Sie sind auf Seite 1von 5

Enhancing the integration of renewable sources in

distribution systems using DC-links


A. Marano-Marcolini, E. Romero-Ramos, A. Gomez-Exposito, J.M. Maza-Ortega, J.L. Martinez-Ramos
alejandromm@us.es, eromero@us.es, age@us.es, jmmaza@us.es, jlmr@esi.us.es
Department of Electrical Engineering
University of Seville
Camino de los Descubrimientos s/n, 41092 Sevilla, Spain

HV MV
Abstract—This work proposes the use of DC-links between
existing radial distribution feeders so that full advantage is taken
of their maximum capacity, which is particularly important
when new distributed generators are to be connected at the Substation 1
distribution level. An optimization tool is developed to determine
DC-link NO
the optimum settings for the DC-link control variables, in order
location
to maximize the penetration of dispersed sources subject to HV MV
customary electrical constraints. Simulation results regarding two
real networks are presented to illustrate the former advantages.
Also, improvements in system operation resulting from the
presence of DC-links are studied and discussed. A real loss model Substation 2
for the DC link is considered in all test cases.

I. I NTRODUCTION Fig. 1. Typical distribution topology


Traditionally, medium voltage systems are operated in a
radial arrangement. The low degree of reliability for these
radial networks is improved by adding emergency ties that imply the substitution of the existing circuit breakers [3],
provide alternative routes for power supply in case of outages and secondly, important inversions in new advanced protection
or scheduled interruptions. These links end with an open schemes would have to be carried out.
switch so that radial structure is maintained during normal A way to circumvent these problems, at the same time
conditions. The improvements that would result by closing the the improvements associated with a meshed operation are
normally open links are well known [1]: power lines would be achieved, consists of using DC-links [4]. DC-links take ad-
less loaded, better voltage profile, a reduction of power losses, vantage of the full capacity of the distribution system so that
etc. However, radial systems are easier to operate and more new distributed generators can be connected to the system [5].
economic to protect, at least when all the customers connected Such DC-links connect two or more feeders, usually at remote
to the network are passive loads, as the direction of power points, making power flows from each other possible and
flows, from distribution substations to consumers, is known offering new alternative routes to carry all the supplied power
in advance in those cases. This advantage has prevailed over from generators to distribution substations. Additionally, the
the former limitations until now, but the increasing presence reactive power supplied or consumed by the inverters of the
of distributed generators at the medium voltage starts to argue DC-link constitute new control variables to take into account in
this radial operation topology. order to incorporate as much dispersed power as possible [6].
Distribution companies are increasingly facing the problem This work formulates the optimization problem that re-
that a great deal of distributed generators (DG) apply to sults by considering this new control devices in order to
connect to their radial systems. In many occasions, the utility improve as much as possible the operation of the system
is forced not to admit all of them due to the fact that, with from different points of view. The mathematical problem is
all the generators supplying power, the system would be at formulated rigorously including a real model for the DC-link
“full load”, that is, some power line ampacities or voltage in section II-A. Simulation results regarding two real networks
magnitudes would be out of limits. If the system was operated are presented in section III, while section IV summarizes the
in a meshed way, a higher DG penetration would result as has main contributions of this work.
already been proved [2]. This mesh operation would allow the
deferment of investments needed to meet with the growing II. S YSTEM MODEL
energy demand, no matter if it is consumed or supplied. Figure 1 shows the classical topology for medium voltage
Unfortunately, actual radial systems are very difficult to systems. Only two feeders have been detailed for the sake of
operate as a network due mainly to two reasons: firstly, a simplicity, while other feeders from distribution substations
rise in the short circuit current in each node that would are represented by concentrated spot loads. The two feeders,
Pl Pm where constants a, b and c depend on the IGBTs manufacturer,
Ql Qm and I is the RMS current through each IGBT. A 3 MVA
AC DC
VSC is considered by using one of the most powerful existing
IGBTs from ABB. Notice that equation (2) represents the loss
DC AC of each IGBT in the VSC, six in total for the adopted two-level
l m bridge.
Note that these active power losses must appear in the
balance equation (1), that is, the active power flow leaving
Fig. 2. Back-to-back converter one feeder must be equal to the active power flow arriving to
the other one plus the power losses in both converters. So, if
this more realistic model is considered for the back-to-back
leaving different distribution substations, extend along a cer-
arrangement, the equation (1) must be rewritten as,
tain number of secondary distribution transformers where dis-
persed generators and consumers are connected. Both feeders
Pl + Pm + Plloss + Pm
loss
=0 (3)
could be connected through a switch that should be normally
open to force the radial system topology. A meshed system can where Plloss and Pm
loss
denote power losses at each VSC.
be easily constructed by installing a DC-link between those
two points where the open switches are located. B. Optimization formulation
Next, the mathematical model for all the components of Any optimization problem has a general formulation con-
the power system and the resulting optimization problem are sisting of a scalar objective function f to minimize or maxi-
formulated. mize, along with equality and inequality constraints that must
be enforced during the solution process. The mathematical
A. DC-link problem can be formulated in a general way as,
DC-links can have different configurations. The one consid-
ered here is based on the so-called voltage source converter min f (x, u)
(VSC). Figure 2 shows a back-to-back device consisting of 
g(x, u) = 0
two VSCs sharing a common DC bus. The most economical s.t.
topology has been considered, namely the 6-pulse two-level h(x, u) ≤ 0
bridge. where vector x comprises the dependent variables and vector
When this converter is installed between two feeders via u consists of control variables.
transformers, the power flow thorough the link can be adjusted In this work, complex node voltages are the dependent
to system requirements in both directions. Note that the sum variables, that is, voltage magnitude and angle at each node,
of the active powers injected by the converters on each feeder Vi and θi . The control vector u is composed of the previously
must be zero as no storage of generation is connected to presented control variables related to the DC-link: reactive
the DC bus. Additionally, thanks to the self-commutating power injected or consumed from each VSC and the active
technology, reactive power at both sides of the converter can power flow trough the DC-link.
be controlled, that is, voltage magnitudes at the furthest buses The equality constraints are obtained imposing at each node
of both feeders can be regulated independently, which provides the power balance equations,
an important operational advantage, as will be seen later.
Consequently, the DC-link contributes with three new control  
variables: both reactive powers injected at each terminal bus Pig − PiL = Vi Vj Gij cos(θij ) + Vj Bij sin(θij ) (4)
of the converter, Ql and Qm , and one of the active power j
 
injections, Pl or Pm . These two last quantities are related by: Qgi − QL
i = Vi Vj Gij sin(θij ) − Vj Bij cos(θij ) (5)
j
Pl + Pm = 0 (1)
where Sig= Pig
+ jQgi is the total complex power injected
Additional equations must be added to properly model the by dispersed generators connected to bus i, SiL = PiL + jQL i
internal losses of IGBTs and diodes, that is, the switching is the total complex load power demanded at bus i, and Yij =
elements in the VSC. Both the switching and conduction losses Gij + jBij is the ij-element of the bus admittance matrix.
of the IGBTs and diodes are considered. The losses model HV/MV transformers at distribution substations are equipped
used in this work is the one proposed in [7]. A quadratic with automatic tap changers to keep the voltage magnitudes at
polynomial equation is used to compute the IGBTs losses. head nodes of feeders to a constant value. So, equations (4) and
The polynomial coefficients are determined using information (5) are replaced by Ui = Visp |αi for each node i from 1 to ng ,
from manufacturers’ computing programs. The loss equation ng being the total number of distribution substations. Finally,
formula for each IGBT-diode pair is: equation (1), forcing the null generation or consumption of
active power from the converter, is the last equality constraint
Ploss = aI 2 + bI + c (2) to consider in the optimization problem.
Inequality equations h are defined by the following electri- Feeder 1
10000
cal constraints,

S (kVA)
1) Conductor ampacity: thermal limit of overhead or un- 5000

derground lines,
0
5 10 15 20
max
0 ≤ Iij ≤ Iij (6) Feeder 2
300

S (kVA)
2) Voltage limits: voltage magnitudes must be in a specified 200
range, 100
V min ≤ Vi ≤ V max (7) 0
5 10 15 20
Hour
3) Converter capacity: The apparent power through each
voltage source converter can not exceed its rated power,
Smax , 3 MVA for the simulated DC-link, Fig. 3. Loading evolution for both feeders of Network A.

0 ≤ Sl ≤ Smax 1500

P (kW)
(8)
0 ≤ Sm ≤ Smax 1000

l, m being the DC-link terminal buses. The total appar- 500


5 10 15 20
ent power associated to each converter must include the
1500
power losses defined by (2).
Q (kvar)
At this point, it is worth remembering that the main benefits 1000
of the installation of DC-links are, on the one hand, the
500
increased feeder loadability by sharing spare capacity between 5 10 15 20
the interconnected feeders and, on the other hand, the addi- Hour
tional control of voltage magnitudes by taking advantage of the
capacity of the converters to absorb or supply reactive power. Fig. 4. Active and Reactive power through the DC-link. The total reactive
Different objective functions can be considered depending on power is presented, i.e., the addition of the injections in both terminals.
the main control goal. If the goal is to increase the penetration
of DG, then the objective function of the optimization problem
is, A. Power loss minimization
 The benefits of using a DC-link on a real network, from
max Pig , (9) now on Network A, are shown in this section. This network
i consists of two feeders connected to the same substation. One
where Pig represents the new power injected by distributed of them is a long line connecting rural loads (feeder 1), while
generator at bus i, and the sum is extended to those nodes the other is a short urban line (feeder 2). The first one has
where distributed generators are to be connected. higher loads than the second one for the whole day long.
An alternative objective function could be to minimize The rural feeder is working at about half its nominal capacity,
power losses for a fixed generation and load profile, that is, meanwhile the urban feeder has most of its capacity unused.
minimizing the total active power injected by each distribution Figure 3 shows the power demand evolution during a whole
substation to the medium voltage network under study, day for both feeders. A representative winter day is used to
obtain the results, as this is the season with highest power
ng demand for this network.

min Pj (10) The OPF described in Section II-B is used with the objective
j=1 function defined in equation (5), that is, the objective is to
reduce the active power injected from the substation. The
ng being the total number of distribution substations.
results of this problem provide the proper control variables
Those objective functions will be considered in the next
for the back-to-back device (Pl , Ql and Qm ).
section by applying them to real distribution systems.
The DC-link must be rescheduled each time the loading
conditions vary. In this case, the aforementioned OPF is
III. R ESULTS
executed 24 times, one for each hour of the considered day.
This section is structured in two subsections. The first one Doing so, the active power through the DC-link and reactive
presents the results concerning the power loss minimization. power injection at each bus, Ql and Qm , that reduce the system
And the second one presents the results obtained with the looses are obtained. Figure 4 shows these results. The lower
objective of increasing the DG penetration. plot represents the total reactive power managed by the DC-
30 kWh using the DC-link, representing a reduction of about 20
%.
Loss reduction (%)

20

B. Increasing the DG penetration


10
The case evaluated here represents a real problem in a
0 distribution network, Network B. This system consists of two
feeders departing from the same substation. Both feeders are
−10 equally loaded, but one of them (feeder 1) has a DG connected
0 5 10 15 20 25
Hour
to its further terminal. This DG is currently providing 3 MW
of power without providing reactive power support. If new
DG want to connect by a near point, the voltage of the further
Fig. 5. Loss reduction in percentage using a DC-link. terminal reach the highest allowed value.
The benefits obtained inserting a DC-link between the
TABLE I
E VOLUTION OF THE DC- LINK LOSSES IN K W FOR THE 24 HOURS .
problematic feeder and the other one are assessed here.
In the initial conditions, slightly increasing the power in-
Hour Loss Hour Loss jected from the DG leads to infeasible high voltage in the DG
1 23.37 13 25.57
2 23.37 14 25.71
connection bus. If a reactive compensator is connected to that
3 23.37 15 25.85 bus the allowed injected power is 4.18 MW. Now, the limit is
4 23.37 16 25.45 caused by the ampacity of the feeder.
5 23.37 17 25.30
6 23.37 18 25.26 Using a DC-link leads to further improvement. Not only is
7 23.37 19 26.06 the back-to-back device able to absorb the required reactive
8 23.37 20 26.57 power, but also allows the active power generated from the DG
9 23.89 21 27.47
10 25.78 22 28.04 to be transferred to the second feeder, increasing the capacity
11 25.70 23 27.07 of the whole system.
12 25.52 24 25.64 Using the OPF with the objective defined by (9), the
following results are obtained for a 3 MVA DC-link:
• Active power flow through the device (Pl ): 2.69 MW
link, that is, the result of adding |Ql | to |Qm |.
Note that the active power flow through the DC-link follows • Reactive power injection to feeder 1 (Ql ): -1.25 Mvar
the trend of feeder 1 demand. It is minimum for hour 7 and
• Reactive power injection to feeder 2 (Qm ): 0.49 Mvar
maximum for hour 22, coinciding with the lowest and highest
demand values, respectively. The DC-link force a higher power With these controls the feeder 1 could accept 6.9 MW till the
flow through the unloaded feeder 2, which implies a decreased ampacity of its lines is reached. This represents an increment
power flow through feeder 1. of a 130 % with respect to the original case, and 65% with
Active power losses depend on line resistances and the respect to the case of adding a reactive power compensator.
squared currents, assuming that the line resistances of both
feeders have equivalent magnitudes, the best situation will be IV. C ONCLUSIONS
achieved when both currents are approximately the same. In an
ideal situation, this is what the DC-link should do. Two facts This paper presents the benefits of using DC-links in dis-
affect, however, this condition. One is the power limitation tribution networks. A detailed model of the IGBT losses is
of the back-to-back device, forced by the IGBTs maximum implemented for taking into account the losses of the back-to-
current flow, 3 MVA in this case. The other is the power losses back selected configuration. The formulation of the OPF used
of the DC-link itself, which according to (2) increase following to obtain the proper DC-link control values is also depicted.
a second order polynomial equation of the current. Results concerning two real networks are presented. The
The achieved loss reduction, in percentage, can be seen in first network is used to demonstrate the DC-link influence
Figure 5. Notice that for certain hours, particularly those in in decreasing the total active losses. This goal is achieved
which the feeder 1 is less stressed, no reduction is achieved, leveling the load of the feeders using the DC-link properly.
instead a small increment happens. This is due to the DC-link The second network is used to show that, using DC-links,
power losses. For these hours, the benefit obtained transferring the DG connected to distribution networks can be increased
power from one feeder to the other is counteracted with the allowing the efficient use of the full capacity of the feeders.
DC-link losses. Table I depicts the active power losses of the The proposed OPF, working with different objective func-
back-to-back device. tions, can be used to adjust the DC-link controls to the loading
Despite the DC-link losses, the global effect is extremely conditions of the distribution network, searching for low losses
positive as the total power losses (computed for the whole when few DG is connected and increasing the penetration
day) decrease from an initial value of 1247.8 kWh to 992.5 capacity when needed.
A PPENDIX
This appendix is a short abstract of [7] depicting the method Ploss = aI 2 + bI + c, (11)
used to compute the IGBT losses. where the parameters a, b and c are unknowns that have to be
The exact computation of IGBT losses involves the use of adjusted using the power loss computation programs provided
precise physical models requiring transient simulation tools by the manufacturers. Once the operational conditions of the
with integration time steps lower than 0.1 μs. This compu- converter (such as DC bus voltage, switching frequency, power
tation technique is clearly not suitable for the optimization factor, ambient temperature, etc.) have been established, a
problem presented in this paper, so that a simplified compu- number of simulations varying the converter phase current,
tation of losses is proposed. The alternative chosen in this I, from no load to its rated value are carried out, yielding the
work to model power losses is to use the data given by the simulated power loss curve as a function of the phase current.
manufacturer of power electronic components. Then, the parameters a, b and c are adjusted using a least-
Nowadays, manufacturers such as Semikron, ABB or Mit- mean-square technique.
subishi offer free software for computing the power losses
of their products. ABB has developed a program for the ACKNOWLEDGMENTS
computation of power losses called SimulationTool for the The authors would like to thank the support provided by
high power range of IGBTs [8]. Semikron has a web-based Endesa under the Novare research program, and by the Spanish
power loss computation program called SemiSel for the low MEC under research projects ENE-2007-63306 and ENE-
and medium power IGBTs [9]. Finally, Mitsubishi offers the 2007-68032-C04-02.
software Power Module Loss Simulator for the high power
range of IGBTs [10]. R EFERENCES
In spite of being different programs for computing device [1] T-H Chen et all. Feasibility Study of Upgrading Primary Feeders from
power losses, all of them are based on the same assumptions, Radial and Open-Loop to Normally Closed-Loop Arrangement. IEEE
Transactions on Power Systems, Vol. 19, No. 3, pp 1308-1316, Aug. 2004.
being possible to calculate the power losses for a particular [2] G. Celi et all. Meshed vs Radial MV Distribution Network in Presence
application such as a single-phase or three-phase VSC con- of Large Amount of DG. IEEE PES Power Systems Conference and
sidering several factors like rated power, rated current, rated Exposition, Vol. 2, pp 709 - 714, 10-13 Oct. 2004
[3] T.H. Chen y W.T. Huang. Evaluation of the Variations of Short-Circuit
power factor, ambient temperature, etc. It is important to know Capacities along a Feeder due to a Distribution Network Upgrading.
which of these factors are the key parameters affecting the IEEE Power Engineering Society General Meeting, pp 1714 - 1719, 12-
power losses in order to derive a simplified model from the 16 June 2005
[4] B. Buchholz, I. Huebel, D. Povh, M. Weinhold. Medium Voltage Direct
data given by the manufacturers throughout these simulation Current (MVDC) Coupling in the liberalised market. CIGRE 2002; paper
tools. In this sense, a huge number of simulations have been 14-203.
carried out in order to asses the power loss dependance on [5] R.A.A. de Graaff, J.M.A. Myrzik, W.L. Kling, J.H.R. Enslin. Series
controllers in distribution systems - Faciliting increased loading and
the operational parameters used to compute the power losses. higher DG penetration. Power Systems Conference and Exposition pp.
As a result, the commutation frequency, the DC voltage and 1926-1930. October 29-November 1, 2006, Atlanta, Georgia, USA.
the RMS value of the current have been determined as the [6] N. Okada. Verification of Control Method for a Loop Distribution System
using Loop Power Flow Controller. Power Systems Conference and
key parameters affecting the IGBT power losses. However, Exposition, IEEE PSCE, pp. 2116-2123. October, 2006.
it is important to note that for the given application the [7] J.M. Mauricio, J.M. Maza-Ortega and A. Gómez-Expósito. Simplified
commutation frequency and the DC voltage are kept constant. Switch Model for Power Loss Computation in Electromagnetic Transient
Programs, International Conference on Power Systems Transients (IPST
As a consequence, the IGBT power losses depend exclusively 2009), 2-6 June 2009, Kyoto (Japan).
on the RMS value of the current. The following quadratic [8] ABB, Simulation Tool for IGBT, November 2008, http://library.abb.com.
equation is proposed to relate the power losses given by the [9] Semikron, Semikron Semiconductor Selection software. November 2008,
http://semisel.semikron.com.
manufacturer, Ploss , to the RMS phase current of the VSC, I: [10] Mitsubishi, Power Module Loss Simulator Ver. 4.02, November 2008,
http://www.mitsubishichips.com.

Das könnte Ihnen auch gefallen