected
systems in order to analyse their characteristics and
technical performance:
• Inverter efficiency versus power,
• Quality of the signal,
• Self-consuption,
• Protections against ac and dc disconnections,
• Accuracy of the displayed parameters,
• Productivity.
The objective of the study is to provide a more
“service oriented approach”, with guaranty of results in
terms of power output and power quality. Figure 1: Schematic of the testing bench
2.2 Scope Measurements (see Figure 2) are made on both AC and
DC sides:

17/124
• DC voltage, current and power, ripple voltage, • The behaviour of the inverter during the start-up
• AC voltage, current, power and harmonics, power phase.
factor.
5 RESULTS

5.1 Main characteristics

• Efficiency: measurements were made at an indoor
temperature of 20 °C; a time delay of at least 2
minutes is observed between each measurement of
the power. Curves of the efficiency are presented in
Figure 4:

Figure 2: Testing bench

The photovoltaic are set on a movable supporting

structure enabling to shift the inclination to make the
power vary (see Figure 3).

Figure 4: Efficiency of inverters

Maximum and European (IEA*) values are shown in

Table 2:

Table 2: Inverter efficiencies

ηmax (%) ηEU (%)
Inverter 1 94.8 93.5
Inverter 2 92.5 91.4
Inverter 3 91.9 89.9
Inverter 4 91.8 90.9
Inverter 5 90.8 89.2
Inverter 6 93.2 87.8
Figure 3: PV modules
*The European efficiency is defined by the formula:
4.2 Test methodology ηEU = 0.03 ηps 5 + 0.06 ηps 10 + 0.13 ηps 20 + 0.10 ηps 30 +
The various tests intend to validate the main 0.48 ηps 50 + 0.20 ηps 100
characteristics of the inverters:
• Efficiency: evaluation of the ratio Pac/Pac nominal In general, measured values are about 1 % below
versus the ration Pac/Pdc, the values given by the manufacturers.
• Quality of the signal: the non linearity of the signal
generated may cause some disturbances in the • Quality of the signal: the power factor is checked to
operation of some appliances and premature ageing. be higher than 0.85 as soon as Pac > 25 % Pac nom;
The power factor and the harmonic components of current harmonics are measured at 25, 50 and 100 %
the current are measured, Pnom (components should not be above given
• Self-consumption: measurement of the ac power thresholds) and the harmonic distortion rate is
consumed from the grid when the inverter is off (no measured at several power levels (see Table 3).
dc/ac conversion): sleeping and stand-by modes.
Measurement of the minimum required voltage for Table 3: Inverter power factor (Prated=25 %) and current
operation, harmonic distortion rate (Prated>50 %)
• Protections: detection of loss of dc supply or loss of Power factor distortion rate (%)
ac line; measurement of the disconnection time, Inverter 1 0.98 4-6
• LCD screen display accuracy: comparison of values Inverter 2 0.95 10-15
displayed and measurement to evaluate the precision. Inverter 3 0.85 10-15
Inverter 4 0.99 3-6
Extra tests have also been performed to analyse: Inverter 5 0.99 10-20
• The productivity of the system (PV modules + Inverter 6 0.87 4-10
inverter): efficiency versus the daily irradiance
• The influence of the input voltage level, Figure 5 shows inverter 2’s current THD:
• The ripples on the dc input signal,

18/124
Figure 5: Current THD
About the power factor, all the inverters comply
with the Standard requirements. Concerning the
distortion, values are higher than expected thus
leading to a certain signal deformation.
• Self-consumption: ac power consumption is
measured in sleeping mode (no dc/ac conversion) and
in stand-by mode (MPPT inactive), see Table 4.
Figure 6: dc disconnect
5.2 Extra tests
• Productivity of the system: the amount of energy
produced by two of the systems (PV modules +
inverter) has been recorded during several days to
evaluate the efficiency in real use (see Figure 7):

Table 4: Inverter self-consumption

Sleeping mode (W) Stand-by mode (W)
Inverter 1 0.10 1.5
Inverter 2 0.33 2.4
Inverter 3 0.16 0.16
Inverter 4 1.24 1.5
Inverter 5 1.25 1.25
Inverter 6 0.62 4.73

Values are higher than the ones given by the

manufacturers; consumption doesn’t seem to be
related to the inverter power.
Figure 7: Daily efficiency of system
• Display accuracy: only two of the six inverters had
such a display (see Table 5). In real use, the efficiency is close to the European
efficiency calculated for the inverter or even
Table 5: Inverter disconnect time slightly higher (inverter 6).
Vdc (%) Vac (%) Idc (%) Iac (%)
Inverter 2 < 0.7 < 2 < 10 • Influence of the dc voltage level: performance of the
Inverter 6 < 0.3 < 0.7 <5 < 15 inverter depends on the voltage level as it can be seen
in Figure 8:
Voltage indications are rather precise, current
indications are less accurate.

• Protections: the behaviour of the devices has been

tested in the case of loss of dc supply and loss of ac
line at 15 and 100 % Pnom. The devices should stop
operation within 200 ms.
Table 6 shows the results at 100 % Pnom:

Table 6: Inverter disconnect time

dc disconnect (ms) ac disconnect (ms)
Inverter 1 40 80
Inverter 2 40 90 Figure 8: Influence of voltage on efficiency
Inverter 3 25 20
Inverter 4 80 60 • Ripples on dc input signal: a large amplitude of the
Inverter 5 160 60 fluctuations of the dc voltage shows the bad ability of
Inverter 6 100 50 the MPPT in the search of the maximum power point
Figure 6 shows the impact on the ac signals when the dc (see Table 7).
supply is disconnected (100 % Pnom, disconnect at t=70
ms): Table 7: Inverter Vdc ripple

19/124
 Maximum Vdc ripple (%)
Inverter 1 3.3
Inverter 2 4
Inverter 3 21
Inverter 4 4.1
Inverter 5 19.4
Inverter 6 13

Figure 9 presents the Vdc ripple (Vdc average/Vdc

max) of inverter 3:

Figure 11: Efficiency of 3 samples (zoom)

6 CONCLUSIONS

6.1 Main conclusion

Figure 9: Ripples on dc voltage
• Behaviour during start-up phase: procedures are
miscellaneous. In some inverters, start-up begins with
the charge of an internal capacity (see Figure 10);
some make a grid impedance measurement. The delay
for the start-up of the MPPT ranges from 3 to 50
seconds for the inverters tested.
Results show that the devices are in accordance with
the international standards concerning the safety
(disconnection delay in case of loss of PV supply or AC
line) and power quality issues (power factor and current
harmonic distortion). Measured values of the static
efficiency seem always to be slightly lower (about 1 % or
more) than indicated (data provided by manufacturers).
About the current harmonic distortion for powers higher
than 50 % Prated, measurement values are higher than
the values given in the technical leaflets. Concerning the
consumption of the devices in sleep and stand-by modes,
results depend on the tested inverter, but always seem to
be higher than expected.

6.2 Harmonisation
In the field of electrical requirements, among all the
national and international codes or standards, the
comparison of the performance of miscellaneous
inverters is a contribution to elaborate common
recommendations for a European harmonisation.

6.3 Further work

Figure 10: Start-up sequence
5.3 Sampling test
Some of the inverters were available in many samples
and a comparison of the test results enable to check if
there is no appreciable variation. Figure 11 shows the
efficiency of three samples of inverter 4:
Some complementary test are planned to be done
dealing with:
• The influence of the inverter temperature on
performance,
• The MPPT efficiency,
• The influence of the PV cell type on performance,
• The dynamic efficiency.
The study is also going to be completed by an
analysis of the whole system (PV modules and inverter)
operation to evaluate the efficiency and energy produced
in accordance with the system configuration
(architecture, voltage, etc) and the interferences between
several inverters when they are simultaneously connected
to the grid.

Acknowledgment
The authors wish to thank the ADEME for co-
financing this work. References

[1] H. Colin, M. Vervaart, Tests d’onduleurs pour le

raccordement au réseau électrique de modules
photovoltaïques, technical report DTEN/DR/2003/45.

20/124
 Design optimization methodology for power converters based on global
energy requirement criteria. Application to a DC-DC flyback structure

Cédric JAOUEN(1,2), Bernard MULTON(2), Franck BARRUEL(1)

CEA-INES(1)
50 Av. du lac Léman - Le Bourget du lac, France
SATIE, ENS Cachan Bretagne, CNRS, UEB(2)
Av. Robert. Schuman - Bruz, France
Tel.: +33 / (0) – 4.79.44.46.89
E-Mail: cedric.jaouen@cea.fr, bernard.multon@bretagne.ens-cachan.fr,
franck.barruel@cea.fr
URL: www.cea.fr, www.satie.ens-cachan.fr

Keywords
« DC power supply», « Design», « Life Cycle Analysis (LCA)», «Device modeling».

Abstract
This paper discusses a design methodology for power electronic converters according to the Global
Energy Requirement (GER) criterion, i.e. the primary energy consumption over their entire life cycle.
For given specifications of the converter, each power component of a self-oscillating flyback converter
and the main control parameters are optimized using scaling laws for given technologies. Models
linking sizes of components, loss parameters and embodied energy parameters are presented. Finally a
comparison is realized between a GER-designed converters and a classical one (designed from thermal
criteria) which concludes to an improvement of efficiency for a given consumption profile.

Introduction
Nowadays, converters are currently designed to present the best cost and/or volume under thermal
constraints, or to present the best efficiency (at rated power level) under cost and/or volume
constraints [1]. However, we think that optimizing the design should be done on the overall life cycle
by taking cumulated losses and embodied energy into account. More and more manufacturers
communicate about their environmental friendly product through an Environmental Product
Declarations (EPD) [2]. This EPD is based on a Life Cycle Assessment (LCA) which evaluates all the
environmental impacts caused by their product during their whole life cycle, i.e from raw material
extraction to its recycling or disposal. Contrary to the LCA approach which is an a posteriori
demarche, this paper proposes to eco-design converters in order to take its environmental impacts
(here, only its global primary energy consumption) into account a priori. Such demarche has already
been used to design transformers and single phase induction motors in [3]. The proposed approach
takes place in the building DC distribution scheme context, but the methodology could easily be
extended to other ones. Indeed, considering the DC distribution scheme appears to be more and more
relevant due to the development of distributed renewable energy sources, which may produce direct
current, and most existing electric loads could be powered with direct currents [4]. Moreover,
electronic loads seem to be at present ready to be DC-powered because they intrinsically work with
direct current; consequently, we choose to consider their power supply first. As shown in [4], the DC-
DC stage of electronic load power supply is widely made with flyback converters controlled in self-
oscillating mode. That is why this article focuses on a flyback DC-DC converter in order to develop
the design optimization methodology of power converters with environmental concern in mind.

First, we will explain our overall approach for the design optimization of converters. Then we will
expose the models used to assess the converters environmental impacts. Based on these models, we
will carry out a parametric study to evaluate the impact of component dimensions and control

21/124
parameters (frequency, duty cycle) on the converters impact. Finally, we will compare an eco-
designed flyback converter with a “classically designed” one (components sized according to thermal
constraints and control parameters chosen to minimize the losses cumulated overall life).

I. Converter design approach and hypothesis

1. Energetic impact and methodology
We propose a converter design methodology according to its environmental impacts. However, a full
LCA approach focuses on numerous impacts, from primary energy consumption to Emissions to
water, air or land etc... Since directives, like REACH and RoHS, already regulate the use of chemical
product in order to limit the emissions, we propose to focus only on the primary energy consumption
impact also called the Global Energy Requirement (GER). The GER represents the amount of primary
energy consumed by the system over its entire life cycle, i.e. the losses during its use phase (converted
into primary energy using the efficiency of electricity production) and the embodied energy related to
the other steps of its life cycle. The losses are linked to the component loss characteristic parameters
(like the Equivalent Series Resistance (ESR) for capacitors) which are related to the component
dimensions. The embodied energy also depends on component size. For instance, increasing the
component sizes generally induces a decrease of its losses, but an increase of its embodied energy, and
vice versa. Therefore, we have chosen to modify the dimensions of a reference component using
homothetics factors (Hmag, Hdiode, Hcapa, Hswitch) in order to find the optimal component sizes which
minimize the flyback converter GER, as shown in Fig. 1. These homothetic factors will be applied to
transformer and capacitor volume (VCore; Vcapa) whereas it will be applied only to diode and MOSFET
area (Adiode, Aswitch).

Fig. 1: Design optimization method based on scaling laws and scheme of the considered Flyback
(component represented in dotted line are not taken into account)

Moreover, the losses being related to the control parameters (FSw, α) too, we propose to study their
effect on the flyback converter GER. Since those parameters have an influence on components
constraints we choose to consider one reference component for one set of control parameters. For each
component, a reference component will be choosen according to the following criteria:
• For magnetic component, the reference component corresponds to the smallest component
respecting the thermal constraints (θcopper<120°C, θferrite<100°C).
• For semiconductors, the reference component is the component for which breakdown voltage
and current rating are equal to voltage and current constraints.
• For filtering capacitors, the reference component is chosen according to the maximum
acceptable voltage ripple and acceptable current rms value.

22/124
All those reference components are defined at the rated power level and will represent the unitary
homothetic factors.

Fig. 2 represents the proposed synopsis to design the flyback converter which minimizes its GER for a
given set of control parameters. For each set of control parameters, the constraints on the components
allow us to determine the reference components dimensions using a pre-sizing step at the rated power
level. Then, the homothetic factor modifies those dimensions and changes its losses characteristic
parameters and its embodied energy. Finally we use an optimization algorithm to find the optimal
homothetic factor (according to the consumption profile and the life span) for each control parameters,
considering the following hypothesis:
• the voltage drop effect from one component to another is neglected
• the flyback is controlled in self-oscillating mode

We can then consider that the optimal flyback converter is made of components which have been
optimized independently. All the relations which are necessary to determine dimension from electrical
constraints and losses characteristic parameters from dimension are presented in the next chapter.

Fig. 2: Eco-design synopsis for flyback converters

2. Scope of the study

As presented in Fig. 1, we propose to design a flyback converter controlled in self-oscillating mode.
Fundamentally, a flyback DC-DC converter is mainly composed of an inductor, a switch, a diode and
several capacitors. Moreover, snubbers and heatsinks are currently added to the switch and the diode.
However, for this first study, these last elements are not taken into account. We considered
multicellular planar MOSFETs for the switch, Schottky diodes, aluminum electrolytic capacitors and
inductor realized with two “E core” made with N97 ferrite materials.

Only use
phase

Charge
and
use
phase

Fig. 3: (Left) Example of assumed daily load profile and its statistical distribution (Right)

23/124
As a first example, we propose to consider the case of a laptop power supply. In this case, the flyback
converter will be optimized according to a 70W rated power level, a 20V output voltage level and a
320V input voltage level (output voltage level for a diode bridge rectifier connected on the French
utility grid). Moreover, we will arbitrarily consider a 5 years life span with a 90 minutes daily use
according to a measured profile (divided in two phases: use and charge and only use) presented Fig. 3

II. Evaluation of flyback converters GER: losses and embodied energy

models
1. Magnetic component GER models
Losses inside magnetic component (inductor with two coils) are divided into copper losses and
magnetic losses in the core. In this study we have neglected the magnetic losses in the windings.
Ferrite materials magnetic losses are documented inside datasheets by curves representing losses by
unit of volume versus induction and frequency for sinusoidal induction waveforms. However, the
induction is not sinusoidal inside flyback converters. [5] proposes a model for triangular induction
based on Double Natural Steinmetz Equation (DNSE). The DNSE coefficient is determined from
datasheet curves with an average accuracy of 6%. This model is established according to 18 points
with an induction from 12mT to 200mT and a frequency from 100kHz to 1MHz. We give the obtained
model for the N97 ferrite materials at the equation (1) where D is the duty cycle and Pr is a reference
volume losses for a reference induction Br and frequency fr extracted from the datasheet.
β α γ
f ⎛ Bˆ ⎞ ⎛ f ⎞ ⎛ Bˆ ⎞
( )
Pvol f , Bˆ = Pr × δ × × ⎜
f r ⎜⎝ Bˆr ⎠ ⎝ f r ⎠ B
⎝ ⎠
r
(
⎟⎟ + Pr × (1 − δ ) × ⎜ ⎟ × ⎜⎜ ˆ ⎟⎟ × 2 × D + (1 − D )
α 1−α 1−α
)
× K (α )

1 (1)
K (α ) = 2π
α
( 2π )α × ∫ cos (ϑ ) dϑ
0

with α = 2.31 β = 3.4 γ = 2.15 δ = 0.19 Pr = 300kW / m3 Br = 200mT f r = 100kHz

Now, for the evaluation of embodied energy, we consider a transformer made by ferrite and enamel
coated copper wire wound on a PVC coil former (due to lack of data for other glass reinforced
polymer materials). Our evaluation is based on following data:
• [6] estimates at 12.6 kWhp/kg the amount of primary energy necessary to produce one kg of
copper wire without insulation.
• [7] estimates at 3.6 kWhp/kg the amount of primary energy needed to coat one kg of wire with
insulating enamel.
• [8] estimates at 8 kWhp/kg the amount of primary energy to produce one kg of ferrite material.
• [9] estimates at 16.5 kWhp/kg the amount of primary energy necessary to produce one kg of
PVC.

Since losses and embodied energy are linked with the magnetic component volume, the homothetic
factor will be applied on all core dimensions. Therefore, the component volume increases with the
cubed homothetic factor. The ratio between all core dimensions will be then considered as constant
and based on a reference magnetic circuit E10/5.5/5 from EPCOS. Only the air gap and the number of
turns (primary and secondary coils) will be adjustable regardless of other dimensions.

2. Semiconductor component GER models

For semiconductor component losses, due to ZCS (due to self-oscillating mode), diode switching and
MOSFET turn-off switching losses will be neglected. The loss assessment is based on the hypothesis
proposed in [10] for the MOSFET switching losses and on classical expression for the conduction
losses inside both diode and MOSFET. All these expressions are given in equation (2).

24/124
 2 1 ⎛ BVds × I sw _ max × QG ⎞
Psw _ cond = Rdson × I sw Psw _ com = ⎜ ⎟ × f sw
_ rms
2⎝ IG ⎠
(2)
Pdiode = V f 0 I D + RF I D2 _ rms

For the gate current IG, we choose to consider a value of 1A independent from the current rating, the
breakdown voltage and the control parameters. Other scenarios have been studied but are not
presented here. The drain source resistor Rdson and the total gate charge QG will be calculated with
relations associating the Factor Of Merit (FOM in Ω.C) and the specific resistance (Ω.m²) with the
breakdown voltage BVds [11]. This relation and the one which links the die area with the breakdown
voltage and current rating were established according to the HyperFET IXYS MOSFET Family [12].
This family contains MOSFET with breakdown voltage from 200V to 1200V and current rating from
3A to 180A. Those relations present respectively a 16%, 10% and 10% average accuracy for 31
references and are given in equation (3).
FOM (mΩ.nC ) = 0.3 × BVds1.83 (V ) Rdson * S Switch (mΩ.mm ²) = 0.086 × BVds1.95 (V )
(3)
S*Switch (mm ²) = (5 × 10−6 + 5.7 × 10−3 × BVds (V )) I max ( A)
Concerning the diode, the threshold voltage Vf0 is considered constant, whatever the current rating IF,
with a value of 0.1V. The dynamic forward resistor RF has been linked with the die area and the
breakdown voltage VRM. The die area is associated with the current rating IF. This relation was
established according to the Vishay Shottky diode catalog containing 37 references [13]. Those
relations present respectively a 26% and 15% average accuracy and are given in equation (4).
0.527
34 × VRM (V )
S Diode (mm ²) = 0.68 × I F ( A) RF (mΩ) = (4)
S Diode (mm ²)
The assessment of embodied energy for semiconductor is based on data coming from [14]. This data
(EGTO220) estimates at 522 Whp the amount of primary energy which is necessary to produce one
transistor encapsulated inside a TO220 package. Other data coming from a semiconductor
manufacturer (MOSFET on 6’’wafer) allow us to estimate at about 400 Wh the cost of final wafer
processing (for a 20 mm² area and 10 mils thickness die). Then it appears that the embodied energy for
MOSFET is mainly due to wafer processing. For a given MOSFET or diode die area, we divide it with
the maximum die area acceptable inside a TO220 package (202 x 150 mils) [15] and we apply the
result to the data EGTO220, as shown equation (5).
S Diode S Switch
EGDiode = EGTO 220 EGSwitch = EGTO 220 (5)
STO 220 STO 220
Since the die thickness remains constant for a given series of components, we have chosen to apply the
homothetic factor only on the die and case set. Therefore, the component area increases with the
squared homothetic factor.

3. Electrochemical capacitor GER models

The losses of power capacitors filtering the input and output voltage are represented by the Equivalent
Series Resistance (ESRin, ESRout) according the expression (6) where IC_in_rms and IC_out_rms represent
the RMS value of the capacitor current.
Pcapa _ in = ESRin I C2 _ in _ rms Pcapa _ out = ESRout I C2 _ out _ rms (6)

Studying the B43851 and B41851 electrolytic capacitor series from EPCOS manufacturer [16], allows
us to associate capacitor ESR with capacitor volume and breakdown voltage. The models presented in
equation (7) give those two relations; the first one is valid for small voltage level corresponding to
B41851 series (6,3V Æ 100V) and the other one for higher voltage level (160V Æ 450V). The first
one will be apply for output capacitor and the second one for the input capacitor. Those relations
present respectively an average accuracy of 12% and 18%.

25/124
 −1.09
ESRout = 85 × U out 0.60 × Volout ESRin = 6.1 × U in1.43 × Volin−1.2 (7)
The assessment of embodied energy is based on [17], which estimates at 261 kWhp/kg the primary
energy needed to produce one kg of through-hole mounting capacitor. This analysis of electrolytic
capacitor series allows us to set an average density of 1460 kg/m3 for “low voltage” capacitor and 722
kg/m3 for “high voltage” capacitor.

Since ESR and embodied energy are linked with the capacitor volume, we have chosen to apply the
homothetic factor on the capacitor volume. Therefore, the capacitor volume increase, like the magnetic
component volume, with the cubed homothetic factor.

III. Homothetic factor effects on the magnetic component GER

This section brings some results about the effect of the homothetic factor on the magnetic component
GER. For this analysis, the control parameters will be fixed, as a first example, to 100kHz for the rated
switching frequency and 0.5 for the duty cycle. Fig. 4 represents the methodology for the evaluation of
the magnetic component GER for given control parameters and homothetic factor. A pre-design is
made for a given set of control parameters at the rated power level. This pre-design determines the
reference magnetic component which corresponds to the smallest core respecting the thermal
constraints (a simplified steady state thermal model is used to compute the ferrite and copper
temperature). Then, for each evaluation of homothetic factor, the dimensions and the loss
characteristic parameters are computed. However, those elements depend on the air gap value which is
optimized in order to minimize the losses on the case study power profile (highlighted part of
synopsis). The range of air gap will be limited by a maximum flux density constraint. It is important to
notice that, due to the choice of the self-oscillating mode, the switching frequency change with the
power level (as shown in Fig. 4), unlike duty-cycle which is constant according to the assumptions on
voltage losses, for the evaluation of losses during use. Finally the GER is evaluated according to the
component optimized dimensions and the corresponding losses.

Fig. 4: (Left) Magnetic component eco-design methodology (Right) Ratio between switching
frequency and rated switching frequency versus time for the Fig. 3consumption profile described in
Fig. 3 and Evolution of magnetic component GER according to the homothetic factor for given control
parameters (FSw=100kHz & α=0.5)

26/124
The bottom right graph of Fig. 4 represents the GER evolution according the homothetic factor value.
We can first remark that an optimal homothetic factor exists due to the decrease of losses and the
increase of embodied energy with the increase of homothetic factor. We can also observe that this
optimal homothetic factor is bigger than one. In other words, for our assumption on load power profile
and life duration, the optimal magnetic component is bigger than the smallest one respecting only the
thermal constraints. It is important to notice that longer the use duration is, higher the optimal
homothetic factor is, and inversely.

Fig. 5: Repartition inside GER, losses in use and embodied energy for the optimal homothetic factor

Fig. 5 represents the repartition of losses and embodied energy for the GER, as well as the repartition
of copper and magnetic losses during use, and finally the details of the embodied energy for the
optimal homothetic factor corresponding to this example. This repartition will be the same for other
case studies, on condition that the optimal air gap is different from the minimum one imposed by
induction constraints and thermal constraints. For the embodied energy repartition, we can see that the
wiring (composed by copper and enamel) will be preponderant comparing with the core and the coil
former, even if the reference component imposes a copper volume smaller than the core one.

For the other components, the approach is the same as the one presented for the magnetic component.
A pre-sizing has been carried out at the rated power level, and then homothetic factors modify the
main component dimensions to assess cumulated losses during use and embodied energy for
evaluating its GER. The same study realized for the different components points out that they present
an optimum GER for a scaling factor upper than one. In other words, according the GER point of
view, it is greater to oversize the components compared to a basic sizing method which is based on the
thermal and electric constraints. However, it’s important to notice that longer the running duration is,
bigger the homothetic factor is and inversely.

IV. Influence of control parameters (FSw, α) on flyback GER

As we have already explained, for the different set of control parameters, the homothetic factor of each
component will be optimized independently. Fig. 6 represents the evolution of the flyback converter
GER, according to the rated control parameters. We can remark that an optimal set of rated control
parameters minimizes the converter GER. The optimal parameters are 30 kHz for the rated switching
frequency (at the rated power point), and 0.35 for the duty cycle. For these parameters the optimal
GER is 23 kWhp.

In order to explain the location of the optimal set of parameters which minimizes the converter GER,
we have analysed the position of optimal sets of parameters for each component independently, as
shown in Fig. 6. The fact that the minimum GER is independent from the frequency for the diode is
due to assumption of neglecting diode switching losses. For the capacitors, we have neglected the ESR
variation according to the frequency, and then the minimum GER is independent from the frequency
too. In fact, the frequency modifies the magnetic and switching losses, whereas the duty cycle alters
the current rms values and the rated voltage constraints. We can also conclude that only the transistor
and the magnetic component have an influence on the rated frequency choice. However, the transistor
GER is bigger than the magnetic component one for each own optimal set of control parameters. Then
the flyback optimal frequency tends to the MOSFET optimal frequency. For the duty cycle, it tends to

27/124
the closest value to the one which minimizes the GER of the most GER-important components which
are input capacitor, diode and MOSFET.

Fig. 6: (Left) Evolution of the flyback converter GER according to the control parameters. (Right)
location of optimal set of control parameters for each component independently and for the flyback
converter

V. Comparison between an eco-designed flyback converter and a

“classical” one
As a first comparison, we propose to compare the eco-designed flyback with a “classical” one. For the
classical one, the control parameters have been optimized in order to reduce the cumulated losses. For
each control parameters, the components have been chosen according to thermal constraints using the
pre-sizing step. Table I gives the main characteristics of the two flyback converters. Firstly we can
remark that the optimal control parameters with a classical sizing method are 76 kHz for the rated
frequency and 0.37 for the duty cycle. Since classical converter efficiency takes only the rated point
into account, we have compared the two converters according to their energetic efficiency. The
energetic efficiency is defined as the ratio between the load energy need and the energy which has
been really provided to the system for a given power profile. The “classical” converter presents an
energetic efficiency of 88% with the Fig. 3 power profile and a GER of 87 kWhp, whereas the eco-
designed one presents an energetic efficiency of 97% and a GER of 23 kWhp. In fact, an eco-designed
converter tends to increase the component dimensions, as we have already noted, and then reduces
component losses.

Now look at the GER for the two converters. Fig. 7 represents the repartition of GER, cumulated
losses during use and embodied energy corresponding to each component of the classical converter.
We can remark that the losses during use are the main contributor inside the GER. They are
approximately equally shared among the transistor, the diode and the capacitors. For the embodied
energy, the output capacitor is dominant compared to the input capacitor due to the difference of mass
density and volume. The semiconductor components are negligible compared to the other ones, due to
the low mass of silicon relatively to the other component mass.

Fig. 7: Classical design case: component contributions inside flyback GER, embodied energy and
cumulated losses

28/124
Fig. 8 represents the repartition of GER, losses in use and embodied energy, between the different
components of the eco-designed converter. Since the components are bigger in the eco-designed
converter than in the classical one, the proportion of embodied energy inside the GER is bigger. In this
case the GER is approximately equally shared between all the components. Inside the embodied
energy, the diode part is not negligible due to an important increase of die area comparing with the
pre-sizing diode; however the output capacitor is the main contributor again.

Fig. 8: Eco-design case: component contributions inside flyback GER, embodied energy and
cumulated losses
Table I: Main characteristics of two compared flyback converters

Eco-designed flyback converters Classically designed flyback converters

Rated switching
frequency / duty 30 kHz / 0.35 76kHz / 0.37
cycle

GER / on cycle
23 kWhp / 97% 87 kWhp / 88%
efficiency

Components Losses characteristic parameters & dimensions

Magnetic e=0.9mm / Rprim=0.37Ω / Rsec=7 mΩ / e=0.92mm / Rprim=0.67Ω / Rsec=11 mΩ /

component N1=150 / m=0.12 / AeAw=17400 mm4 N1=157 / m=0.11 / AeAw= 2514 mm4

Rdson=1.9 Ω / Qg=13 nC / Ids=2.2 A / Rdson=11 Ω / Qg=2.4 nC / Ids=0.5A /

Transistor
Ig=1A / Adie=8 mm² Ig=1A / Adie=1.5 mm²

Diode If=52A / Rf=8 mΩ / Adie=36mm² If=3.4A / Rf=120 mΩ / Adie=2.3mm²

Input capacitor ESR=1.6 Ω / Vcapa=10 cm3 ESR=22.5 Ω / Vcapa=1 cm3

Output
ESR=46 mΩ / Vcapa=7cm3 ESR=180 mΩ / Vcapa=2.5cm3
capacitor

Conclusion
In this paper, we have proposed a new methodology in order to size converters based on the
minimization of the primary energy which will be necessary along its whole life cycle. For this first
approach of this very complex problem, the design problem has been greatly simplified and we are
well aware that it would need more complex models. Indeed, this methodology uses conventional loss
models associated with incomplete LCA data to assess the global energy requirement (GER) of a
flyback converter. Since cumulated losses and embodied energy are generally linked with component
dimensions, we have proposed models which link loss characteristic parameters, component
dimensions and voltage and current ratings. Applying our approach to a flyback converter, we have
shown that an optimal set of control parameter which minimizes the converter GER exists. Finally, we

29/124
have compared this eco-designed converter with a classically designed one, and we logically
concluded that the eco-designed converter presents a better energetic efficiency in addition to a better
GER. It is important to notice that the results, presented in the last section, depend on the assumptions
which have been made on the power profile and the use duration. Despite the needs of assumption on
the consumption profile and use, we think that our global energetic approach should be applied to any
energetic system to reduce its life cycle primary energy consumption. This approach could be easily
used with more complex losses models, and extended to other environmental impact.

The continuation of this paper will be the study about the effect of input voltage level on the converter
GER. Then, this approach will be applied to other current converters in order to conclude about the
best voltage level for a DC distribution scheme inside building.

References
[1] Schanen JL., Ferrieux JP., Guichon JM., Barbaroux J., Keradec JP., Maurel A.: Flyback Converter Surface
Minimization: Design Procedure and Formulas, IEEE Power Electronics Specialists Conference, 2008.
PESC 2008, Rhodes, Greece, 7p.
[2] Tolvanen CJ., Miettinen T.: A worthwhile investment, ABB Review: Green for growth, Feb. 2009.
[3] Debusschere V., Multon B., Ben Ahmed H., Cavarec P.E., Life cycle design of a single-phase induction
motor, Electric Power Applications IET, Vol. 4 n°5, may 2010, pp.348 - 356.
[4] Jaouen C., Multon B., Barruel F.: Investigation of DC distribution by measuring and modeling power
supply devices for buildings with PV production, EVER’10, Monaco, Monte Carlo, March 2010, 7p.
[5] Van den Bossche A.P., Van de Sype D.M., Valchev V.C.: Ferrite loss measurement and models in half
bridge and full bridge waveforms, IEEE 36th Power Electronics Specialists Conference PESC’05, Recife,
Brazil: 2005, 8p.
[6] Tikana L., Sievers H., Klassert A.: Life cycle assessment of copper product, European Copper Institute
report, 2005
[7] Boughamni W., Roger D., Manata J.P., Brudny J.F, Frezel PH.: Comparative analysis of wire impregnating
and different solutions of Thermo bonding (In French), CONFREGE, Toulouse: 2010, 7p.
[8] Hischier R., Classen M., Lehmann M., Scharnhorst W.: Life cycle invetories of electric and electronic
equipment : Product use and disposal Part III §5, Eco-invent report n°18, Dec 2007
[9] Hischier R.: Life cycle inventories of packaging & graphical papers, part 2: Plastics, Ecoinvent report n°11,
Dec. 2007
[10] John Shen Z., Xiong Y., Cheng X., Fu Y., Kumar P.: Power MOSFET switching loss analysis : a new
insight, Conference Record of the 2006 IEEE Industry Applications Conference, Tampa, Finland: Oct
2006, 5p.
[11] Morancho F.: New limits for compromise between specific resistance and rated voltage for unipolar power
components (In French), HDR, Toulouse, 2004
[12] MOSFET HyperFET dies catalog, Ixys website (http://www.ixys.com/), visited in Oct 2010.
[13] Vishay high performance Schottky diode dies catalog, Vishay website (http://www.vishay.com/die-
wafer/schottky_die/), visited in Nov 2010.
[14] EIME database V11.0, 2009, EIME Website (http://www.codde.fr/page.php?rubrique=20)
[15] TO220 package dimension, PSI website (http://www.psitechnologies.com/products/todo220.php), visited in
Nov 2010
[16] B41851 and B43851 electrolytic capacitor series datasheet, EPCOS website (http://www.epcos.com),
visited in Nov 2010.
[17] « Capacitor for hole mounting assembly » LCA data in Chalmers university LCA database
(http://www.cpm.chalmers.se/CPMDatabase/Scripts/sheet.asp?ActId=ECOP3207), March 2000, visited in
Dec 2011.

30/124
 WIRING DESIGN BASED ON GLOBAL ENERGY REQUIREMENT CRITERIA : A FIRST STEP TOWARDS OPTIMIZATION OF DC
DISTRIBUTION VOLTAGE I NTERNATIONAL C ONFERENCE ON R ENEWABLE E NERGY AND E CO -D ESIGN IN E LECTRICAL
ENGINEERING - LILLE , 23-24 MARCH 2011

Cédric JAOUEN†‡, Bernard MULTON‡ and Franck BARRUEL†

CEA/LITEN INES Institut National de l’Energie Solaire, Laboratoire des Systèmes Solaires (L2S),
50, avenue du Lac Léman - 73377 Le Bourget du Lac, France. Email: name.first name@cea.fr
SATIE, ENS Cachan Bretagne, CNRS, UEB, av. Robert. Schuman F-35170 Bruz, France
Tel: +33(0)4 79 44 46 89 – Fax: +33(0)4 79 68 80 49 –

Topics: Eco-design

1. Introduction
Whereas a full-scale Life Cycle Assessment (LCA) is
Nowadays, the global energy context leads to a need for
focused on the different methods for balancing the
the development of distributed Renewable Energy
different impacts, we chose to focus only on the Global
Sources (RES). Since most of them produce direct current
Energy Requirement (GER). The GER is defined as the
while electricity flows in alternative current in the
total primary energy consumed by the system along its life
network, the energy produced by RES has to be converted.
cycle, i.e. the losses during its use phase, and the energetic
As modern electrical loads and storage systems can be
cost related to the other steps of its life cycle.
potentially fed with direct current, the DC distribution
scheme need to be evaluated. Then, we need to determine
its specification (topology, voltage level, protection
device…). Some papers have already shown the reduction
of conduction losses induced by a DC distribution [1],
others presented some distribution schemes [2] or
efficiency analysis [3]-[4] for different voltage levels in a
given case study. This paper proposes the first steps of a
method for the optimization of the voltage level for a DC
distribution scheme according to environmental issues,
and focuses on the wire eco sizing. In a first part, we will
present our global approach, i.e. the constituent which
will be taken into account and the criteria which has been
selected before focusing on the wiring design. In a second
time, we will present the models and the case studies
which have been considered to design wires. And finally
we will present our results, comparing our approach with
the recommendation made by French standards in our
case studies and carrying out a sensitivity study about our Fig. 1: Life cycle steps of a device [5]
working hypothesis. 2.2. General optimization synopsis
In the DC distribution network, we take electronic
2. Sizing optimization methodology based on the converters, which compose the device power supply,
primary energy environmental impact wirings and protection devices into account. Then, we
2.1. The Global Energy Requirement (GER) have to evaluate the GER of all these elements at a fixed
criteria voltage level; on the basis of this criterion we optimize the
Searching global criteria in order to evaluate the DC voltage level of the global distribution network. But still,
distribution scheme pertinence, we choose to consider changing the voltage level will induce a change in
environmental impacts on the whole life cycle. The constituent sizing. The GER criterion will then once again
environmental impact evaluation for a system is be used at a local scale to optimize each element of the
normalized by the International Standards Organization distribution. As shown in Fig. 2, the global optimization
according to the standards ISO14040. This analysis is process of voltage level contains local optimization loops
known as Life Cycle Assessment (LCA). The LCA is for each component. Local optimizations can be treated
based on the analysis of many environmental impacts, like independently. In this paper we present the local
greenhouse gases emissions, water consumption, energy optimization for wiring only.
consumption,

etc. Assessment is made for all life cycle steps presented

in Fig. 1.

31/124

Fig. 2 : Optimization process synopsis: Focus on wiring
design
3. Wiring optimization : Models and methods
3.1. Optimization process
As we have already explained, the GER is composed by
losses during uses WLosses and the embodied energy
WEmbodied. The energy consumed during the product use is
directly link with its life duration. This duration could be
chosen according to the ageing of the device or according
to the replacement of the device by a better one whereas it
is still usable. The first approach needs to understand the
ageing phenomena which are responsible for device wear
and tear. The second one could be based on statistic
approach and linked to our consumption ways or to
standard evolution. Since wire replacement is currently
due to building renovation, we have considered a 20 years
life span. As presented Fig. 3, which represents the wiring
optimization synopsis, the wiring losses are also related to
the wiring map via the wiring resistance, to the
consumption profile and to the voltage level which fixes
the current flowing into the wiring. The embodied energy
is linked to the wiring map too, which fixes the wiring
length, and to the LCA data, extracted from LCA
assessment.
Since, an increase of conductor section induces an
increase of the wiring embodied energy and a decrease of
its losses and vice versa, we choose the conductor section
as the optimization parameter. For each section which will
be tested by the algorithm, the losses and the embodied
energy will be evaluated in order to find the better one for
a given voltage level. Now, we will present the models
used to evaluate the GER and the constraints apply to the
optimization.
3.2. GER evaluation
For this study we have considered a wire made of
copper for the conducting material and PVC for the
insulating material. We will explain our assumptions
made to evaluate the GER WLCA which is then expressed
by the following expression:
WLCA = WLosses + WEmbodied (1)
Fig. 3 : Wiring optimization synopsis
Loss and thermal model
The loss model is simply based on Joule losses.
However, we have taken resistivity variation according to
thermal effect into account. We considered a resistance
variation of +0.4%/°C and a uniform copper temperature.
Then, for each profile power point, the wire temperature is
evaluated according to a steady state thermal model and
the scheme presented in Fig. 4.
θ air
l
Rconv / rad
φ isu θ copper φ Cu
PJ
Rcond
θ copper
Fig. 4 : Geometric and thermal wire parameters
Thermal resistances evaluation is based on a convection
and radiation heat transfer coefficient of 7W/K/m²
associated with a conduction heat transfer coefficient of
0.17 W/K/m for the PVC insulator [6]. Then, the losses
WLosses are expressed by the following expression:
R (θ copper ( Pload ) ) ×
1 Pload
WLosses = × ∫ dt (2)
ηelec life span
U dc
Where ηelec represents the conversion efficiency to
transform primary energy (Fuel, Nuclear …) into
electrical energy. In fact, the embodied energy is currently
expressed in primary energy, and then losses shall be
expressed in the same units. For our study we consider an
efficiency of 0.28 related to the French energy mix [7].
Embodied energy evaluation
The embodied energy is defined as the primary energy
that was used in the work to make any product, bring it to
market, and dispose of it. Due to few available LCA data
we have considered only the extraction of raw materials
and the wire manufacturing without the PVC extrusion
around the conductor.
Our evaluation is based on two data:
• The primary energy which is necessary to
manufacture 1 meter of 1mm² wire including
raw material extraction and production effort:
WCuWire= 12.5 kWhp/kg 0
32/124
 • The primary energy which is necessary to
produce 1 kg of PVC: WPVC= 16.5 kWhp/kg
[9]
However, we need to find a model of the PVC section
according to the copper one. This model is based on
standard [10] which defines the insulating thickness for
450/750V rated voltage wire related to the conductor
section. Fig. 5 present the data extracted from standard
and the model taken into account for our study.
Fig. 5 : Insulator section versus Conductor section
according to [10]
Then for a given conductor section ACu, the insulator
section APVC will be evaluated and multiplied by the cable
length LCable related to the wiring map. Finally, the
embodied energy WEmbodied will be estimated by the
following expressions:
M Cu = ρCu × ACu × LCable
M PVC = ρ PVC × APVC ( ACu ) × LCable
WEmbodied = M Cu × WCuWire + M PVC × WPVC
3.3. Optimization constraints
Temperature
The temperature constraint is related to material
degradation. According to [11], the maximum PVC
temperature to perform a wire life span of 30 years is 70°C.
The maximum PVC temperature is presented at the
interface between insulator and conductor. Then, we have
chosen to exclude all conductor section which induces a
copper temperature beyond 70°C.
Voltage losses
Standard [12] give some recommendations about
maximum voltage losses related to appliances: 3% for
lighting appliances and 5% for the others. Our voltage
constraints are based on these rules.
3.4. Case studies
We chose to consider two appliances in the commercial
and residential building power range.
Wiring between a feeder and an electrical vehicle
We propose to consider a 1 meter line dedicated to the
charging of an electric vehicle battery. The maximum
power charge is 3.3 kW. The electric vehicle is assumed to
have a 20 km/day use, which is approximately the French
median commuting distance [13], and a 160 Wh/km
energy consumption (based on manufacturer information).
The load profile is based on measures realized at INES on
a Peugeot 106 electric vehicle and shown Fig. 6. This
profile is composed of a first phase which really brings the
required energy, and a second phase for the battery
equalization.
Fig. 6 : Daily electrical vehicle charging profile for a 20
km day trip with a 160Wh/km consumption.
The simulated profile is defined on this one-day time
basis profile repeated five days a week and 48 weeks a
year over twenty years corresponding to a commuting use.
Then, the total energy flowing in the cable is
approximately 18 MWh.
Lighting circuit
(3)
Fig. 7 : Geometric lighting circuit parameters
We propose to consider a lighting circuit composed of
four 20 W compact fluorescent lamps. The circuit and its
dimensions are presented Fig. 7. For each bulb, we
assume the same daily load profile, shown in
Fig. 8, for all the days of the simulation. This profile is a
typical one for a small flat made of a bedroom, a bathroom,
a lounge and a kitchen. The simulated
profile is defined on this one-day time basis profile
repeated each day and each month over 20 years. Then,
the total energy flowing in the circuit is approximately 1
MWh. Compared with the first application, this case study
allows to consider the constraints due to the consumption
simultaneity on a same part of the distribution scheme. All
the lighting circuit parts will have the same section.
33/124

Fig. 8 : Daily power profile
4. Results
4.1. Optimization at given voltage level
First look at Fig. 9 which represents the GER, the
embodied energy and the losses in use versus conductor
section at 230 VDC for the charging electric vehicle case
study. We remark the existence of a unique minimum. For
a conductor section smaller than the optimum one, the
losses are preponderant compared with the embodied
energy, then the GER curve looks like a current “losses
versus conductor section” curve and is in inverse
proportion to the conductor section. However, for bigger
conductor section, the embodied energy becomes
preponderant then the GER curve becomes directly
proportional to conductor section. We could also remark
that the GER shape is really smooth around the optimal
section, i.e. we could choose a smaller conductor section
without an important GER variation in order to reduce the
quantity of materials. The quantity of materials could be
an additional criterion to take into account for completing
this study. In this case, we should solve a multi objective
optimization as it is proposed in [14] for a single-phase
induction motor.
Fig. 9 : GER, embodied energy and losses versus
conductor section at 230VDC for the charging electric
vehicle case study: 1 and 2 meters cable length
Now, compare the two GER curves. We can point out
that the optimum conductor section is the same
independently of cable length. Moreover, it’s important to
notice that this study realized for DC distribution scheme
is viable for AC distribution scheme with a power factor
equal to one, and could be easily extended to different
power factors.
4.2. Comparison between optimized wiring and
standard wiring
We now propose to compare the GER-optimized
conductor section with the standard ones given in [12] for
a 230V voltage level.
For the electric vehicle charging appliance, the
standardized section is 2.5 mm² because the maximum
power level is smaller than 3.6 kW. We also evaluate the
GER of a 1 meter 2.5 mm² conductor section wire subject
to the electric vehicle case study. According to the
optimization process the optimal conductor section is
21mm² with a GER of 11 kWhp whereas the 2.5 mm²
standardized section GER is of 64 kWhp. Since the
standardized conductor section is smaller than the
optimized one, the loss weight is then bigger than the
embodied energy one in the GER. In other words, this
wiring is under-sized according to the GER criteria. We
are aware that 21 mm² represent a very important
conductor section according to current standards.
However, we don’t take the whole life cycle, and into
account then, the optimal section could be smaller if all
the steps were included. In all cases, we could set a
conductor section bigger than 2.5 mm² usual standards for
supplying electrical vehicle (for instance 6 mm² like oven
dedicated wiring) to improve the wiring GER without
spending too much raw materials.
For the lighting circuit appliance, the standardized
section is 1.5 mm². We also evaluate the GER of a 1.5
mm² conductor section circuit subject to the lightning case
study. According to the optimization process the optimal
conductor section is 0.6 mm² with a GER of 1.5 kWhp
whereas the 1.5 mm² standardized section GER is of 2
kWhp. Since the standardized conductor section is bigger
than the optimized one, the loss weight is then smaller
than the embodied energy one in the GER. In other words,
this wiring is over sizing according to the GER criteria.
The standard [10] is based on lighting circuit made with
incandescent bulbs, whereas they are replaced by more
energy efficient bulbs (considered in this study). Our
study shows that the lighting circuit conductor section
could also be reduced in the future.
Fig. 10 shows the evolution of the optimal conductor
section and the associated GER, for each case study, with
the voltage level. It points out that bigger the voltage level
is, smaller the GER is. Then, from the wiring point of
view, the high voltage DC distribution scheme is greater
according to the GER criteria. However, higher the
voltage level is, more important the protection devices are.
More studies need to be performed to affirm this trend.
34/124
 study. In this case study, an error of +/-20% on the life
span induces a GER variation of +-10%. However, by
comparing the GER obtained in the real life span with the
real life span optimized one, we can conclude that the
difference between the two GER is minimal for a +/-20%
life span estimation error. In other word, the variability of
life span around the average value doesn’t induce
significant change on the optimal wiring area.

Fig. 10 : Optimal section and associated GER versus

voltage level: electric vehicle case study (top) and lighting
case study (bottom)

4.3. Sensitivity study

As we have presented sooner, the life span, the LCA
data and the consumption profile could be different from
one study to another. Consequently, we also realize a Fig. 12 : Consequence of a bad estimation of life span on
sensitivity study for these parameters. the GER (blue curve) and comparison with the optimized
one (red curve) for the charging electric car case study.
Life span sensitivity
The life span assumption has been chosen according to LCA data sensitivity
an estimation of a building renovation every 20 years.
As we explain sooner, all the life cycle steps are not
However we do not have any study to confirm this
taken into account. This may change the optimal
assumption. In a first time, we look at the influence of the
conductor sector. Moreover, LCA data depend on local
life span estimation on the optimal section for the
and technological conditions. Then we compare the
supplying of an electric car. We plot the graph presenting
different optimal conductor sections according to a
the optimal conductor section versus voltage level for a
+/-20% variation of WCuWire in the charging electrical
+/-20% life span variation around 20 years (Fig. 11).
vehicle case study (Fig. 13).

Fig. 11: Effect of life span estimation on optimal

Fig. 13 : Impact of the copper embodied energy on the
conductor section: charging electric vehicle case study
optimal conductor section for the charging electric car
First, we can remark that the effect of life span is case study.
independent from the voltage level. Then, a life span
Like in the life span sensitivity study, the impact is
increase induces an increase of optimal conductor section.
independent of the voltage level; however, in this case, the
In fact, a life span increase corresponds to a loss increase
GER variation is opposed to the copper embodied energy
due to a bigger energy transfer; the section can also be
one. In fact, the wiring embodied energy is directly linked
bigger to reduce these losses. In the Fig. 11 case study, a
to the copper embodied energy. Then, the wire section
+-20% life span variation produces a +-10% optimal
should be reduced to reach the optimum. In this case study,
conductor section. But, as we remark on Fig. 9, the GER
the optimal conductor variation due to copper embodied
variation is very smooth around the optimal conductor
variation is similar to the variation due the life span
section, which explains why this potential error is not so
variation: a +/- 20% variation induces a +/- 10% GER
important.
variation.
Now, look at the impact of an error of wiring life span
estimation, on the GER. We also determine the optimal
Consumption profile sensitivity
section for a 20 years life span and evaluate the GER for
this wire used during other life span in a 230V distribution In this part, we will quantify the impact of the
scheme (blue curve). Fig. 12 presents the results of this simultaneity in the lighting case study and the impact

35/124
of the maximum power level on the optimal conductor
section for supplying an electric vehicle.
The impact of simultaneity will be assessed comparing
two lighting case study: one with the
Fig. 8 power profile and the other with the Fig. 14
one-day base power profile at 230 Vdc. These profiles are
built concentrating the consumption at the beginning of
the day.
Fig. 14 : Simultaneous lighting power profile
According to the optimization process, the optimal
conductor section for the Fig. 14 day profile is 0.65 mm²
with a GER of 1.65 kWhp whereas the former study
concluded to an optimal conductor section of 0.6 mm²
with a GER of 1.5 kWhp. This study points out that the
consumption repartition may play a role in the
determination of the optimal conductor section. In fact,
the consumption superposition inside several circuit parts
increases losses and consequently the optimal conductor
section too. But this effect is independent of the designer
will.
Fig. 15 :3.3kW and 2.2kW charging 20 km commuting
electric vehicle power profile
Now look at the effect of maximum power level in the
charging electric vehicle case study. We will then compare
two charging profile needed to bring the energy for a 20
km commuting at 230 Vdc: one with a 3.3 kW maximum
power level and another with a 2.2kW one, as shown Fig.
15
According to the optimization process the optimal
conductor section for the 2.2kW charging profile is 17
mm² with a GER of 9 kWhp whereas the former study
resulted to a 21 mm² optimal conductor section with a
References
[1] A. Sannino, G. Postaglione, H.J. Bollen, Feasability
of a DC network for commercial facilities, IEEE
GER of 11 kWhp. This study points out that the maximum
power level effect is the same that the effect of
simultaneity; however, the designer can have an influence
on the desired power level.
Sensitivity study conclusion
In this chapter, we have shown the complexity to give
quantitative results after an eco-design study due to
multiple uncertainties. The first one is due to LCA data
variation depending on manufacturing location and
process but it could be evaluated. Others are due to
consumer behavior (life span and simultaneity
parameters) and are difficult to assess. However, they may
be not always significant. Finally, the designer could
choose this specification in order to reduce the
distribution scheme impact for a given energetic need
according the power level point of view.
5. Conclusion
In this paper, we have proposed an approach based on
the LCA to evaluate about the pertinence of the DC
distribution scheme. Our method introduces the
eco-sizing of DC distribution scheme’s constituents
according to the GER criteria. This paper has focused on
the wiring eco-design for two appliances: a lighting circuit
and a wiring for charging electric vehicle. After
presenting our model to assess losses and embedded
energy, we have shown the existence of a unique
minimum at a given voltage level and the fact that bigger
the voltage level is, smaller the GER is, (i.e. a minimum
doesn’t exist according to voltage level). Then, we have
compared the GER of an optimized wire with the
standardized wire one. We concluded about the need to
increase the conductor section in the charging electric
vehicle case study and the need to decrease the conductor
section in the lighting circuit case study. Finally, we have
made a sensitivity study to evaluate the effect of the
profile property (simultaneity and maximum power level)
and the effect of the uncertainties related to LCA data and
life span estimation. It is important to notice that, due to
these uncertainties, an eco-design study can nowadays
only give trends for wiring design but the smooth
variation around the optimal conductor section allows
accepting some tolerance about our hypothesis. Moreover,
this study could be completed by taking the
non-renewable resources consumption into account in a
multi-objective optimization.
Appendix
Tableau 1: Definition of constant
Symbol Quantity Value
ρCu Copper density 8960 kg/m3
ρPVC PVC density 1400 kg/m3
Transactions on Industry applications, vol. 39, Oct.
2003.
36/124
[2] V. Azbe, R. Mihalic, Distributed generation from
renewable sources in an isolated DC network,
Renewable Energy, vol. 31, Nov. 2006.
[3] M. Sechilariu, AC vs DC: Hybrid distribution scheme
for building with renewable energy production (In
French), EF 2007, Toulouse: 2007.
[4] C. Jaouen, B.Multon, F. Barruel, Investigation of DC
distribution by measuring and modeling power
supply devices for buildings with PV production,
EVER’10, Monaco, 2010.
[5] A. Helias et al “Life Cycle Approaches: The road
from analysis to practice”, UNEP-SETAC Life Cycle
Initiative, 2005.
[6] CSTB, RT 2005, appendix 2/5 Materials : Useful
thermal characteristics (In French), 2005
[7] Software Bilan Produit 2008, ADEME,
http://www.ademe.fr/internet/bilan_produit/login.asp,
accessed September 2009
[8] L. Tikana., Sievers H., Klassert A., Life cycle
assessment of copper product, European Copper
Institute, 2005
[9] R. Hischier, “Life cycle inventories of packaging &
graphical papers, part 2: Plastics”, Ecoinvent report
n°11, Dec. 2007
[10] Standard ”NF C 32-201”, PVC insulated wires and
cables for a 450/750V rated voltage (In French), 1998,
Last updated 2009
[11] D. Serre, Low voltage installations: wiring design
and implementation (In French), “Techniques de
l’ingénieur”, Ref. D 5 046, feb. 2007
[12] Standards “NF C 15 100”, Low voltage installations
(In French), 2002, Last updated 2010
[13] B. Baccaïni, F. Sémécurbe, G. Thomas, “Commuting
amplification due to suburbanization”, INSEE
Premiere, N° 1129 - MARS 2007
[14] V. Debusschere, B. Multon, H. Ben Ahmed, P.E.
Cavarec, « Life cycle design of a single-phase
induction motor », Electric Power Applications IET,
Vol. 4 n°5, may 2010, pp.348 - 356.
37/124
38/124
1.2 PV SYSTEM MONITORING
AND DIAGNOSTICS

39/124
40/124
 IN-SITU I-V CURVE SCANNER FOR PV MODULES AND PV ARRAYS

Jérémy MARTIN, Nicolas CHAINTREUIL, Franck BARRUEL, Jens MERTEN

CEA-LITEN INES Institut National de l’Energie Solaire, Laboratoire des Systèmes Solaires (L2S)
50 Avenue du Lac Léman, BP 332, F-73370 Le Bourget du Lac, France
Phone: +33 (0)4 79 44 45 46 - Fax: +33 (0)4 79 68 80 49
Email: name.first name@cea.fr

ABSTRACT:

Nowadays, there is no commercial product which enables to anticipate and localise possible faults on a PV installation.
The CEA-INES investigates reliable and economical tools for PV application diagnostic methods. In this paper, the
authors propose a system for fast I-V measurement which is realized without interruption of the power plant operation.
The data-analysis of these I-V characteristics allows us to detect possible faults like shading, mismatch, ground, bypass
diode and series resistance faults. I-V measured by this device are exposed and analyzed for several photovoltaic modules
technologies.

Keywords: PV System, PV Module, PV Array, Defects, Safety

1 INTRODUCTION one diagnostic system one diagnostic system integration into the inverter
per module per module

Aging of PV power plants is of critical importance. Diagnostic

device
Serial Diode

Defects may appear after a while and can be for example

responsible of fires caused by a DC arc fault. PV system PV
Module

safety should be taken seriously, the best way in order to Diagnostic

prevent incidents is to anticipate faults and localise them. device Chopper / Inverter

Link inductor
Diagnostic
One solution to check the PV plant health is to analyse I- device
DC
V curves data’s in order to deduce which parameters are Vond Vgrid

critical while the PV power plant is still operating. This AC

system is adapted to two configurations: one system per

panel or one system for a whole string.

2 THE DIAGNOSTIC SYSTEM & OPERATION

CONFIGURATION
The principle of the proposed system is to switch a
MOSFET parallel connected to the PV panel in order to
sweep the PV panel operating point from the short circuit
current Isc to the open circuit voltage Voc (Fig.1). As each
PV module technology has it’s own response time around
a few tens of microseconds, an additional circuitry is
required to slow down the MOSFET switching time. The
current and the voltage are measured and curves are
analysed by a micro processor unit. A model of the
system is built at CEA-INES and experimental results are
presented.
Photovoltaic Bypass Ip MOSFET
module Diode Enhancement mode
Figure 2: Three possibilities for the implementation of
the in-situ I-V curve scanner in a PV system.
3 THE MODEL
The model is presented in figure 3. It is mainly
composed of a N-channel enhancement mode MOSFET,
and a gate drive circuitry where a RC circuit is adjusted
in order to adapt the commutation last to the panel
response time. The model operates in real conditions.
Panel voltage and current probe are connected to an
oscilloscope and panel temperature, irradiation are
measured.

Vp

Ip Vp
Micro processor Unit
Figure 1: In-situ I-V curve scanner

The proposed system can be used in three

configurations (Fig.2):
-One diagnostic system by module configuration
-One diagnostic system per string configuration
-Integration into the inverter configuration
Figure 3: Laboratory device for in-situ I-V curve scanner

4 I-V CURVES PLOTTING

41/124
 Experimental results of the fast I-V curves data 2 1cell total shaded and 1 cell
65% shaded G=562W/m²
acquisition system are given in figure 4.The I-V curves Classical I-V curve tracer G=501W/m²
with the considered system are very close to those 1 cell total shaded
measured with a classical I-V curve tracer. For these G=531W/m²
results, bypass diodes are disconnected from the cell, Normal conditions G = 494W/m²
that’s why there are no inflexion points on the curves. Ip(A) 3

6 cells Shaded G=591W/m² 1

Normal conditions without bypass diode
G=1061W/m²
Classical I-V curve tracer
G=990W/m²
Module 20% shaded Vp(V)
Ip(A)
G=1055W/m² Module 10% Shaded Figure 6: I-V curves: normal situations and three shading
G=1054W/m² situations (MOSFET commutation last: 1ms)
Module 50% Shaded
G=1068W/m² Module with a leaf
G = 1045W/m²
5.2 increase of the PV module response time
Vp(V)
Figure 4: Experimental results for the in-situ I-V curve A shading situation modified the PV panel response
scanner for four shading situations (Duration: 1ms) time. This behavior can be combined with the
deformation off I-V curves in order to get redundancy on
5 SHADING FAULT shading detection. Voltage response time of the PV panel
is given in figure 7 for a Polycrystalline silicon
5.1 I-V Curves modification technology.

1cell total shaded

Polycrystalline silicon technology panel I-V curves and 1 cell 65% shaded
3 1 cell total shaded
are given in figure 5 for normal situations and for three G=562W/m² G=531W/m²
shading situations (Fig.5, Fig.6). The curves given by the
system are closed to those given by a classical I-V curve 2
tracer. Shading can be observed by the inflexion points Ip(A) Normal conditions
on the curves which reveal the conduction of a bypass G = 494W/m²
diode. 1
6 cells total shaded
G=591W/m²

V p(V)
t(s)
Figure 7: PV panel response time modification for three
shading situations (MOSFET commutation last: 1ms)

Bypass diode Response time analysis can be extended to several

technologies. Each technology as a response time
1 2 increased when shading situation.

6 PV PANEL SERIES RESISTOR FAULT

3
In order to study the case of fault of increase PV
module series resistance increasing, a rheostat is
Figure 5: PV Module in shading situation connected between the PV module and the fast I-V
curves data acquisition system. The series resistance is
adjusted from 100mΩ to 50Ω.
Shading situations correspond to the inflexion point
on the I-V curves (Fig.7); it reveals the turn-on of a
bypass diode. For the considered module, the bypass
diode equips 18 cells. In the situation number 2, two
groups of 18 cells are shaded, so there is 2 inflexion
points. The situation number 1 is the worst case where all
bypass diodes are on-state.

42/124

Figure 8 shows for a polycrystalline technology the
increasing of the δIp/δVp when the series resistor
decrease, analysis of the δIp/δVp rate will guarantee the
detection of a series resistor fault.
8 CONCLUSIONS
Experimental results are encouraging: curves
acquired by the system appear coherent for all PV
module technologies. In a second way all the tests have to
be performed under standard test condition in order to
evaluation the accuracy of the display.

R=1.5Ω, G=642W/m² R=0.6Ω In order to integrate this solution, I-V curve data
G=587W/m²
analysis algorithms are investigated in addition with this
study; it should determine finally shading, mismatch,
R=10.2Ω R=2.3Ω
Ip(A) G=587W/m²
ground, bypass diode and series resistance faults from the
G=709W/m²
Normal conditions I-V curves data.
G=567W/m²
δIp The major advantage of this system is to require a
low cost and very compact hardware so the integration in
R=50.4Ω
G=729W/m²
the junction box is easy.
δVp
Vp(V)
Figure 8: Series resistor fault
(MOSFET commutation last: 1ms)

7 BYPASS DIODE FAULT

The Bypass diode is reverse connected to a 18 cells

group. The diode is on state when no aging on the 18
cells. The diode is turned off when aging. In the case of
reverse connected bypass diode: the open circuit voltage
is decreased by 9 volts which corresponds to the loss of a
18 cells group (Fig.9).

As seen before, for the groups of cells with bypass

diode right connected, the inflexion point appears when
the bypass diode is turned off.

Normal conditions without bypass diode

G=744W/m²
1 cell 50% shaded
G=1040W/m²

Normal conditions
Ip(A) with reverse bypass diode
G=1027W/m²
1 cell total shaded
1 cell total shaded G=796W/m²
G=763W/m²

Vp(V)

Figure 9: Bypass diode fault

(MOSFET commutation last: 1ms)

43/124
44/124
 EFFECTS OF SHADOW ON A GRID CONNECTED PV SYSTEM

N. Chaintreuil, F. Barruel, X. Le Pivert, H. Buttin, J. Merten

INES R.D.I., Laboratory for Solar Systems (L2S)
BP 332, 50 avenue du lac Léman, F-73377 Le Bourget-du-lac, France
Phone : +33(0)4 79 44 46 11 ; mail : nicolas.chaintreuil@cea.fr

ABSTRACT: The aim of this paper is the study of the impact of partial shading on a grid connected photovoltaic
(PV) plant. Two technologies of PV modules are examined: multi-crystalline silicon and amorphous silicon.
Two types of studies are made: On one hand, shadowing losses are assessed instantaneously under stable solar
irradiance conditions. On the other hand, these losses are summed up during a five weeks period. Different types of
shadows are tested.
This study shows better results for amorphous silicon systems face to shadows. Partial shading may lead to
complete system failures when the inverter stops operating when the Maximum Power Point voltage gets out of
range. The results of these two cases are compared.

Keywords: grid connected, shading, system performance

1 INTRODUCTION The modules can be connected in series or parallel

(see Figure 3). Each array has one PV inverter SMA
Today, development of grid connected PV systems Sunny Boy SB1100 connected to the grid. These arrays
due to good purchase price of energy requires having are positioned at south.
losses the lowest as possible. It is important to have a
good ROI (Return On Investment). In France, there is a
special prime for building integrated PV (BIPV) systems.
For these systems, however, partial shading is difficult to
be avoided. Importance of shadows on a grid connected
PV system was known but they have not been quantified
yet. Problem is that there are many types of partial
shadows on a PV plant. This study analyzes six types
which are “the most representative” of the realistic cases.
Furthermore, the thermal impact of a shadow on the
PV cells is poorly known. This information would permit
to draw a vision of long term risks of degradation due to
overheating.
Tests are made on two technologies of PV modules:
multi-crystalline silicon and amorphous silicon. It was
based on measurements of two identical grid connected Figure 1: Multi-crystalline array structure.
PV systems. Steady state tests are performed to compare
the influence of different types of shadow on different
electrical architectures of the PV array. Furthermore, a 2.2 Amorphous PV arrays
long period measurement of five weeks is performed to
determine the energy losses suffered by the two grid Two identical amorphous silicon PV arrays of 744Wc
connected installations of different module technologies were used. They consist of six modules Uni-Solar ES-
under the shadow of a pole. 124. Each cell of these modules has a surface of 0.085m². It
This study has been realised within the framework of represents a rectangle of 35.6cm (14in) per 23.9cm (9.4in).
a project named SOLUTION PV which has the goal to The modules are connected in series. Each array has
research an optimisation of PV system architecture. one PV inverter SMA Sunny Boy SB1100 connected to
the grid. These arrays are positioned at south but slightly
shifted 6° to the east.
2 DESCRIPTION OF PV ARRAYS STUDIED

The PV arrays examined are located at the INES
headquarter close to Chambéry in the Rhône-Alpes
region in France.
2.1 Multi-crystalline PV arrays
3 STEADY STATE MEASUREMENTS
In these tests, we study the impact of different types
of shadows on a PV plant (without inverter). We compare
different electrical architectures and quantify the losses.

Two identical Multi-crystalline PV arrays of 1.5kWc
were used. They consist of nine modules Photowatt
PW1650 each. The cells used in these modules measure
12.55cm (4.9in) per 12.5cm (4.9in). There surface is
0.016m².
For this part, we used I-V curves tracers which are
connected directly to the complete PV arrays. The tracer
type is PVPM6020C with a reference cell named Solar
Radiation Sensor (calibrated). One array was exposed to
partial shading whereas the other served as a reference. I-

45/124
V curves tracers permitted to determine maximum power
point (MPP) of each array. The difference between the
two arrays has been found to be less than 0.5%. This test
was made at 829W/m² and at a temperature of 45°C in
back of module.
3.1 Multi-crystalline PV arrays
3.1.1 Shadowing losses on array level
In this part, six types of shadows were tested, which
are shown in Figure 2.
The parallel architecture presents better results for
shadows 1, 4, 5 and 6.
For the shadowing 2 and 6, the difference between
series and parallel may be up to a factor 2.
There is no clear relation between the surface shaded
and the corresponding losses because of the series
connection of the cells. This explains the drastic impact
of shadow 1, which only covers 1 single cell: the power
loss is 25 times the loss of active surface in the case of
the series architecture, and 11 in the case of the 3x3
architecture. Note that this loss is limited due to the
bypass diodes - losses would be more important and
could lead to the destruction of the modules due to the
hotspot effect.

3.1.2 Shadowing losses on system level

Shadowing can cause additional losses on system

Figure 2: The Six types of shadows examined.
Two architectures were tested with these shadows:
level, as it has important effects on the Maximum Power
Point of the array; two points are discussed here.
Firstly, the Maximum Power Point may be difficult to
be found by Maximum Power Point Tracker, as the
shadowing causes the presence of two Maximum Power
Points, which is shown in Figure 4.

Figure 3: Electrical architectures examined: series (left)

and 3x3 (right).

I-V curves tracers are placed on plus and minus terminals

of the complete array.
Figure 3 illustrates the deformation of the array I/V
curves due to shadowing. The steady state power losses Figure 4: Example of two MPP curve for parallel
are synthesized in table I: architecture and shadow 2 (red: no shading, blue:
shading).
Degree of Array Power loss
module In this case, the Maximum Power Point Tracker can stop
Series 3x3
shading on the local maximum point at 92V and forget real
Shadow 1 0,15% 3,7% 1,7% maximum power point at 68V. Should the Maximum
(1 cell) Power Point Tracker bet on the wrong Maximum Power
Shadow 2 (72 11,1% 12,6% 29,2% Point, the loss due to shading would be 31.5% instead of
cells) 29.2%.
Shadow 3 (90 13,9% 22,2% 36,8% Secondly, the voltage of the Maximum Power Point
cells) can vary significantly due to shading and going out of
Shadow 4 (81 12,5% 18,3% 17% operating voltage range of inverter, which is shown in
cells) Figure 4, where the Maximum Power Point voltage is
Shadow 5 (72 11,1% 36,5% 30,5% reduced by 38%.due to shadow 5. In this case, if
cells) operating voltage range of inverter starts at 180V, the
Shadow 6 2,62% 16,7% 7% inverter will stop operating and the shadowing losses are
(≈ 17 cells) 100%.
Table I: Steady state shadowing losses for the multi-
crystalline PV array.

As we can see in this table I, the series architecture is

better for shadows 2 and 3. However, in this case the
orientation of the strings is very important and can
significantly influence results.

46/124
 (error of 5%) and losses on Shadow 2 are too small
compared measure precision, so it explains difference.
700

600

500

400

P d c (W )
300

200

100

0
Figure 5: Maximum power point moving out of the 0 50 100 150 200 250
operating range due to shadow 5 (series architecture, red: Vdc (V)

no shading, blue: shading). Figure 7: Example of it can be an out of range voltage

3.2 Amorphous PV arrays
Amorphous PV modules are interesting to study
because of their conception. They have one diode per cell
and present a good immunity to face shading. Tests are
made on 6 types of shadows:
curve for shadow 5 (red: no shading, blue: shading).
Figure 7 shows a reduction of the Maximum Power
Point voltage under shadow to 123V, which represents a
reduction in voltage of 22%, whereas the power is reduced
by 26%. With the multi-crystalline modules, we did not
observe such a close relation of voltage and power loss
under shading. The reason is the presence of a by pass
diode in each cell of this module: the cells shadowed are
by-passed, which favors a reduction of the voltage and not
of the module current under shading.
These variations of voltage can stop running inverter
for same reasons as said for multi-crystalline PV arrays.

4 LONG TERM MEASUREMENTS

Figure 6: Amorphous PV array shadows. 4.1 Multi-crystalline PV array
In this case, only series architecture was tested. As done These measures have been made in January 2008 at
with the multi-crystalline PV arrays, I-V tracers are placed INES headquarter close to Chambéry. Two identical PV
on plus and minus terminals of the complete array. arrays of nine PW1650 modules with a total array power
The precision of the power measurement is about of 1500W have been examined. Inclination of the array
0.91%. It’s the difference between the two I-V tracers and was 33°, the array was oriented south. The pole with a
modules dispersion. We consider this difference in these diameter of 5 cm was used to create partial shading of the
measurements because some results values are near 1%.. array. Identical inverters are used to connect these arrays
to the grid. One array has been exposed to shadowing
with a pole as shown in Figure 8, and one array served as
Degree of Array a reference.
module shading
Power
Loss
Shadow 1 0.7% 1.3%
Shadow 2 0.8% 0.3%
Shadow 3 3.3% 4. 9%
Shadow 4 4.3% 7.6%
Shadow 5 14 % 25%.
Shadow 6 1.4% 2%
Table II: Steady state shadowing losses for the
amorphous PV array. Figure 8: Shadowing by a pole of the Multi-crystalline
array. The arrow indicates the daily coverage of the
The results are shown in Table II. This table shows a shadow on the array.
better correlation between shading surface and losses of
amorphous PV array than multi-crystalline PV array. This In this case shadow is out of array after 2 P.M. (in
is due to the presence of one bypass diode per each cell. January). We measured both DC power and AC power.
The factor between surface shaded and the loss in array To eliminate inverter losses we compare only DC power
power is around 1.6 in most cases. Only shadows 1 and 2 is here.
different because Shadow 1 shading surface is an estimate Measure difference between the two channels is about

47/124
about 1% without shadow, it is the measure precision.
Results indicate losses and DC production per day. It
permits to verify if there is a link between losses and
production (linked to sunshine).
Figure 9: Daily shadowing losses (a) and DC-production
of the reference array (b) for the Multi-crystalline array.
Figure 9 indicates that losses are linked to DC
Production (linked to sunshine) due to dispersion of
shadow when sunshine is low. Effectively, if sunshine is
very important (without indirect sunshine), shadow was
very net and contrasted. It results bypass of shading cells.
If sunshine reduces or indirect sunshine becomes
important, shadow will be less net and contrasted.
Shading cells will not be automatically bypassed and
losses of power (compared to a not shading array) will be
lower (see Figure 10).
Figure 10: Multi-crystalline one cloudy day results (12th
January 2008) (red: no shading, blue: shading, green:
shading losses.
On Figure 10 difference between the two channels is
directly linked to power production. When production is
strong, the difference increases.
The DC losses on January 2008 sum up to 4.5%
(Figure 9). This result is not negligible in view of the
small surface covered (2% in the worst case). It is
important to note that shadow is out off array near a half
day. So, losses, if array is shading throughout the day,
can be almost doubled..
For a good day, we observe the curve on Figure 11:
Figure 11: Multi-crystalline one “good day” results in
December 2007 (red: no shading, blue: shading, green:
shading losses).
In this case, day losses sum up to about 7%. It is easy
to see when shadow of the pole goes out of PV array (at
13H on horizontal scale).
4.2 Amorphous PV array
This part of the study was made from 1st to 30th June
2008. Two identical PV arrays are used. Each array
represents 744Wc with 6 ES-124 modules. Inclination of
the array was 30°, the array was oriented south. The pole
with a diameter of 5 cm was used to create partial shading
of the array (Figure 12). Two identical inverters are used
to connect these arrays to the grid.
Figure 12: Shadowing by a pole of the amorphous array.
The arrow indicates the daily coverage of the shadow on
the array.
In this case, the shadow covers the whole array
during the whole day. Measure precision is about 1% and
includes PV modules dispersion and Maximum Power
Point Tracker differences. Inverters are identical.
Results are presented as done in the previous section.
The relation between production and losses is the same as
multi-crystalline tests. Dispersion of shadow depends of
intensity of sunshine and direct or indirect sunshine.
48/124
Figure 13: Amorphous long period results.

In this case, shading losses sum up to 1% throughout

test period. This results is very close to the measurement
uncertainty. This means that no significant shading losses
of the amorphous silicon array have been detected.

5 HOT SPOTS

It is clear that shaded cells do not generate current,

but have to pass through the current produced by the rest
of the string. The corresponding cell heating has been
observed with thermal camera (see Figure 12):

Figure 14: Thermal behavior of multi-crystalline PV array

under the shadow of a pole.

Figure 14 shows thermal impact on cells shading by a

pole. Difference of temperature between overheating cells
(under shadow) and not shading cells is about 15°C in
this case. Air temperature is around -1°C and sunshine
around 200W/m² (morning of a sunshine day). This
results permits to conclude that a repeating shadow on a
multi-crystalline PV array can generate degradations on
heating cells.

49/124
6 CONCLUSION AND OUTLOOK

This study has helped to quantify the losses suffered by

partial shading of PV power plants based on multi-
crystalline and amorphous silicon technology.
Several mechanisms have been identified which
reduce the performance of PV generators under shading:
• Reduction of current generation
• Blocking of current by shaded cells
• Creation of local maxima on the array I/V curve
which may lead to operation of the array far away
from the optimum operation voltage
• Reduction of the Maximum Power Point voltage until
running out of the operation range of the inverter In
this case production is stopped!
Steady state test have shown that amorphous silicon
arrays are less sensitive to shading than the multi-
crystalline arrays. This is due the presence of a bypass
diode for each cell of the amorphous silicon module
employed.
Shadowing losses are higher, when all modules of the
array are connected in series.
Long term tests with a pole shadow on the array also
show that amorphous PV systems are less affected by
partial shadowing than multi-crystalline systems.
Long term studies on different sites are needed in
order to quantify the annual productivity loss under
shading of different module technologies.

50/124
 PERFORMANCE ANALYSIS OF THE FIRST MEDIUM SIZE PV-PLANT IN BURKINA FASO ON THE
CITY GRID OF OUAGADOUGOU

Mark Vervaart, Jens Merten

CEA/LITEN INES, Laboratory for Solar Systems (L2S), BP 332, 50 avenue du lac Léman,
F-73377 Le Bourget-du-lac, France. Phone/fax : +33 (0)4 79 44 45 46 / +33 (0)4 79 68 80 49
Mark.vervaart@cea.fr

R. Cool
SGBB, 248 Rue de l’Hotel de ville, 01 BP 585 Ouagadougou 01, Burkina Faso
Phone : (+226) 50 32 32 32

Tanguy Cadin
Photalia SAS, 160 Rue des Sables de Sary, 45770 Saran, France
Phone : +33 (0) 2 38 72 80 35, t.cadin@photalia.fr

ABSTRACT: The ‘Société Générale de Banques au Burkina’ (SGBB) installed the first 80kWp grid-connected
system on the roof of their headquarters in Ouagadougou in January 2010. The system was developed by Photalia
(Groupe Vergnet, France) and was inaugurated on the 16th of April. Since in Burkina Faso no feed-in tariffs are
applicable (it is not even allowed to inject energy back into the grid!) the system is rather particular in energy
management and is exposed to multiple extreme circumstances. The chosen solution is direct consumption of the
instantaneous production without storage batteries. On the other hand, the extreme circumstances of climate and
supply of electricity in Burkina Faso have led to put in place an appropriate monitoring system. This study
provides the results of the first month of monitoring data and an overview of system performance.

Keywords: Developing countries, Grid connected, System performance

1 INTRODUCTION The PV-system is spread over multiple buildings of

different heights of the complex, with a slightly different
tilt. There are even some flat V-shaped roofs facing
The grid in Burkina Faso is considered as a weak grid,
North AND South. The PV-array is divided into 14 sub-
powered by big raw petrol generator sets out of town. In
systems with each their own inverter. A profound 3D
fact, it is a very big stand alone grid for all the major
analysis has been performed by Photalia (figure 1) in
cities in the country. The grid frequency is fluctuating
order to prevent shading of part of the array by
rather much with a rising energy demand, especially in
surrounding buildings of different height.
April when air conditioners are used in large quantities
all over town. During those peak hours, the grid regularly
trips and switches off during shorter or longer periods. A
local gen-set at the SGBB functions as UPS and delivers
the energy for the SGBB building. The 80kWp array
should not feed in to the gen-set when the grid is not
present so in order to avoid problems, the PV-system is
shut down in those circumstances. During the weekend,
when only stand-by power is used, the inverters will be
shut down as soon as the building consumption
approaches the PV-yield in order to avoid feeding back
into the grid. So in this case, system yield will have a
relation to the energy consumption of the building, which
is rather unusual for a grid-connected system.

2 ENVIRONMENTAL CONDITIONS

Circumstances in Ouagadougou are rather extreme

for PV-systems. Outside temperature and so module
temperature can be very high in April. Much dust is
present in the city which demands regular cleaning of the
modules. Even in one day, one can visually see the
accumulation of dust on the modules. This needs to be
kept in mind while analyzing system performance, while
the reference cell will suffer the same circumstances.
Modules are placed at a very low tilt because of the
geological situation close to the equator. In August the
sun is situated North, in January South! This will Figure 1: PV-installation at the roof of the SGBB
surely have a negative influence on dust accumulation.
The switching of the air conditioners in the building

51/124
causes important voltage spikes on the grid, so good
filters are needed for PV installations as protection of the
inverter outputs. Frequency windows of the inverters are
enlarged in order to be able to respond to the fluctuating
grid frequency during peak hours.
Due to the geographical position near the equator,
most irradiance is received in April and October. August
is the month with least irradiance, which should be
considered during this study, where monitoring started in
August.
Figure 2: Average irradiance per day and temperature in
Ouagadougou (source: meteonorm)
3 ROLE AND MOTIVATION OF THE PARTNERS
SGBB
Societe Generale Bank of Burkina (SGBB), a subsidiary
of SOCIETE GENERALE Group (International Retail
Division) is investor and owner of the system. With this
project, at a time when the climate change issue is
topical, SGBB positions itself as a pioneer company in
the renewable energy sector in Burkina Faso and West
Africa. This approach is fully in accordance with the
Group environmental policy and was supported as such
by SOCIETE GENERALE. The SG is committed in
limiting significantly its global CO2 emissions since
2008. The Group encourages its entities to reduce their
energy consumptions and to use renewable energy
technologies when they have the opportunity.
Photalia
Photalia is system designer and realized the
implementation of the system. Photalia belongs to the
Vergnet Group, an internationally recognised energy (and
water) supply specialist for remote or complex locations.
As an innovating company, Photalia brings to Africa the
best of the western solar technologies and projects. It
adapts them for the benefit of the rural population and
now the urban ones. Through the SGBB project, Photalia
shows that the western conception of solar power could
respond to the African needs. Such installations present
advantages at three stages:
- At the short term level, the solar energy produced can
represent a significant economy on the electricity bill.
- At the midterm level, the energy saved from the
national grid remains available for others uses and
consumers.
- At the long term level, the energy from the renewable
source has a positive impact on the CO2 reject balance.
INES
The role of INES is the definition of the monitoring
system and the data analysis. INES is the French national
institute of solar energy which has started operations end
2005. Actually, INES counts on 250 employees
developing innovative solutions to increase the use of
solar energy.
4 SYSTEM DESCRIPTION
4.1 System topology
The system is divided into 14 sub-systems of 6kWp
with each a separate inverter. All inverters are SMA.
Each module array consists of 3 strings of 12 modules of
170Wp in series. The modules come from BP-Solar. The
14 sub-systems are divided over the three phases of the
electricity network.
In case of grid failure, the complete building is
switched to a generator set. The main PCs and servers are
provided with an individual UPS system. The choice was
made not to have battery storage. The installation and
maintenance costs of a battery pack were considered as
unnecessary and too expensive.
Figure 3: System topology
Each group of inverters is connected to the network
via a central relay. A control unit is installed, which
steers the relays in order to switch-off groups of inverters
if needed. The control unit measures the power of the
three phases of the grid connection.
4.2 System functioning
The control unit measures the three phases of the
connection with the utility grid individually. If one of the
phases has the tendency to inject energy, a corresponding
group of inverters is switched off. The unit also is
informed if the generator set is switched on in case of
grid failure. The choice was made to switch off the whole
PV-installation in case of grid failure, in order to be
absolutely sure that the PV-system will not inject energy
into the generator set. In that case, the control unit opens
the relays of all inverters.
52/124
4.3 Monitoring system
Besides the standard SMA Sunny web box from the
side of Photalia, INES installed a more extensive
monitoring system for a detailed study on the behavior of
the installation. From all 14 inverters, the AC side is
measured. From inverter 1 to 3 the DC side is measured. A
reference cell is mounted in module plain with the modules
facing South. Ambient temperature is measured, as well as
one module temperature.

The three phases of the connection point with the grid

provider are monitored on voltage, current (3P + N), active
power and reactive power. The connection point of the
complete PV-system to the building network is monitoring
on every phase as well for voltage, current, active and Figure 5: Standard power consumption profile of the
reactive power. building and PV generation during the week

An industrial server is scanning the sensors and stores

the data on its internal hard disk. Via an independent
Internet connection the data is downloaded every morning
to the server of INES in France. In case of Internet failure,
the monitoring system continues as usual. As soon as the
connection is reestablished, the recorded days are
transferred.
5.2 PV-generation
The 80kWp system has a global performance of around
70% if all inverters are switched on. This performance
changes slightly with dust accumulation and temperature.
In the morning and evening, some small parts of the PV-
array are unavoidably shaded by the higher parts of the
building, and the shade of the high antenna in the middle
(figure 1) sweeps over a part of the array over the day.
The maximum output power of the PV-system is
therefore around 65kW at midday.

Photalia installed a water connection at the roof for

regular cleaning in order to avoid dust accumulation.
Access between the arrays is provided and the SGBB has
a maintenance contract with a local company.

5.3 Energy management

On a normal day, like in figure 5, no inverters are being
switched off, and the PV-generation follows the potential
of the PV-plant (reference yield, here corrected for
Figure 4: Monitoring and analysis software
system efficiency in order to visualize the abnormal
situations).
Monitoring and analysis software has been developed for
this specific project, because of the non-standard character
However, on other days, the inverters are being switched
of the system management. Values like “the number of
off while the energy consumption is enough to let them
switched-off inverters” and the corresponding energy loss
connected (figure 6).
had to be implemented. The 14 inverters need to be
analyzed individually, as well as the three individual phases
of the connection with the grid provider.

5 MONITORING RESULTS

5.1 Power consumption

The SGBB in Ouagadougou is the head office for the
African continent. Therefore it is equipped with an
extensive server system and its cooling. This installation
has a permanent base-consuption of 60kW. On Sundays
this is the only consumption present.

During opening hours, the consumption inceases up to

250kW. During lunch break consumption decreases to Figure 6: Unexpected Inverter switch-off
around 150 kW, and from 14:00 on it comes back up to
250 kW.

53/124
And on a Sunday, most inverters are switched off all day
(figure 7).
Figure 7: Inverter switch-off on a Sunday, with
apparently a power margin of about 30kW.
The reason behind this is the problem of the high phase
imbalance of the building consumption. The groups of
inverters are connected to a specific phase, and
equipment used during the day is not balanced over the
three phases. Regularly it happens that the consumption
on one phase is instantly decreasing, so inverters on that
phase are shutting down, while the other phases are still
under normal load. This means that 30% of the PV-plant
is shut down, while other phases can use that power
without any problem.
If we take a closer look at the building consumption in
the (not averaged) raw data, this surprising phase
imbalance is visible. For not injecting into the grid,
inverters are being switched of regularly.
Figure 8: Power consumption on the three separate
phases.
5.4 Sub-system inter-comparison
Due to the V-shaped roof and the sun moving from North
to South, part of the array is optimized for April and part
of the array for October.
Since inverters are actively switched of by the control
unit, daily performance per system does not give an
objective view. In order to visualize the difference in
performance between the two orientations, the maximum
AC-power during a day of every sub-system has been
displayed over time and normalized between them. Here
the difference becomes clear between the arrays facing
North and those facing South. Over the coming year this
image should inverse, because the modules facing South
will have the optimum orientation in April.
Figure 8: Normalized maximum power output of the
individual sub-systems during the day.
There is a structural difference visible between strings 6,
9 and 10 compared to the others. On average, the
difference in performance ration between the sub-systems
facing North and those facing South is around 8%. Some
inverters are in the middle yield, having strings North and
stings South. The points with very low performance are
days where an inverter was switched off most part of the
day.
5.5 Grid failures
Over the recording period, only 3 times a period of
around 20 minutes was detected where the grid failed and
the generator set started. During the hot season this will
be significantly more frequent, because the grid in
Ouagadougou is not able to handle all the air
conditioners in town.
Besides the grid failures, no situations were found where
the grid came out of the voltage and frequency windows
of the individual inverters!! Photalia has programmed
these windows extra large, otherwise the inverters would
have been off all day. On the other hand, August is also
the holiday month in Burkina so many people are out of
town. The grid is relatively calm.
6 CONCLUSIONS
First of all, the monitoring session needs to continue at
least over a year in order to draw conclusions on system
performance. During the hot season in March, much
more dust accumulation takes place, and the city grid
shows regular blackouts, which was not often the case in
August.
However, the main energy management of this non-
typical system can be analyzed over this period. From the
global system performance and the building
consumption, one can see that the basic idea of cutting
inverters when the corresponding phase has the tendency
to inject is also a cause of significant energy loss, because
that energy could potentially be used on one of the other
phases.
Due to a significant imbalance of the building
consumption between the phases, the optimal solution for
a maximum PV-yield must be found in an active way of
phase equilibration. This can be achieved by switching
54/124
inverters from one phase to another. The hard-switching
of inverters by relays however is not preferable as far as
lifetime concerned. Another, more gentle option is phase
equilibration via common battery storage and three
bidirectional inverters for the exchange of energy
between the phases. On one hand, a significant energy
gain can be achieved; on the other hand, this adds
complexity to the system and so reduces reliability. Also
the financial viability of such an investment must be
considered.

Regarding the total power consumption of the building

and PV-generation, only sometimes energy would be
injected for a short period around noon, especially on
Sundays. At present, it is not allowed to inject energy.
This limitation imposed by the grid operator leads to
rather complex solutions, as phase equilibration of the
whole building, energy transfer between phases, and
eventually inverter shutdowns if needed. This limitation
therefore also leads to significant energy loss and stress
on inverters by the regular switching of the output.

In the first study of this installation, the maximum power

possibly injected is 20kW per phase if no phase
equilibration is used, while over the same line 250kW is
used during the day by the same building. In order to
stimulate solar energy in Burkina and not to make
installations unnecessarily complicated and expensive,
regulations should allow the injection of PV-energy into
the grid. ◄

55/124
56/124
 Author's personal copy

Renewable Energy 36 (2011) 2334e2340

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Reliability of large-scale grid-connected photovoltaic systems

Gabriele Zini*, Christophe Mangeant, Jens Merten
Institut National de l’Energie Solaire, Commissariat á l’Energie Atomique et aux Energies Alternatives, 50 Av. du Lac Léman, 73377 Le Bourget du Lac, France

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a method for assessing the reliability of large-scale grid-connected photovoltaic
Received 7 April 2010 systems. Fault tree and probability analysis are used to compute the reliability equation and the
Accepted 30 January 2011 developed model is applied on military-standard data and on data taken from scientific literature.
Available online 26 February 2011
The method provides a tool useful to single out the different impacts that the large number of
components belonging to the photovoltaic field and the BOS (Balance of System) chain have on system
Keywords:
overall reliability, hence granting the possibility to design and implement more effective monitoring/
Reliability
diagnostic strategies and maintenance plans.
Photovoltaics
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction
Reliability is an important issue in large-scale grid-connected
photovoltaic (PV) systems as their operations rely on business plans
developed over periods of time of at least twenty years which often
assume fault-free functioning. Only rarely are faults factored in cash
flow budgeting, and maintenance plans represent a cost that
should be kept the lowest possible to improve economic appeal of
such large-scale plants.
Not many papers discussing PV systems reliability are available
in literature. For instance, [1] analyzes simple stand-alone PV
systems using failure mode effect analysis (FMEA) and fault tree
analysis (FTA). Failure rates estimates are also given assuming that
time to failure is exponentially distributed. The failure rate for a PV
array is hypothesized as being 33.3
106 failures/month, while
inverter failure rate is assumed to be 342:5$106 failures/month.
[2] derives reliability equations from the application of FTA on
stand-alone, grid-connected and utility-interactive simple block-
diagrammed PV systems. No statistical analysis is implemented,
and no estimates of reliability probability functions are computed.
[3,4] study the optimal interconnection of PV modules and use of
bypass diodes to maximize reliability of PV arrays. Architecture,
failure modes and failure probability are the three parameters
found to impact on array reliability. [5] presents a reliability eval-
uation method of an electric power generation system including PV
sub-systems, but reliability is considered here as the capability of
the PV system to provide power to the load depending on the
variability of meteo conditions, given all system components never
* Corresponding author.
E-mail addresses: gabriele.zini@cea.fr (G. Zini), christophe.mangeant@cea.fr
(C. Mangeant), jens.merten@cea.fr (J. Merten).
fail. The authors calculate a Loss of Load Expectation index as the
probability of the generation system not being able of meeting the
load demand. [6] studies the reliability of battery voltage regulators
(BVRs) used in PV systems; it considers a constant failure rate and
calculates the reliability of the overall system as the joint proba-
bility of the reliabilities for a series of system components, whose
failure behavior is independent from one another. The BVRs are
used in small-scale systems, and this contribution does not
consider a complete PV system. [7] models the reliability in a way
similar to the approach followed in this paper but with no FTA,
using an exponential distribution and estimating the single solar
cell as being equal to 0.0042 failures/(25 years). The focus is on
verifying the reliability of several different connection modes of PV
modules (series, parallel, series-parallel, total-cross-tied, bridge-
linked, and their different combinations), not on the complete PV
system. [8] employs highly accelerated life tests (HALTs) to deter-
mine potential failure modes for PV inverters. All results are
obtained experimentally by means of laboratory measures. The
paper constitutes a very interesting and complete essay of tests on
the main component of the BOS chain, but does not provide fault
probability density functions for inverters. HALTs are indeed not
intended to supply information on component life expectancies,
but rather to find failure modes which can be useful to engineers to
understand possible design or manufacturing issues in very hard or
even extreme testing conditions. [9] reports data from a field study
in Japan, outlining failure and maintenance data in mean time
between failure (MTBF) and mean time to repair (MTTR), with
failure rates computed as the inverse of the MTBFs and expressed in
failures/year. From the field study, it’s evident that none of the
monitored PV systems has reached the end of its intended life time
without failures, achieving a failure rate under 0.0032 failures/year,
with data taken for 1242 plants (with an average PV system power

0960-1481/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.renene.2011.01.036

57/124
 Author's personal copy

G. Zini et al. / Renewable Energy 36 (2011) 2334e2340 2335

of 29 kWp) over 7 years of operations. In this paper, 52 faults Table 1

relative to PV modules are reported. [10] analyzes the reliability of Characteristics of the PV module and inverter used in the design of the large-scale
PV systems.
stand-alone small-scale systems, suggesting that higher system
reliability is achieved by using module-integrated inverters. Some PV Module (230 Wp) Isc ¼ 8.24 A Voc ¼ 37.2 V Impp ¼ 7.60 A
estimates of module and inverter failure rates are also given (0.04 Vmpp ¼ 30.2 V mI ¼ 3.3 mA/ C mV ¼ 120 mV/ C
Inverter (100 kW) Vmpp,min ¼ 450 V Vmpp,max ¼ 820 V Vmax ¼ 1000 V
failures/year for PV modules, 0.1 failures/year for inverters). [11] IDC,max ¼ 235 A
describes several problems arising in a grid-connected PV system,
quickly touching some inverter reliability issues from a qualitative
point of view.
None of the papers hereby briefly reviewed though analyze the
as a result of i.e. partial shading. In general, the protection can be:
specificities and actual electrical architecture of modern large-scale
a blocking diode (as represented in Fig. 1), a fuse or a circuit breaker.
grid-connected PV systems.
It is fair to note that such protection could also be omitted (and in
This paper presents a method to analyze and quantify the reli-
some real world installations is omitted) when the value of the
ability of large-scale grid-connected PV systems. This methodology
maximum inverse current resulting from the N1 number of irra-
is based on FTA [12e14] using an exponential distribution to model
diated strings, being N the total number of strings connected in
the fault probability density function.
parallel, is considered not to be dangerous for the string which is
It must be pointed out that using an exponential distribution is
not, at that time, operational and hence becomes a load for the
not necessarily the correct ex-ante choice: the real probability
other strings which inject their current in it. Furthermore, blocking
density function should be derived from experiments like acceler-
diodes tend not to be used since they introduce a power loss due to
ated life tests (ALTs), with Log-normal, Weibull or mixed-Weibull
their inherent voltage drop.
distributions being the likeliest outputs. Changing the function
After the string protections, DC circuit breakers give the user the
from exponential to a different one is not a big issue, the application
possibility to disconnect the PV field even under solar irradiation,
of the method staying the same, but running ALTs to correctly
for maintenance or safety purposes.
estimate pdfs should be considered as a fundamental step to be
The inverter is then added to perform the conversion from DC to
taken in order to obtain realistic reliability estimates.
AC. Surge protection devices (SPDs) protect each inverter from inlet
The paper relies on assumptions that need to be refined in
and outlet surges resulting i.e. from lightings hitting the installation
further development: failures are non repairable and components
directly or indirectly.
sustain no degradation during operation. Furthermore, as soon as
Downstream the inverter, on the AC side, a series of circuit
a component fails, the overall system is considered to be in a failure
breakers protect the AC lines as per normal electrical design
state. Consequently, also the evaluation of energy and economic
practice.
losses, tied to component degradation and the time needed to
The PV system reliability analysis ends at the point where the
restore full plant functioning, needs to be refined in following
system is connected to the transformer, not considered in this
contributions.
study.
However, results are drawn and discussed to provide the reader
Seven large-scale PV systems, with nominal power ranging from
with a possible interpretation and use of the developed method-
100 kWp to 2500 kWp , are designed in order to evaluate their
ology. Reliability data are taken from military standard manuals
overall reliability. To compute the total number of components
[15] and scientific literature (other reliability databases are avail-
needed for each system, the PV module and inverter with the
able in literature [16]). The military standard we employ [15] tends
characteristics shown in Table 1 are used in all systems. The
to estimate higher failure rates than real ones, positioning this
resulting number of components per each PV system is shown in
study towards a more conservative evaluation of real failure rates.
Table 2. It is possible to note that the number of components
But as already pointed out, ALTs would provide reliability data that
increases with the PV system intended nominal power output. Of
can be readily changed in the model in order to get more precise
course, using inverters or PV modules with different nominal
results useful when designing real-life installations.
power can change significantly the number of components in each
system.
2. Large-scale grid-connected PV systems The total number of components is computed by considering
only one main field electrical cabinet (or distribution board) in the
The electrical architecture of a generic PV system is shown in DC side. No sub-field cabinets have been added in order to keep the
Fig. 1. installation free from site specificities and maintain the lowest
The PV module strings are connected to the inverter by means of possible number of components. This way, only one level of
a protection to assure that no inverse current is running in a string distribution boards is considered. For instance, a 1 MWp installation

PV Module String DC Inverter AC Grid AC Differential Transformer

Protection Switch Circuit Protection Switch Circuit Breaker
Breaker

SPD SPD

kWh Load

PE PE

Fig. 1. Electrical architecture for the PV system.

58/124
 Author's personal copy

2336 G. Zini et al. / Renewable Energy 36 (2011) 2334e2340

Table 2
Number of components for each PV system.

Power (kWp) 100 200 500 1000 1500 2000 2500

PV modules 437 874 2166 4351 6517 8702 10868
String Protection 23 46 114 229 343 458 572
DC switch 3 6 15 27 42 57 72
Inverter 1 2 5 9 14 19 24
AC circuit breaker 1 2 5 9 14 19 24
Grid protection 1 1 1 1 1 1 1
AC switch 1 1 1 1 1 1 1
Differential circuit breaker 1 1 1 1 1 1 1
Connector (couple) 874 1748 4332 8702 13,034 17,404 21,736

Fig. 3. Failure probability for the exponential distribution (where q ¼ l). When
t ¼ q; FðtÞ ¼ 0:632 [14].
must resort to 229 string protections (whether diodes, fuses or MT
switches) which are placed in only one field cabinet per inverter.
Each inverter is equipped with 3 DC switches with DC fuses to
connect strings to the inverter. In case a 2-level distribution were to
be introduced in the PV field design, the number of DC switches
would increase by the number of sub-fields that connect the strings
together and relay them to their relevant inverter.
The analysis assumes that cables do not introduce failure modes
and that system design and installation are flawless, this way
granting the possibility to focus only on electrical/electronic
components failures. It is also assumed that the SPDs never fail in
short-circuit mode and that the measuring equipment is not
opening the circuit in case of failure.
The study analyzes the reliability of seven PV systems over
a period of time of twenty years, with an average of 8.5 h operations Fig. 4. Reliability probability for the exponential distribution (where q ¼ l). When t ¼
q; RðtÞ ¼ 0:368 [14].
a day. The failure rate unit is hence failures/hour.

3. Quantitative reliability analysis The reliability probability function R(t) indicates the population
fraction surviving time t, and is derived from Equation (1) knowing
3.1. Probability functions and fault distributions that it is the complement to 1 of the F(t) (the probability of success
being one minus the probability of failing):
The following concepts are needed to describe the basics of the
probability analysis of PV systems reliability. More details are ZN
available in [14]. RðtÞ ¼ 1  FðtÞ ¼ PrðT  tÞ ¼ f ðtÞdt (2)
The probability density function (pdf) f(t) represents the failure t
distribution of the component population over the entire time
range; the larger its value in t, the more failures will take place in The failure rate h(t) (also called hazard rate) is the rate of change
a infinitesimal interval around t. in the probability that a surviving product will fail in the next small
The failure probability function f(t) is the probability that interval of time, given by:
a component will fail by a specified time t; it is defined as the
f ðtÞ
cumulative distribution function (cdf) of the pdf: hðtÞ ¼ (3)
RðtÞ
Zt
The mean time to failure (MTTF), measured in the chosen units of
FðtÞ ¼ PrðT  tÞ ¼ f ðtÞdt (1) time, is:
N
ZN
and can be also interpreted as the population fraction failing before
MTTF ¼ RðtÞdt (4)
or at time t.
0

Table 3
Component adopted failure rates and bibliographic reference.

Component Failure Rate (106 failures/hour) Reference

PV modules 0.0152 [9]
String Protection (Diode) 0.313 [15] Sect. 6e2
DC switch 0.2 [15] Sect. 22e1
Inverter 40.29 e
AC circuit breaker 5.712 [15] Sect. 14e5
Grid protection 5.712 [15] Sect. 14e5
AC switch 0.034 [15] Sect. 14e1
Differential circuit breaker 5.712 [15] Sect. 14e5
Connector (couple) 0.00024 [15] Sect. 17e1
Fig. 2. Probability density function for the exponential distribution (where q ¼ l) [14].

59/124
 Author's personal copy

G. Zini et al. / Renewable Energy 36 (2011) 2334e2340 2337

Energy null or reduced

if PV irradiated (TopEvent)

Differential No input
Circuit Breaker to Differential
rupture Circuit Breaker

DCB

No input to AC Switch
AC Switch rupture

ACS

No input to Grid Protection

Grid Protection rupture

GP

No input to
AC Circuit SPD
AC Circuit
Breaker rupture short-circuited
Breaker

CBac SPD

No input to Inverter
Inverter Fault

INV

DC Switch SPD No input to

rupture short-circuited DC Switch

DCS SPD

No or reduced
Blocking Diode No input to
output from
rupture Blocking Diode
PV cells

BD CON PV

Fig. 5. Fault tree for the PV system in Fig. 1.

60/124
 Author's personal copy

2338 G. Zini et al. / Renewable Energy 36 (2011) 2334e2340

Table 4 Table 6
Fault tree basic event codes. Total component reliability [in %] for PV systems after twenty years of operations
(Note that a reliability of 0% means that there will be at least one component with
DCB: differential circuit ACS, DCS: AC or DC switch GP: grid protection a failure; it does not necessarily mean that the overall PV system will completely
breaker stop its energy conversion).
CB: AC circuit breaker SPD: surge protection device INV: inverter
BD: blocking diode CON: connector PV: photovoltaic cell Power (kWp) 100 200 500 1000 1500 2000 2500
PV modules 66.06 43.64 12.81 1.61 0.21 0.03 0.00
String Protection 63.96 40.90 10.91 1.17 0.13 0.01 0.00
DC switch 96.35 92.82 83.01 71.53 59.38 49.29 40.92
The time distribution of faults is assumed to be exponentially Inverter 8.21 0.67 0.00 0.00 0.00 0.00 0.00
distributed. The pdf for the exponential distribution is (see Fig. 2): AC circuit breaker 70.16 49.22 17.00 4.12 0.70 0.12 0.02
Grid protection 70.16 70.16 70.16 70.16 70.16 70.16 70.16
f ðtÞ ¼ lexpðltÞ (5) AC switch 99.79 99.79 99.79 99.79 99.79 99.79 99.79
Differential circuit breaker 70.16 70.16 70.16 70.16 70.16 70.16 70.16
and from Equations (1), (2), (3) and (4) (see Figs. 3 and 4): Connector (couple) 98.71 97.43 93.75 87.85 82.36 77.17 72.35

FðtÞ ¼ 1  expðltÞ (6)

RðtÞ ¼ 1  FðtÞ ¼ expðltÞ
hðtÞ ¼ l
MTTF ¼ 1=l
To obtain systems reliability estimates, Equations (6), (7), (8)
and (9) will be used.
As already mentioned, the exponential distribution is not
necessarily the correct distribution for faults occurring in electrical
or electronic components, as is the case for PV technology. The
analysis makes use of this distribution since in most reliability
databases only the failure rate l is reported or estimated, while
finding other distribution characteristic parameters (like Weibull’s
shape and scale factors) is not at all common. The real probability
density function for each PV component should be estimated by
means of long and relatively expensive ALTs.
Furthermore, from Equation (8) the exponential distribution
failure rate is independent of time, the exponential being also indi-
cated as the memoryless distribution. This confirms, again, that the
exponential distribution is not realistic if applied to the description of
life-long behavior of PV systems due to last many years.
This choice does not undermine the validity of the proposed
methodology, but at the same time poses serious questions over the
correctness of the results that occur by applying the methodology
to exponential distributions. This path is followed knowing that, as
soon as real distribution characteristic parameters will be made
available, adapting the methodology to the correct distribution will
be fairly straightforward.
The failure rate for the PV modules is obtained from [9]. The
inverter failure rate is obtained by considering a change-over time
of 1 in 8 years; we consider this as implying 1 failure in 8 years,
hence the failure rate becomes l ¼ 1=8=365=8:5 ¼ 40:29$106
failures/hour. All other failure rates are taken from [15]. All rates are
reported in Table 3.
3.2. Fault tree analysis
(7)
The techniques pertaining to FTA are used to understand the
(8) interdependencies between all the components belonging to the
overall PV system. A fault tree is indeed very useful to comprehend
the behavior of a complex system by analyzing the relationships
(9)
between the single components that, together, make up the whole
system.
All components in the PV system are connected in series,
meaning that if a single component fails, the overall system fails as
a consequence. A series connection, as far as reliability is concerned,
is not necessarily a series connection in the physical world; it is
a representation of the direct effect that a component has along the
chain of all other components constituting the overall system. For
comparison, a system with two components operating in parallel
might be more reliable than a two-component series system since, if
one component fails, the other paralleled component might operate
as a back-up (the well known concept of redundancy).
In this study, system failure is intended not only as a complete
shut-down, but even as a small power loss due to a single cell in
a single module being damaged. This consists in a very strong
constraint, since a small power loss due to a single module can not
even be spotted in a large-scale PV system.
Other assumptions for the FTA quantitative analysis of large-
scale grid-connected PV systems are:

top event binary state;

hard failure (on-off, degradation of components not
considered);

non repairable failures;

independent events;

non mutually exclusive events;

good design (components are adequate and correctly
installed);

always on mode;

constant failure rate (exponential distribution).

For the assumptions taken, the system has only two modes of
Table 5 functioning: ON or OFF (top event binary state); it is constituted by
Total component reliability [in %] for PV systems after one year of operations.

Power (kWp) 100 200 500 1000 1500 2000 2500 Table 7
PV modules 97.95 95.94 90.24 81.35 73.41 66.18 59.72 System failure rates and reliabilities after one and twenty years of operations (Note
String Protection 97.79 95.63 89.51 80.05 71.66 64.08 57.36 that a reliability of 0% means that there will be at least one component with a failure;
DC switch 99.81 99.63 99.07 98.34 97.43 96.52 95.63 it does not necessarily mean that the overall PV system will completely stop its
Inverter 88.25 77.88 53.53 32.47 17.38 9.30 4.98 energy conversion).
AC circuit breaker 98.24 96.52 91.52 85.26 78.03 71.41 65.36
Grid protection 98.24 98.24 98.24 98.24 98.24 98.24 98.24 Power (kWp) 100 200 500 1000 1500 2000 2500
AC switch 99.99 99.99 99.99 99.99 99.99 99.99 99.99 System failure rate (,104 ) 0:72$102 1.32 3.14 5.71 8.74 11.77 14.80
Differential circuit breaker 98.24 98.24 98.24 98.24 98.24 98.24 98.24 Reliability (in %, 1 year) 79.94 66.22 37.71 17.0 6.64 2.59 1.01
Connector (couple) 99.93 99.87 99.68 99.35 99.03 98.71 98.39 Reliability (in %, 20 years) 1.14 0.03 0.00 0.00 0.00 0.00 0.00

61/124
 Author's personal copy

G. Zini et al. / Renewable Energy 36 (2011) 2334e2340 2339

Table 8 “Energy null or reduced if PV irradiated” causing the top event to

FusseleVesely relative importance measures for a period of twenty years of occur in case any basic event occurs.
operations.

Power (kWp) 100 200 500 1000 1500 2000 2500

PV modules 34.3 56.4 87.2 98.4 99.8 100 100 3.3. Minimal cut set calculation
String Protection 36.5 59.1 89.1 98.8 99.9 100 100
DC switch 3.7 7.2 17.0 28.5 40.6 50.7 59.1 To achieve a quantitative evaluation, the fault tree must be
Inverter 92.8 99.4 100 100 100 100 100 converted into a boolean expression and then into a probabilistic
AC circuit breaker 30.2 50.8 83.0 95.9 99.3 99.9 100
Grid protection 30.2 29.9 29.8 29.8 29.8 29.8 29.8
equation. Since basic events are connected by OR gates, it is quite
AC switch 0.2 0.2 0.2 0.2 0.2 0.2 0.2 straightforward to build the boolean equation of the basic events:
Differential circuit breaker 30.2 29.9 29.8 29.8 29.8 29.8 29.8
Connector (couple) 1.3 2.6 6.2 12.2 17.6 22.8 27.7
TopEvent ¼ ðððððððPV þ CON þ BDÞ þ ðSPD þ DCSÞÞ þ INVÞ
þ þðCB þ SPDÞÞ þ GPÞ þ ACSÞ þ DCBÞ (10)
non degradable components but can break (hard failure) without Equation (10) is the logical equivalent of the fault tree in Fig. 5.
being repairable (non repairable failures); the events that happen Since the assumption that the SPDs never fail in short-circuit mode,
during system functioning are statistically independent events and SPD ¼ 0(no effects whatsoever on the TopEvent) and Equation (10)
events are not preventing one another to happen (non mutually is reduced to:
exclusive events). The assumption that the components are not
degradable can be viewed as not realistic, but if we consider the
idea that a decrease in productivity due to PV modules degradation TopEvent ¼ PV þ CON þ BD þ DCS þ INV þ CB
is already considered in all large-scale PV systems business plans, þ GP þ ACS þ DCB ð11Þ
only a sharp performance decrease or malfunctioning will be
considered as a system failure. These hypothesis pair with the two A cut set is defined as a collection of basic events whose
already introduced assumptions of good design and constant failure occurrence will cause the top event to occur. A minimal cut set is
rate. defined as the smallest combination of basic events which, if they
The techniques used to draw the fault tree for large-scale grid- all occur, will cause the top event to occur [14]. Equation (10) shows
connected PV systems can be found in [12e14], a discussion of such that the fault tree can be expressed as the union of the nine
techniques being outside the scope of this paper. The fault tree for minimal cut sets; each cut set is equivalent to each basic event.
the architecture shown in Fig. 1 is depicted in Fig. 5. From probability theory, the probability of failure of the top
To help interpreting the symbols in Fig. 5, it is useful to know event, in case of union of the minimal cut sets, is given by the total
that the connections between the events are visualized in the form probability of the minimal cut set:
of a bi-dimensional diagram where:
PrðTopEventÞ ¼ PrðE1 þ E2 þ . þ En Þ (12)

a Circle symbol represents a basic event, with no downstream Since the events are assumed to be independent and non
fault analysis development. For a basic event, we need to know mutually exclusive, using the inclusion-exclusion rule, Equation
its pdf to perform a reliability analysis on the overall system; (12) can be demonstrated to be equivalent to:

a Square/Rectangle symbol represents an intermediate event
that can be developed into a combination of other intermediate 1  PrðTopEventÞ ¼ ½1  PrðE1 Þ$½1  Prð1  E2 Þ.½1
or basic events;  Prð1  En Þ (13)

an OR gate represents a logic gate whose output occurs in case
any of the inputs occur. Since the event probability Pr(E) is the failure probability,1-Pr(E) is
the reliability probability. The total system reliability is hence given by
Of course, many other symbols and gates are used in general FTA the product of the reliability of each event Ei where i ¼ 1; .; n being
theory. The codes used in the basic event circles in Fig. 1 are clar- n the total number of events in the fault tree cut set:
ified in Table 4.
The fault tree in Fig. 5 shows that all the components that are Y
n
RTot ¼ RðEi Þ (14)
outlined in the schematic in Fig. 1 constitute a series system (both i¼1
logical and electrical) which creates many issues from the point of
you of the overall system reliability: all basic and intermediate This product relationship entails that the more the components
events are indeed connected by means of OR gates, hence every in the series system, the less the total system reliability. In this
fault that is generated within the tree is propagated to the top event analysis, the total system reliability probability RTot is given by the
product of the components reliabilities:

Table 9 RTot ¼ RðPVÞ$RðCONÞ$RðBDÞ$RðDCSÞ$RðINVÞ$RðCBÞ$RðGPÞ$

Rank of relative importance and tentative solutions.

Rank Component Solution Effort RðACSÞ$RðDCBÞ (15)

1) Inverter Preventive maintenance, monitoring Normal Since an exponential distribution has been chosen, the total
2) String Protection (Diode) Preventive maintenance, monitoring High
system reliability becomes, using Equation (7):
3) PV module Preventive maintenance, monitoring High
4) AC Circuit Breaker Monitoring Normal !
5) Grid Protection Monitoring Normal X
n

5) Differential Circuit Breaker Monitoring Normal RTot ¼ exp  mi li t (16)

7) DC Switch Monitoring Normal i¼1
8) Connector Monitoring Normal
9) AC Switch Monitoring Normal
where mi is the total number of the same kind of component i (i.e. 24
inverters in the 2.5 MWp configuration), li are the failure rates of each

62/124
 Author's personal copy
2340 G. Zini et al. / Renewable Energy 36 (2011) 2334e2340
kind of component i, n is the total number of different components,
and t is the time frame chosen for the reliability analysis.
For example, the total reliability of the set of 24 inverters over
one year (3102.5 h) is given by (see Table 5):
 
RInv;Tot ¼ exp 24$40:29$106 $3102:5 ¼ 0:0498 (17)
The system total reliability is then the product of all component
reliabilities computed as in Equation (17) by changing relevant
component numbers and failure rates.
4. Results and discussion
Applying the failure rates in Table 3 to Equation (16) yields the
results summarized in Table 5 and Table 6 for one and twenty years
of operations.
It is straightforward to notice how reliabilities decrease with
system power. After one year, for a 100 kWp system the 23 string
protections (considered as a single sub-system) have 97.79% proba-
bilities of functioning without failures, while the inverter only 88.25%;
for a 2.5 MWp system, the 572 string protections (again considered as
a single sub-system) have 57.36% probabilities of functioning without
failures, while the 24 inverters only 4.98% (Table 5).
If considering twenty years of operations, the reliability esti-
mates drop radically. For a 2.5 MWp system, faults will occur with
more than 99% probability to the PV modules, string protections,
inverters, and AC circuit breakers (Table 6).
The overall system failure rates and reliability probability in one
and twenty years are reported in Table 7.
With the failure rates and pdf adopted (see Table 3), in twenty
years of operations at least 11 out of 10,868 modules would fail. To
compute the energy loss due to a module fault, two fundamental
parameters are needed: the time to detect the fault and the time
needed to change the faulty component (known as the Mean Time
To Repair or MTTR). In case the 11 modules failed on year 10, they
were not replaced and were all placed in different strings, 11 strings
out of 572 strings would indeed stop converting energy for the
remaining 10 years. Assuming that 10 years would provide 50% of
total production (this way neglecting the performance decay of the
PV plant due to aging of components that would shift a higher
percentage of energy conversion in the first 10 years rather than in
the final 10 years), the energy loss would amount approximately to
only 1% of the overall production (i.e. 11=572$50% ¼ 0:96%).
In case the module failure rate was found (through more precise
accelerated life tests) to be ten times higher than the estimation
used here (i.e. 1.52E-07 instead of 1.52E-08), the number of faulty
modules would climb up to around 103. Always in the previous case
scenario, the energy loss would climb to an unsustainable 9% of the
total potential production.
As far as the inverters are concerned, using the failure rate in
Table 3, 23 inverters out 24 would have a fault over the twenty years
period. The energy loss caused by one inverter would be easily
traceable, but for a 2.5 MWp PV system, two weeks of lost production
per each inverter (assuming two week as the MTTR) would entail
a loss of more than 4% of the overall system production.
To understand the impact of each single component on the
reliability of the overall system, the FusseleVesely relative impor-
tance measures are calculated by means of Equation (18):
PrðE1 þ E2 þ . þ Eni Þ
IFV ðijtÞ ¼ (18)
FðtÞ
where Eni is the event that the components in the minimal cut set
containing component i are all failed, with ni the total number of
the minimal cut sets containing component i, and F(t) is the failure
probability of the system at time t [14]. The results for a period of
twenty years of operations are outlined in Table 8.
The FusseleVesely measures confirm the impact of the inverters,
string protections, PV modules and AC circuit breakers on the
reliability of large-scale grid-connected PV systems.
5. Conclusions
In order to extract useful information from this reliability
analysis, Table 9 reports the list of components ranked by relative
importance, and proposes some possible strategies to cope with the
issues introduced by large numbers of components in PV systems.
The only likely way to figure out faults occurring in PV modules
and string protections is to use automatic monitoring and diag-
nostic systems to capture reduced power output from small
defaults which can result, if not detected, in potential sources of
serious economical loss.
Periodical verification and politics of preventive substitution of
string protections (if present) and inverters can greatly improve
energy conversion output. Inverters can also be easily monitored
automatically.
Finally, circuit breakers can be equipped with automatic
switches that trip in case of malfunctioning.
A tentative interpretation of the degree of effort needed to
proceed with preventive maintenance or monitoring is also given
in Table 9. Advances in monitoring and diagnostic equipment will
greatly reduce these issues related with the use of large number of
components especially in very large PV power systems.
Further research on the subject will have to relax the assump-
tions set by considering, for instance, repairable failures or degra-
dation of components, and evaluate related energy and economic
losses in order to find means to improve real-life PV power plant
efficiency.
References
[1] Stember LH. Reliability considerations in the design of solar photovoltaic
power systems. Solar Cells 1981;3(3):269e85.
[2] Hamdy MA, Beshir ME, Elmasry SE. Reliability analysis of photovoltaics
systems. Applied Energy 1989;33(4):253e63.
[3] Al-Rawi NA, Al-Kaisi MM, Asfer D. Reliability of photovoltaic modules I.
Theoretical considerations. Solar Energy Materials and Solar Cells 1994;31
(4):455e68.
[4] Al-Rawi NA, Al-Kaisi MM, Asfer D. Reliability of photovoltaic modules II.
Interconnections and bypass diodes effects. Solar Energy Materials and Solar
Cells 1994;31(4):469e80.
[5] Jain A, Tripathy SC, Balasubramanian R. Reliabiity and economic analysis of
a power generation system including a photovoltaic system. Energy Conver-
sion and Management 1995;36(3):183e9.
[6] Mishra PR, Joshi JC. Reliability estimation for components of photovoltaic
systems. Energy Conversion and Management 1996;37(9):1371e82.
[7] Gautam NK, Kaushika ND. Reliability evaluation of solar photovoltaic arrays.
Solar Energy 2002;72(2):129e41.
[8] Islam S, Woyte A, Belmans R, Heskes PJM, Roiji PM. Investigating performance,
reliability and safety parameters of photovoltaic module inverter: test results
and compliances with the standards. Renewable Energy 2006;31:1157e81.
[9] Oozeki T, Yamada T, Kato K, Yamamoto T. An analysis of reliability for
photovoltaic systems on the field test project for photovoltaic in Japan, In:
Proc. ISES Solar World Congress; 2007. 1628e1632.
[10] Rohouma WM, Molokhiab IM, Esuri AH. Comparative study of different PV
modules configuration reliability. Desalinisation 2007;209:122e8.
[11] Eltawil MA, Zhao Z. Grid-connected photovoltaic power systems: technical
and potential problems - A review. Renewable and Sustainable Energy
Reviews 2010;14:112e29.
[12] Vesely WE, Goldberg FF, Roberts NH, Haasl DF. Fault tree handbook. U.S.
Nuclear Regulatory Commission; 1981. NUREGe0492.
[13] Electronic reliability design handbook, MIL-HDBK-338B. Washington DC USA:
Departement of Defense; 1998.
[14] Yang G. Life cycle reliability engineering. J. Wiley & Sons; 2007.
[15] Reliability prediction of electronic equipment, MIL-HDBK-217F, 217F Notice 1,
217F Notice 2. Washington DC USA: Departement of Defense; 1991e92e95.
[16] Akhmedjanov FM. Reliability databases: state-of-the-art and perspectives.
Pitney Bowes Management Services Danmark A/S. Roskilde: Ris National
Laboratory, ISBN 87-550-2809-8; 2001.
63/124
64/124
 II.
MANAGING THE FLUCTUATION
OUT PUT THAN PV-SYSTEMS

2.1 GUARANTIED PV: COUPLING WITH

STORAGE SYSTEMS

2.2 ENERGY MANAGEMENT

65/124
66/124
2.1 GUARANTIED PV: COUPLING
WITH STORAGE SYSTEMS

67/124
68/124
 MANAGEMENT OF PHOTOVOLTAIC POWER GENERATION WITH
A VANADIUM REDOX FLOW ENERGY STORAGE SYSTEM

Franck BOURRY1, Pierre BESSON1, Fabien CANAL2, Nicolas MARTIN2, Xavier LE PIVERT1, Marion PERRIN2
1
National Institute for Solar Energy (INES) / CEA-LITEN/DTS/Laboratory for Solar Systems (L2S)
2
National Institute for Solar Energy (INES) / CEA-LITEN/DTS/ Laboratory for Electricity Storage (LSE)
INES, 50 Avenue du Lac Léman, BP 332, F-73370 Le Bourget du Lac, France
Phone: +33 (0)4 79 60 19 17 - Fax: +33 (0)4 79 68 80 49
franck.bourry@cea.fr

ABSTRACT: Photovoltaic (PV) generation is characterized by a low predictability, a high variability and a low
controllability, which may be critical for the large scale integration of PV generation into power systems.
Consequently, PV systems can be combined with energy storage devices in order to reduce these integration issues.
This paper presents the operational results obtained from the combination of a 18 kWp PV system combined with a
10 kVA – 100 kWh Vanadium RedOx flow energy storage system. The combined system has been operated in order
to respect grid integration constraints proposed by the French energy regulation authority. Two energy management
strategies are proposed and implemented. Both strategies lead to a reduction of fluctuations of PV generation, which
enable large scale integration of PV generation into electricity networks.
Keywords: PV System, Storage, Grid Integration, Strategy;

1 INTRODUCTION

Photovoltaic (PV) generation is characterized by a

low predictability, a high variability and a low
controllability. Consequently, the large scale integration
of photovoltaic generation into power systems may be
critical regarding the secure operation of the power
systems. In order to reduce such impact, the PV power
units can be combined with energy storage devices in
order to limit their variability and increase their
controllability.
This work shows first operation results obtained by
combining a 18kWp PV power plant with a 10kVA-
100kWh Vanadium RedOx flow energy storage system.
This type of system is compatible with the French Figure 1: Description of the system operated at INES
regulatory framework. Indeed, the French Energy
Ministry published a “Call for tender” for building PV 2.2 Monitoring and control
power plants spread all over France, with a total amount Specific monitoring applications have been
of 5 MW. In this tender, PV plants built in French developed in order to collect real-time data from the PV
oversea “territories” must include an ESS for smoothing generation and from the storage system. Regarding PV
the production along the day. generation, the collected data include AC active and
The paper is organized as follows: section 2 describes reactive power, as well as the current and voltage, from
the system, from the PV and storage systems to the the 3 phases. The same data is collected at the storage
monitoring and energy management applications. Then connection point. Also, DC data from the battery
section 3 gives some details about the PV generation controller, such as State Of Charge, are also collected.
variation constraint and the proposed energy management The monitoring applications are Labview® programs
strategies for respecting such constraint. Section 4 which are run in remote mode. The communication
presents the results of the system operation for the two between the acquisition program and the PV and storage
strategies, and, finally, some conclusions and systems relies on TCP/IP protocol. A screenshot of the
perspectives are given in section 5. monitoring application for the storage DC data is given in
Figure 2.

2 DESCRIPTION OF THE SYSTEM

2.1 PV and storage systems

The PV system in the proposed demonstration field
test is a 18 kWp roof installation at INES facilities, close
to Chambéry (France). Such PV system, including its
inverter, is connected to the INES AC microgrid, called
PRISMES, as it is described in Figure 1. Also, the storage
system, including its bi-directional inverters, is connected
to the same microgrid.

Figure 2: Screenshot of the monitoring application for

the storage DC data.

69/124

2.3 Energy Management System (EMS)
The Energy Management System (EMS) is organized
as described in Figure 3. The database collects data from
the PV and storage monitoring applications, as well as
day-ahead forecasts of the PV generation. Such forecasts
will be used as inputs for providing a production plan for
the system, as detailed in section 3. Based on such
database, the EMS computes the operating setpoints for
the storage system. The main storage control parameter is
in this case the current delivered by the bi-directional
storage inverters. Such command is set to the inverters
through TCP/IP communication protocol.
Figure 3: Architecture of the Energy Management
System
3 ENERGY MANAGEMENT STRATEGIES
3.1 Operational constraints relative to grid connection
In 2010, the French regulation agency (CRE in
French), published for the French Energy Ministry a
“Call for tender” for building PV power plants spread all
over France, with a total amount of 5 MW. In this tender,
PV plants built in French oversea “territories” should
include a storage system for smoothing the production
along the day. Even if this tender was finally declared as
unsuccessful, the grid constraints proposed in this
document take into account current PV integration issues
in French overseas territories. Consequently, proposing
energy management strategies respecting these
constraints is interesting for future installations.
Technically, the grid integration constraints consist in
a limitation of the PV generation variability: the variation
of PV energy production between two consecutive 30-
minute-periods has to be lower than the energy
corresponding to 15% of the PV nominal power for 30
minutes. Such constraint is illustrated in the next Figure4.
Figure 4: Two consecutive 30-minute-periods for
illustrating the PV generation variability constraint
In this figure, the grey areas represent the energy
delivered during the T1 and T2 periods, which are two
consecutive 30-minute-periods. The constraint can then
be formulated as:
| ET2 - ET1 | ≤ Elim,
where Elim = 15 % x PPV x 30 min, and PPV is the
installed PV power.
In order to respect this constraint, we propose two
energy management strategies which are detailed in the
next two sub-sections.
3.2 Strategy 1: Continuous limitation of PV variability
In this strategy, the storage setpoints computation is
based on the real-time variation between the energy
delivered by the PV-Storage system during the two last
consecutive 30 minutes periods which are before the
control time. The strategy is illustrated in Figure 5. At a
given control time tc, the storage setpoint for the period [tc,
tc+∆t] is calculated from the energy difference between the
2 periods T1 and T2’, so that the variability constraint
given in section 3.1 is respected.
Figure 5: Real-time computation of the storage setpoint
in the case of the continuous limitation of PV variability
strategy
3.3 Strategy 2: Day-ahead scheduling
This second strategy consists in limiting the
variability of PV production by planning the combined
PV-Storage system generation one day ahead. Then the
storage power setpoint is calculated in real-time
according to the difference between the real PV
production power and the PV-Storage system production
plan. The strategy is illustrated in the figure below. At
18h00 the day before the operation, a day-ahead
production plan is computed by the EMS for the next
day, for the combined PV-storage system. During the
operation (day d), and more precisely at a control time tc,
the storage setpoint is calculated according to the
difference between the day-ahead production plan and the
actual delivered PV generation. It is important to note
that in this case, the variability limit is imposed to the
production plan. Then, the system will be operated in
order to be respect as much as possible the production
plan, and consequently, the variability constraint will be
respected by the system output.
Figure 6: Real-time computation of the storage setpoint
in the case of the day-ahead scheduling strategy
4 RESULTS
4.1 Strategy 1: Continuous limitation of PV variability
The strategy based on continuous limitation of PV
70/124
variability, which has been detailed in section 3.2, has
been implemented in the EMS; figure 7 shows the
operational results which have been measured during the
day of the 16th of November 2010. The dashed orange
line represent the instantaneous PV power generation.
The orange bars correspond to the energy delivered by
the PV system during 30-minute periods, without
considering the storage system. Alternatively, the green
bars illustrate the energy delivered by the combined PV
and storage system, which is the energy delivered to the
grid. Consequently, the variation limit criterion has to be
respected for the green bars. A focus is given to a period
of sharp increase of PV generation, represented in the
black rectangle in Figure 7. During this period (between
12:00 and 13:00), the storage is first charged and then
discharged in order to reduce the sharp increase of PV
generation. In that way, the energy variation between two
30-minute periods is reduced. Figure 8: Results of the operation of the storage system
for the day-ahead scheduling strategy

5 CONCLUSIONS AND PERSPECTIVES

This paper shows operational results from a 18 kWp

Figure 7: Results from the operation of the ESS for the
continuous limitation of PV variability strategy
PV system combined with a 10 kVA – 100 kWh storage
system. Also, energy management algorithms have been
developed and implemented. The energy management
strategies enable the reduction of the energy variations
from PV generation, and consequently may contribute to
a reduction of the negative impacts related to the large
scale integration of PV generation into electricity
networks.
Moreover, the energy management system could be
further improved, for instance by integrating short term
PV production forecasts for computing the storage
setpoints.

4.2 Strategy 2: Day-ahead scheduling

Also, the strategy based on day-ahead scheduling,
which has been detailed in section 3.3, has been
implemented in the EMS; figure 8 shows the operational
results which have been measured during the day of the
20th of October 2010. The red line represents the
production plan for the system, which has been computed
by the EMS. The PV power forecasts used by the EMS
are given by an operational PV generation forecasting
tool developed at INES. The PV generation corresponds
to the orange curve, and the green curve represents the
power delivered by the combined PV-storage system.
The results show that the production plan is respected
during the operation. Since the variability constraint is
respected by the production plan, this constraint is also
respected by the combined system output. Remaining fast
fluctuations, which we can observe from the green curve,
could be removed by further improvements of the speed
of the control loop.

71/124
72/124
PRACTICAL IMPLEMENTATION OF GRID CONNECTED ENERGY STORAGE SYSTEMS WITH LITHIUM-
ION BATTERY: THE GROW-DERS PROJECT

H. Colin1, X. Le Pivert1,J. Merten1, A. Barona2, JF Cousseau3, P. De Boer-Meulman4, G. Bloemhof4, J. Bozelie5, H.

Dietschmann6, G. Kourtis7, M. Okasinki8, E. Raaijen9

(1) CEA/LITEN INES, 50 avenue du Lac Léman 73377, BP332, Le Bourget du Lac, France,
Email : name.first name@cea.fr
(2) IBERDROLA, Menorca 19, E-46023, Valencia, Spain, abarona@iberdrola.es
(3) SAFT, 111 bd Alfred Daney, 33074 Bordeaux, France, jean-francois.cousseau@saftbatteries.com
(4) KEMA, P.O. Box 9035, 6800 ET Arnhem, The Netherlands, petra.deboer@kema.com, gabriel.bloemhof@kema.com
(5) LIANDON, Dijkgraaf 4 Duiven, The Netherlands, jan.bozelie@alliander.com
(6) MVV, Olbrichstr.50 D-60488 Frankfurt, Germany, dietschmann@t-online.de
(7) Electricity Authority of Cyprus, 11 Amfipoleos st, 2025 Strovolos, Nicosia, Cyprus, GeorgeKourti@Eac.com.cy
(8) Institute of power Engineering, Mikolaja Reja 27, 80-870 Gdansk, Poland, m.okasinski@ien.gdn.pl
(9) EXENDIS, P.O. Box 56 6710 BB Ede Keesomstraat 4, 6716 AB Ede, The Netherlands, e.raaijen@exendis.com

ABSTRACT: Development and use of Renewable Energy Sources is one of the key elements in European
Research. However connecting these sources to the electricity distribution networks causes significant
effects on the management of these networks. Bottlenecks are stability, security, peaks in supply &
demand and overall management.
Electricity storage systems provide means to overcome technical and economic hurdles for large-scale
introduction of PV and other distributed sustainable energy sources. The costs of different storage
technologies have dropped dramatically during the last years making introduction of storage systems
feasible. However, unfamiliarity with the techniques, lack of confidence and uncertainty with costs and
benefits cause very large inertia by end users (utilities, investors). Moreover, network components are
designed to be technical and economic effective for a (minimum) period of 40 years, causing conservative
behavior.
The latest up-to-date results of the laboratory tests of the first integrated system (Li-ion battery, inverter,
overall energy management system and the connection to the local distribution grid) are presented.

Keywords: Electricity storage; Grid connected system; guaranteed power; Lithium-ion battery; GROW-
DERS (Grid Reliability and Operability with Distributed Generation using Flexible Storage).

1 INTRODUCTION including intelligent prediction software. The

Transportable storage provides a flexible asset
for network management introducing possibilities
for deferral of investments in expensive switchgear
avoids stranded assets and creates possibilities to
enable economically viable operation of (smart)
distribution grids with large scale penetration of
DER/SER. The project GROW-DERS provides the
solution by using transportable (container) and
flexible storage systems in networks, with high
penetration of DER/SER, made possible by new
developments in power electronics. In this
innovative demonstration project operational
experience is gained and the technical and
economic feasibility of storage systems will be
demonstrated. More information, the project plan
and current results can be found in [1].
2 APPROACH
On four sites at different international utilities,
three different storage systems, one flywheel and
two Li-ion batteries, and the combination of all
storage systems will be monitored and evaluated
with a technical-economic assessment tool
storage systems have a magnitude of around
100 kW, able to support significant parts of the
network, whether grid connected or as an island.
Practical experiments will be monitored online, data
will involve all network loads, voltages and
currents, but also relevant external conditions like
solar radiation, temperature, wind speed, etc.
In parallel a tool is developed to assess the
benefits of the application of storage systems in
generic distribution systems. Storage is used for
keeping technical constraints and to support
commercial trade of energy. The tool will therefore
need to predict the expected generation and load of
the network. A distribution network operator will
validate the tool. Trainings with the tool will be
organized to improve the implementation of
connecting Sustainable Energy sources like Solar
Photovoltaic Systems to the electricity distribution
networks.
3 LITHIUM BATTERY BASED SYSTEM
DESCRIPTION

73/124
 The system is made of an ESI (energy storage
inverter) from Exendis (figure 1). It is a bi-
directional galvanic isolated inverter/battery charger
offering a window from 230 to 400 Vac 50 Hz for
the mains voltage and 336 to 448 Vdc for the
battery voltage. Its power level is 60 kW with water
cooling and 30 kW with build in forced air cooling
able to drain 90 kW during 10 seconds. The typical
efficiency of the ESI is 94.5 %. The ESI can run in
parallel with the grid, without the grid, with a
generator, and with other ESI’s. The functions
currently built into the ESI are: battery charge,
inverter (to island grid but also to supply active
power), peak shaving/buffering with generator.
Figure 1 – Energy Storage Inverter from Exendis
The Saft lithium-ion battery (figures 2 and 3)
offers 34 kWh of energy in a voltage window from
336 to 448 Vdc. The battery is able to be charged
under 8.5 kW (20 A) in less than 5 hours and to be
discharged continuously under 40 kW (100 A) or
under 75 kW for a short duration.
The battery is made of sixteen 24 V cell
modules and two battery management modules
(BMM) dispatched in two cabinets (figure 2).
Figure 2 – Li-ion battery cabinets from Saft
Each cell module contains fourteen VL45E Li-
ion cells stacks in two strings of seven cells in series
(2p7S) (figure 3). The BMM contains a circuit
breaker, a contactor, a human machine interface
(HMI) and a battery management controller (BMC)
for operational and safety functions. The
communication between the module and outside the
battery works by can-BUS.
Figure 3 – Li-ion battery cell modules
The communication between the ESI and the
battery is done via MODBUS. The communication
between the system and the central management is
done via GPRS (figure 4):
Figure 4 – Communication scheme
4 ENERGY MANAGEMENT SYSTEM
The management of the storage system is
performed at 3 levels:
A first level, called ‘planning’, prepares a day
ahead planning for the use of storage. The objective
of the algorithm is to maximize profit through an
application of 'trading' type.
It is based on:
- The characteristics and limits of the ESI
and batteries,
74/124
 - The state of charge provided at midnight, management system (EMS).
- The price of electricity provided for the
following day, with a half-hour time step, 5.1 Tests performed
- The cost per kWh for battery cycle.
The schedule is then used by the second layer: Several tests were identified and realised:
real-time management. - Test of communication between
The algorithm could also integrate some specific components,
models, as for example an evaluation function of - Storage capacity test: measurement of the
the ageing of the battery depending on usage available real capacity,
patterns, the depth of discharge… - Active and reactive power tests: reaction to
various active/reactive power set points,
The second level is that of real time management - AC Voltage variation test: reaction voltage
of the storage system. It defines, with a periodicity increase/decrease during charge and
of about a minute, the instructions to be sent to the discharge of the battery,
inverter. These are the active and reactive power to - Islanding mode test: detection of islanding,
be produced or consumed until the next set point. disconnection or operation as UPS,
It is based on: - Power quality test (under process):
- The characteristics and limits of the ESI reaction to voltage dips, voltage increase.
and batteries,
- The state of batteries (SOC, Vcellmax, 5.2 Communication test
Vcellmin, Tmax, Tmin),
- The minor alarms from the BMS, The aim was to verify that the communication
- The AC voltage (measured by the ESI). between storage system and control unit is working.
The management of the system (battery + inverter)
The third level is the real-time control of is done with a computer directly connected to the
components: the ESI and batteries. Control of the system (cable link): the implemented software is
inverter applies the set points given by the higher used to define the set points (voltage, power…) that
level while providing various other security features are sent to the inverter (ESI).
or internal diagnostic. The battery BMS aims at
avoiding premature ageing and operation outside
the normal range.

Planning

24h - Power
production/consumption SOC

Real time management

SOC, Alarm,
Vbatt, Ibatt,
Vcellmax, Vcellmin,
P, Q, Vdcmax, Idcmax, Pmeas, Qmeas, Vacmeas Tmax,Tmin

Real time control Real time control

ESI BATT
Figure 6 –Set point definition

5.3 Storage capacity test

Figure 5 –Energy management diagram
Three cycles (battery voltage between Vmin and
5 SYSTEM LABORATORY TESTING Vmax) of successive charge/discharge are applied
with set points of active power (10 kW for charge
Before the field demonstration, the system and -10 kW in discharge):
including Li-ion battery has been tested in the Lab
of Ines at Chambery in France. The objective was to
validate the main functionalities of Growders
system and check good operation. Thanks to the
BMM of the battery data like battery voltage,
discharge current and remaining capacity (state of
charge) is recorded. Set points of active power and
reactive power are sent to the ESI by the energy
75/124
 Battery Capacity Voltage variation test (case 4)

30 240 1

20 235
0

230
10
-1
225
0
0 10000 20000 30000 40000 50000 60000 70000 80000 90000
Capacity (Ah)

-2

Power (kW)
Voltage (V)
220
-10

215
-3
-20
U AC (V)
210
P (kW)
-4
-30
205

-40 -5
200

-50
195 -6
Time (s)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time (s)

Figure 7 –Battery capacity

Figure 9 – Negative then positive variations of Vac
during battery discharge
The measured battery capacity was in
accordance with the characteristics of the battery
Voltage variation test (case 4)
(84-90 Ah at C/3). 420 80

U DC (V)
415 70
SOC (%)

5.4 Active and reactive power tests

410 60

The objective was to verify the value and 405 50

DC voltage (V)

SOC (%)
stability of active (resp. reactive) power compared 400 40

to the set points, and to evaluate the current quality, 395 30

the balance between phases, the reactive (resp. 390 20

active) power produced. 385 10

Active Power Test

380 0
20 0 2000 4000 6000 8000 10000 12000
Voltage Limitation (DC voltage) Time (s)

15

Figure 10 – DC Voltage and SOC variations

10

The test showed that the battery was able to

Power (kW)

-5
0 1000 2000 3000 4000 5000 6000 7000 8000
withstand AC voltage variations.
-10

-15
5.6 Islanding test
Prior discharge to SOC of 50%

-20

-25
+ 5 kW/ - 5 kW + 10 kW/ - 10 kW + 15 kW/ - 15 kW + 20 kW/ - 20 kW The objective was to test the detection of the
Time (s)
islanding situation and the operation of the inverter
Figure 8 –Active power in normal and UPS modes.
In the case of the active power test, the active
power remains stable in comparison to the set Islanding test (normal mode)

points and measured reactive power remained very 300 0

low; a slight imbalance between phases was also 250 -1

stated. Results are similar with reactive power 200 -2

(without any imbalance).

AC voltage (V)

Power (kW)

Grid is
disconnected
150 -3

Grid is
reconnected

5.5 AC Voltage variation test 100 -4

50 -5

The objective was to test the behaviour of the Grid on

U AC (V)
P (kW)
Grid off Grid on

0 -6
storage system when facing AC voltage variation 200 220 240 260 280 300
Time (s)
320 340 360 380 400

(increase and decrease) during charge and

Figure 11 – Islanding in normal mode
discharge.
In case of islanding in normal mode, the inverter
is connected and the loads are not supplied with
energy any more.

76/124
 Islanding test (UPS mode)
economic benefits etc... An assessment tool will be
U AC (V)
260
P (kW)
1
developed to select the specifications of the most
255 0
suitable storage technology for a specific location
250 -1
and to determine the economic feasibility of this
Inverter start-up
electricity storage system application.
AC voltage (V)

245 -2

Power (kW)
240

Grid is off
-3
8 REFERENCE
235 -4

[1] www.growders.eu for general information

230 -5

on the GROW-DERs project.

Grid on Grid off
225 -6
0 50 100 150 200 250 300
Time (s)

Figure 12 - Islanding in UPS mode 9 ABBREVIATIONS AND ACRONYMS

A ampere
In case of islanding in UPS mode, the inverter Ah ampere-hour
power is reduced to the power level consumed by
BMC battery management controller
the loads.
BMM battery management module
BMS battery management system
5.7 Conclusion
EMS energy management system
All the tests performed during this initial phase ESI energy storage inverter
of lab tests show the system operates appropriately. HMI human machine interface
kWh kilowatt hour
6 FIELD TESTING PV photovoltaic
SOC state of charge
In the coming months field tests with Li-ion V volt
batteries will commence in parallel on the second Vdc continuous voltage
battery system, in Spain at Iberdrola facilities. Vac alternating voltage
The flywheel system will be field tested at a site
of Liander in the Netherlands. The combination of 10 ACKNOWLEDGMENTS
Li-ion batteries and the flywheel will be tested
together at another site from MVV in Germany in This program is conducted with the support of
2010. In 2010 the assessment tool will be finalized the European Commission.
and the overall findings will be disseminated.
Recommendations from GROW-DERS may help
to fine-tune regulation to overcome practical
hurdles. The practical experience collected during
the GROW-DERS project will thus help to speed up
the implementation of energy storage and help to
facilitate the use of dispersed generation.

7 CONCLUSION

In 2009 the start of the lab test of the integrated

first system (Li-ion battery, inverter, overall energy
management system and the connection to the local
distribution grid) was successful. The tests
demonstrated the compliance of the system with the
specification for a good operation in grid-connected
mode.
After this first validation of the battery system,
the lab test will continue at Ines on the second
Lithium-ion battery based system prototype with the
same objective to demonstrate the functionalities
brought by the GROW-DERS system.
The flywheel based system is also tested in
parallel in laboratory before installation in the field.
Besides technical results the GROW-DERS
project also considers organizational and other
experiences, like transportation, regulation,

77/124
78/124
 INNOVATIVE ENERGY STORAGE SYSTEMS FOR PV GRID-CONNECTED APPLICATIONS

H. Colin1, J. Merten1, A.Graillot2, X.Vallvé2, G. Sarre3, A. fedzin4, P. Gaillard5, JP. Smaha6

(1) CEA/LITEN INES, 50 avenue du Lac Léman 73377, BP332, Le Bourget du Lac, France – Tel. +33 4 79 44 45 48
Fax +33 4 79 68 80 49 – herve.colin@cea.fr , jens.merten@cea.fr
(2) Trama TecnoAmbiental S.L. (TTA), c/Ripollès, 46, 08026 Barcelona, Spain - Tel. +34 93 446 32 34 - Fax +34 93 456
69 48 - tta@tramatecnoambiental.es
(3) SAFT, 111 bd Alfred Daney, 33074 Bordeaux, France - Tel. +33 5 57 10 64 09- Fax +33 5 57 10 64 12
79 44 45 48 – guy.sarre@saftbatteries.com
(4) ENERSYS, ul. Leszczynska 73, 43-301 Bielsko-Biala, Poland - Tel. +48 33 822 52 23- Fax +48 33 822 52 24
79 44 45 48 – agata.fedzin@pl.enersysinc.com
(5) MAXWELL Technologies sa, route de Montena 65, 1728 Rossens, Switzerland - Tel. +41 26 411 8539- Fax +41 26 411
8505 – pgaillard@maxwell.com
(6) HAWKER sarl, rue A. Fleming, ZI Est, BP965, 62033 Arras, France - Tel. +33 3 21 60 24 43- Fax +33 3 21 60 25 74–
jean-patrick.smaha@fr.enersysinc.com

ABSTRACT: The number of decentralised electricity supply systems based on renewable energies has grown
exponentially during the last years. These systems improve the energy efficiency of the electric system as they are
installed closer to the location of the consumption. However, they are sometimes connected at the end of grid lines where
the grid may be weak. Therefore, these systems may have a significant impact on the electric system or be victim of
malfunctions.
INES-CEA, TTA, SAFT, ENERSYS, MAXWELL and HAWKER have launched a project cofinanced by the European
Commission in order to develop an inverter, dedicated to the injection of photovoltaic energy into low voltage grids, with
special features so that:
• The impact on the grid of the PV system is minimised and even more, the system provides grid support on demand,
• The performance of this PV system is increased,
• The end user is protected against poor power quality and outages of the grid.
The article describes the benefits of the inverter, the sizing of the components, the use of innovative technologies for the
storage system and the field validation of the concept.

Keywords: Small grid-connected PV systems, Battery storage and control, lithium, supercapacitor

1 INTRODUCTION lasting few seconds to blackouts such as the ones

In order to make PV electricity generation more
attractive from a technical point of view and increase its
acceptability, it is necessary to demonstrate its ability to
supply high-quality service, reliability and profitability.
Power electronic devices that decrease the impact of PV
generation on the grid, that provide additional services
such as power quality to the end-user and the support of
the grid for the utility are enabling technologies for this
increased penetration.
The present paper presents an analysis of the quality
of the grid and the interaction between grid and PV
systems in developed countries. These considerations
have led to the implementation of a project aiming at
designing and developing an innovative inverter fulfilling
the functionalities previously mentioned. Data about the
sizing of the different components, including the inverter
itself and innovative storage systems are presented in this
paper.
2 CONSIDERATIONS ABOUT THE GRID
experienced in the USA or in Europe on 5th of
November.
The frequency of occurrence of these events depends
on the strength of the electrical system. The situation is
quite heterogeneous in Europe: they are very scarce in
centralised and oversized grids, such as in Germany or
France, whereas they happen relatively often in weak
grids. For instance:
• A study made in 2004 indicates that costumers in
Eastern Europe are facing up to nine interruptions per
year with a total duration over five hours [1],
• A publication of the French Utility EDF reported an
average cumulated duration of interruptions in 2005 of
less than 1 hour,
• A case study in Spain has evaluated the interruption
duration according to the density of connections to the
grid. In urban zones, with high density, the repartition
is illustrated by Figure 1, which shows that a device
able to supply electricity for at least 3 hours would
cover about 90% of the interruptions.

2.1 Energy supply instability

Disturbances in the voltage supply can cause tripping
or even damage to sensitive equipments. These
disturbances include voltage sags, dips, transients, swells,
harmonics as well as short interruptions. They are
generally caused by weather, accidents or utility
equipment failures.
The interruptions can range from only minor events

79/124
 Repartition of interruption by duration in urban zone
phenomena is of great importance for electric systems
managers and producers.
40
The origin of these disconnections is not yet well
35
30
known as very few studies have been conducted on this
25
topic. Nevertheless some possible criteria are:
(% ) 20 • Voltage (increase of voltage as mentioned in
15 clause a, voltage sags),
10 • Frequency,
5 • Impedance.
0
0-3 3-60 60-120 120-180 180-240 >240
These disturbances or fluctuations on the grid are
Interruption duration (minutes) detected by the inverters, which disconnect the PV array
very rapidly.
Figure 1 – Repartition of interruption by duration in In given configurations [2], the loss of production
urban zone can reach some days more than 50 %.

2.2 Interaction between grid and PV systems c) Solution

The cumulative installed PV capacity has been
expanding this last decade by about 30% per year. This
increase is related to the share of grid-connected systems,
whose market is driven by feed-in tariffs in industrialized
countries.
Some studies related to PV systems connected to the
grid already show the mutual interaction of PV systems
and the grid.
a) The impact of PV systems on the grid
From a quality point of view, utility specialists’
opinion is that PV private installations do not have any
impact on the grid due to the limited level of power
produced.
But in the case of concentrated PV systems, the
situation may be more critical as it has been shown in
Oota City, Japan [2]: under certain circumstances like
high level of irradiance or low level of consumption -for
instance during week ends- the voltage at end of line, far
from the transformer, increases above the security
threshold and leads to disconnections and energy loss of
the conventional PV system.
From a network planning point of view, a large
coverage of the energy demand by PV systems can have
an influence on the network operation as the production
of green electricity enables to smooth the load curve: the
daytime peak of consumption may disappear, in a case of
pure grid-connection, as it has been illustrated (see
Figure 2) by the study [3]:
no PV penetration
20% PV penetration
900
800
700
600
500
W
These two types of interaction between grid and PV
systems are strong arguments in favor of a grid-connected
PV system including a storage function.
The interest is threefold:
Firstly a battery can store the energy in excess
• In case of inverter disconnection, the PV
electricity that cannot be fed into the grid is
stored; this avoids losses of energy, and thus
leads to an improved performance ratio of the
system,
• Or when the level is consumption is low,
storing energy instead of injecting on the grid
can limit the increase of voltage,
Secondly the stored energy can be supplied to the
loads during the evening peak of consumption so as
to reduce the amount of energy requested from the
grid, and thus smooth the load profile.
Finally in case of grid shortage, which can last from
few seconds up to few hours, the stored energy can
be supplied to the end user during this interruption.
We saw that autonomy of 3 hours would cover 90%
of the interruptions in urban configurations in
developed countries.
3 SOS-PV project
These considerations have led to the project of
developing an multi-functional grid-connected PV
inverter, including a storage function, dedicated to the
10% PV penetration injection of photovoltaic energy into low voltage grids,
with special features so that:
30% PV penetration
• The PV system provides grid support on
demand,
• The end user is protected against poor power
quality and outages of the grid.

400 3.1 Market study

300 To achieve this overall objective, the first step in the
200 project consisted in an analysis of the potential market
100 for such a system. This market can be split between short
0 term and long term applications.
•
00:00 02:00 04:00 06:00 08:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 00:00
Day hours
The short term application market is deduced
Figure 2- Urban load curve in Spain assuming different from the present quality of the networks, the
rates of PV coverage of the energy demands. existing market for PV systems and for
uninterrupted power supply systems,
b) The impact of the grid on PV systems • The long term application market implies a
The grid disturbances that PV systems suffer are change in markets and regulations so that the
mainly their disconnections leading to a loss of owner of a SoS-PV system receives a monetary
productivity. The understanding of these disconnection advantage for supporting the network by

80/124
 providing energy or by shedding from the grid
on demand.
3.2 Choice of scenario
In the short term scenario, a single feed-in tariff is
applied and the storage is used for UPS purposes and for
storing the PV energy in case of disconnection from the
inverter. In this case, the size of the array is determined
so as to cover the annual consumption and the inverter
should be able to feed the totality of the PV power to the
grid.
In the long term scenario, there are two possibilities:
• Either the system is designed for grid support
purpose and for securing the house. In this case,
the inverter does not need to be of the size of the
PV array since injection will happen only when
the grids requires support. The storage system
should have a larger size in order to store the PV
energy for later feeding,
• Or the system is operating with real time pricing
and it is of interest either to feed as much energy
as possible during the peak times or to shed
from the network.
For the study it was decided to focus on the long term
scenario.
3.3 Sizing of components
The sizing depends basically on the general
requirements for the system, the energy demand, the solar
radiation and the feed-in tariffs in each country.
a) PV array
We made the assumption that feed-in tariffs may
decrease and disappear in the long term. In this situation
there is no interest to have a large PV array: the right size
is the one enabling to cover the energy consumption.
According to the evaluation of the yearly
consumption of southern and northern countries in
Europe, and the optimum yearly productivity of 1 kWp of
PV modules, it was possible to determine the range of
power to cover the needs of a household: between 4 and
6 kWp.
b) Inverter
In a grid-connected system, the inverter only runs at
100% in few occasions (high irradiance), but has a low
efficiency the rest of the time.
Having a battery in the system permits to undersize it
as it is possible to store the PV energy during the peak of
generation and use it afterwards.
An analysis of the consumption profiles in Spain of
residential users and mixed type (i.e. residential and
SME) users in winter and summer times showed that an
inverter of 1,5 kW would be sufficient to secure the
critical loads (size for grid support and house secure
mode).
In the real time pricing mode, the inverter should
have the same size as the PV array to supply not only
critical loads. In this case connecting in parallel two 2,4
kVA inverters can be an optimum (as it falls below the
single phase limit of the German regulation) and gives
high modularity.
c) Battery
The battery has to be sized in order to fulfill two
ways of operation:
• Normal daily operation where the battery is used
to delay the injection of energy into the grid,
• Operation in case of grid shortage where the
battery should provide energy to critical loads.
The previous consumption profiles were also used to
evaluate the size of the battery bank in the first mode.
The most constraining scenario led to a daily need of
11000 Wh. As said before 90% of interruptions would be
covered by a storage system with 3 hours of autonomy;
the need for critical loads is therefore 4500 Wh (3 hours
at 1,5 kW).
This makes a total capacity need of 15,5 kWh.
4 INNOVATIVE STORAGE SYSTEMS
The energy for the power quality and UPS functions as
well as for the grid support will be provided by 2 types of
storage systems that proved to be most adapted to this
application after evaluation in the INVESTIRE thematic
network.
• A lithium-ion based system,
• A hybrid system combining a lead-acid battery
and supercapacitors.
Both types of storage systems have never been used yet
in this application. They have the following advantages:
• Maintenance free operations,
• Long life duration,
• Positive impact on environment.
4.1 Innovative aspects
The first innovation in this project is to associate in
parallel a valve-regulated lead-acid (VRLA) battery with
supercapacitors as a hybrid system in order to provide the
following advantages for the lead-acid battery:
• Decrease the effect of the “coup de fouet”,
• Limit the depth of discharge,
• Absorb the peak power pulses in discharge,
• Allow to use a lower capacity thus decrease the
cost,
• Decrease the weight and volume.
The second innovation in terms of storage is to use a
lithium battery of large size in an application where it
was never used yet. This type of battery provides :
• High energetic efficiency (> 95 %),
• Operation whatever the state of charge of the
battery,
• Low weight and volume,
• Low life cycle cost.
4.2 Lithium-ion based system
The battery bank is based on Saft’s Li-ion VL45E
elements, which can supply a maximum of energy within
a compact and light packaging. It has a low self-discharge
and gives an excellent reliability during its whole
lifespan. It is highly suited to any charge/discharge
cycling application that demands a battery with
drastically reduced weight and volume.
81/124
 Elements are gathered by group of 14 cells in a
module; each module includes an electronic board for
safety management, cells balancing and data acquisition
(thermal and electrical state of the storage system).
According to the energy needs, 8 modules are assembled
in a cabinet according to a rack configuration (see Figure
3Erreur ! Source du renvoi introuvable.).
Figure 3 – Li-ion modules in cabinet
During normal operation the battery voltage will stay
between 380 V and 448 V; thus, there is no need of a
DC/DC converter to connect it to the 400 V DC bus
linked to the inverter. In case of emergency, during an
interruption of supply from the grid, this voltage may
drop down to 336 V.
The 112 Li-ion cells are able to provide the requested
15,5 kWh especially at the end of life of the battery (20
years).
4.3 Hybrid storage system
The hybrid storage system is composed of a lead-acid
battery and a supercapacitor, which will cover the power
peaks. The goal of this parallel construction is to extend
the lifetime of the battery by suppressing the high
currents.
a) Lead-acid battery
The battery bank is based on Enersys’ VRLA
batteries with AGM construction. Internal
electrochemical design has been adapted to the cycling
requirement in order to achieve a long cycle life (7-10
years).
To ensure this long life, daily cycles are made
between 50 and 90 % of the state of charge, which
statement leads to a global energy capacity of near 28
kWh. The reserve autonomy, in case of grid shortage, is
higher than 5 hours (between 50 and 20% SOC).
A prototype of the VRLA battery is designed to
achieve a low cost battery with longer life and better
reliability. This prototype is based on a 12 V module in
an existing box and lid with revised electrochemistry:
adapted grid design, adapted paste formulation, specific
AGM separator (see Figure 4 ).
Figure 4 – Lead-acid battery
The battery bank is made of 32 cells of 2 V put in
series, giving a nominal voltage of 64 V. A DC/DC
converter steps up the voltage up to the 400 V of the DC
bus. The battery management system is included in the
inverter.
b) Supercapacitor
The supercapacitor is based on Maxwell’s
BOOSTCAP® cells. It is sized to level off the peak
current from the battery and thus ensures the function of
a low pass filter.
The role of the supercapacitor is to supply current to
the loads if the demand exceeds the maximum tolerable
for the battery (i.e. maximum battery current allowed).
During the charge and discharge phases the voltage
across the BOOSTCAP® terminals changes from a
maximum value to a minimum value. In order to use the
available stored energy at high efficiency a voltage
variation of 50% should be set. The unit is coupled to the
VRLA battery DC-bus through a bidirectional DC/DC
converter.
One aim concerning this component was the increase
of the cell voltage of the capacitor, which results in an
increased power as well as energy density via appropriate
selection of the electrode/electrolyte combination and
improved electrode design, and the increase of the
electrochemical stability regarding temperature domain
and cycle life via the selection of electrode materials with
tailored pore structure, novel electrode design, and a
matching electrolyte.
The supercapacitor is made of 2 modules of 18 cells
of 2,7 V put in series, giving a operating voltage of 97,2
V. The capacity of each module is 165 F (see Figure 5).
Figure 5 – Supercapacitor
The module includes cell voltage management
82/124
electronics where each cell is monitored; this leads to
improved efficiency and operation.
c) Operation
The principle is that the battery delivers the energy needs
and the supercapacitor supplies the power needs.
In the case the storage system has to provide the
loads with energy, if the demanded peak current exceeds
the maximum current allowed for the lead-acid battery,
the supercapacitor is requested to supply this current;
otherwise the battery is discharged accordingly.
The supercapacitor is recharged by the battery itself
or the PV generator according to the load demand. The
battery is recharged by the PV generator (normal daily
charge and equalization charge regularly or when the low
state of charge threshold is reached).
5 VALIDATION
The individual components are tested individually to
check their characteristics and performance before
integration, and then the whole system is tested in the
field.
5.1 Battery testing
Tests are performed to check the initial characteristics
of the storage systems (capacity, efficiency, internal
resistance, self-discharge) and their cycling ability
according to the load profiles defined previously.
The test of the storage system (whole unit that is
connected to the 400V DC bus, i.e. storage device +
converter + BMS) are done independently of the
technology and all procedures are applied to one system
of each technology.
5.2 Inverter testing
Tests are performed to check the characteristics and
performance of the inverter: static power efficiency, total
harmonic distortion, power factor, start-up sequence,
losses, reaction to disconnections, and behavior of
MPPT.
Some tests related to the inverter with the storage
function will also be performed according to the standard
concerning UPS installations IEC 62040.
Electromagnetic compatibility tests are also foreseen.
5.3 System testing
Four PV systems (two with the Lithium based storage
and two with the hybrid storage) are going to be tested in
the field to validate the new functions: 3 sites (2 in Spain
and 1 in France) have been selected which exhibit
different classes of grid weaknesses and different ratios
between PV generation, storage, and consumption.
Typical single household system with a PV power of 2 to
3 kWp will be considered. The installations and site
management will be executed including monitoring
equipments. The systems will be tested (reaction to grid
interruptions and voltage perturbations) and monitored
during at least 5 months of operation for what concerns
the energy flows and the grid stability parameters. The
data resulting from the field tests and the field operation
will then be analysed for quantifying grid stabilisation
services provided and the efficiency of the systems
components as well as PV production comparing to
conventional PV inverters.
6 CONCLUSION
The present paper deals with the integration of
innovative storage systems in the PV grid-connected
installations in order to address the different problems
encountered in the low voltage distribution grids.
The interest of a storage function in the PV grid-
connected systems is to improve the mutual impact
between grid and PV systems, and increase the
performance ratio of the PV installations.
The modular architecture of the system enables to
select two different storage technologies using the same
system components: either on lithium-ion or a
combination of lead-acid batteries and supercapacitors.
Based on a long term scenario, in which the system is
designed for grid support purposes and the security of
supply for the house needs, sizing has led to a large
storage system in order to store PV energy for a deferred
feeding (11 kWh for daily cycling and 4,5 kWh as
reserve for emergency use) and an inverter power (2,4
kVA) much lower than the PV generator power (4 to 6
kWp).
Test procedures have been defined on component and
system level in order to validate the concept in the field
and assess the different storage solutions selected.
7 REFERENCES
[1] ERRA, EU accession Countries Working Group,
Quality of Electricity Supply – Comparative Survey,
April 2004
[2] Y. Ueda, K. Kurosawa, Performance Analyses of
battery integrated grid-connected residential PV
systems, 21st European Photovoltaic Solar Energy
Conference, Dresden, 2006
[3] A. Graillot, X. Vallvé, M. Perrin, E. Bosch, Interest
of a storage system in PV grid-connected
installations, 21st European Photovoltaic Solar
Energy Conference, Dresden, 2006
Acknowledgments:
This program has been conducting with the support of the
European Commission.
83/124
84/124
2.2 ENERGY MANAGEMENT

85/124
86/124
ENERGY MANAGEMENT STRATEGIES FOR OPTIMAL CHARGING OF ELECTRIC VEHICULES WITH
PHOTOVOLTAIC PRODUCTION

H. GUILLOU*, V.D. CUNG**, D.L. Ha*, M. JACOMINO**, J. Merten*

Herve.guillou@cea.fr,
Van-Dat.Cung@grenoble-inp.fr, Duy-Long.Ha@cea.fr,
Mireille.jacomino@grenoble-inp.fr, Jens.Merten@cea.fr
*CEA/LITEN INES Laboratory for Solar Systems (L2S),
50, avenue du Lac Léman - 73377, Le Bourget du Lac - France
Phone: +33 (0)4 79 44 52 60 - Fax: +33 (0)4 79 68 80 49
** G-SCOP/INPG,
46, avenue Félix Viallet - 38031 Grenoble Cedex 1 - France

ABSTRACT: Pushed by the need of clean energy sources as well as means of environment-friendly
transportation, photovoltaic energy and the electrical vehicles are two emerging vectors towards a more
sustainable future. For this purpose, the solar mobility concept has been developed: Aim is to facilitate the
massive grid integration of PV and to minimize the impact of electric vehicles on the grid. These vehicles can be
charge with PV to reduce greenhouse gas emissions, but they can also their battery system to provide grid
services. Key element for the solar mobility is the energy management system, which is discussed in this paper.
The first works will concern the charge the electrical vehicles thanks to the photovoltaic with the implementation
of the concept of shifting charge. Two models in Mixed Integer Programming have been proposed. The results
obtained by this optimization are presented.

Keywords: Solar mobility, Electric vehicle, shifting charge concept, Energy management.

1 INTRODUCTION approach is developed in section 4. Then numerical

Many international decisions highlight the need to
develop new ways of consuming and producing
environmentally friendly energy. In France, the
Commissioner General Sustainable Development [1]
argues that the photovoltaic (PV) and electric vehicles
(EV) are two priority areas to develop. When the electric
cars will be widely spread and become common, the
main difficulty to face is the electric power required to
charge their batteries. This increasing number of electric
cars may lead to a large increase of demand on the power
grid.
The PV resource is one of the local production being
developed. It is widely accepted that PV systems can
provide some benefit to the grid [2], [3]. In short term,
PV may limit the environmental impact, especially
during peak consumption. In the medium term, it
increases the global production capacity. These benefits
are particularly strong if such facilities are located in
urban areas or close to consumption places.
PV production is an intermittent resource; the
fluctuation depends on the weather but especially the
cloudy moment [4]. Indeed, it is possible to observe large
drops productions in one of these periods. In addition,
local PV sources and the large scale PV power have
significant effect on the network [5] [6].
It is therefore necessary to develop an intelligent use
of this new energy source to facilitate its integration into
the grid [7]. Therefore, the solar mobility can be a
solution to facilitate this integration.
In this article we will focus on the development of an
algorithm for vehicle charging system with the aim of
maximizing the use of PV production. The paper is
organized as follows: the problem of solar mobility is
introduced in section 2. Section 3 presents the
assumptions and the problem addressed. The solving
results obtained with the energy management strategy are
analysed in section 5. Finally, concluding remarks and
future works are addressed in section 6.
2 SOLAR MOBILITY CONCEPT
The new energy requirement involved by the electric
vehicles is not negligible compared to the current energy
consumption. Let us consider the example of France. One
can wonder what would be the consumption curve with a
35% penetration rate of electric vehicles. We will make
two assumptions on the use of electrical vehicles:
1) the vehicles are used only for commuting;
2) they are charged both when arriving in the
morning at work and in the evening at home.
The blue curve (Grid) in Figure 1 corresponds to the
current daily energy consumption from the grid in
France. The red dotted curve (GRID +VE) corresponds to
the daily energy consumption in France with the
integration grid charge EV with a 35% penetration rate.
The comparison of these two curves in Figure 1
shows that the integration of electric vehicles as new
loads in the grid leads to an energy need and a peak
power demand on the grid which is very significant.
Moreover instantaneous power consumption higher than
7 megawatts has to be satisfied. This peak corresponds to
a 10% increase in power called on the grid. The presence
of new peaks in consumption will imply a strong carbon
footprint. Indeed, in order to respond to such calls of
power, coal plants have to be set on. These plants are
responsible for a large amount of greenhouse gas
emissions. Consequently, the problem of gas pollution
due to the vehicles has simply been shifted from the cities
to the power plants.
To clear the power peak in the morning, a proposed
solution is to use renewable energy source such as
photovoltaic in order to charge the electric vehicles. In
Figure 1 PV production is assumed. The size of PV

87/124
system can meet the energy needs of electric vehicles. MW
This new hypothesis corresponds to a 2% penetration rate 90000

of PV.
85000
The green curve (GRID+VE+PV) in Figure 1 shows
the demand side curve to the grid including the electric 80000

vehicles as well as the photovoltaic production. One can 75000

notice that the solar production reduces the calls to the
grid but it is not available during the consumption peaks. 70000

The solar production is more intense at midday. This 65000

paper promotes the solar mobility concept. It’s is to
Grid
match the need of electric vehicles with the solar 60000

production. This may be possible by managing the 55000

Grid + VE + PV

vehicles charging. This solution would allow to reduce

50000
the peak consumption generated by electric vehicles

0
charging in the morning.

:3

:0

:3

:0

:3

:0

:3

:0

:3

:0

:3

:0

:3

:0

:3

:0
00

02

03

05

06

08

09

11

12

14

15

17

18

20

21

23
Figure 2 : Energy daily consumption in France with
MW
90000
electric vehicles, photovoltaic production and
management system
85000
7 MW
7 MW In order to apply this solar mobility concept, we do
80000
interest in using local resources PV energy. If we can
75000
completely charge the EV by using the PV production,
while the impact of the electric vehicles integration will
70000 then be reduced. Throughout the rest of the article, this
case of study will be focus.
65000

60000
3 ENERGY MANAGEMENT: PROBLEM
Grid
Grid + VE
DESCRIPTION
55000 Grid + VE + PV
In this issue of management of the electric vehicles
50000
charging, we will focus on the development of the
shifting charge concept. We assume that the only
:30
:00
:30
:00
:30
:00
:30
:00
:30
:00
:30
:00
:30
:00
:30
:00
00
02
03
05
06
08
09
11
12
14
15
17
18
20
21
23

parameter, on which we can act, is the start time of the

Figure 1: Daily load curve to the grid in France with
electric vehicles and photovoltaic production
Electric vehicles are also equipped with batteries, we
can consider that the electric vehicle is not only a
consumer of energy but it can also be considered as
energy storage. The electric vehicle can then be used as a
storage resource that helps to achieve the synchronous
property of the consumption needs and the production
abilities. This storage can also allow to switch the midday
solar production to match the evening needs. This is to
use a produced energy that is lost otherwise. .
As grid operator, the main goal is to ensure that the
production is always equal to the demand. The electric
vehicles can provide energy to the grid during peak
consumption, but it can also recover energy when there is
low consumption, for example at night. Figure 2
summarizes all these use cases. Comparing the current
curve (GRID) with the green curve (GRID+VE+PV) that
takes into account the integration of electric vehicles on
the grid and the solar production charging of the electric
vehicles, the peak power consumption of vehicles is
cleared with a load management system. In addition,
electric vehicles provide energy to the grid around 19:00
to clear the evening peak consumption. Finally the
battery is charged at night to minimize its impact. The
scenario depicted in Figure 2 can be possible if there are
the energy management systems which able to control the
vehicles charging.
vehicle charging. For this, it is necessary to know a
number of data which are: the arrival dates of the
vehicles at the charging station, the departures dates and
the battery state of charge.
In the absence of such data, a usage forecast of the
vehicles which includes knowledge can
be used for the load charging requirements and
arrival time. The vehicles are assumed to arrive later in
the morning and to stay locked on the site during a time
window. These assumptions are very well adapted to a
professional car fleet. However the mathematical
formulations that will be presented can be used and
adapted for large ranges of data values and then
application field.
Also in order to achieve a good planning for vehicles
charging, it is necessary to know the energy requirements
of vehicles and to have a good assessment of the PV
production.
In this issue, the first main energy source is the
photovoltaic power supply. Unlike the grid power,
photovoltaic power cannot be controlled and varied
during the day. The variability of this resource leads to
new difficulties for planning issues where we try to use a
maximum PV energy during the day.
The second power source is the grid. This resource is
used here as a complement to photovoltaic production.
This energy source is constant over time, whose available
with maximum power at any moment is limited. This
energy source is primarily used to ensure the vehicles

88/124
charging for solving the variability of PV energy
production problem. In addition, when PV power is not
consumed by a vehicle, it will be sold automatically to
the grid operator.
The addressed problem is presented in Figure 3. The
source variation is illustrated as well as the shape of
recharge.
P We set up two Mixed Integer Programming (MIP)
T (h)
Figure 3: Charging electric vehicles with solar
energy.
In order to implement the Solar Mobility concept, we
are interested in setting up a solar charging infrastructure
into parking equipped with sun shelters. Indeed a parking
space is a floor area of approximately 15m². If the
parking is equipped with the same solar panel surface,
then the infrastructure is able to produce the equivalent
energy of a 30km vehicle course, each day. This value
can be compared to behavioral studies on commuting
showing that daily average use of a vehicle is about
20km. These facts show that developing such a PV
charging architecture, with a smart energy management
system, can be a solution to reduce the morning energy
consumption peak and provides a “storage ability” to face
the evening peak. This second part of the application is
not addressed in this paper.
A prototype has been developed including all the
necessary elements and described in Figure 4.
Figure 4: Infrastructure of a solar car charging station
using Solar Mobility
In this system, the principal element is the Energy
Management System (Figure 4). This system will use
different mathematical models for managing EV charging
according to the PV production forecast. In order to
respect the daily use of system and to minimize the
responding time of the charging system the algorithm
maximum execution time will be limited to one minute.
4 SOLVING APPROACH
This optimization problem is solved as a scheduling
problem with the following given data and constraints: 1)
several vehicles are assumed with one battery per vehicle
2) we have one charging station for all the vehicles with a
PV power production forecast. Moreover the arrival dates
of the vehicles at the charging station, the departure dates
and the battery characteristics are known.
models of this problem based on a given sampling time:
• Energy Planning model: The vehicle charging
is scheduled continuously all over the planning horizon.
The energy consumption over each sampling period is
optimized [8].
• Power Planning model: The vehicle charging
is synchronized with sampling periods [9].
4.1 Energy Planning Model
In this first case, we consider that the signal of an
electric vehicle charging is composed of a single phase
with a constant power. Therefore the charging time varies
depending on the state of charge of the battery. In
addition, a vehicle charging cannot be stopped and
resumed later.
A first approach aim is to ignore the time
discretization, planning vehicle charges continuously
over time. Indeed, if we manage to not take the time
discretization in the planning process into account, we
can get results with high accuracy
Then we associate with each period of the
discretization, a part of the energy required for the charge
(Figure 4). We can then distinguish two cases:
• Considering the period in which the charge begins
or ends, then it determines the power charging as E1 =
E1’ and E3 = E3’.
• If the period is in the middle of the charge, then the
power charging during this phase does not change .
In practice, the establishment of such a mixed integer
programming is extremely expensive in terms of number
of variables and number of constraints. Thus it implies
the use of a large discrete period (about 2 hours) making
lost of accuracy.
Figure 5 : Energy planing model
It is then found that when this optimization is
preferred, the two critical moments are at the beginning
and at the end of a vehicle charging. This can be
explained by the lost of information about the real power
retrieved from the network. The handle the instantaneous
peak power is not possible by using this approach (Figure
89/124
5). There is also a lost of information on production by can use smaller discrete period (about 10 minutes).
the PV discretization. The longer the time discretization
is, the more information will be lost.
This method allows us to obtain solutions whose
accuracy does not depend on the discretization used
initially. Thus, this model allows us to obtain a good
solution with high accuracy on our single variable
parameter: the starting charging time.
In the context of the implementation of the
management system, we can solve problems containing
up to 10 vehicles in less than a minute.

4.2 Power Planning Model

Works on the characterization of the batteries show

that their charging activity can be modelled by two
activity steps. The first step corresponds to a battery
charging phase (phase 1 in Figure 6). This first step
duration depends on the vehicle charge state at arrival.
The second step corresponds to an energy balancing
phase (phase 2 in Figure 6) and its duration is not
influenced by the vehicle charge state. The second step
requires less energy and less power than the first step.
These two steps will constrain the planning of the
vehicles charging. In addition, charging a vehicle will not
be interrupted for reasons previously detailed.
Figure 7 : Power Planing model
We note here that the accuracy of the solution
entirely depends on the discretization adopted in the
optimization because the charging is synchronized with
the period. In addition, increased discretization can get
closer to PV production curve.
Because of the decrease in the discretization period,
this model allows us to increase the planning accuracy
with respect to production forecasts of the day. In
addition, it also allows a better consideration of power
peak during the scheduling.
In the integration management system, we can solve
up to 20 vehicles within 1 minute.

5 RESULTS

Case studies of daily use of electric vehicles have

been solved and optimal charging of a fleet of vehicles
has been computed. Experience has established a pattern
of arrival of vehicles, established after 11 months of
operation. Figure 8 illustrates the distribution function of
arrivals.
In order to simulate the arrival time of a vehicle, we
draw a random value between 0 and 1. Using the
distribution function in Figure 8, it is possible to deduce
Figure 6 : Charging Activity of the battery the time at which the vehicle is parked. The range of the
vehicle battery on arrival in the resort is set randomly
In this model, charges are synchronized with the time between 0 and 50%.
period and the power is optimized in respect to the
available power at each time period. The accuracy of the
model depends on the time discretization. In fact if the
period discretization is too large, the solution will be far
from the optimal solution.
Moreover, the duration of the charge curve is
approximated to a whole number of periods (Figure 7).
To obtain such a curve, we have to model the duration of
each phase of the signal, in respect to the discretization
period. We first determine the new duration of “phase 1’
” to be a whole number of sampling periods. We always
approximate to the upper bound. Then we determine the
length of the “phase 2’ “to be a whole number of periods.
We always approximate this value in order to verify that
the total duration of the “phase 1' ” and “phase 2' ” is an
upper bound of the real charge duration. So, it is Hour
important to determine the needed sample size if we
don’t want to plan too much energy in the optimisation. Figure 8 : Distribution function of the vehicles arrival
In practice, the establishment of such a solution is time
less expensive than the Energy Planning Model thus we

90/124
 Earliest Charging
45
Power Consume
40
PV
35

30 38 %
25
kW

20

15

10

0
04/23 04/24 04/25

Time (days)
45
Power consume
40
Optimized charging Management PV
35

30

25
kW

20

15
95 %
10

0
04/23 04/24 04/25
Time (days)
Figure 9 : Grid power demand (blue) and solar power (red). Without energy management (upper Figure), there is a strong
power demand to the grid, which is reduced once energy management system has been activiated (lower Figure).
Furthermore, the energy management system increases the solar coverage of the vehicles’ recharge.

Fifty vehicles have been simulated for this operation,
distinguishing two cases:
• 40 users per day making a charge: their early
arrival range is between 7:30 and 10:30. Their leaving
time is set at 18:30.
• 10 users per day making two charges: for the
first charging, they go to work between 7:30 and 10:30
and leave at 12:15. For the second one, they join the work
place between 13:00 and 15:00 and end their day at
18:30.
The aim is to maximise the local energy
consumption. The optimisation procedure improves the
solar cover from 38% to 95%. This is illustrated by the
following graphs (Figure 9) in which “PV” means
photovoltaic production and “Power Consume” is the
consumption of the EV.
6 CONCLUSION
This study was motivated by the industrial interest in
energy management systems for the solar mobility
concept. The control mechanism optimizes the use of the
PV system output energy by anticipating the predictable
events and the users’ needs. Two models of the problem
have been developed to better manage the charge of the
EV. The simulation, in this paper, shows that we are able
to improve the use of PV production to charge EV. The
management system for EV with PV production is in
progress and is implemented in the real system in
CEA/INES.
commissionership in the sustainable development”
(in French). Octobre 2009.
[2] Perez R, Seals R, Herig C. “Photovoltaics can add
capacity to the grid”. National Renewable Energy
LaboratoryGolden, 1996.
[3] International Energy Agency “Photovoltaic Power
Systems Program”. Task 5. Grid connected
photovoltaic power systems: power value and
capacity value of PV systems, IEA-PVPS T5-11,
2002.4] A. Mills “Understanding Variability and
Uncertainty of Photovoltaics for Integration with
electric Power System”, Lawrence Berkley National
Laboratory, 2010.
[5] M. Thomson, D.G. Infield “Impact of widespread
photovoltaics generation on distribution systems”,
The Institution of Engineering and Technology 2007.
[6] J.V. Paatero, P.D. Lund “Effects of large-scale
photovoltaic power integration on electricity
distribution networks”, ScienceDirect, 2006
[7] E Caamano-Martin et al. “Integration between
photovoltaic Distributed Generation and Electricity
Network”, Wiley Interscience, 2008
[8] D. L Ha, “An advanced management system of
energy in the building to coordinate production and
consumption” (in French) PhD Grenoble University
2007.
[9] H. Guillou, D. L Ha, C. V. Cung, M. Jacomino,
“Power Allocation Problem in Charging Electric
Vehicles with Photovoltaic Production,” SCMIS,
2010.

REFERENCES :

[1] Study “Green sectors: the industrial sector strategies

of the green growth. Report of the general

91/124
92/124
HOW MUCH PV ENERGY WILL I PRODUCE TOMORROW? A FORECASTING TOOL WHICH FITS THE
FUTURE CONDITIONS ON THE FRENCH ELECTRICITY MARKET.

Sylvain Lespinats 1,*, Fanny Cugnet 1, Hervé Guillou 1, Xavier Le Pivert 1

1
CEA/LITEN INES, Laboratory for Solar Systems (L2S), BP 332, 50 avenue du Lac Léman
F-73377 Le-Bourget-du-lac, France.
*
corresponding author.
Contact : phone: +33 4 79 44 45 46, Email: name.first name@cea.fr

ABSTRACT:

The contribution of solar energy generation assets to the French electrical grid is fast growing. Because the weather is
intrinsically fluctuating and chaotic and photovoltaic production is obviously connected to weather conditions, the
supply of photovoltaic energy is fluctuating and chaotic as well. But the electrical grid must be balanced at any time.
To control the solar energy penetration into the network while keeping it balanced, photovoltaic production pattern
must be anticipated [1, 2]. That is how weather forecasts become very helpful to manage solar plants. Photovoltaic is
actually exempt from establishing a production guideline within French rules (unlike other energy plants), so as to
support the growing of renewable energy. However, when the percentage of unforeseeable energy exceeds a
threshold of 30% on a network, the energy administrator is allowed to disconnect a producer without any financial
compensation. Such a threshold is actually frequently reached in several French insular departments and territories
(French Caribbean islands, Réunion, Corse, …). Unfortunately, this threshold will not promote investments in
photovoltaic plants.
Our Vision - Lately, wind farms are asked to provide a production guideline as any other energy producer. We
believe that, sooner or later, a similar rule will be applied on solar farms, which will allow the removal of the
threshold on the share of unpredictable energy injected on the network. Otherwise, there is no doubt that a tool to
forecast the production will help photovoltaic producers and network managers.

1 AIM & APPROACH
We developed a forecasting system to predict the
photovoltaic production. Our system takes advantage of
weather forecasts supplied by specialised providers (the
actual system uses data from NOOA – US National
Oceanic and Atmospheric Administration –). Our aim is
to predict in a given day D (typically at 16h) the pattern
of production of a plant for day D+1 from 0h to 24h (our
time-step is 1-minute). This choice comes from the
electricity regulation rules in France.
We set up a semi-physical system: each considered
element corresponds to a physical quantity, whereas
models are adjusted thanks to machine-learning
procedures. Weather forecasts (including solar radiation
and temperature) are considered as well as electrical
losses. A remedial model has also been implemented.
Each day, the performed forecasts are stored to compare
them to real measures. In order to improve the efficiency
of our tool, learning algorithms are thereafter used to
adapt the system to particularities of each considered
plant.
2 TOOL DESIGN
evaluation can be very different according to the method
used to calculate the error between what was the forecast
and what has been measured (see [1-7]). Our calculus is
based on the mean absolute error (MAE). The electrical
power (forecasted as well as measured) along the day is
smoothed using half-an-hour windows and is normalised
by the power-peak. This methodology has been chosen
for its proximity with French rules for penalty calculus
when a power plant fails to follow its production
guideline. Night is disregarded. One MAE is calculated
for each day; all these MAE are then averaged in order to
provide a unique performance value for each solar
station.
The D-1 persistence model to forecast production in
day D+1 is chosen as reference forecasting system.
Indeed let us suppose that we want to use the day D
persistence. The forecast for day D+1 would need the
whole day D data (from sunrise to sunset). But, in
accordance with the French electricity market, this
forecast has to be provided before 4pm on day D (before
sunset), which cannot be done in time!
4 OUR SYSTEM IS MORE EFFICIENT THAN
PERSISTENCE

The forecasting system has two types of operating
modes: 1) Forecasting mode where the next day
production is predicted; 2) the training mode based on the
difference between what has really been produced the
day before and what was expected (figure 1).
3 EVALUATION
A special attention has been given to the proper
evaluation of the forecasting tool. Indeed, we observe this
Our system is actually tested on 11 solar plants in
France. For each one, we compare MAE for forecasts
with the persistence model and with our system. On the
considered plants, the persistence forecasts are clearly
less precise than our system (figure 2).
Moreover, as expected, the forecasting system
improves its performances while the time is slipping by.
Indeed, best results are observed for plants watched by
the system since a long time; and a clear link between the
number of available days of data and the efficiency of the

93/124
forecasting tool can be observed (figure
3).

Solar irradiation forecast Local irradiation Local solar irradiation

on several sites around model forecast on the plant
the plant for day D+1. for day D+1.
Solar irradiation
measured at day D-1.

Temperature forecast Temperature Local temperature

on several sites around model forecast on the plant
the plant for day D+1. for day D+1.

Temperature measured
at day D-1.

Theoretical production
forecast for day D+1.
Forecasts for day D+1
Inputs, day D
(output)

Losses model

Remedial model
Inputs, day D-1
Production measured
at day D-1.

Models are used to forecast

what will be produce in day D+1.
Models Models are modified according
to what appended in day D-1.
Production forecast
for day D+1.

Figure 1: Map of functions of the forecasting program. Rectangles correspond to inputs, smooth shapes correspond to
forecasts and arrows correspond to models. Let us consider the tool on day D. The next day (day D+1) weather and
production are forecasted (green elements). On day D, we also get the real production acquired the previous day (day D-1)
in order to improve models (orange elements).

Figure 2: Demonstrating the precision of the forecasting algorithm. Each circle corresponds to a solar plant. The size and
the colour of a circle correspond to the reduction of persistence error (noted as a percentage) corrected by our forecasting
system in the corresponding plant (values inside circles display these percentages).

94/124
Figure 3: Map: number of available days of data for each PV plant. The graph shows the reduction of persistence error by
our forecasting system efficiency as a function over the time elapsed since the plant is watched.
5 EVALUATION OF LEARTNING PROCEDURES
It is worth to notice that the usual risk for machine-
learning procedures is the over-fitting phenomenon. This
corresponds to cases where the output of the algorithm is
excellent regarding the learning dataset, but cannot be
generalized to new data points (test dataset). This often
occurs with there are too many degrees of freedom (too
many parameters to be learned) in comparison to the
number of learning data points. To avoid this drawback,
it is common practice to evaluate the algorithm on a
dataset (the so-called test dataset) which must be
different and statistically independent from the learning
dataset.
However, the present framework is somewhat original
because our tool should obviously be tested on its ability
to forecast the following day. And the fact is that
productions on one day and the next one are not
statistically independent (as reflected by the efficiency of
the persistence model). Thus, we can tolerate some over-
fitting, as long as it does not emphasise too specific
factors (knowing these risks, the choice of the learning
algorithms was challenging).
6 CONCLUSION
Our forecasting tool is based on learning algorithms.
We observed that forecasts become clearly better after
several days (it seems that 15 days generally allow
enough time to train algorithms enough to get acceptable
performances). However, we have to provide the best
forecast as possible from the first days. For that purpose,
we plan to use different strategies depending on the
quantity of available data for training: for the first days,
the best model is a priori the persistence; the actual
system should be used thereafter. We are now developing
more sophisticate algorithms for cases where several
years of data are available in order to reduce the forecast
error.
The current forecasting system is working satisfyingly.
Each day, forecasts are displayed in the control room at
INES for 8 solar plants (3 plants among 11 are not
followed anymore). The plant managers can upload
forecasts and use it as production plant.
However, three points should be investigated
1) It is well known that weather forecasts
sometimes make mistakes. The impact of these
errors on our electrical power forecasts must be
analysed. The related uncertainty would
probably have an impact on the choice of the
production plan of the PV plant.
2) Remaining difference between production plan
of the PV plant and real output due to high
frequency fluctuations or small defaults can be
compensated by involving an electrical storage
system.
3) Moreover, if photovoltaic becomes balance-
responsible, we will be able to define the
optimal strategy to draw production plans soon,
by combining production forecasts and
hypotheses on prices on energy market.
The final goal of the developed tool is to help solar plant
managers to optimize the photovoltaic energy distribution
by providing production plan that match the French
legislation. Thanks to these plans, the need of electrical
storage systems associated to PV plants can be
minimized or avoided.
95/124
References
[1] Ernst, B., Oakleaf, B., Ahlstrom, M.L., Lange, M.,
Moehrlen, C., Lange, B., Focken, U. and Rohrig, K.
“Predicting the Wind” Power and Energy Magazine,
IEEE , 5(6): 78-89. 2007.
[2] Lorenz, E., Scheidsteger, T., Hurka, J., Heinemann,
D. and Kurz, C. “Regional PV power prediction for
improved grid integration” Progress in
Photovoltaics: Research and Applications.
doi: 10.1002/pip.1033. 2011.
[3] Bacher, P., Madsen, H. and Nielsen, H.A. “Online
short-term solar power forecasting” Solar Energy,
83(10): 1772-1783, 2009.
[4] Breitkreuz; H., Schroedter-Homscheidt, M., Holzer-
Popp, T., and Dech, S. “Short range direct and diffuse
irradiance forecasts for solar energy applications
based on aerosol chemical transport and numerical
weather modeling” Journal of Applied Meteorology
and Climatology, 48(9): 1766–1779, 2009.
[5] Lara-Fanego, V., Pozo-Vazquez, A.D., Ruiz-Arias,
J., Santos-Alamillos, F., Alsamamra, H. and Tovar-
Pescador, J. “Forecasting solar irradiance using NWP
models: an evaluation study in Andalusia (Southern
Spain)” Proceedings of SolarPACES Symposium,
2009. Berlin, Germany.
[6] Perez, R., Moore, K., Wilcox, S., Renne, D. and
Zelenka, A. “Forecasting solar radiation—
preliminary evaluation of an approach based upon the
national forecast database” Solar Energy, 81(6): 809–
812. 2007.
[7] Lorenz, E., Remund, J., Müller, S.C., Traunmüller,
W., Steinmaurer, G., Pozo, D., Ruiz-Arias, J.A.,
Fanego, V.L., Ramirez, L., Romeo, M.G., Kurz, C.,
Pomares, L.M. and Guerrero, C.G. “Benchmarking of
different approaches to forecast solar irradiance”
Proceedings of 24th European Photovoltaic and
Solar Energy Conference and Exhibition, 2009.
Hamburg, Germany.

96/124
 DESIGN OF A MANAGEMENT AND SIMULATION ’ S TOOL FOR SOLAR CAR PARK

Thang VU - Franck BARRUEL – Jens MERTEN

CEA/LITEN INES (Institut National de l'Energie Solaire) – Laboratory of Solar System
Savoie Technolac – 50 avenue du lac Léman – 73377 Le Bourget du Lac - FRANCE
Email :name.first name@cea.fr

ABSTRACT: The concept of Solar Mobility consists of making the maximum use of the synergy of the fluctuating
energy flow of PV systems and the grid coupled storage inherent to electrical cars. This paper shows that this concept
does make sense and that negative impacts on the electricity grid due to massive penetration of PV systems and
massive demand form electrical vehicles can be balanced.
It is the energy management system which has to assure this balance. This paper presents a simple tool to
evaluate the grid impact of the energy management strategy, which has to be designed considering a global vision
taking into account the state of the electricity grid, the state of charge of the vehicles, and the state of the PV
production .For demonstration, a sample strategy is simulated as a real case on the INES microgrid.
Keywords: solar energy, electric vehicle, Solar Mobility

1 INTRODUCTION 3 MANAGEMENT ALGORITHM CONCEPT

Recently, the electric car returns in the foreground.
The success or not of this novel mode of transport will be
strongly related on the new performances of storage’s
elements … and to the (re)definition of the real need of
the user. The objective of this article is to show an
interest (and feasibility) on convergence between solar
energy and the electric vehicles. It is not a question to
work on a concept of car with integrated PV panels but of
a “parking lot” equipped with PV panels supplying the
cars. For that, the first part is the assumptions with the
existing statistics on transport and the PV.
The second part will relate to the development of
management tools of the EV batteries connected to a
solar installation and the network.
2 TOWARDS A CONVERGENCE ELECTRIC
VEHICLE AND PHOTOVOLTAIC.
According to INSEE, by including the people who reside
and work in the same commune (27% employee) for
which the distance between home and office and the time
to work are conventionally regarded as null, the average
distance of residence-work is 25,9 km. For half of the
employees, the distance is lower than 7.9 km.
After calculations, for the route home-office (7.9 km),
half of French would require 3 kWh per day to go to
work and back with an electric vehicle.
It is noticed that 3 kWh/day is a value very compatible
with the daily production energy of a small sized solar
installation.
3 kWh/day = 1080kWh/year ≈ 1kWp ≈ 7m² of PV
However, if this solution is applied at a private
individual, that poses some disadvantages. The solar
installations produce only in the course of the day, when
the private individual left to work! If we think storing this
energy in battery the day, and charging the electric
vehicles the evening, the energetic output decreases
enormously. Thus it is necessary to create an operating
process where we can directly transfer solar energy
towards the electric vehicle’s batteries.
In our study we considered that the vehicles are
reloaded the day on the work place. Indeed most of the
time, the vehicle remains motionless and useless from
morning at the evening.
3.1 Proposed scenario
It is considered that:
•The solar panels are installed on the work place.
•Maximum 8 m2 of PV can be installed for each
vehicle in car park (the surface of a parking space is
11m²).
•The morning, starting from 8:00, the EV will be
charged during the work hour.
It is supposed then that there are two possible modes of
refill:
•Normal mode (or slow charge) of which the duration
is several hours
•Quick charge mode in 1 hour for the emergencies.
In the evening, when the employee returns in his home, it
connects the EV to the house socket. The battery is not
obliged to be immediately charged, it can provide part of
its capacity to decrease the consumption of energy of the
house at the rush hour for example (typically around
20:00). At 1:00 to 6:00 am, in off-peak hour, it can be
charged in order to reach the level necessary to ensure the
route home-office
The SOC is determined by its remaining energy. In
reality, we never discharge or charge the batteries at
100%, We put a limit for the SOC from 10 to 90%. We
put also a coefficient (to be fixed by user) for the output
efficiency of battery. This model is quite simple.
Figure 1 Daily Evolution of SOC
3.2 Algorithm
First of all, it is decided that the battery must reach a
certain level of SOC at the end of day's work. Thus if it is
the case of an electric vehicle, the battery must have
enough time to be recharged completely (the slow
recharge lasts only 6 hours). Then the first priority must
be the guarantee to reach the SOC required in end-of-
day.
Then, if many electric vehicles require at the same
time a quick recharge, it can create a strong demand of
instantaneous power. In this situation, the cables can be
congested even not supported. Thus if we do not want to

97/124
remake all the installation or to support a great voltage
drop, we propose the following solution: to discharge the
other batteries if they have enough energy or to suspend
their recharge. Although the solution of transfer of
battery energy from one to another does not give a good
energetic efficiency, it does not remain less interesting. In
the majority of the cases, the quick recharge never exceed
1 hour. The energy loss is not much. In comparison with
the cost of reinstall, this loss is reasonable.
Lastly, if the difference in tariff between peak hour
and off-peak hour can vary from 2 to 3 times it is
necessary to charge the batteries in the off-peak hours,
and during the peak hours, to make them discharge or
suspend the recharge. There are 2 interests. First, we
reduce the cost of total electricity in the course of the
day. Second, we can reduce the maximum consumption
power. That makes possible to sign a weaker subscription
of electricity.
Here the summary of the principles by set of
priorities:
• If there is enough time, the required SOC must be
satisfied.
• If there is a quick recharge, it is priority. Thus either
the non priority batteries recharge the priority battery, or
they are disconnected to decrease the total required
power.
• At peak hours, the batteries can discharge to limit the
maximum power, or to stop charging.
4 DEVELOPMENT OF MANAGEMENT TOOLS
It is advisable to have a simulation tool with which
we can test management strategies for the vehicles
batteries refill by measuring the impact on the low
tension network. With the current version of PSAT
(Matlab® toolbox for electric power system), we can
only make static simulations. The first step has been to
update the PSAT to include hourly load and production
profiles.
The second step has been to develop a man-machine
interface which includes all objectives named before.
5 ALGORITHM TESTED FOR SOLAR CAR PARK
CONCEPT
The example study is a solar station with 6 electrical
cars, each of 30kWh connected to the INES microgrid.
Two of these are service cars with a booster charge
option. A PV plant of 5kWp (about 6m²/car) is simulated
with meteorological data at Le Bourget du Lac.
5.1 Quick recharge without management tool
Now it is considered that there are 2 cars (5 and 6)
charging in booster mode (30kW each). The powers
distribution of the batteries from 1 to 4 is the same (5kW
each). For batteries 5 and 6, the quick recharge mode last
only a half an hour.
With this configuration, the total power request from
the car park is approximately 80kW. This one exceeded
the acceptable maximum power of the line. An important
voltage drop is then inevitable. This problem also appears
when we want to increase the number of EV to the car
park.
5.2 Quick recharge with management tool
The simple energy production by PV cannot cover all
the recharge instantaneously. Under these conditions here
the result of simulations with the strategy of management
defined by our algorithm.
The batteries recharge functions like the proposed
algorithm. The red arrow (8-8h30) indicates the quick
recharge, the green arrows (8h30-11h and 13h-17h) the
normal recharge and the blue arrow (11h-13h) the
discharge of the batteries during the peak hours.
Figure 2: Car park consumption power
For better distinguishing operation from each battery,
we look at the following figure:
Figure 3: Recharge powers distributions
When batteries 5 and 6 are in quick recharge, the
batteries from 1 to 4 discharge so that the maximum
power does not exceed 60kW. After half an hour, when
the quick recharge is finished, the non-priority batteries
(1 to 4) return in normal recharge. If we fixe the peak
hours from 11:00 to 13:00, batteries 1 to 4 discharge this
times. After 13:00, the batteries recharge up to the
required SOC.
6 CONCLUSION AND FACILITIES AT INES
We just showed that the concept electric vehicle for
home-work travel is interesting. Energy necessary to
ensure the travel is compatible with the daily production
of PV panels.
The connection of the EV to the network does not
have only advantages. The quick recharge can involve
voltage drops of lines. The idea charging the batteries by
batteries during this time can be an answer, even between
cars (case study in this paper) or with stationary batteries
under the solar car park.
However, it should not be forgotten that the lifespan
of the batteries is not unlimited. It is inversely
proportional with the depth of discharge. But we know
that with an adequate management, we can decrease this
effect.
The management tools are far from being perfect.
However we have a good base to imagine scenarios
which would not be limited to individual but collective
displacements. Overall, we sought to validate the broad
outlines of a concept which still requires advanced
technologies where the battery will remain critical. Li-ion
technologies are promising on the increase in the cycle
number and possible depths of discharge.
To validate the solar mobility concept, a solar park
will be installed and plugged at INES with 3kWp, 2
electric cars and some electric bicycles. This solar park is
connected to INES micro-grid (PRISMES) and
98/124
monitoring to control the power management system.
Finally beyond technological advances, there remains
one about it, and not the least; will of the people to
change their means of transport and their behavior.
And then which cannot be sulky the pleasure of
taking one day its pump assistant, “full of sun, please!”

99/124
100/124
 Energy Management System for a photovoltaic grid-connected buildiNG

D.L. Ha*, S. Ploix**, F. Wurtz***, P. Perichon*, J. Merten*

*CEA/LITEN INES - Laboratory for Solar Systems (L2S),
50, avenue du Lac Léman - 73377, Le Bourget du Lac - France
Email :name.first name@cea.fr
** G-SCOP/INPG, 46, avenue Félix Viallet - 38031 Grenoble Cedex 1 - France
*** G2Elab/INPG, Grenoble, France
ENSE3 bat D, 961, rue Houille Blanche- BP 46 -38402 St Martin d'Hères Cedex - France

ABSTRACT: This paper proposes a system for the global home energy management, which adjusts the electric energy
consumption and production for the building. The available flexibilities of the services provided by domestic appliances are
used to compute optimal planning for appliance controls. Three optimization criteria are used: the user's comfort, the
economic cost and the CO2 equivalent rejection. A three layers control approach is used to first compute
consumption/production coordination plans and then to dynamically adjust the computed plan to actual
consumption/production data. Application examples are given.

Keywords: Energy management system, Building Integrated PV, Grid-Connected

1 INTRODUCTION solar photovoltaic, supervision and automatic control. To

Load management allows inhabitants to adjust power
consumption according to expected comfort, energy price
variation and CO2 equivalent rejection. For instance,
during the consumption peak periods when power plants
rejecting higher quantities of CO2 are used and when
energy price is high, it could be possible to decide to
delay some services, to reduce some heater set points or
to run requested services even so according to weather
forecasts and inhabitant requests. Load management is all
the more interesting that local storage and production
means exist. Indeed, battery, photovoltaic panels or wind
mills provide additional flexibilities. Combining all these
elements lead to systems with many degrees of freedom
that are very complex to manage by users.
The objective of this study is to setup a general
mechanism that makes it possible to design optimized
building electric energy management systems able to
determine the best energy assignment plan, according to
given criteria. A building energy management system
consists in two aspects: the load management and the
local energy production management. [1] and [2] have
proposed optimal control strategies for HVAC (Home
Ventilation and Air Conditioning) system taking into
account the natural thermal storage capacity of buildings
that shift the HVAC consumption from peak-period to
off-peak period. [2] has shown that this control strategy
can save up to 10% of the electricity cost of a building.
However, these approaches do not take into account the
energy resource constraints, which generally depend on
the autonomy needs of off-grid systems [3] or on the total
power production limits of the suppliers in grid
connected systems.
The Multisol project focuses on photovoltaic grid-
connected buildings with a medium term vision when
tariffs of produced electricity by the PV are favourable.
The project adopts also a long term vision when tariffs
can be reduced or even disappear. The produced energy
will rather be consumed intelligently by coordinating the
energy consumption and energy production of this
building or by sharing with other buildings and the grid-
network when there is surplus. The consortium on this
project brings together complementary skills: electrical,
reinforce all, this project builds on the technical
demonstration at the Frecnh National Institute for Solar
Energy (INES).
In this paper, the hardware and software architecture
of MultiSol Energy Management System will be
presented (MEMS). The energy management problem is
formulated in mixed integer programming and will be
solved by a specific solver.
The MultiSol Energy Management System optimises
the use of PV system output energy by anticipating the
predictable events and the user’s needs. A study case of
PV housing is also presented for illustrating the control
mechanism.
2 MULTISOL ENERY MANAGEMENT SYTEM
2.1 Hardware architecture:
Hardware Architecture according to the building
standard is proposed. A load switchboard connects the
whole of loads (HVAC system, water heating washing
machine…). Furthermore, a resource switchboard
connects with the PV system, the grid and others energy
sources. (see fig 1)
Figure 1: Hardware Architecture of MEMS
2.2 Software architecture:
This architecture composes of different software
components (prediction component, control mechanism
component and interaction human-machine component).

101/124
 Multi layer control is the heart of this system and it
bases on a three layers control (Anticipative layer,
Reactive Layer and Local Layer). Anticipative Layer
deals with the predictable events for computing the plan
of energy consumption/prediction for responding to
user’s needs. The objective of the Reactive Layer is to
manage adjustment of this plan in spite of unpredicted
events and perturbations. The Local Layer composes all
embedded controllers inside the different devices.
The anticipative layer is responsible for scheduling
end-user, intermediate and support services taking into
account predicted events and costs in order to avoid as
much as possible the use of the reactive layer. The
prediction procedure forecasts various information about
future user requests but also about available power
resources and costs. Therefore, it uses information from
predictable services and manages continuously
modifiable and shiftable services. This layer has slow
dynamics and includes predictive models with learning
mechanisms, including models dealing with inhabitant
behaviours. This layer also contains a predictive control
mechanism that schedules energy consumption and
Figure 2: Schema of the 3 layers control mechanism
production of end-user services several hours in advance.
This layer computes plans according to available
predictions. The sampling period of the anticipative layer
2.3 Solver component
is denoted ∆. This layer relies on the most abstract
Anticipative control in home energy management can
models.
be formulated as a multicriteria mixed-linear
The reactive layer has been detailed in [4]. Its
programming problem represented by a set of constraints
objective is to manage adjustments of energy assignment
and optimization criteria. The detail of this formulation is
in order to follow up a plan computed by the upper
presented in [5]. The solver component aims to solving
anticipative layer in spite of unpredicted events and
the optimization problem of energy management. After
perturbations. Therefore, this layer manages modifiable
solving of the problem, the optimal control will be
services and uses information from observable services
obtained and send to the household device via
(comfort for end-user services and power for others).
communication network.
This layer is responsible for decision-making in case of
There are different ways of modifying services.
violation of predefined constraints dealing with energy
Sometimes, modifiable services can be considered as
and inhabitant comfort expectations: it performs
continuously modifiable such as the temperature set
arbitrations between services. The set-points determined
points in room heating services or the shift of a washing.
by the plan computed by the upper anticipative layer are
Some other services may be modified discretely such as
dynamically adjusted in order to avoid user
the interruption of a washing service. The different ways
dissatisfaction. The control actions may be dichotomic in
of modifying services can be combined: for instance, a
enabling/disabling services or more gradual in adjusting
washing service can be considered both as interruptible
set-points such as reducing temperature set point in room
and as continuously shiftable. A service modeled as
heating services or delaying a temporary service. Actions
discretely modifiable contains discrete decision variables
of the reactive layer have to remain transparent for the
in its model whereas a continuously modifiable service
plan computed by the anticipative layer: it can be
contains continuous decision variables. Of course, a
considered as a fast dynamic unbalancing system taking
service may contain both discrete and continuous
into account actual housing state, including unpredicted
decision variables. That’s why the mixed integer
disturbances, to satisfy energy, comfort and cost
programming is used here for converting the different
constraints. If the current state is too far from the
elements of the problem in the same formulation.
computed plan, the anticipative layer has to re-compute
Three criteria are used here:
it.
• Comfort criterion: it aims to minimize the
The local layer is composed of devices together
dissatisfaction of the energy management plant
with their existing local control systems generally
which is generated by the availability of the
embedded into appliances by manufacturers. It is
service (e.g. the time delay of washing
responsible for adjusting device controls in order to reach
machine) or thermal sensation of heating/Air
given set points in spite of perturbations. This layer
conditioner service.
abstracts devices and services for upper layers: fast
• Cost criterion: this financial aspect is generated
dynamics are hidden by the controllers of this level. This
by the cost of imported energy from grid and
layer is considered as embedded into devices: it is not
the income from exporting energy to grid
detailed into this paper.
network.
• Ecological criterion: is the CO2 equivalent
rejection by energy consumption

102/124
3 PROTOTYPE
Figure 3: The technical demonstration of MEMS
The technical demonstration and Life Simulator were
realized in INES/CEA in order to validate this
mechanism. This prototype composes a hardware
architecture which is presented section 2.
The building loads are simulated by resistors
connected to servo amplifiers which are controlled
according to the consumer profile of different equipment.
Energy sources: network, photovoltaic, etc. ... when are
they in real. It is also possible to replace these sources of
power supplies to reproduce at will and generate
scenarios of disruption of operation.
These devices are then controlled by software "life
simulation of house, which reproduces the profile of
energy consumption/production (instantaneous power
and energy stored on the actual equipment for example)
and possible sources (if they are emulated). You can
easily test different equipment types (different classes of
washing machine for example).
These devices are placed in service after 24 hours on
a schedule that mimics what a user would do in a day.
This scenario is planning a trial: for example operation of
the machine washing of 12 at 13.30, range from 12h to
12h30 ... etc.
The control system of MEMS will take control
several service of housing (e.g HVAC service, washing
machine etc…). Different scenarios are testing for the
comparison.
4 STUDY CASE
Description of the study case: The study taken into
account is a simulation of a grid-connected housing (see
figure 2). The tariff of peak period is 0.1074€\kWh and
off-peak period is 0.06540.1€\kWh. In this housing the
heating room service, washing service is controlled by
Mulisol system (i.e. the set point temperature of heating
service can be varied and the execution of washing
service can be shifted).. This housing has 30m² PV
installations. The PV output energy is chosen to be
consumed locally and the surplus will be exported to grid
(long-term vision). The feed-in tariff is 0.1€\kWh.
A room heating service whose model is given in [5]
is used. According to the inhabitant programming, the
room is occupied from 6pm to 6am. Out of the
occupation periods, the thermal inhabitant dissatisfaction
is not taken into account. The washing service can be
shifted between 6am to 2pm.
The results of energy management system of Multisol are
illustrated by energy consumption/production plan for
duration of 24h in advance. The weather and energy price
forecast are illustrated the figure 4. Two configurations
are proposed for the comparison: the “without Multisol”
configuration and “with Multisol” one. “Without
Multisol” mode suppose that the user try to manage his
consumption by himself (turn on/off the device). The
second configuration bases on the optimal control of
MEMS.
Figure 4: The weather and electricity price forecast (top-
down: solar radiant, outdoor temperature, electricity
price)
The “Without Multisol” Mode the heating service is
turned off when the user go out and turn on when he
comes home (fig. 5). The important consumption of
heating service in the evening is observed for reaching
the heating set-point temperature. The washing service is
started at 6am when his leaving (fig. 7). The PV
production is presented in fig. 11. The local PV
consumption is tiny (only 2%) in comparison with the
grid feed-in amount (98%). The main raison that the
washing machine is started too soon when the PV
production is still weak. The criterion consumption cost
of 0.7€ is obtained and the thermal comfort criterion
0.16. This dissatisfaction is caused by the indoor
temperature is only 20.2 °C when the user comes home.
The value of the CO2 equivalent rejection criterion is
3452.2g.
The “With Multisol” Mode : The fig. 10 shows that
the Multisol system favours the consumption during the
period where the price of electricity is less expensive and
PV output energy is max. At noon, the heating service
overheats the room then the temperature reaches to
22.5°C then it goes down to 21.3°C when user comes
home. So the electrical energy is then stored in the form
of thermal energy in the envelope of building and in the
air. As a result, consumption of room heating services is
reduced considerably in the evening. There is also the
execution date of washing service is set to 2pm. Thanks
103/124
to this shifting, the local PV production is improved
(31%) (see fig. 13).
The value of the cost criterion is reduced 22%: 0.52€.
The value of the CO2 equivalent rejection criterion is
almost divided by two: 1216.2g. However, the comfort
criterion has deteriorated: it is equal to 0.2. Degradation
comfort translates into an increase of temperature in the
room in the evening.
By using the Energy Management System of Multisol,
the user can improve significatively the energy
consumption/production of the housing according to his
needs.
Figure 7: Washing service of “without Multisol”

Figure 8: Washing service of “with Multisol”

Figure 5: Heating room service of “without Multisol”

Figure 9: Grid-network energy production of “without

Multisol”

Figure 10: Grid-network energy production of “with

Multisol”
.
Figure 6: Heating room service of “with Multisol”

104/124

Figure 11: PV production and local consumption of
“without Multisol”
Figure 12: PV production and local consumption of
“with Multisol”
Figure 13: Proportion of PV production uses in two case
Figure 14: Comparison between two modes
5 CONCLUSION
This paper proposes an energy management system for
the global home electricity management problem, which
consists in adjusting the electric energy
consumption/production for habitations. A 3-layer
control mechanism has been proposed. The paper focuses
on the anticipative layer, which computes optimal
planning to control appliances according to inhabitant
request and weather forecasts. These planning are
computed using service models that include behavioural,
comfort and cost models. The technical demonstration
and Life Simulator were realized at INES in order to
validate this system.
References
[1] House, J. M., Smith, T. F., 1995. Optimal control of
building and HVAC systems. In : Proceddings of the
American Control Conference, Seattle, Washington.
[2] Zhou, G., Krarti, M., 2005. Parametric analysis of
active and passive building thermal storage
utilization. Journal of Solar Energy Engineering 127,
37-46.
[3] Muselli, M., Notton, G., Poggi, P., Louche, A.,
2000. Pv-hybrid powersystem sizing incorporating
battery storage : an analysis via simulation
calculations. Renewable Energy 20, 1-7.
[4] Abras, S., Ploix, S., Pesty, S., Jacomino, M., 2006. A
multi-agent home automation system for power
management. In : The 3rd International Conference
on Informatics in Control, Automation and Robotics.
Setubal, Portugal.
[5] Ha, D. L., 2007. Un système avacé de gestion de
l’énergie dans le bâtiment pour coordonner
production et consommation. Ph.D. thesis, Grenoble
Institute of Technology, Grenoble, France.
105/124
106/124
 ADVANCED SIMULATION PLATFORM M2C FOR THE DEVELOPMENT OF HOME ENERGY
M ANAGEMENT S YSTEM
Duy Long Ha, Benjamin Cinquin Lapierre, Pierre Besson, Franck Bourry
CEA/LITEN INES 50 avenue du Lac Léman, BP 377 – 73377 Le Bourget du Lac FRANCE
Email :name.first name@cea.fr

ABSTRACT- The development of Home Energy Management Systems is one of the contribution to the smart grid
deployment. This study presents the general concept relative to the Home Energy Management problem, and describes a
simulation platform called M2C, which is adequate to such problem. The aim of this platform is to be a support for the
development and the comparison between different energy management strategies. The Home Energy Management Problem
(HEMP) is illustrated through a study case, and the results obtained with the M2C platform show the high potential for
energy management.

KEYWORDS: Home Energy Management Problem (HEMP), optimization, simulation platform, model predictive
control, smart grid, home automation system.

INTRODUCTION

A home automation system basically consists of household appliances linked via a communication network allowing interactions
for control purposes (Palensky & Posta, 1997). Thanks to this network, a load management mechanism can be carried out: it is
called distributed control in (Wacks, 1993). Load management allows inhabitants to adjust power consumption according to
expected comfort, energy price variations and CO2 equivalent rejection. For instance, during the consumption peak periods when
power plants rejecting higher quantities of CO2 are used and when energy price is high, it is possible to decide to delay some
services, to reduce some heater set points or to run requested services even so according to weather forecasts and inhabitants’
requests. Load management is all the more interesting that local storage and production means exist. Indeed, battery, photovoltaic
panels or wind mills provide additional alternatives for power generation. Combining all these elements lead to systems with many
degrees of freedom that are very complex to manage by users.

A building energy management system consists in two aspects: the load management and the local energy production
management. (House & Smith, 1995) and (Zhou & Krarti, 2005) have proposed optimal control strategies for HVAC (Heat
Ventilation and Air Conditioning) systems taking into account the natural thermal storage capacity of buildings that shift the
HVAC consumption from peak periods to off-peak periods. (Zhou& Krarti 2005) have shown that this control strategy can save up
to 10% of the electricity cost of a building. However, these approaches do not take into account the energy resource constraints,
which generally depend on the autonomy needs of off-grid systems (Muselli et al., 2000) or on the total power production limits of
the suppliers in grid connected systems.

The household load management problem can be formulated as an assignment problem where energy is considered as a resource
shared by appliances and tasks are energy consumptions of appliances. (Ha et al. 2006a) presents a three-layer household energy
control system that is both able to satisfy the maximum available electrical power constraint and to maximize user satisfaction
criteria. This approach carries out more reactivity to adapt consumption. The study in (Ha, D.L 2007) brings a new global
formulation for home energy management problem by using the mixed linear programming and logical transformation. Several
efficient resolution approaches are also presented. In parallel, (Abras S. 2006) presents a multi-agent approach for HEMP; this
alternative approach could also handle the HEMP in power management problem.

In the literature, the HEMP is managed with different approaches offering different technical solutions. There are many choices of
the strategies for controling of energy consumption and energy production. The HEMP can be treated as well as an optimization
problem, or as a classic control problem, and more generally as an artificial intelligence problem. The real question is how to
compare the different energy management strategies? How can we justify the fact that we need intelligence integrated into the
Home Energy Management System? Also, how such improvements about energy management systems can be taken into account
for system sizing? The objective of this study is to bring some elements for answering theses questions. The paper is organized as
follows: in the next section, the general energy management problem will be presented. Section III presents the general simulation
platform called M2C (Multi Models, Multi Components) that makes it possible to design optimized home electric energy
management systems. This platform can help to determine the best energy strategy, according to given criteria. In section IV, a
study case will be presented. Concluding remarks and future works will be addressed in Section V.

HOME ENERGY MANAGEMENT PROBLEM (HEMP)

In (Ha,D.L et al, 2010) the definition of HEMP was introduced. Housing with appliances aims at providing comfort to
inhabitants thanks to services which can be decomposed into three kinds: the end-user services which produce directly
comfort to inhabitants, the intermediate services which manage energy storage and the support services which produce
electrical power to intermediate and end-user services. Support services deal with electric power supplying thanks to
conversion from a primary energy to electricity. Fuel cells based generators, photovoltaic power suppliers, grid power
suppliers such as EDF in France, belong to this class. Intermediate services are generally achieved by electrochemical

107/124
batteries. Among the end-user services, well-known services such as clothes washing, water heating, specific room heating,
cooking in oven and lighting can be found.
In this Home Energy Management Problem, we assume that we have several controllable loads in the housing. There are
several possibilities for modifying the load curve:

• Shifting services: load can be shifted in the time;

• Modifiable services: modifiable loads can be considered as continuously modifiable such as the temperature set
points in room heating services;
• Interruption services: those services offer the possibility to be cut-off and to be restarted later. Water heating
service is an example for this category.
• Accumulative services: theses services can handle the energy storage, for reusing it later, such as for example, a
water heating or a battery;
A service can also be characterized by the way it is known by a home automation system. The consumed or produced
power may be observable or not. Moreover, for end-user services, the impact of a service on the inhabitants comfort may be
known or not. Obviously a service can be taken into account by a home automation system only if it is at least observable.
Some services are indirectly observable.
An important issue in HEMP is the uncertainties in the model data. For instance, solar radiation, outdoor temperature or
services requested by inhabitants may not be predicted with accuracy. In order to solve this issue, in (Ha,D.L et al, 2010) a
three-layer architecture was proposed: a local layer, a reactive layer and an anticipative layer. The anticipative layer is
responsible for scheduling end-user, intermediate and support services, by taking into account predicted events and costs, in
order to avoid as much as possible the use of the reactive layer. The prediction procedure can forecast various information
about future user requests and also about available power resources and costs. Therefore it uses information from predictable
services and manages continuously modifiable and shifting services. This layer also contains a predictive control mechanism
that schedules energy consumption and production of end-user services several hours in advance. This layer computes plans
according to available predictions. The reactive layer’s objective is to manage adjustments of energy assignments in order to
follow up a plan computed by the upper anticipative layer in spite of unpredicted events and perturbations. Therefore this
layer manages modifiable services and uses information from observable services (comfort for end-user services and power
for others). The local layer is composed of devices with their existing local control systems, which are generally embedded
into appliances by manufacturers.
A strategy of energy management consists in configuring the anticipative and reactive layers. Inside the anticipative
layer, a configuration has to determine the service’s degree of freedom and the optimization method that will be used. The
reactive layer is also configured by determining the rules which allow to make a decision when there are deviations between
the reality and the plan computed by the upper anticipative layer. The number of configurations for a HEMS strategy is large,
so we need a tool that helps decision making for selecting the more adequate strategy. In the next section, the M2C platform
will be presented as a tool for designing and selecting energy management strategies.

M2C PLATFORM
This section gives a brief presentation about the simulation platform which has been used for carrying out the different
simulations in this study. Such simulation platform is called M2C, standing for Multi Models and Multi Components, and has
been developed by INES –CEA.

Figure 1: Description of the combined systems taken into account by the software;

108/124
 In a general way, the aim of the M2C software is to stand as a simulation platform for the development of advanced energy
management strategies for combined systems including photovoltaic generation. Such tool can be used for sizing and
simulating the different components of the system.
• Multi-component aspect: M2C is able to consider different power system configurations. As a result, it offers the user
the possibility to build its own system by defining its components from a models library, the simulation and control time
frames, as well as the desired energy management strategy. The software is developed based on an Object Oriented
Programming technique.
• Multi-model aspect: in a single simulation run, M2C is able to consider different models for the components of the same
system. Simplified component models can be used for computing the system control in a reasonable calculation time,
while advanced models are used for simulating the operation of the different components, taking those calculated
controls into account.

In order to guarantee the possibility to consider a wide range of system configurations, the energy management platform
M2C has been developed based on an acausal approach: the components’ state variables are determined at every time steps
by solving an equation system modeling the energy transfer. Moreover, the equation system for the components’ state
variables is automatically built from the system configuration given by the software user. In addition to such acausal
approach, the system control is taken into account into the system operation as a set of additional constraints or objectives, in
a causal way. Such system control is determined from the energy management strategy given by the user, and may take into
account advanced forecasts of the photovoltaic generation.
Both simulation and control time frames have been carefully integrated into the platform. The following figure describes
the interaction between the control time frame and the simulation time frame, for a system including shifting loads. This
illustration shows two control times TC1 and TC2, when the starting time for the different loads of the system are optimized.
The starting time schedule is determined in this example at 17h00 and 7h00 for the next 13 hours. Then, the simulation of the
system operation takes these starting set points into account, and simulates the operation of the given load. In that way, the
time management in M2C makes possible the implementation of the M2C software for real-time control applications.

CONTROL
Control time-
TC1 step ∆tC TC2

17h 7h 17h
Simulation time-
step ∆tS SIMULATION
Figure 2: Interaction between the simulation and the optimization time frame in the M2C software;

The software architecture is detailed in the Figure 3. On this figure, the flow chart on the left describes the different steps
for the operation of the software, while the right side gives the M2C libraries associated to the different steps.

1. The first step consists in defining the system, i.e. the components, the simulation and control time frames, as well as the
selection of the adequate strategy. This definition is based on real-world data included in the data library;
2. In the second step, the system control is computed from solving the optimization problem defined in the selected
strategy. Such strategy is taken from the strategy library, and the mathematical models used for the optimization are
described in the component library for control;
3. Then, the simulation is carried out, from the components model implemented in the adequate component library;
4. Finally, Graphical User Interfaces have been developed for the construction and the visualization of the results, using
generic libraries.
It is important to note that the output of the optimization-based control can be directly used as an input the real-time
control of real equipments. This functionality is under development.

109/124
 Figure 3: Architecture of the developed software and associated libraries;

STUDY CASE
To illustrate the software’s ability to compare the effects of different energy management strategies on the same system, we
present here a simplified example of a housing. It consists of the following components:
• energy sources :
o a 3 kWp photovoltaic power system,
o a grid connection with a variable electricity price,
• a component which manages the sale of the PV power surplus (price = 0,43€/kWh),
• loads which can be shiftable:
o an electric car,
o a washing machine,
o a dishwasher.
The other loads of the house are not supposed to be shiftable so they are not taken into account in the example.

Two management strategies are tested:

1. Loads occur at specified dates, which are not modifiable; self-consumption of photovoltaic power is privileged (with a
self-consumption bonus, as in Germany).
2. Loads can be shiftable: they are now defined by a time interval during which they can occur. An optimization process is
carried out to determine their optimal starting dates; the objective is to minimize the cost of the purchased energy
(assuming that PV power is free) and to limit the maximum power on the grid (<= 3kW). Self-consumption is also
encouraged.

As this second strategy is an anticipation process, it would require in a real case some forecast data of photovoltaic
production. But in this example we supposed those forecast data to be equal to the effective PV production, for a
simplification purpose.

The time parameters of the simulation are described below:

• duration: from 17:00 to 19:00 on next day, to see differences between night and day,
• time step : 15 min,
• loads’ starting times/intervals :

1st strategy 2nd strategy

Electric vehicle
20:00 10:00 [20:00 3:30] [10:00 17:30]

Washing
machine 20:00 10:00 [20:00 23:30] [10:00 13:30]

Dish-washer
20:00 10:00 [20:00 00:00] [10:00 14:00]

• the optimization process is carried out at 2 times per simulation : 17:00 and 7:00, each time for the 13 following hours,
as described in the Figure 2.

110/124
Graphic results of the comparison are shown on Figure 4 and Figure 5, energy and economic balance are presented in Table
1.

Figure 4 : Simulation’s results with 1st strategy

Figure 5 : Simulation’s results with 2nd strategy

1st strategy 2nd strategy

Loads energy 15.9 kWh
PV production 12.6 kWh
Energy from grid 12.6 kWh 10.8 kWh
Energy to grid 9.3 kWh 7.5 kWh
Self-consumption 3.3 kWh 5.1 kWh
Self-cons. bonus 0.8 € 1.3 €
Energy purchase 2.7 € 1.7 €
Energy sale 4.0 € 3.2 €
Profits 2.1 € 2.8 €
Table 1 : energy and economic balance comparison

Analysis of the values in Table 1 allows us to quantify the improvements made by the 2nd strategy. The final result is an
increase in profits (by 30%) due to a lower consumption of grid electricity and a greater self-consumption of photovoltaic

111/124
power. Thanks to a greater distribution of loads to times when electricity is cheaper, there is an improvement in the mean
purchase price of electricity from grid, from 0.21 to 0.16€/kWh.

CONCLUSIONS

In this paper, we have presented our software solution for HEMS development. The HEMP is a complex problem which has been
treated thought different approaches in the literature. In this context, the M2C platform has been developed in order to compare the
different strategies of HEMS. M2C is based on an acausal approach with two layers: the simulation layer and control layer. A
particular attention has been paid to the time management for combining the control and simulation time frames. We focused in a
study case for illustrating the

Software’s ability for comparing the performances of different energy management strategies. In the next step, the software will be
deployed in real systems. The library of M2C will be completed and enriched;

REFERENCES
1. Ha, D.L.; Ploix, S. ; Zamaï, E. ; Jacomino M., “Tabu Search for the optimization of household energy consumption,” The
2006 IEEE Internationnal Conference on Information Reuse and Integration IEEE IRI 2006, Heuristic Systems Engineering,
September 16-18 2006, Waikoloa Hawaii, USA.

2. Palensky, P. & Posta, R. (1997). Demand side management in private home using lonworks,Proceedings.1997 IEEE
International Workshop on Factory Communication Systems.

3. Wacks, K. (1993). The impact of home automation on power electronics, Applied Power Electronics Conference and
Exposition, pp. 3 – 9.

4. House, J. M. & Smith, T. F. (1995). Optimal control of building and hvac systems, Proceddingsof the American Control
Conference, Seattle, Washington.

5. Zhou, G. & Krarti, M. (2005). Parametric analysis of active and passive building thermal storage utilization, Journal of Solar
Energy Engineering 127: 37–46.

6. Muselli, M., Notton, G., Poggi, P. & Louche, A. (2000). Pv-hybrid power system sizing incorporating battery storage: an
analysis via simulation calculations, Renewable Energy 20: 1–7.

7. Ha, D.L. ; Ploix S. ; Wurtz, F. ; Perichon P. and Merten J., Energy Management System for a photovoltaic grid-connected
building 24th EU PVSEC and 4th World Conference on Photovoltaic Energy Conversion, 21-26 September 2009, Hamburg,
Germany

8. Ha, D.L. Un système avancé de gestion d'énergie dans le bâtiment pour coordonner production et consommation, PhD
Automation-Manufacturing, Grenoble, Septembre 2007.

9. Abras S. ; Ploix, S. ; Pesty, S. & Jacomino, M. (2006). A multi-agent home automation system for power management, The
3rd International Conference on Informatics in Control, Automation and Robotics, Setubal, Portugal.

10. Ha D.L.; Ploix, S. ; Jacomino, M. and Le M.H. (2010) Chaper 5 Home energy management problem:towards an optimal and
robust solution, ISBN 978-953-307-065-0, Publisher: InTech

112/124
TSTE-00141-2010 1

OPTIMAL POWER FLOW MANAGEMENT FOR GRID CONNECTED PV SYSTEMS

WITH BATTERIES

Y.Riffonneau, S.Bacha, Member IEEE; F.Barruel and S. Ploix

CEA/LITEN INES 50 avenue du Lac Léman, BP 377 – 73377 Le Bourget du Lac FRANCE
Email: name.first name@cea.fr

[3], PV production into the grid is considered as limited.

14000
Abstract—This paper presents an optimal power management Centralized grid connected
mechanism for grid connected photovoltaic systems with storage. 12000 Distributed grid connected
The objective is to help intensive penetration of photovoltaic (PV) Off grid
production into the grid by proposing peak shaving service at the

Cumulative power (MW)

10000
lowest cost. The structure of a power supervisor based on an
optimal predictive power scheduling algorithm is proposed. 8000
Optimization is performed using Dynamic Programming and is
compared with a simple ruled-based management. The 6000
particularity of this study remains first in the consideration of
batteries ageing into the optimization process, and secondly in the 4000
“day-ahead” approach of power management. Simulations and
real conditions application are carried out over one exemplary 2000

day. In simulation, it points out that that peak shaving is realized

with the minimal cost, but especially that power fluctuations on
the grid are reduced which matches with the initial objective of
helping PV penetration into the grid. In real condition efficiency
of the predictive schedule depends of accuracy of the forecasts,
which leads to future works about optimal reactive power
management.
Index Terms— Batteries, Photovoltaic power systems, Energy
management, Optimization, Dynamic Programming, Storage
I. INTRODUCTION
T o face the increase of the electricity demand, the
reduction of fossil fuel resources and the need of reducing
CO2 emissions, grid connected renewable power systems have
gained outstanding interest. In this context, wind and
photovoltaic generation appears as the most promising issues.
Especially, photovoltaic is highly developing in building
energy sector for which it is particularly relevant. This
progression clearly appears on Fig. 1 through distributed grid
connected systems increase. According to the incentive
policies to help PV development and taking into account the
reduction of the costs, expansion of grid connected PV
systems is expected to continue in the next decay. Because of
its intermittent and irregular nature, PV generation makes the
grid management harder. Consequently, for some authors [2]
Manuscript received August 1, 2010. This work was supported in part by
the ADEME (Agency of Environement & Energie management, France)
Y.Riffonneau & S.Bacha are with Univeristy Joseph Fourier at the
Grenoble Electrical Engineering Laboratory (G2ELAB), 961, rue Houille
Blanche 38402 St Martin d'Hères, France.
(yann.riffonneau@g2elab.grenoble-inp.fr)
S. Ploix is at the Grenoble Institute of Technology in the G-SCOP lab, 46,
avenue Félix Viallet - 38031 Grenoble, France
F.Barruel is with the National Solar Energy Institute (INES), 50 av. du Lac
Léman, Bâtiment Lynx 73377 Le Bourget du Lac, France.
0
2003 2004 2005 2006 2007 2008
Year
Fig. 1: Cumulative PV systems installed in the IEA country [1]
One of the major challenges for PV systems remains in the
matching of the intermittent energy production with the
dynamic power demand. A solution is to add a storage element
to these non conventional and intermittent power sources [4]
[5]. In this case, the hybrid system, composed of a PV
generator, local loads, electricity storage and the grid, can
perform many applications [6]. However, the integration of
grid connected storage is currently limited by two constraints:
1) Regulations:
At the moment, grid connected storage in an end-user
installation is forbidden to prevent abusive benefit from
electricity purchase of incentive feed-in tariff attributed to PV
power. However, it is expected that regulations will becomes
more flexible as the example of the new German feed in law
from EGG [7].
2) Power flow management:
An electrical storage element generates expensive
investment and operation costs with strong operating
constraints. In this context, and considering subventions are
restricted in a short-term future, the objective is to reduce
operation costs by managing the power flows in the system. It
is an optimization problem that consists of optimizing the use
of the storage, the use of the PV source and to match the local
production with the local consumption.
This paper deals with power flows management for grid
connected PV systems with storage (GPVS) with a focus on
optimal scheduling. Section II presents the analyzed system. It
describes the structure of the power flow supervisor proposed

113/124
TSTE-00141-2010 2

and introduces the power management and optimization tools. Performing peak shaving with the PV generator and the
Section III presents the modeling of the components of the storage helps the penetration of PV electricity into the grid.
system. Section IV describes the optimization method and its Firstly, it makes easier the grid management as the load profile
application to solve the power management problem. is smoothed. Secondly, it reduces the CO2 emissions as the
Section V carries out results from simulations in a specific peaks power are generally supplied by rapid start-up power
context. Section VI presents experimental application in real plants that are important CO2 producer.
condition and section VII concludes the study and introduces
B. Power flow supervisor
the reactive management concept.
To perform optimal power flow management, a supervisor
II. SYSTEM ARCHITECTURE whose structure is presented in Fig. 3 is proposed. It is
composed of 3 stages that are shortly described below. The
A. Presentation of the system input data and the time range of the output data of each stage
A scheme of the studied system is presented in Fig. 2. The are shown.
Past
different power directions are represented. The sign IPV, ICH, Tariffs
1 Forecasting stage Month to hours

convention in Fig. 2 is used as a reference throughout the SOC, SOH Predictions and forecast of input variables, learning from the past

whole paper. The main components of the hybrid system are

the PV generator, the batteries energy storage, the user loads, 2 Predictive optimization stage Hours to 10min

the distribution grid and the electronics power converters. The Predictive setting

parameters « SOC » and « SOH » are respectively the values

of the State of Charge and State of Health of the batteries. 3 Local command stage 1min to 1ms

Whose calculations are detailed in section III. Local command of the batteries converter. Physical link with the system

IBAT*(t)
PPV (t) < 0 Bidirectional PGRID (t) < 0
PV converter inverter = =
=
= =
= =
Grid =
PGRID (t) > 0
PBAT (t) < 0 PBAT (t) > 0
PLOADS (t) > 0
PV generator
Fig. 3: Proposed structure of the power flow management supervisor
Batteries =
converter =
The forecasting stage provides the necessary predicted input
data to the optimization stage. Because the more reliable the
User Loads forecasts are, the more efficient the power management is,
Batteries
SOC(t), SOH(t) many techniques to minimize the prediction error have been
Fig. 2: Power direction and sign convention in the system studied developed [8] and [9]. The forecasts are calculated from
months to hours.
According to the sign convention, the laws of physics During the predictive optimization stage, the power flow
require the power balance in the system described by (1). management problem is solved. It is called predictive
optimization as the resolution is built on the a priori
PGRID ( t ) = PPV ( t ) + PBAT ( t ) + PLOADS ( t ) (1) knowledge of the future from the forecasts. The step time of
this stage is from hours to 10 minutes.
The first limitations on the operation of the system are the The local control stage is in charge of assigning the
physical constraints formulated by (1), (2), (3) and (4). These command to the power electronic elements in order to apply
constraints are set to limit the batteries degradation and ageing. the power flow schedule from the optimization stage. This
stage operates within a time scale from 1 minute to 1
SOC min ≤ SOC( t ) ≤ SOC max (2) millisecond.
≤ PBAT ( t ) ≤ PBAT
min max
PBAT (3)
This study focuses on the predictive optimization stage. The
SOH ( t ) ≥ SOH min
(4)
forecast data (irradiance (GT [W/m²]), ambient temperature
(Tamb[°C]), user loads consumption profile (PLOADS[W]) and
Another limitation is imposed by the usage of the system: as electricity grid price [€/kWh]) are considered known. To find
mentioned before, where the target is peak shaving such as the out how to solve the power management problem, a quick
power exchange with the grid is limited to a maximum overview of the suitable tools is proposed in the next
threshold value as formulated by (5). subsection.

PGRID ( t ) ≤ PGRID
max
(5) C. Power management and optimization tools
The method to perform power management optimization is
chosen according to the nature of the problem (component,

114/124
TSTE-00141-2010
constraints and performance index).The main purpose of this
work is to find the power flow scheduling that minimizes the
energy bill of the owner of the system over the studied period.
The input data (irradiance, ambient temperature, user’s loads
consumption profile and electricity grid price) are variable and
their prediction is uncertain. It is a constrained problem in a
finite horizon because the length of the studied period is
known. The components (powers in the system) are “a priori”
continuous and the constraints and objective function are not
particularly derivable or linear. The proposed overview is
inspired from the fields of stand alone PV systems, grid
connected PV systems and Hybrid Electric Vehicles (HEV).
A ruled-based management may be applied. Power flows
are managed according to a set of case-triggered rules built
from heuristic knowledge [10]. Operation of the system is
restricted to the developed rules that have to ensure that the
constraints are satisfied, but optimization is not guarantee [11]
[12] [13]. A first step to optimization is to determine the rules
from a fuzzy logic controller such as in [14] & [15]. It is also
important to notice that the development of a set of triggered
rules becomes quickly fastidious because of the number of
possibilities.
As a first step, a non-optimal ruled-based algorithm has
been developed. It guarantees operation of the system with
respect to the constraints. The ruled-based powers schedule
will be used as the reference to compare the performances of
the optimization algorithms.
The most commonly used techniques for optimization are
Linear Programming (LP) [16] [17] [18], Dynamic
Programming (DP) [18] [19] [20] and quadratic programming
by formulating the problem in relaxing form with the
Lagrangian multipliers [18] [21] [23].
According to its name, Linear Programming implies that the
problem is linear. In our case, to formulate the problem on
linear form, it is necessary to integrate a binary variable and so
a discrete value variable [17]. In this case, LP cannot solve the
problem and Mix Integer Linear Programming (MILP) has to
be used, such as in [16]. MILP gives very good results as the
problem is correctly formulated and demands low computing
resources. However, the main limitation is the need of a
specific mathematic solver [16]. In perspectives of futures
work, it is interesting to notice that LP and MILP become also
very unsuitable techniques for reactive optimization
(modification) of the predictive strategy according to actual
measurement Actually in this case the problem has to be
solved again over the period left at each unpredicted
disturbances with the forecast (important computation time).
Dynamic Programming (DP) is a graph based technique
corresponding to the shorter path algorithms. Advantage is that
the performance index and the constraints can hold all the
natures (linear or not, differential or not, convex or concave
etc) and no specific mathematic solver is needed. DP can also
be used for reactive optimization by correcting the predictive
3
strategy at each unpredicted disturbance according to the real
values. As DP works on discrete or sequential problems,
evolution of the system has to be decomposed into several
steps [20] [21]. A similar approach has been used in [22] for
deterioration and maintenance model of wind turbines. The
weakness of this technique is its high memory needs when the
studied period is long and discretized with a small time step
[10]. However, it is not problematic if the computation
parameters are well chosen. In addition, the computation time
can be reduced by appropriate modifications.
Quadratic programming gives very good results since the
problem is formulated in relaxing form through the Lagrangian
multiplier. However, this method needs the objective function
to be convex (or concave), that generally implies
simplifications of the problem [18] [21]. In [21], it is carried
out that these simplifications are the main origins of the errors
between the ideal and the calculated solution. Also, quadratic
programming is suitable only for small problem (less than 50
variables) and works only with continuous variables. This
technique is a strong candidate for reactive optimization and
the most used in HEV application as in [23] and [24].
From this overview, Dynamic programming has been
chosen to perform optimal power flow management. The
Bellman algorithm, detailed in section IV, has been chosen.
III. SYSTEM MODELING
In this section, we present the modeling of the PV generator,
the batteries and the power converter. The grid is considered
as a perfect power source or power sink.
A. PV generator
The PV generator has been modeled by a linear power
source according to the ambient temperature and the irradiance
level [25] [26]. Output power of the total generator at the
maximum power point is obtained from equations (6) and (7),
where parameters are gotten from information available from
the manufacturer data sheets (Nominal Operating Cell
Temperature (NOCT) and Standard Test Condition (STC)).
 G
[ ( )]

PPV =  PPV ,STC × T × 1 − γ × T j − 25  × N PVs × N PVp (6)
 1000 
where PPV, PPV,STC, GT, γ, Tj, NPV_S, NPV_P are the generator
output power at the maximum power point (MPP), the rated
PV power at the MPP and STC, the irradiance level at STC,
the power temperature coefficient at MPP, the cell temperature
and the number of modules in series and in parallel that
composed the generator, respectively. The STC measures
conditions are Tj,STC = 25 °C, GT,STC = 1000 W/m² and wind
speed of 1m/s.
The cell temperature Tj is obtained from (7), where Tamb and
NOCT are the ambient air temperature and the nominal
operating cell temperature. The NOCT measurement
conditions are Tamb,NOCT = 20°C, GT,NOCT = 800 W/m² and
115/124
TSTE-00141-2010 4

wind speed of 1m/s. capacity losses calculated with (12).

T j = Tamb +
GT
× ( NOCT − 20 ) (7)
C ref ( t ) = C ref ( t − ∆t ) − ∆C ref ( t ) (11)
800 ∆C ref ( t ) = C ref ,nom × Z × [SOC( t − ∆t ) − SOC( t )] (12)
In this paper, the PV generator is composed of
polycrhistalline modules with 72 cells in series that are all Replacing Cref (t) in (10) by expressions (11) and (12) and
considered to be at the same temperature. The temperature the SOH according to the SOC variation is expressed by
coefficient ‘γ’ of the modules is 0.043% /°C, the ‘NOCT’ is (13).
45.5°C the ‘PPV,STC’ of each module is 165 W.
C ref ( t − ∆t )
B. Batteries SOH ( t ) = − Z × [SOC( t − ∆t ) − SOC( t )] (13)
C ref ,nom
This study has been performed with flat plate lead acid
batteries. The model presented corresponds to this technology.
Equation (12) is always positive as it is computed only in
discharge. The new capacity of reference calculated by (11) is
State Of Charge (SOC)
replaced into the SOC estimation (8) such as ageing involves
Estimation of lead acid batteries SOC is not obvious and it
performance modification.
is still the subject of many studies [27] [28] [29]. The specific
SOC calculation from [30] and [31] has been chosen for this
Voltage (“VBAT”)
study. It takes into account the variation of the quantity of
As a simple representation of the global behavior, the
charge in the process as a function of the current rates and the
voltage of batteries has been considered linear in charge and in
ambient temperature. As this model has already been used and
discharge as a function of the SOC (calculated as time
described in [32], presentation of calculation in this paper are
integration of the current). From linear interpolation of
restricted to equation (8) and (9), where “C(t)”is the batteries
experimental results carried out at INES institute, the voltage
capacity at each instant, “Cre f(t)” is the capacity of reference,
of batteries in charge and discharge is determined by equations
Q(t0) is the initial quantity of charge, Qc(t) and Qd(t) are
(14) & (15), where ‘NBAT_S’ is the number of batteries of 12V
respectively the quantity of charges exchange during the
connected in series. Experimental results and linear
charge and discharge process. The capacity of reference
interpolation are presented on Fig. 4.
“Cref (t)” varies according to the ageing process express by
(11) and (12). For more details, the reader is encouraged to
VBAT ( t ) = [12.94 + 1.46 × SOC( t )]× N BAT _ S (14)
read references [30] [31] [32].
VBAT ( t ) = [12.13 − 1.54 × (1 − SOC( t ))]× N BAT _ S (15)
C( t )
SOC = (8)
C ref ( t )
14.5
C ( t ) = Q( t 0 ) + Qc ( t ) + Qd ( t ) (9)
14

13.5
Ageing process VBAT (IBAT >0) = 12.94 + 1.46 x SOC
State Of Health (SOH) of batteries is defined by (10) where 13
Voltage (V)

Mesures charge
Cref,nom is the nominal capacity of reference available from the Model charge
12.5
manufacturer datasheet [31]. Mesures discharge
Model discharge
C ref ( t ) 12
SOH ( t ) = (10)
VBAT (IBAT < 0) = 12.13 - 1.54 x SOC

C ref ,nom 11.5

11

Degradations of the performances of batteries during the 10.5

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ageing process have been modeled by [31] and [33]. It is SOC in charge & (1-SOC) in discharge
represented as losses of capacity of reference that are Fig. 4: Experimental measures and linear modeling of the charge and
considered linear according to the Depth Of Discharge of discharge voltage as a function of SOC
batteries [34] [35]. From experimental results performed at (12V lead acid battery and IBAT is constant at a rate of 5h)
INES institute and presented in [33], linear ageing coefficients
‘Z” for different batteries technologies have been carried out. This modeling is elementary but appropriate for this
The coefficient ‘Z’ has a value of 3.10-4 for lead acid application as it is simple and significantly representative of
technology. the behavior in the range of operation conditions of the system.
At each step size, if the batteries operation is a discharge, The batteries power “PBAT” in (1) and (3) is obtained by
the new capacity of reference is obtained by (11) from multiplying the batteries voltage modeled here by the input or
output batteries current.

116/124
TSTE-00141-2010 5

C. Converter efficiency
The power electronics converters have been modeled
according to their efficiency as a function of the input
normalized power, where losses are assumed to be a quadratic
function. The conversion efficiency formulation (16) is carried
out from a quadratic interpolation of an experimental curve
generated at INES institute [36].
Constraint (17) comes from the law of physics, constraints
(18) to (20) are physical constraints to limit the batteries
degradations and ageing, and constraint (21) corresponds to
the peak shaving application.
It is assumed that the loads are not controllable but have to
be supplied at any time. Forecast on load consumption are

( )
1 used to find the power management strategy.
ηconv = 1 − × 0.0094 + 0.043 × In + 0.04 × In 2 (16) The objective function of supplying the user loads at the
In
where “In” is the normalized input power of the converter best cost is expressed by (22). It amounts to minimize the final
(Pin / Pconv_rated). value of the cash flow “CF” over the entire studied period. The
cash flow is composed of the cash received “CR” and the cash
Fig. 5 shows the efficiency curve measured and modeled paid “CP”.
from identification. The average error is 0.17 %. Calculation
Min (CF) = Min ∑t = t [CR ( t ) + CP ( t )]
T
of efficiency with (16) is applied on the PV converter, the (22)
0
batteries converter, and the DA/AC and AC/DC conversion of
the bidirectional converter. The cash received is the benefits made from any electricity
92 operation related to an agreement with the grid operator.
Generally, it corresponds to the benefits from selling PV
electricity. However, some new agreements, such as [7],
Convertion efficiency η conv (%)

90
Measures enables to make benefits from the self-consumption of energy
Model from the PV generator. In this case, cash received corresponds
88 to both benefits. The cash paid corresponds to all the operation
2
y = 0.0094 + 0.043.x + 0.04.x costs of the system. In our case, it is composed of the cost of
86
ε = 0.17% electricity purchase from the grid and of the ageing cost of
batteries. As a very important cost, ageing of batteries is taken
into account in the optimization process through the
84 calculation of the SOH. Details of ageing cost calculation are
presented in subsection below. The cash received or paid is a
82 power (kW) multiplied by a price (€/kW) so its value is
0 20 40 60 80 100 120
Pin / Pnom(% ) directly related to the power flow management. The power
Fig. 5: Measures and identified converter efficiency curves [36]
sign convention in Fig. 2 has been defined such as the cash
received is negative and the cash paid is positive.
It is important to notice that the constraint (21) is not
IV. DYNAMIC PROGRAMMING AND POWER FLOW necessary strict such as if it is not satisfied, the solution is not
MANAGEMENT rejected but a penalty cost is applied. In this way, all the
solutions are taken into account which gives more flexibility to
This section details the optimization stage procedure of the
the system.
supervisor presented in Fig. 3. The first subsection summarizes
the problem to be solved with the constraints and the objective B. Dynamic programming
function. The second subsection presents the application of the Dynamic programming (DP) has been used to solve the
dynamic programming to solve the problem. optimization problem. Definitions and fundamental principles
A. Problem formulation may be found in [37]. The problem is formulated as a system
evolution divided in a multi-stage decision process. At any
The system studied is presented in Fig. 2 in section II and
time “t”, the state of the system is determined by a set of
the operating constraints, defined by (1), (2), (3), (4) & (5) are
quantity called states variables. In our case, the state of the
summarized below.
system corresponds to the SOC of the batteries, discretized
with a step size of ‘δSOC’. At each step time “∆t”, the
PGRID ( t ) = PPV ( t ) + PBAT ( t ) + PLOADS ( t ) (17)
batteries can take any of the discretized value δSOC in a range
SOC min
≤ SOC( t ) ≤ SOC max
(18) that guarantees the constraint (18). Fig. 6 illustrates an
PBAT ≤ PBAT ( t ) ≤ PBAT
min max
(19) example of a SOC space for three time steps. All possible
trajectories from the initial state at the initial time to all the
SOH ( t ) ≥ SOH min (20)
allowed states at the final time are represented.
PGRID ( t ) ≤ PGRID
max
(21)

117/124
TSTE-00141-2010 6

t 0 1.∆t 2.∆t 3.∆t make the BrC positive.

SOC SOCmin SOCmin SOCmin
∆SOC01  − ∆SOH ( xi , xj , t ) 
+ δSOC
BrC( xi , xj , t ) = BiC ×   (27)
? ∆SOC02  1 − SOH min 
∆t)
SOC1(1.∆ ∆t)
SOC1(2.∆ ∆t)
SOC1(3.∆
? t such as SOH = SOH min

∑BrC( t ) = BiC
SOC0
+ δSOC
? ∆SOC03
N states (28)
? t such as SOH =1
∆t)
SOC2(1.∆ ∆t)
SOC2(2.∆ ∆t)
SOC2(3.∆

∆SOC0N
Formulated as above, the problem is to find the optimal
sequence of SOC transition from the initial time to the final
SOCmax SOCmax SOCmax
time that achieves the final state from the initial state with the
Fig. 6: Batteries SOC space and the possible trajectories for the case of 3 steps lower CF value. To solve this shorter path problem, the
time Bellman’s algorithm, described in [37] and [38], has been
used. Fig. 7 shows the flowchart of the algorithm with the
details computation of the weight of the arcs. Example and
A transition between two states during one time step results are presented in the next section.
corresponds to a SOC variation written ‘∆SOC’. As each Legend
Initial State
State « xj » = 0, CF(xj)* = 0

∆SOC corresponds to a batteries power value, the constraint

N = maximum number of state
• pxj* is the optimal predecessor state of the state xj
(the path through this state reach the state xj with the lower CF value)

(19) is translated in terms of SOC variation as presented by •xi* is the optimal state to pass through to build the shortest path Initialization of the CF of all the path to infinite
∀ j int . ∈ {1, j max }, CF ( xj )* = ∞
(23). This reduces the computation time to check the
satisfaction of the constraints. Computation of the weight of each arc j = j+1
i=0

u(xi,xj)

∆SOC min ≤ ∆SOC( xi , xj , t ) ≤ ∆SOC max ∆SOC(xi,xj,t) = SOCxj(t+ ∆t) - SOCxi(t) CF ( xi )* ≠ ∞

(23) and
N
∃ u( xi , xj ) such as
I ( u ) = xi and T ( u ) = xj

For each ∆SOC, first is calculated ‘PBAT’ and ‘PGRID’ N

Y
∆SOCmin< ∆SOC(xi,xj,t) < ∆SOCmax

according to the a priori knowledge of the loads consumption u(xi,xj)

Computation of w(u)
‘PLOADS’ and the PV availability ‘PPV’, and secondly is w(u(xi,xj))
Y

obtained the corresponding CF value. As an example, if the

Computation of SOH & ∆SOH Computation of PBAT N
cash received comes only from feed-in PV power, the Eq. (23) & (39) Batteries model
CF(xj)* > w(u(xi,xj)) + CF(xi)*

corresponding CF of one SOC transition is expressed by (24), PBAT(xi,xj,t)

Y

where ‘FiT’’ is the Feed in Tarif, ‘EgP’ is the Electricity grid SOH(xi,xj,t)
∆SOH(xi,xj,t)
PPV(t)
PLOADS(t)
Calculation of PGRID
Power balance (29) CF(xj)* > w(u(xi,xj)) + CF(xi)*
pxj* = xi

Price and ‘BrC’ is the Batteries replacement Cost defined

below. Computation of BrC
Eq. (40) to (41)
PGRIDmin< PGRID(xi,xj,t) < PGRIDmax
N
i = i +1

Y N

CF ( ∆t ) = CR( ∆t ) + CP( ∆t )
i≠j

Y
BrC(xi,xj,t) PGRID(xi,xj,t)

 
N
i≥N

=  PGRID ( ∆t ) × FIT ( ∆t ) × ∆t  CF calculation & Weight of the arc Y

 ∆ ≤ 
Eq. (37)
PGRID ( t ) 0 (24) w(u(xi,xj)) j≥N
N

  Y

+  PGRID ( ∆t ) × EgP( ∆t ) × ∆t + BrC( ∆t ) Final state = N

xn*= p(N)*

 PGRID ( ∆t ) > 0 
xn-1* = p(xn*)*

Shortest path
X* = {N,xn*,xn-1*,…,0}

The replacement cost is the translation of the degradation of

the batteries. It is calculated by (25) according to the variation
of SOH during each time step. Replacing SOH by its
formulation (13); the SOH variation is expressed by (26) and
is a linear function according to the SOC variation and the
ageing coefficient. The SOH decreases only during discharge
process so expressions (25) and (26) are negative as it is
calculated only in discharge.
∆SOH ( xi , xj , t ) = SOH xi ( t − ∆t ) − SOH xj ( t )
∆SOH ( xi , xj , t ) = Z × [SOC xi ( t − ∆t ) − SOC xj ( t )]
The replacement cost BrC is calculated by (27) such as it
satisfies (28). The sign of the SOH variation is inverted to
End
Fig. 7: Flowchart of DP algorithm developed for power management
V. SIMULATIONS AND RESULTS
A. Context and parameters values
The optimization has been performed in the economical
context of a French standard electricity price. For instance, to
illustrate the upcoming economical context, no subvention for
(25) the PV power is considered: The energy fed into the grid is
(26) purchased at the same price as the one electricity is sold to the
final consumer. It is assumed it is allowed to feed the grid with
power from batteries, but it is not paid. However, it is allowed
to charge the batteries from the grid. Fig. 8 is a scheme of the

118/124
TSTE-00141-2010 7

system where the energy meters. forecasts on load consumption are measurements achieved in
Chambéry (France) for a family of three people with an
PV production Feed in Consuption off peak/on peak electricity agreement and an electrical
meter meter meter
heating system. According to this agreement, the loads profile
C1 C2 C3 carries out high fluctuation and power peaks. This profile has
Grid
been chosen because it is a standard situation in France. As the
PV generator load profile is already optimized according to off peak/on peak
electricity prices, the simulation and optimization presented
below have been performed in a context of a single fixed
Batteries electricity price. Size of the system is set to the value provided
User Loads
by Table II and has been determined with the sizing method
Fig. 8: Scheme of the system with the energy meters proposed in [39].

Table II: Size of the elements of the system studied

In this context, the cash flow paid and received are
PLOADSmax 4 (kW)
expressed by (29), (30) and (31) where ‘FiT’ is the feed in 1.6 (kW), PV peak power of the generator
PPVp
tariff, ‘EgP’ is the electricity grid price and ‘Eci’ is the energy NPV_S 5 Number of PV modules in series
measured with the meter “Ci”. Calculation of the cash received NPV_P 2 Number of PV modules in parallel
by (30) and (31) guaranties to purchase only the power fed Cref,nom 100 (Ah) Capacity of the total storage
Vbat,nom 12 (V) Voltage of 1 battery
into the grid from the PV generator. According to the power’s
Ns 10 Number of batteries in series
signs convention defined in Fig. 2, the cash received expressed Np 1 Number of batteries in parallel
by (29) is negative.
B. Results and graph
CP( t ) = EC 3 ( t ) × EgP ( t ) + BrC ( t ) (29) The simulation has been carried out on 24 hours for the 22th
Since EC2 ≤ EC1 of February as an exemplary day to figure out optimization
CR( t ) = EC 2 ( t ) × FiT ( t ) (30) results. The dynamic programming algorithm whose flowchart
Else is shown in Fig. 6 has been developed with the Matlab
(31) software. Regarding the results presented here, the
CR( t ) = EC 1( t ) × FiT ( t )
computation time is less than one minute.
The constraints are expressed by (17) to (21) and the The power scheduling obtained with the dynamic
programming optimization algorithm is presented in Fig. 9. In
objective function is (22), according to the values of Table I.
this figure, positive batteries power corresponds to a charge
Table I: Simulation parameters values and negative power to a discharge. Positive grid power
T 24 (h) Period studied corresponds to a consumption and negative grid power to feed
∆t 10 (min) in power. Fig. 11 shows the SOC variation corresponding to
δSOC 0.01 the power schedule of Fig. 9.
SOCmin / SOCmax 0.2 / 0.9
SOC(t0) 0.5 As expected, the power exchanged with the grid is
∆SOCmin / ∆SOCmax -0.7 / 0.7 (during 1h) maintained to the maximum limit. The batteries operation is
PGRIDmax 3 (kW) managed to satisfy the constraints regarding the lower cost for
BiC 150 (€/kWh) the owner of the system.
EgP 0.11 (€/kWh)
FiT 0.11 (€/kWh)
As a SOC of 50% is not enough to guarantee peak shaving
GpF 10 later in the day, the batteries are charged in the morning.
Charge is completed from the grid because there is no PV
The constraint (19) is expressed in terms of ‘∆SOC’ as (23). production. As shown in Fig. 11, the batteries are fully
The constraint (21) is not strict, such as if it is not satisfied, the discharged after the first peak loads. The charge process is not
grid penalty factor ‘GpF’ is applied to the over power directly launch as there is a period to wait more PV
according to (32). production. The batteries are charged as much as possible with
the PV power as it is the optimal operation according to the
If PGRID(t) > PGRIDmax: economical context. In this case, the use of the PV power is

[
CP( t ) = EgP × PGRID
max
(
+ PGRID ( t ) − PGRID
max
)× GpF ]× ∆t (32)
considered to be optimal. The second peak load is in priority
shaved with PV power and batteries are used only when the
solar production is not sufficient.
The input values are forecasts coming from meteorological With this management, it is observed in Fig. 9 that the feed
data (‘GT’ & ‘Tamb’), forecasts of the loads consumption profile in power is minimized, which reduces the power flow
(‘PLOADS’) and the electricity prices (‘EgP’ & ‘FiT’). fluctuation on the grid.As shown in Fig. 11 the SOC of
Experimental measurements performed at the INES institute in batteries returns to its initial value at the end of the day. The
2007 are used as meteorological forecasts. Similarly, the

119/124
TSTE-00141-2010 8

initial SOC of 50% (and final also) has been chosen such as it the next day and so in a day-ahead supervision.
gives the maximum of flexibility to the power management of

4 4

Pm ax Pm ax
GRID
GRID
3 3

2 2
Power (kW)

Power (kW)
1 1

0 0
PV PV
Loads Loads
Grid
-1 -1 Grid
Bat Bat

0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time (h) Time (h)
Fig. 9: Power schedule with DP optimization Fig. 10: Power schedule with simple ruled-based management

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
SOC

SOC

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time (h) Time (h)
Fig. 11: SOC schedule of batteries with DP optimization Fig. 12: SOC schedule of batteries with ruled-based management

The results obtained with DP are compared with the ones
obtained with a simple ruled-based algorithm presented in Fig.
10 and Fig. 12. The rules are defined such as the batteries are
maintained to there maximum SOC to guarantee peak shaving.
As there is no forecasting, charge of batteries is performed
with the first available power source (PV or grid) according to
the constraints: the PV power is in priority used to supply the
loads.
The ruled-based algorithm works with continuous values of
the components as the DP needs discrete values, which
explains the differences on power profiles, on SOC variation
and the ‘steps’ of batteries power that appear in Fig. 9.
The ruled based algorithm is interesting because the power
exchange with the grid is maintained to the maximum bound
and the constraints are satisfied. The two figures carry out that
the batteries are charged as soon as possible with the first
available source. Consequently the solar energy produced
during the most powerful period is not locally used but is
injected into the grid and creates power fluctuations. In this
case, it is considered that the use of the solar power is not
optimized. Because there is no forecasting, the charge at the
beginning of the day is the consequence of the main rule and
not a prediction of future needs. As the rule is to keep the
batteries as full as possible, the final SOC is maximum and
different that the initial one. In this condition, the flexibility of
the management of the next day is restricted as the batteries
are full and the charge operation won’t be possible. By this
way, it is considered that the ruled-based management is not
the most suitable for the day-ahead supervision.
The final value of the objective function (cash flow) for the
DP optimization and the ruled based management are
presented in Table III. The DP algorithm gives a lower value
as it is an optimization. Gains with DP optimization are around
13% higher than with the ruled-based management but it is
very dependent of the economical context.

120/124
TSTE-00141-2010
Table III: Final value of the objective functions for each power management
DP Ruled-based
Final CF (€) 9.50 10.95
VI. EXPERIMENTATION OF THE MANAGEMENT IN REAL
CONDITIONS
The optimal management proposed in this study is based on
the forecasts on the meteorological conditions (“Tamb” and “E”
for the PV production) and on the loads profile consumption.
According to this important assumption, this section studies
the power flows in the system when the predictive power
scheduling, from the predictive optimization, is applied in real
conditions. In this objective, the dynamic programming
algorithm has been implemented into a microcontroller and
tested in real time operation. This has been realized on the
experimental bench “RTLab” that is presented in the
subsection below and in [40] and [41]. To evaluate the
interest of optimal predictive scheduling, the results are
compared with the ruled based strategy, also implemented in
the microcontroller and applied in the same conditions. At the
moment, studies in real conditions are limited to the case of a
different PV power production from the forecasts. The loads
power profile is considered a-priori perfectly known. The real
conditions have been simulated as representative perturbations
on the forecast according to the real behavior of the
meteorological parameters.
A. Presentation of RTLab and implementation of the
management
The model of the system presented in the previous sections
has been implemented in RT-Lab HILBox 4U digital system.
This system provides tools for running simulations of highly
complex models on a multi-processor architecture
communicating via ultra low-latency technologies to achieve
high speed computations [40]. RT-Lab handles
synchronization and real-world interfacing using fast I/O
boards and data exchanges for HIL/PHIL applications. Its
architecture consists in an eight processor (two Intel Xeon
quad-core 2.33 GHz) machine exchanging data on a shared
memory. Their PCI bus is connected to the digital and analog
I/O system via an FPGA controller. There are 2x16, 16-bit
analog inputs and outputs, with high update rate of 1µs. The
digital interface consists of 2x16 I/O allowing a data rate of
1 MHz. The model of the system has been implemented in
Matlab/Simulink software where specific blocks for RTLab
operation have been added. These blocks are necessary to
perform fixed time step simulation involving dynamics and
discrete events asynchronous with respect to the simulation
clock. The optimization algorithm is integrated in Simulink as
an S-function such it can be implemented in the
microcontroller. The complete model is then compiled, loaded
into the above-described hardware and executed in real time
[41].
9
B. Application in real time and real conditions
When the model is executed in the microcontroller, it first
computes the predictive optimization stage (the DP algorithm)
to obtain the anticipative strategy to apply over the next 24
hours (schedule of the power to exchange with the batteries).
Then, the anticipative strategy is applied in the simulated real
conditions. The power exchanged with the grid on these new
conditions is obtained from the power balance (1). The
anticipated scheduling on “PBAT” is applied whatever the real
conditions such as physical constraints (3) and (4) are
guarantee even if the peak shaving constraint (5) on “PGRID” is
not. A scheme of the application of the predictive strategy in
simulated real conditions is presented on Fig. 13. As shown on
this figure, the “artificial” perturbations on PV are considered
of 2 natures:
- Global error on the daily production according to the
forecasts. This error is due from the imperfect and
approximate forecasts on the meteorological data used to
predict the PV production over the day. This is a “low
frequency” perturbation. In real cases, this error is generally
in the range from 10 % to 20 %. Generally the error on the
daily PV production is low but the pattern of the production
profile is different than expected. This perturbation is
created by multiplying the predictive PV production by a
factor that varies from 0.6 to 1.4 as a sinus function during
the daily PV production period. In this way the pattern of
the PV production profile is modify but the error on the
daily production is in the range of the considered
perturbation.
- Local error on the power with rapid variations. This
perturbation is due from the clouds or such other
inescapable and fast perturbations on the PV power. This is
a “high frequency” perturbation. This error is created by
multiplying the predictive PV power by a random factor
that varies as a Gaussian distribution around 1 with a
variance of 0.01. The sample time for this perturbation is 10
seconds.
Simulated real conditions
Sinus signal
Low variations
Rapid variations
X
PPV (t)
+
Random number
PGRID (t)
PLOAD (t) + PGRID real that guarantee the
PPV * (T) control on PBAT in real conditions
PLOAD * (T) PBAT * (t)
Min [Σ CF (T)]
Anticipated strategy
+
Tariffs * (T)
Predictive optimization
Forecasts with dynamic programming
Fig. 13: Scheme of the application of the predictive strategy in real conditions
The real time application has been performed for the 22th of
February as the same day as the simulations results above. For
a question of time, as the bench test operates in real time and
121/124
TSTE-00141-2010 10

the perturbations are related only to the PV power, the

3.5
experiment has been performed only for the duration of 12 P max
GRID
hours from hours 6 to 18. Fig. 14 compares the forecast of the 3

PV power used to calculate the predictive strategy and the real 2.5
PV power after applying the artificial perturbations. On this 2
figure, the rapid and local variations as the low variations on 1.5

Power (kW)
the global production from the sinus signal are clearly visible.
1
1.8 0.5
PV forcast
1.6 PV real 0

1.4 -0.5 Predictive

-1 Ruled Based
1.2
-1.5
Power (kW)

1
6 8 10 12 14 16 18
0.8 Time (h)
Fig. 15: Profiles of power exchanged with the grid in real conditions for
0.6 application of the predictive strategy and ruled based strategy for the 22th of
February
0.4

0.2
0.9
0
6 8 10 12 14 16 18
0.8 Predictive
Time (h)
Ruled based
Fig. 14: Comparison between the forecasted and the real PV power for the 0.7
22th of February
0.6
C. Experimental results and discussions
SOC

An important remark is that all the algorithm and strategies 0.5

presented in this paper, including the optimal predictive
algorithm, can be implemented into a microcontroller and 0.4

executed in real time and in real conditions. The modifications 0.3

are to translate the algorithms, initially in Matlab format, into a
lower level language understandable by the microcontroller.
This is a significant result as an issue to a rapid industrial
development.
The power exchanges with the grid from application of the
predictive strategy and ruled based strategy in real conditions
and real time are presented on Fig. 15. The SOC of batteries
are compared on Fig. 16. According to the method of
application, the SOC profile for the predictive strategy in real
condition is the same as the one anticipated and presented on
Fig. 11. The load profile is not presented as the same as the
one on Fig. 9.
As expected when the predictive strategy is applied in real
conditions, the anticipated schedule is followed since there is
no perturbation (Fig. 15). As the predictive power of batteries
is imposed, in real conditions the perturbations are balanced
on the power exchanged with the grid. As observed on Fig. 15,
all the rapid variations appear on the power exchanged with
the grid and when the PV power is higher than predicted more
power is injected into the grid. In the same way, when the PV
power is lower than anticipated, the grid supplies more power.
In real conditions, this mode of operation does not guaranty
the peak shaving constraint (5) on “PGRID”, as observed on Fig.
15 around 14:30.
0.2
0 5 10 15 20 25
Time (h)
Fig. 16: Profiles of SOC of batteries in real conditions for application of the
predictive strategy and ruled based strategy for the 22th of February
For the ruled based strategy, the constraint on the power
exchanged with the grid in real conditions is always
guaranteed. However and as expected, the equality constraint
on the initial and final SOC of batteries is not respected. As
shown on Fig. 16, batteries are fully charged at the end of the
day which restricts the flexibility of the management of the
next day. Also and as remarked before, use of the PV power is
not optimized as it is mainly used to feed the grid instead of
charging the batteries. In this way, the objective function is
also not optimized. The ruled based algorithm is the easiest
one to implement and to develop at an industrial scale but it
does not reach the objective and it limits the flexibility of a
day to day management. In this consideration, the ruled based
algorithm does not present interests for our study and has been
used only as a reference for comparisons with the strategy
proposed.
Theses results and comparisons lead to the reflection about
the type of control to apply in real conditions to guaranty all
the constraints and in the ideal case why not also to guaranty

122/124
TSTE-00141-2010
optimality according to the real conditions and the objective
function. The question of performing a new predictive
optimization from updates forecast can also be asked.
Finally, it is carried out that the management in real
conditions is strongly dependent on the accuracy of the
forecasts and on the reactive mode of operation that is applied.
However, predictive optimization is a necessary stage to reach
the final objective of optimal management in real conditions.
In this consideration, a significant step has been achieved
implementing the predictive algorithm into a microcontroller
to real time execution.
VII. CONCLUSION AND FUTURE WORK
In this paper, a predictive control system based on a
dynamic programming approach, that optimizes the power
flow management into a grid connected PV system with
storage, has been presented. According to the power
supervisor proposed in section II.C, the study focuses on
predictive optimization from “a-priori” known forecasts. The
objective was to perform peak shaving with the lower cost for
the owner of the system. The particularity of the approach lies
in the consideration of batteries ageing into the day-ahead
power management.
Simulations and real time operation in simulated real
conditions have been performed and results compared with a
ruled-based management. In simulation over 24 hours,
predictive optimization provides around 13% of gain on the
electricity bill for the economical context of the study.
Implementation of the predictive optimization algorithm into a
microcontroller to perform real time management figured out
the industrial potentiality of the method proposed. Operation
of the management in real conditions has validated the results
and assumptions from simulations. Performances of the
management in real conditions strongly depend of the accuracy
of the forecasts and of the mode of operation. This important
conclusion leads to many questions about reactive power
management without denying the importance and the necessity
of the predictive optimization stage.
The management developed helps integration of PV power
into the grid as peak loads are shaved. Depending of the
reactive management in real conditions, the power fluctuation
of the PV production is balanced to the power exchanged with
the grid or with the batteries. In this context, next and future
works will deal with reactive management for real conditions
operations. From the conclusion of this study, different
reactive management will be developed and compared in real
conditions operations. Specifically, an optimal reactive
management, developed from the DP algorithm of the
predictive optimization stage, will be proposed. In this way the
sum of the works will achieved to a complete power
management system that is reactive such as optimal
management is guaranteed according to the real conditions.
11
REFERENCES
[1] IEA Photovoltaic power systems program, IEA PVPS T1-17, “Trends in
photovoltaic applications. Survey report of selected IEA countries
between 1992 and 2007”, 2008
[2] Achim Woyte, Vu Van Thong, Ronnie Belmans, and Johan Nijs,
'Voltage Fluctuations on Distribution Level Introduced by Photovoltaic
Systems', IEEE Transactions on energy conversion, Vol. 21, no 1, 2006,
pp.202-209
[3] Paul Denholma, Robert M. Margolis, “Evaluating the limits of solar
photovoltaics next term (PV) in traditional electric power systems”,
Energy Policy, 2007, Vol.35, pp.2852–2861
[4] Paul Denholma, Robert M. Margolisb, “Evaluating the limits of solar
photovoltaics (PV) in electric power systems utilizing energy storage
and other enabling technologies”, Energy Policy 35 pp.4424–4433,
2007
[5] M. Perrin, Y.M. Saint-Drenan, F. Mattera, P. Malbranche, 'Lead–acid
batteries in stationary applications: competitors and new markets for
large penetration of renewable energies', Journal of Power Sources,
Vol.144, 2005, pp.402–410
[6] J.M.Eeyer, J.J.Jannucci, G.P.Corey, “Energy Storage Benefits and
Market Analysis Handbook”, A study for the DOE Energy Storage
Systems Program, Sandia report, Sandia National Laboratories, 2004
[7] German Bundestag, “Update of feed in tariff for renewable energies in
Germany “, Feed in international corporation, 2008
[8] Adel Mellita,1, Soteris A. Kalogirou, “Artificial intelligence techniques
for photovoltaic applications: A review”, Progress in energy and
combustion science,Vol.38, 2008, pp.574-632
[9] F.Fusco and J.V. Ringwood, 'Short-Term Wave Forecasting for Real-
Time Control of Wave Energy Converters', IEEE Transactions on
Sustainable Energy, Vol.1, n°2, 2010
[10] M.Ceraolo, A.Di Donato, and G.Franceschi, “A General Approach to
Energy Optimization of Hybrid Electric Vehicles”, IEEE Transactions
on vehicular technology, Vol.57, n°3, pp.1433-1441, 2008
[11] C.Wang, and M.Hashem Nehrir, “Power Management of a Stand-Alone
Wind/Photovoltaic/Fuel Cell Energy System”, IEEE Transaction on
Energy Conversion, Vol. 23, n°3, 2008
[12] K.Agbossou, M.Kolhe, J.Hamelin, and T.K.Bose, “Performance of a
Stand-Alone Renewable Energy System Based on Energy Storage as
Hydrogen”, IEEE Transactions on energy conversion, Vol.19, n°3,
pp.633-640, 2004
[13] S.Jain and V.Agarwal, “An integrated Hybrid Power supply for
distributed generation application by nonconventional energy sources”,
IEEE Transactions on energy conversion, Vol.23, n°2, 2008
[14] Michael Urbina, Zuyi Li, 'A Fuzzy Optimization Approach to
PV/Battery Scheduling with Uncertainty in PV Generation', IEEE Power
Symposium,. NAPS 2006. 38th North American , 2006
[15] AWang, W.Yang, “Design of Energy Management Strategy in Hybrid
Vehicles by Evolutionary Fuzzy System Part I: Fuzzy Logic Controller
Development”, Proceedings of the 6th World Congress on Intelligent
Control and Automation, June 21 - 23, 2006, Dalian, China, 2006
[16] A.Borghetti, C.D’Ambrosio, A.Lodi, and S.Martello, 'An MILP
Approach for Short-Term Hydro Scheduling and Unit Commitment
With Head-Dependent Reservoir', IEEE Transaction on power systems,
Vol.23,n°3, 2008
[17] T.T. Ha Pham, F. Wurtz, S. Bacha, “Optimal operation of a PV based
multi-source system and energy management for household
application”, IEEE International Conference on Industrial Technology,
ICIT, 2009
[18] Bo Lu, M. Shahidehpour, 'Short term scheduling of battery in a grid
connected PV/battery system', IEEE Transaction on power systems, Vol.
20, No.2, pp.1053-1061, 2005
[19] A. G. Bakirtzis, P. S. Dokopoulos, 'Short term generation scheduling in
a small autonomous system with unconventional energy sources', IEEE
Transactions on Power Systems, Vol. 3, No. 3, pp.1230-1236, 1988
[20] Q.Gong, Y.Li, and Z.-R.Peng, “'Trip-Based Optimal Power
Management of Plug-in Hybrid Electric Vehicles”, IEEE Transactions
on vehicular technology, Vol.57, n°6, pp.3393-3401, 2008
[21] M.Koot, J.T.B.A. Kessels, Bram de Jager, W.P.M.H.Heemels, P.P.J.Van
Den Bosch and M.Steinbuch, “Energy Management Strategies for
Vehicular Electric Power Systems”, IEEE Transactions on vehicular
technology, Vol.54, n°3, 2005
123/124
TSTE-00141-2010
[22] F.Besnard and L.Bertling, 'An Approach for Condition-Based
Maintenance Optimization Applied to Wind Turbine Blades', IEEE
Transactions on Sustainable Energy, Vol.1, n°2, 2010
[23] J.T.B.A.Kessels, M.Koot, Bram de Jager, P.P.J.Van Den Bosch, N.
(Edo) P.I.Aneke, and D.B.Kok, “Energy Management for the Electric
Powernet in Vehicles With a Conventional Drivetrain”, IEEE
Transactions on control systems technology, Vol.15, n°3,pp.494-505,
2007
[24] J.T.B.A.Kessels, M.W.T.Koot, P.P.J.Van Den Bosch, and D.B.Kok,
‘Online Energy Management for Hybrid Electric Vehicles’, IEEE
Transactions on vehicular technology, Vol. 57, n°6, pp.3428-3440,
2008
[25] E. Skoplaki, J.A. Palyvos, 'On the temperature dependence of
photovoltaic module electrical performance: A review of
efficiency/power correlations', Solar Energy, In press
[26] A. Luque and S. Hegedus, “Handbook of Photovoltaic Science and
Engineering”, John Wiley & Sons, 2003, p.943 & 951
[27] M.Ceraolo, “New Dynamical Models of Lead–Acid Batteries”, IEEE
Transactions on Power Systems, vol. 15, n°.4, pp.1184-1190, 2000
[28] C. hA. Shepherd, “Design of Primary and Secondary Ceils - II. An
Equation Describing Battery Discharge”, Journal of the electrochemistry
society, Vol.112, N°7, pp.657-664, 1965
[29] J.B.Copetti, E.Lorenzo & F.Chenlo, “A general Battery Model for PV
systm Simulation”, Progress in Photovoltaics: Research and
Applications, Vol.1, pp.283-292, 1993
[30] A. Delaille, F. Huet, E. Lemaire, F. Mattera, M. Perrin, M. Vervaart,
'Development of a battery fuel gauge based on ampere-hour counting',
21st European Photovoltaic Solar Energy Conference, 4-8 September
2006, Dresden, Germany, 2006
[31] A.Delaille, “Development of new state-of-charge and state-of-health
criteria for batteries used in photovoltaic systems“, PhD report (french),
University Pierre et Marie Curie, 2006
[32] Y.Riffonneau, A.Delaille, F.Barruel, S.Bacha, “ System modeling and
energy management for grid connected PV systems with storage”, 24th
EU Photovoltaic Solar Energy Conference, 2008
[33] E.Lemaire-Potteau, F. Mattera, A. Delaille, P. Malbranche, “Assessment
of storage ageing in different types of PV systems technical and
economical aspects”, 24th EU Photovoltaic Solar Energy Conference,
2008
[34] Y. Guoa, S. Tang, G. Meng, S. Yang, “Failure modes of valve regulated
lead acid batteires for electric bicle application in deep discharge”,
Article in press, Journal of Power Sources, 2008
[35] S. Hua, Q. Zhou, D. Kong, J. Ma, “Application of valve-regulated lead-
acid batteries for storage of solar electricity in stand-alone photovoltaic
systems in the northwest areas of China”, Journal of Power Sources, Vol
158, pp.1178–1185, 2006
[36] M.Vervaart, “Test of a photovoltaic inverter of a rated power of
1100 W, INES institute, 2007.
[37] R.Bellman, “The theory of dynamic programming”, RAND
Corporation, Proceedings of the National Academy of Sciences, pp.503-
715, 1952
[38] A.Kauffman, “Méthodes et modèle de la recherche opérationnelle”, tom
1 & 2, L’économie d’entreprise Vol.10, Dunod, 1968
[39] C.Venu, Y.Riffonneau, S.Bacha, Y.Baghzouz, 'Battery Storage System
Sizing in Distribution Feeders with Distributed Photovoltaic Systems',
IEEE Powertech conference, Roumania, Bucarest, 2009
[40] RT-Lab website available on “ http://www.opal-rt.com/product/rt-lab-
professional ”, 2009
[41] O.Crăciun, A.Florescu, S.Bacha, I.Munteanu and A.Iuliana Bratcu
“ Hardware-in-the-loop testing of PV control systems using RT-Lab
simulator”, 14th International EPE Power Electronics and Motion
Control Conference (EPE-PEMC), Ohrid, Republic of Macedonia,
september 2010
12
VIII. BIOGRAPHIES
Yann Riffonneau was born in Tours, France, in
1983. He received his Master Degree in renewable
energy in 2006 at Savoir University and a PhD in
electrical engineering at Joseph Fourirer University
of Grenoble in 2009. His is currently doing a post doc
at G2Elab on energy management in multi power
sources building at G2Elab. His fields of interest
include energy management, optimization, quality of
power and renewable energy.
Seddik Bacha (M’08) received the Engineer and
Magister degrees from École Nationale Polytechnique
de Algiers, Algeria, in 1982 and 1990 respectively,
and the Ph.D. degree from Institut National
Polytechnique de Grenoble, France in 1993. He joined
the Laboratoire d’Electrotechnique de Grenoble
(LEG) in 1990 and in 1998 he was habilitated to
conduct research. He is currently Manager of the
Power Systems Group with Grenoble Electrical Engineering Laboratory
(G2ELab) and a Professor with the Joseph Fourier University of
Grenoble. His research interests include power electronics system
modelling and control, power quality and renewable energy integration.
Stephane Ploix is Maître de Conférences at the Grenoble
Institute of Technology in the G-SCOP lab. After an
engineer degree in Mechanics and Electricity, he
obtained in 1998, a PhD from the Institut National
Polytechnique de Lorraine in Control engineering and
Signal processing. He is a specialist in supervision,
monitoring and diagnosis and his studies focus on
human-machine cooperative mechanisms. He is involved
in different industrial projects dealing with supervision of distributed plants,
diagnosis of human skills, iterative diagnosis tool for companies and power
management in buildings.
Franck Barruel Franck Barruel received the Engineer
and Magister degrees from Ecole Nationale Supérieure
d'Ingéieurs Ilectriciens de Grenoble-France, in 2002, and
the Ph.D. degree from Joseph Fourier University, in the
Laboratoire d'Electrotechnique de Grenoble (LEG)- in
2005. He joined the Commissariat à l'Energie Atomique
et aux Energies Alternatives (CEA) in the French
National Solar Institut (INES) in 2006. He's specialist in solar system
modelling, energy flow management and involved in the Solar Mobility
concept.
124/124