Applied Energy 87 (2010) 23402351

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Daily energy planning of a household photovoltaic panel

Mohsen Ben Ammar a,b, Maher Chaabene a,*, Ahmed Elhajjaji b
a
Unit de commande de machines et rseaux de puissance CMERP-ENIS-TUNISIA, Tunisia
b
Laboratoire Modlisation, Information et Systmes MIS-UPJV, Amiens, France

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

Article history: This paper puts forward an energy planning approach which offers a daily optimum management of a
Received 25 May 2009 household photovoltaic panel generation (PVG) without using storage equipment. The approach consid-
Received in revised form 8 October 2009 ers the PVG of the last 10 days to estimate the one of the next day, using a Neuro-Fuzzy algorithm. The
Accepted 23 January 2010
estimated PVG is planned according to the consumers needs so as to use the maximum of the generated
Available online 18 February 2010
energy. The algorithm decides by means of fuzzy rules the connection times of appliances, having differ-
ent powers, to the photovoltaic panel (PVP) output during the day. The decision is made on the basis of
Keywords:
optimization criteria with respect to different user operation modes. The approach is validated on a
Photovoltaic panel
Estimation
260 Wp PVP and a set of four appliances of 30 W, 40 W, 60 W and 75 W. The system is installed at the
Energy planning National Engineering School, University of Sfax (ENIS) Tunisia. The daily energetic assessment conrms
Neuro-Fuzzy that the PVG planning makes use of the estimated available energy in between 70% and 80%.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
The integration of renewable energy sources in installations
where the electric grid is available requires a power management
so as to ensure the user needs as when using only the electric grid.
Furthermore, the management should bring maximum energy
prot for the installation. Many strategies are developed so as to
offer an energy management for installations using different
source types: hybrid [1,2], wind [3], photovoltaic [4] and in differ-
ent congurations: autonomous [5,6], isolated [7] and grid con-
nected [79]. These researches use many tools as Fuzzy Logic
[10,11], Genetic Algorithm [11,12], model based [13,14] and evolu-
tionary algorithm [15].
The energetic effectiveness of PVP installation depends on the
adequacy of the generated energy and the load demand. Since
the PVP electric energy is allied to the availability of solar energy,
some researches judge necessary the use of batteries so as to adapt
load needs to PVP generation. However, batteries are costly and re-
quire continuous safeguarding. Hence, different approaches were
adopted to avoid the use of batteries in photovoltaic applications
such as pumping [16] and grid connected PVP [9]. Thus, the PVP
represents the only source of electric energy for the installation
which makes its operation depend only on the suns appearance.
In this eld, research interests are essentially related to the model-
ling [17,18], the optimization [10] and the adaptation system of the
PVP supply according to installation needs. Similarly, other re-
searches are interested in grid connected PVP. Two types of instal-
* Corresponding author.
E-mail address: maherchaabane@yahoo.com (M. Chaabene).
lations are distinguished: either the PVP furnishes totally the
generated energy to the grid which is the case of the large plants,
or the PVP is used as a complementary source for an already grid
connected installation. In the rst design previous researches car-
ried out deal generally with system assessment and characteriza-
tion depending on site climate [10,19]. In the second pattern,
researches developed energy management strategies so as to offer
optimal function to photovoltaic plants where the grid is available
[20]. Most of these researches are interested in a real time power
management [2123] giving an instantaneous decision on the
way to consume the generated energy. The decision consists of
selecting instantaneously the supply source (PVP or grid) for each
appliance function of the available and the needed energies.
Although, this kind of management offers a considerable energetic
efciency for the installation, it involves a high switching fre-
quency of appliances connection between the electric grid and
the PVP output which causes dysfunction essentially when appli-
ances use memories or motors. This problem is due to the instabil-
ity of the generated energy which is sensitive to the climatic
perturbations.
To overcome this inconvenience, this paper proposes an energy
planning of the estimated photovoltaic generation P V b Gd for the
next day. The approach considers the PVG during the last ten days
in order to forecast its behaviour for the following day using a
Neuro Fuzzy estimator ANFIS. Forecasted P V b Gd is optimally
planned to provide energy to appliances of a household installa-
tion. Planning consists of fuzzy rules which determine, according
to different operation modes, the connection times and operation
periods of the household appliances to PVP output considering
their nominal powers, the forecasted P Vb Gd and some optimization

0306-2619/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apenergy.2010.01.016
 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2341

criteria. The planning algorithm is implemented and tested using
an installation located at the National Engineering School, University
of Sfax (ENIS) Tunisia. The approach assessment is shown over
energetic audit.
Next, section two presents the PVG forecasting method. The
planning algorithm is given in section three. Section four shows
the obtained results and introduces a discussion. Finally, a conclu-
sion is presented in section ve.
2. PVP generation forecasting
The systems behaviour is estimated by either symbolic meth-
ods as fuzzy logic and decision trees [20], or adaptive methods
using neural networks [24] and genetic algorithms [12]. Here, the
two methods are combined so as to estimate the PVP generation.
The approach consists of a Takagi and Sugeno model which uses
fuzzy inputs and rules so as to provide a powerful tool for model-
ling complex non-linear problems when combined with a network
structure called ANFIS: Adaptive Network Fuzzy Inference Sys-
tem. ANFIS can be applied to non-linear forecasting where previ-
ous samples are used to forecast the sample ahead.
2.1. Fundamentals of the forecasting method
ANFIS is a class of adaptive multi-layer feed-forward networks.
It generates fuzzy rules from an inputoutput data set. A typical
(a) A1 B1
fuzzy rule in a Takagi and Sugeno fuzzy model has the format
[25,26]:
If x is A and y is B then z f x; y
where A and B are fuzzy sets in the antecedent; z = f(x, y) takes usu-
ally a polynomial form. Consider a model that contains two rules:
Rule 1 : If X is A1 and Y is B1 ; then f 1 p1 x q1 y r 1
Rule 2 : If X is A2 and Y is B2 ; then f 2 p2 x q2 y r 2
Fig. 1a illustrates graphically the fuzzy reasoning mechanism to de-
rive an output f from a given input [x, y]. The ring strengths w1 and
w2 are usually obtained as the product of the membership grades of
the premise part, and the output f is the weighted average of each
rules output. Fig. 1b shows the corresponding ANFIS structure
where nodes within the same layer perform functions of the same
type. Note that Oji denotes the output of the ith node in jth layer
[27,28].
ANFIS has, as the basic learning rule, the back-propagation gra-
dient descent algorithm (the same used in feed-forward Neural
Nets) which calculates the error signals recursively from the out-
put layer backward to the input nodes. From this architecture, it
is seen that given the values of premise parameters, the overall
output function can be expressed as a linear combination of the
consequent parameters [6,15]:
f w1 f1 w2 f2
w1 xp1 w1 yq1 w1 r 1 w2 xp2 w2 yq2 w2 r 2

In time series estimation with ANFIS, two data sets are required so
w1 as to predict one step ahead: training and test data. Sub-cluster-
ing is used to generate ANFIS structure automatically. Training
X Y f1 = p1 x + q1 y + r1
B2 phase involves an iterative procedure, which seeks to calculate opti-
A2
mum values of system parameters by minimizing the sum of
f 2 = p 2 x + q 2 y + r2
squared differences between model predictions and training data
w2
values. Once the training output model (f) is determined, a set of
X Y testing inputoutput data is applied to this model in order to com-
x y pute the ANFIS prediction output. Time series analysis must content
itself with estimating future output values by means of their own
Fig. 1a. Takagi and Sugeno model. past values. The principle consists of using a time series, with a time
step p, of n last measurements up to time t: [m(t  (n  1)p), . . .,
m(t  p), m(t)] as training data and [m(t  (n  2)p), . . ., m(t),
(b) Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 m(t + 1)] as testing data, to estimate during a sliding time interval
x y
the p step ahead future values of the time series. m(t + 1) is xed
A1 w1
x w1 w1 f1 arbitrary at any value so as to achieve the ANFIS testing process.
N The retained output estimated value (p step ahead) is mt e 1 [29].
A2 w1 f1 + w2 f 2
f=
w1 + w2
B1 2.2. Application to PVG forecasting
y N
w2 w2 w2 f 2 = w1 f1 + w2 f 2
B2 During the training phase, the day is considered as time unit
x y (t = d  1: ending day), p = 1 (each day). Since the planning tool is
essentially suitable during hot and moderate seasons where the
Fig. 1b. ANFIS architecture.
climate is stable (as in cold season there is a lack of insolation,

(a) Gd 1 Td 1 M d = [G d Td ]
PVGd
DBd 1 = Gd 2 Td 2 ANFIS PVP model

G
d 10 Td 10
d
Hybrid Learning
Algorithm

DBd
Shift (+1)

Fig. 2a. Neuro-Fuzzy estimator approach.

2342 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351

ten days ago are judged sufcient for the prediction (n = 10). The mated matrix DB b d represents
b d in which the rst line vector M
b Gd for a day d is based on the application of the
estimation of P V the estimated climatic parameters for the day d:
forecasted values of the climatic parameters on the PVP model. h i
Fig. 2a gives the forecasting synoptic schema. bd G
M bd Tb d 1
The estimation of the meteorological parameters Gb d; T
b d char-
acterizing the day d considers the base of measurements DBd1 as where Gb d is the estimated amount of solar radiation and T
b d is the
training data and DBd as testing data: estimated value of the ambient temperature.

2 3 2 3 2.2.1. PVG modelling

Gd1 T d1 Gd Td
6G 7 6G The solar cell is the basic unit of a photovoltaic module. It con-
6 d2 T d2 7 6 d1 T d1 7
7
6 7 U1 DBd 6 verts the suns rays or photons directly into electrical energy. A so-
DBd1 6 .. .. 7 U 2 ; 6 .. .. 7
7
4 . . 5 4 . . 5 lar cell is represented by its equivalent circuit as shown in Fig. 3.
Gd10 T d10 Gd9 T d9 As a function of the input radiation (G) and the ambient temper-
ature (T), the current of a photovoltaic cell (Ipvc) is dened by Eq.
(2) [13,15]:
where U1 and U2 are respectively the measured vector during the
ten previous days to day d of the solar radiation and the ambient Ipv c G; T Ipv c;STC DIG; T 2
     
temperature. The estimation starts with Gd and Td that are supposed G G G
DIG; T a   T  25 Ipv c;STC  1 3
to be the measured values of climatic parameters during the day d; GSTC 40 GSTC
they take any values in order to accomplish the ANFIS testing    
V pv c;STC Rs  Ipv c;STC
process. Ipv c;STC Isc;STC  Isat;STC  exp 1
V T;STC
Considering the Neuro-Fuzzy estimator approach (Section 2.1),
V pv c;STC Rs  Ipv c;STC
the ANFIS architecture is given by Fig. 2b. In order to increase  4
the prediction accuracy, four partitions are xed, which select four
Rp
membership functions, for each input. In view of Fig. 2b, ten rules Isc;STC
Isat;STC   5
V
are delivered by layer 4 to layer 5 in an attempt to determine the exp Voc;STC 1
T;STC
ANFIS output fUi. In Fig. 2b, ANFIS is executed separately for each
n  k  T STC
vector of measure Ui (as a training data) to calculate its relative V T;STC 6
model fUi. Consequently, the estimated value of each climatic q
parameter is computed using its specic testing data vector in Similarly, the output voltage of a photovoltaic cell (Vpvc) is dened
DBd and the ANFIS model fUi. Following the training and the testing by Eq. (7):
process, the combination of the estimated fUi delivers the esti-
V pv c G; T V pv c;STC DVG; T 7
  
G
(b) Layer 1 Layer 2 Layer 3 Layer 4 Layer 5
DVG; T b  T NOCT  T STC  T STC  Rs  DIG; T
GSTC
Ui1 Ui2 Ui10 8
Ai1
wi1
wi1

Ui1
N wi1 f1 where the subscript STC means the value in Standard Test Condi-
Ai4 tions (AM1.5, GSTC = 1000 W/m2, TSTC = 25 C). Isc and Isat are respec-

Ui1 Ui2 Ui10

tively the short-circuit current and the diode reverse saturation
Bi1
Ui2 N fU i current. Vpvc, is the output voltage of the cell and Voc is its value

wi2
wi 2 wi 2 f 2 at open circuit. VT is the thermal voltage, Rs and Rp are respectively

Bi4 the serial and parallel resistances. a and b are the temperature coef-

Ui1 Ui2 Ui10 cients respectively at Isc and Voc, n is the ideality factor (n = 1.62), k
is the Boltzmann constant (k = 1.38  1023 JK1), and q is the elec-
Ji1
wi10 f10 tron charge (q = 1.6  1019 C) and NOCT is the Normal Operating

Ui10 N
wi10
wi10 Cell Temperature (NOCT = 45 C).
Ji4
Since Eq. (4)) is non-linear, the photovoltaic cell model is sim-
plied by neglecting the parallel resistance Rp. By considering data
Fig. 2b. Neuro-Fuzzy architecture applied to meteorological parameter estimation.
provided by the product manufacturers (Table 1) and the different

Solar cell equivalent circuit

Ipvc Boost converter
RS
L

S C
Isc D RP Vpvc Load

MPPT control

Fig. 3. Solar cell schematic.

 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2343

"
Table 1 bd
G
Typical electrical characteristic of TE500CR + PV module. b Gd 0:95  12  4  3:8 0:95  103 
PV
1000
Parameter Variable Value ! !#
bd
G bd
G
Maximum power Pmpp 65 Wp b
 Td  25 3:9  1 12
Voltage at maximum power Vmpp 18 V 40 1000
Current at maximum power Impp 3.6 A
Short circuit current ISC,STC 3.9 A
Open-circuit voltage VOC,STC 22.3 V
Temperature coefcient of open-circuit voltage b 76.32 mV/K 3. Planning the estimated energy
Temperature coefcient of short-circuit current a 0.95 mA/K
Temperature coefcient of power 0.43%/K
Nominal Operating Cell Temperature NOCT 45 C
3.1. Planning strategy
Standard Operating Conditions STC AM 1.5/1000
W/m2/25 C The proposed planning strategy arranges the estimated P V b Gd
during a day by making decision on the optimum connection times
of household appliances to a PVP output. According to one of four
constants values, and by gathering Eqs. (3)(6) in Eq. (2), the cur- distinct operating modes, the decision of the connection time for
rent produced by a photovoltaic cell becomes computed as: each appliance is made by fuzzy rules considering the states of
    the appliance and some planning criteria (Fig. 5).
G G
Ipv c G; T 3:8 0:95  103   T  25
1000 40
 
G 3.1.1. Operating modes
3:9  1 9
1000 Four operating modes are envisaged. The mode selection de-
pends on the appliance functioning manner. The operation period
Considering a PVP composed of Np parallel modules of Ns serial cells
is preset or indifferent; also the starting time is preset or indiffer-
each one, the PVP output power (Ppm) is expressed as (Eq. (10)):
ent. The modes are resumed in Table 2.
Ppv Np  Ipv c  Ns  V pv c Ipv  V pv 10
Mode 0: Considered as the rigorous mode, it is engaged if the
Fig. 4a and b shows the Ppv characteristics function of the output appliance operation period and its starting time are imposed
voltage (Vpv) for different temperatures and irradiances. In order whatever the availability of the PVP provided power.
to make best benet of the available solar energy, the maximum Mode 1: In this mode, the start-up time is preset for an indiffer-
power-points are attempted by a Maximum Power-Point Tracker ent operation period. The appliance functions as long as the
(MPPT) using an appropriate electronic [30]. The MPPT device cali- b Gd is able to supply the appliance.
available P V
brates the PVP current (Impp) and voltage (Vmpp) so as to obtain the Mode 2: Here the start-up time is indifferent but the operation
maximum power (PVG) from the PVP (Fig. 4a and b). Considering period is preset. This operating mode is suitable for most
the MPPT efciency (gMPPT evaluated to 0.95), the PVG is expressed domestic household appliances as washing machines, etc. The
as: starting time is computed as to supply the appliance only by
PVG gMPPT  Impp  V mpp gMPPT  Ppv ;mpp 11 the PVP output.
Mode 3: This mode is considered as the smoothest one. The
where Ppv,mpp denotes the maximum tracked power from the PVP. appliance start-up time and operation period are indifferent.
The model is applied to a PVP (TE500CR+ of Total Energie) com- It permits to get benet of the maximum power from the
posed of four parallel modules (Np = 4). Each module provides PVP. In this mode, there is no need for the electric grid.
65 Wp (Watt peak) as nominal maximum power. Each module is
composed of 36 series-connected (Ns = 36) polycrystalline silicon
cells (Table 1).
Finally, in order to forecast the PVG, the model (Eq. (11)) is com- 3.1.2 Criteria
puted for the estimated solar radiation G b d and ambient tempera- Planning P Vb Gd consists of computing the start and stop connec-
ture Tb d (Eq. (1)). Since the load voltage is xed to 12 V, (Vmpp is tion times of each appliance to PVP output. Nevertheless, multiple
set to 12 V), the estimated P V b Gd is expressed as (Eq. (12)) [19]: planning possibilities can be envisaged for a main situation (mode,

70
(a) 70
Pp v,mp p
(b)
T = 25C 2
G=1000 W/m
60 1 0 0 0 W.m -2 60

50 50
Power (W.m-2)
Power (W.m-2 )

8 0 0 W.m -2

40 40
25 C

Ppv,mpp
35 C

6 0 0 W.m -2
45 C
55 C

30 30
65 C

4 0 0 W.m -2
20 20

10 2 0 0 W.m -2 10

0 0
0 5 10 15 20 23 0 5 10 15 20 23
Voltage (V) Voltage (V)

Fig. 4. (a) PPVVPV characteristics for different irradiations and a xed temperature 25 C. (b) PPVVPV characteristics for different temperatures and a xed irradiation
1000 W/m2.
2344 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351

Maximising connection time

of the PVP connected Reserve a PVGd
Respecting low power
appliances power margin of +10%
appliances priority

CRITERIA
Estimated PVG d
Energy Planning Connections times

OUTPUT
INPUTS
Appliances states of the appliances
Algorithm on PVP
Planning mode

OBJECTIVES

To consume To avoid commutation in To consume a

maximum power supply between power supply
available PVP output and electric stability
PVGd grid.

Fig. 5. The synoptic schema of the proposed approach.

Table 2 3.2.1. The knowledge base of the expert

Operating modes. The approach handles a multi criteria resolution for which three
b Gd and connec-
fuzzy partitions are necessary: appliances states, P V
Starting time Period
tion control (Table 3).
Fixed Indifferent
Where i, l and k are the numbers of the fuzzy subsets, j is the
Fixed Mode 0 Mode 1 appliance reference and lAji ; lBl and lC jk are the membership
Indifferent Mode 2 Mode 3
functions.

PVb Gd and appliances connections states). In this case, planning cri- 3.2.2. The fuzzication
teria (Fig. 5) aid to decide the solution to adopt: The determined fuzzy partitions lead to the calculation of the
membership functions lAji ; lBl and lC jk considering the symmetric
 Give priority to lower power appliance: because it is able to con- triangular type (Fig. 6). These membership functions are expressed
sume weak P V b Gd observed at the beginning and at the end of the as:
day. (
 Maximise connection time of the PVP connected appliances: 1  jxiex 0i j
if jxi  x0i j  ex0i
lAji xi x 0i

since an appliance connected to PVP has the priority to remain 0 otherwise

connected so as to avoid commutation of appliance supply j f1; 2; 3; 4g for appliance states 13
between PVP output and the electric grid. (
 Deduce + 10% of P V b Gd as a reserve power margin in order to 1  jyley 0l j
if jyl  y0l j  ey0l
lBl yl y0l bG
for P V 14
guarantee for PVP connected appliances a stable electric supply 0 otherwise
to face possible climatic perturbations. (
1  jzkez 0k j
if jzk  z0k j  ez0k
lCjk zk z0k

0 otherwise
3.2. Planning algorithm j f1; 2; 3; 4g for connection control 15

The planning fuzzy algorithm is based on four steps: the knowl- where x0i, y0l, z0k are respectively the maximum values of the real
edge base of the expert, the fuzzication, the inference diagram variables xi, yl, zk in their membership domains and e0i, e0l, e0k, are
and the defuzzication [19]. respectively the range values of x0i, y0l, z0k.

Table 3
b Gd and connection control.
The fuzzy partitions details of: appliances states, P V

Fuzzy partition Number of subset Fuzzy subset Fuzzy domain Condition to verify
PN s
Appliance states Ns = 4 Aji = (OFF, ON) X = [0, 1]
i1 lAji x 1
i = {1,2}: number of partition x = ERj e X
j = {1, 2, 3, 4} (ERj: appliance state j)
PN s
PVP power Ns = 9 Bl = {A, B, C, E, F, G, H, I} Y = [0, 260] lBl y 1
l1
l = {1, 2, 3, 4, 5, 6, 7, 8, 9} yl = Pl e Y
(Pl: PVP power)
PN s
Connection control Ns = 2 Cjk = (grid, PVP) Z = [0, 1] lCjk z 1
i1
k = {1, 2} zj = ESj e Z
(ESj: relay state j)
 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2345

(x) 3.2.3. The inference diagram

By means of the obtained membership functions and the appli-
1
ances operating powers which are: 30, 40, 60 and 75 W, a rule base
is established, according to Mamdani [21], using the general rule
(xr) format: Rilk : if X is Ai and yl is Bl then Z is C k . Where X = [x1 x2
x3 x4] is the appliance states vector and Ai = [A1i A2i A3i A4i] its lin-
x guistic values vector, yl the PVP available generation and Bl its lin-
0 guistic value, Z = [z1z2] the command vector of appliances relays
x0r x 0r x r 1
and Ck = [C1k C2k C3k C4k] its linguistic values vector.
It can be seen that for a validated condition (X is Ai and yl is Bl),
Fig. 6. General form of a membership function.
many actions Ck can be affected by the command vector Z. Conse-
quently, a decision making procedure selects the action Ck, which
veries the planning criteria.

(a)

Fig. 7a. The real installation.

(b)

Fig. 7b. Synoptic Schema of installation.

2346 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351


The rules aggregations are given by computing the minimum ESj grid if z0k < 0:5
17
norm conjunction implication of each fuzzy subset of the jth appli- ESj PVP if z0k  0:5
ance relay:
Finally, the fuzzy energy planning algorithm is organized as follows:
j wC 0j1 minwj1 ; lC j1 : for the rst fuzzy subset,
 Initialisation
j wC 0j2 minwj2 ; lC j2 : for the second fuzzy subset. j Select the operating mode for each appliance: 0, 1, 2 or 3.
j Fix the start-up time; if mode 0 or mode 1 is selected.
where wj1 is the minimum norm fuzzy conjunction between the j Fix the operation period; if mode 0 or mode 2 is selected.
rst fuzzy subset of the jth appliance and the lth fuzzy subset of b G estimation
   PV
b Gd : wj1 min l ; l
the P V and wj2 is the minimum norm
Aj1 Bl j Acquisition of the climatic parameters data base: DBd1.
fuzzy conjunction between the second fuzzy subset of the jth j Estimation of the climatic parameter vector: M b d; T
b d G b d .
appliance and the (l + 1)th fuzzy subset of the b Gd .
j Estimation of the photovoltaic generation: P V
 
PVb Gd : wj2 min l ; l .  Energy planning
Aj2 Bl1
j Acquisition of the appliances states: ERj, j = {1, 2, 3, 4}.
Using the maximum T-conorm rule aggregation, the member-
j Fuzzy calculation of the of the relays control: ESj
ship function for an
 operating
 point of the jth appliance relay is gi-
j Decisions: The connection times of appliances to PVP.
ven by lC j max lC 0 ; lC 0 .
j1 j2
 Establishment of the daily energy balance
3.2.4. The defuzzication
Since the rules are aggregated, the defuzzication consists of
calculating the real value z0k of the relay state of each appliance 4. System implementation and assessment
using the centroid method (z0k is the centre of lC j ) (Eq. (16)) [31]:
R1 The system has been installed at the National Engineering
0
zj lC j dzj School, University of Sfax (ENIS) Tunisia. It includes a 260 Wp
z0k R 1 16
lCj dzj PVP made up of four parallel connected arrays (TE500CR+ of Total
0
Energie) and the electric grid as a complementary energy source.
Hence, the connection relay control of the jth appliance is deter- The PVP is equipped with a Maximum Power-Point Tracker
mined by Eq. (17): (MPPT), which is an electronic device that monitors PVP to operate

Measured PVG d
Estimated
.......................................................
PVGd
08/01/08 08/02/08 08/03/08 08/10/08 08/11/08
250

200
Power (W)

150

100

50

0
0 23-0 23-0 23-0 23-0 23-0 23-0 23-0 23-0 23-0 23-0 23
Time (Hour)

250
August 11, 2008 Measured PVGd

200
NMBE = 0.51% Estimated PVG
d

150
Power (W)

100

50

0
6 8 10 12 14 16 18 20
Time (Hour)

Fig. 8. Measured and estimated PVP generation.

 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2347

near its maximum power-point along the IV curve and an inverter tion 2.2). It considers measures taken during the ten past days from
that provides the same output voltage as the electric grid (230 V/ August, 1st to August, 10.
50 Hz). The appliances, chosen as four lamps of 30, 40, 60 and In order to evaluate the estimation performances, the estimated
75 W, are supplied, via a switching relays bloc, either by the PVP and measured data are analysed by computing the Normalized
output or the electric grid. The whole installation is controlled by Mean Bias Error (NMBE) dened by Eqs. (18) and (19) [32,33]:
a PC computer in which the planning algorithm is implemented.
The computer is connected as well to an acquisition system provid- P
N
estimated value  measured value
ing climatic parameter measures (G, T). Fig. 7a shows the real i1
NMBE%  100 18
installation where as Fig. 7b gives its synoptic schema. P
N
measured value
i1
P
N
b Gd;i  PVGd;i
P V
4.1. Validation of PVG forecast
NMBE% i1  100 19
P
N
PVGd;i
The algorithm has been tested, during the year 2008, in view of i1
a household installation composed of four appliances of 30 W,
40 W, 60 W and 75 W as nominal powers. where N is the number of reading points.
The PVG estimation is validated for a given day (August, 11th). The measured and estimated curves are shown by Fig. 8. The
b Gd is computed using the ANFIS approach (Sec-
The estimated P V zoom in of the considered day for estimation demonstrates that

3
Grid
2 Appliance 1 (30W) in mode 0
PVP T'1
1 te 1
OFF T1
0
5 ts1 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid
2 Appliance 2 (40W) in mode 0
PVP
Appliances Connections

1 ts2 te 2
OFF T2
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid
2 Appliance 3 (60W) in mode 0
PVP ts3 te 3
1
OFF T3
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid
2 Appliance 4 (75W) in mode 0
PVP ts'4 T'4
1 ts4 te 4
OFF T4
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

250
PVGd
Augus t 1 1 , 2 0 0 8
225 PVGC
PVGL
200
Grid C
175

150
Powers (W)

125

100

75

50

25

0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

Fig. 9. Appliances connection and power balance: rst case study.

2348 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351

the estimated and the measured PVG present great concordance grid (Fig. 9). Consequently, the algorithm plans appliances 2
(NMBE = 0.51%). and 3 to function with PVP output. As for appliance 1, it starts
on the electric grid during the time T 01 then it is switched on
4.2. Daily assessment PVP output until the end of the programmed operation period.
In addition, appliance 4 starts on PVP output then it is switched
As estimated, the P V b Gd is planned to determine the connection on the electric grid at ts04 and during time T 04 . The times
times and duration of the four appliances either to PVP output or to T 01 and T 04 are calculated so as to use maximum power from
the grid, by taking into account the planning modes and criteria. A PVG. In this case study, the lost PVG is considerable since the
daily assessment is established for the considered day (August, engaged operation mode imposes operation periods and starting
11th 2008). It consists of the switching chronogram of appliance times of appliances. This mode lets no exibility to the planning
between the PVP output and the grid on the one hand and the all algorithm to use the PVP generated energy. Considered as
power curves (P V b Gd ; consumed from PVP: PVGC, lost from PVP: severe, this operating mode does not yield to an optimal use
PVGL, and consumed from grid: GridC) on the second hand. Three of the P V b Gd since it solicits for long time the electrical grid to
cases study are computed for the same day. full the deciency in PVP generation during appliances
operation.
 First case study: Mode 0 is given to all appliances. The operation  Second case study: Mode 3 is programmed for all appliances.
periods (T1, . . . , T4) and start-up times (ts1, . . . , ts4) of appliances The operation periods and starting times are indifferent. The
are imposed. According to the availability of P V b Gd , the algorithm start-up time (ts1, . . . , ts4) and the shutdown times
switches appliances between the PVP output and the electric (te1, . . . , te4) of appliances depend only on the availability of

3
2 Grid Appliance 1 (30W) in mode 3
PVP te 1
1 ts1
OFF T1
0
Appliances Connections

5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
2 Grid
Grid Appliance 2 (40W) in mode 3
PVP ts2 te 2
1
OFF T2
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
2 Grid Appliance 3 (60W) in mode 3
PVP ts3 te 3
1
OFF T3
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
2 Grid Appliance 4 (75W) in mode 3
PVP ts4 te 4
1
OFF T4
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

250
PVGd
Augus t 1 1 , 2 0 0 8
225
PVGC
200 PVGL

175

150
Powers (W)

125

100

75

50

25

0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

Fig. 10. Appliances connection and power balance: second case study.
 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2349

b Gd . This mode makes the best benet of P V

the P V b Gd . In addi- and it has the lowest power, it is planned to function during
tion, there is no need to the electric grid. Fig. 10 shows the the day light which improves the installation efciency.
connection times to PVP output and operation duration of Finally; appliance 4 starts, as imposed at ts4, on the electric
appliances. grid during T 04 then it is switched on PVP output. It remains
 Third case study: Each appliance is programmed for a specic supplied by PVP output until ts04 at which it is switched again
operating mode (Fig. 11). Appliance 1 in mode 3 where ts1 on the electric grid until te4 where it accomplishes the oper-
and T1 are indifferent; appliance 2 must start at ts2 (preset ation period T4.
to 10:00) and for an indifferent operation period T2 (mode
1); appliance 3 in mode 2 where the start-up ts3 is indiffer- Regarding the three cases study, it is obvious that the consumed
ent but the operation period is imposed (T3 = 5 hours), appli- PVP generation (PVGC) tends to reach the estimated P V b Gd . The lost
ance 4 is programmed to operate in mode 0, it must start at energy (PVGL) is caused by the pre-xed appliances nominal pow-
ts4 (preset to 9:00) and during T4 (preset to 7 h). After run- ers and the modes that impose starting times and operation peri-
ning the planning algorithm, it is decided to connect totally ods. In addition, it can be seen that the planning criteria are
appliances 2 and 3 to PVP with respect to the imposed veried since there is no operation disturbance in power supply
start-up times and during their operation periods. Since of appliances and the imposed start-up times and operation peri-
appliance 1 is programmed in the exible mode (mode 3) ods are veried.

3
Grid
2 Appliance 1 (30W) in mode 3
PVP ts1
1 te 1
OFF T1
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid
2 Appliance 2 (40W) in mode 1
Appliances Connections

PVP
1 ts2 te 2
OFF T2
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid
2 Appliance 3 (60W) in mode 2
PVP ts3 te 3
1
OFF T3
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
3
Grid Appliance 4 (75W) in mode 0
2
ts'4 T''4
PVP ts4 T'4 te 4
1
OFF T4
0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

250
PVGd
Augus t 1 1 , 2 0 0 8
225 PVGC
PVGL
200
GridC
175

150
Powers (W)

125

100

75

50

25

0
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Time (hour)

Fig. 11. Appliances connection and power balance: third case study.
2350 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351

Table 4
The monthly audit.

January February March April May June July August September October November December
b Gd (kWh)
PV 25.9 29.6 38.7 40.6 44.5 44.8 46.9 48.4 41.1 37.0 27.8 23.6
b GC (kWh)
PV 18.9 23.2 31.4 32.6 35.2 36.2 37.7 40.0 33.1 29.7 21.5 17.0
GridC (kWh) 17.1 12.8 4.6 3.4 0.8 0.0 0.0 0.0 2.9 6.3 14.5 19.0
b GL (kWh)
PV 7.0 6.4 7.4 8.0 9.3 8.6 9.2 8.4 8.0 7.3 6.3 6.6
g% 73.1 78.3 81.0 80.3 79.0 80.8 80.3 82.7 80.6 80.2 77.4 72.1

95

93

91

89

87
(%)

85

83

81

79

77

75
1 2 3 4 5 6 7 8 9 10 11 12
Month

Fig. 12. Effectiveness coefcient.

4.3. Monthly assessment generally produced around the sunrise and sunset times where
PVG is insufcient (<30 W) to supply energy for even the lowest
The daily assessment is developed during the year 2008 to form power appliance.
a monthly energy balance so as to prove the effectiveness of the A second valorisation consists of the calculation of a monthly
planning approach. All PVP related powers are integrated over day- effectiveness coefcient (g%) dened by [20]:
light (from sunrise to sunset). Obtained energies are cumulated
b Gd  PVGL
PV PVGC
over the month length (ml) to compute PVP monthly energies g%  100  100 23
b Gd , consumed: PVGC, lost: PVGL) as follows (Eqs.
(available: P V PVb Gd b Gd
PV
(20)(22)):
All the monthly PVP energies and effectiveness coefcient (g%) are
ml Z
X sunsetj summarized in Table 4. This coefcient is almost constant through
b Gd
PV b Gtdt
PV 20 the months of the year and it varies between 72% and 82.7%. Fig. 12
sunrisej
j1 shows the effectiveness coefcient behaviour during the year. It
ml Z
X sunsetj conrms that the planning approach can make use of about 80%
PVGC PVGC tdt 21 of the PVP generated energy which justies a continuous energy
j1 sunrisej
savings contribution of the developed solution.
ml Z
X sunsetj
PVGL PVGL tdt 22
j1 sunrisej 5. Conclusion

The consequent energy balance (Table 4) is established by program-
ming all appliances in mode 2. This mode is chosen because it is the
commonly used in household appliances as washing machine, dish-
washer, etc. It can be seen that the PVP is able to supply all appli-
ances needed power during the hot season (JuneAugust). For the
moderate season (MarchMay and October), appliances require
slightly the electric grid supply. Finally, during the cold season
(NovemberFebruary) the PVG is not usually available, the appli-
ances are connected to electric grid for long times. During all
months, there is always some available PVG which is not used.
The lost PVG is due to the imposed operation periods (T1, . . . , T4).
While the electric energy produced by the PVP varies between
23 and 48 kWh/month, the electric energy consumed from the
PVP varies between 17 and 40 kWh/month which involves an un-
used energy between 6 and 8 kWh/month. The lost energy (PVGL) is
A novel energy planning for a household PVP installation is sug-
gested. The approach avoids the use of a battery by planning the
estimated PVG and deciding the optimum connection times and
duration of the installation appliances to the PVP output. Investiga-
tion is developed on a 260 Wp PVP and four appliances of different
powers (30 W, 40 W, 60 W and 75 W). Decisions are taken by fuzzy
rules and on the basis of optimal planning criteria. The algorithm
aims to optimize the utilisation of the PVP generation (PVG) with
respect to appliances operation modes (start-up times and opera-
tion periods). In case of unavailability of PVG, the algorithm con-
nects appliances requiring power to the electric grid so as to
ensure power stability for the installation. A daily and monthly
assessment conrms that the approach effectiveness reaches 80%.
The energy balance should be improved by adding other appliances
of low nominal powers in order to exploit the unplanned PVG.
 M.B. Ammar et al. / Applied Energy 87 (2010) 23402351 2351

References [17] Nfah EM, Ngundam JM, Tchinda R. Modelling of solar/diesel/battery hybrid
power systems for far-north Cameroon. Renew Energy 2007;32:83244.
[18] Li Chun-Hua, Zhu Xin-Jian, Cao Guang-Yi, Sui Sheng, Hu Ming-Ruo. Dynamic
[1] Diaf S, Notton G, Belhamel M, Haddadi M, Louche A. Design and techno-
modeling and sizing optimization of stand-alone photovoltaic power systems
economical optimization for hybrid PV/wind system under various
using hybrid energy storage technology. Renew Energy 2008;33:143543.
meteorological conditions. Appl Energy 2008;85:96887.
[19] Salah Chokri Ben, Chaabene Maher, Ammar Mohsen Ben. Multi-criteria fuzzy
[2] Hongxing Yang, Wei Zhou, Chengzhi Lou. Optimal design and techno-
algorithm for energy management of a domestic photovoltaic pane. Renew
economic analysis of a hybrid solar-wind power generation system. Appl
Energy 2008;33:9931001.
Energy 2009;86:1639.
[20] Chaabene Maher, Ammar Mohsen Ben, Elhajjaji Ahmed. Fuzzy approach for
[3] Elkinton Melissa R, McGowan Jon G, Manwell James F. Wind power systems for
optimal energy-management of a domestic photovoltaic panel. Appl Energy
zero net energy housing in the United States. Renew Energy 2009;34:12708.
2007;84:9921001.
[4] Benghanem M. Low cost management for photovoltaic systems in isolated site
[21] Mamdani EH, Assilian S. An experiment in linguistic synthesis with a fuzzy
with new IV characterization model proposed. Energy Convers Manage
logic controller. Int J Man Mach Stud 1975;7(1):113.
2009;50:74855.
[22] Ipsakis Dimitris, Voutetakis Spyros, Seferlis Panos, Stergiopoulos Fotis,
[5] Rashed Mohamed, Elmitwally A, Kaddah Sahar. New control approach for a PV-
Elmasides Costas. Power management strategies for a stand-alone power
diesel autonomous power system. Electr Power Syst Res 2008;78:949.
system using renewable energy sources and hydrogen storage. Int J Hydrogen
[6] Diaf S, Diaf D, Belhamel M, Haddadi M, Louche A. A methodology for optimal
Energy 2008:115.
sizing of autonomous hybrid PV/wind system. Energy Policy
[23] Maclay James D, Brouwer Jacob, Scott Samuelsen G. Dynamic modeling of
2007;35:570818.
hybrid energy storage systems coupled to photovoltaic generation in
[7] Mondol Jayanta Deb, Yohanis Yigzaw G, Norton Brian. Optimising the
residential applications. J Power Sources 2007;163:91625.
economic viability of grid-connected photovoltaic systems. Appl Energy
[24] Mellit A, Benghanem M, Kalogirou SA. Modeling and simulation of a stand-
2009;86:98599.
alone photovoltaic system using an adaptive articial neural network:
[8] Kim Ju-Young, Jeon Gyu-Yeob, Hong Won-Hwa. The performance and
proposition for a new sizing procedure. Renew Energy 2007;32:285313.
economical analysis of grid-connected photovoltaic systems in Daegu, Korea.
[25] Jang JSR. Ans: adaptive-network-based fuzzy inference systems. IEEE Trans
Appl Energy 2009;86:26572.
Syst Man Cybern 1993;23(3):66585.
[9] Kim Il-Song, Kim Myung-Bok, Youn Myung-Joong. A new maximum power
[26] Jang JSR, Sun CT. Neuro-fuzzy modeling and control. Proc IEEE
point tracker using sliding-mode observer for estimation of solar array current
1995;83(3):378406.
in the grid-connected photovoltaic system. IEEE Trans Ind Electron
[27] Jang JSR, Sun CT, Mizutani E. Neuro-fuzzy and soft computing: a
2006;53(4):102735.
computational approach to learning and machine intelligence. Prentice Hall
[10] Chaabene Maher, Ammar Mohsen Ben, Salah Chokri Ben, Elhajjaji Ahmed.
Inc.; 1997.
Fuzzy based energy management of a domestic photovoltaic panel. In: SSD07,
[28] Yager RR, Zadeh LA. Fuzzy sets neural networks, and soft computing. Van
Fourth international multi-conference on systems, signals & devices,
Nostrand Reinhold; 1994.
Hammamet, Tunisia, March 1922, 2007.
[29] Ammar Mohsen Ben, Chaabene Maher, Rabhi Abdelhamid, Kamoun Mohamed
[11] Galdi V, Piccolo A, Siano P. Exploiting maximum energy from variable speed
Ben Ali. Energy planning of a domestic photovoltaic panel. In: World
wind power generation systems by using an adaptive TakagiSugenoKang
renewable energy congress X and exhibition, Glasgow, Scotland, 1925 July
fuzzy model. Energy Convers Manage 2009;50:41321.
2008.
[12] Yang Hongxing, Zhou Wei, Lu Lin, Fang Zhaohong. Optimal sizing method for
[30] Wai Rong-Jong, Lin Chung-You, Duan Rou-Yong, Chang Yung-Ruei. Efciency
stand-alone hybrid solar-wind system with LPSP technology by using genetic
DCDC converter with high voltage gain and reduced switch stress. IEEE Trans
algorithm. Sol Energy 2008;82:35467.
Ind Electron 2007;54(1):35464.
[13] Uzunoglu M, Onar OC, Alam MS. Modeling, control and simulation of a PV/FC/
[31] Sallem Souhir, Ammar Mohsen Ben, Chaabene Maher, Kamoun MBA. Fuzzy
UC based hybrid power generation system for stand-alone applications. Renew
rules based energy management of a PVP/battery/load system. In: World
Energy 2009;34:50920.
renewable energy congress X and exhibition, Glasgow, Scotland, 1925 July
[14] Li Chun-Hua, Zhu Xin-Jian, Cao Guang-Yi, Sui Sheng, Hu Ming-Ruo. Dynamic
2008.
modeling and sizing optimization of stand-alone photovoltaic power systems
[32] Chaabene M, Ben Ammar M. Neuro-fuzzy dynamic model with Kalman lter to
using hybrid energy storage technology. Renew Energy 2009;34:81526.
forecast irradiance and temperature for solar energy systems. Renew Energy
[15] Bernal-Agustn Jos L, Dufo-Lopez Rodolfo. Multi-objective design and control
2008;33:143543.
of hybrid systems minimizing costs and unmet load. Electr Power Syst Res
[33] Koussa M, Malek A, Haddadi M. Statistical comparison of monthly mean
2009;79:17080.
hourly and daily diffuse and global solar irradiation models and a Simulink
[16] Sallem Souhir, Chaabene Maher, Kamoun MBA. Energy management algorithm
program development for various Algerian climates. Energy Convers Manage
for an optimum control of a photovoltaic water pumping system. Appl Energy
2009;50:122735.
2009.