Electric Power Systems Research 80 (2010) 698–706

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Reactive power control of wind farm made up with doubly fed induction
generators in distribution system
Jingjing Zhao a,∗ , Xin Li a , Jutao Hao b , Jiping Lu a
a
State Key Laboratory of Power Transmission Equipment & System Security and New Technology, Chongqing University, 400044 Chongqing, China
b
School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, 200093 Shanghai, China

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

Article history: In recent years, the number of small size wind farm made up with doubly fed induction generators (DFIG)
Received 7 April 2009 located within the distribution system is rapidly increasing. DFIG can be utilized as the continuous reac-
Received in revised form 8 August 2009 tive power source to support system voltage control by taking advantage of their reactive power control
Accepted 31 October 2009
capability. In this paper, considering both reactive power control and distribution network reconfigura-
Available online 6 December 2009
tion can be used to reduce power losses and improve voltage profile, a joint optimization algorithm of
combining reactive power control of wind farm and network reconfiguration is proposed to obtain the
Keywords:
optimal reactive power output of wind farm and network structure simultaneously. The proposed algo-
DFIG wind turbine
Network reconfiguration
rithm has been successfully implemented on the 16 bus distribution network and the results obtained
Particle swarm optimization demonstrate the efficiency of the algorithm.
Reactive power control © 2009 Elsevier B.V. All rights reserved.
Wind farm

1. Introduction
Wind energy is one of the most important and promising renew-
able energy resources in the world. The penetration of the wind
energy in electrical system is rapidly increasing. Currently, a grow-
ing number of small size wind farms used as DG sources are located
within the distribution system. Installing wind farm in the dis-
tribution system can defer the investments for the distribution
system expansion, but the intermittent and volatile nature of wind
power generation may impact distribution system voltages, fre-
quency and generation adequacy, so the electrical parameters of
the distribution network have to be maintained [1–4]. When wind
energy penetration is high, voltage control in the distribution sys-
tem becomes particularly important. As the consequences, in many
countries, the new established grid codes demand that wind farm
made up with doubly fed induction generators should actively par-
ticipate in improving voltage control in the distribution system
[5].
The variable-speed wind turbine equipped with DFIG is the most
popularly employed generator for the recently built wind farm. The
variable-speed wind turbine has the ability to obtain the maximum
active power from wind speed and control the reactive power inde-
pendently [6,7]. Utilizing DFIG reactive power control capability,
wind farm composed of DFIG can be used as the continuous reactive
∗ Corresponding author.
E-mail addresses: jingjingzhao@cqu.edu.cn, jjzhao sh@163.com (J. Zhao).
power source to support system voltage control with fewer costs
on the reactive power compensation device. Wind farm reactive
power control can reduce power losses and improve the voltage
profile at the user terminal by providing reactive power compen-
sation in distribution systems. Wind farm reactive power output
is controlled by the system optimal operation condition and the
reactive power control capability of each DFIG wind turbine.
There are many previous works on wind farm reactive power
control. Ref. [8] proposes a detailed mathematical model of the
DFIG and two alternative simulation models for the analysis of
both the active and reactive power performances associated with
a wind farm constituted exclusively by DFIG. Ref. [9] proposes an
optimized dispatch control strategy for active and reactive pow-
ers delivered by a doubly fed induction generator in a wind park.
Ref. [10] presents a control strategy developed for the reactive
power regulation of wind farms made up with DFIG, in order to
contribute to the voltage regulation of the electrical grid to which
farms are connected. Ref. [11] describes the relation between active
and reactive power in order to keep each DFIG operating inside the
maximum stator and rotor currents and the steady state stability
limit. Ref. [12] describes a PI-based control algorithm to govern
the net reactive power flowing between wind farms composed of
doubly fed induction generators and the grid.
Network reconfiguration is one of the most significant control
schemes in the distribution system, which alters the topological
structure of distribution feeders by changing open/closed status
of sectionalizing and tie switches. The purpose of the optimal
distribution network reconfiguration problem is to identify an

0378-7796/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2009.10.036
 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706 699

optimal radial operating structure that reduces real power losses

or improves voltage profile while satisfying operating constraints.
Most of the methods used for the network reconfiguration in
the literature are heuristic methods [13–15]. The other class of
approaches applied to network reconfiguration problem is based
on artificial intelligence searching algorithms, such as genetic algo-
rithm, simulated annealing algorithm, tabu search algorithm, etc.
[16–18].
However, the previous studies perform wind farm reactive
power control with no consideration of network reconfiguration,
or perform network reconfiguration with no consideration of wind
farm reactive power control, which cannot find the optimal net-
work structure and wind farm reactive power output at the same
time for system optimal operation condition.
In this paper, a joint optimization algorithm of combining reac-
tive power control of wind farm and network reconfiguration is
proposed to obtain the optimal reactive power output of wind farm
and the optimal network structure simultaneity. To find the optimal Fig. 2. Power curve for a typical 1500 kW DFIG wind turbine.
reactive power output of wind farm, an improved hybrid particle
swarm optimization with wavelet mutation (HPSOWM) algorithm
is utilized, meanwhile a binary particle swarm optimization (BPSO)
algorithm is developed to find the optimal network structure for
each particle updating instance at each iteration of wind farm reac-
tive power output optimization algorithm.

2. System model and control

2.1. DFIG wind turbine model

Fig. 1 shows the model of DFIG wind turbine consisting of a pitch
controlled wind turbine and an induction generator [19]. The sta-
tor of the DFIG is directly connected to the grid, while the rotor
is connected to a converter consisting of two back-to-back PWM
inverters, which allows direct control of the rotor currents. Direct
control of the rotor currents allows for variable-speed operation
and reactive power control thus DFIG can operate at a higher effi-
ciency over a wide range of wind speeds and help provide voltage
support for the grid. These characteristics make the DFIG ideal for
use as a wind generator.
Generally, the reference value of the active power that a DFIG
should generate is established through optimum power curves,
which provide the active power as a function of the generator rota-
tional speed [19]. Fig. 2 shows a power curve for a typical 1500 kW
DFIG wind turbine. Such curves are derived as a result of analysis
of the wind turbine aerodynamics, and by defining the maximum
mechanical power the DFIG can extract from the wind at any angu-
lar speed [8,12].
2.2. DFIG capability limits curve
Fig. 3 shows the single-phase equivalent circuit of the DFIG,
where US is the stator voltage, UR is the rotor voltage, IS is the stator
Fig. 3. DFIG equivalent circuit.
current, IR is the rotor current, RS is the stator resistance, RR is the
rotor resistance, XS is the stator reactance, XR is the rotor reactance,
XM is the mutual reactance, and s is the slip.
The doubly fed asynchronous generator converts the wind tur-
bine mechanical power into electrical power that is fed into the grid
through the stator and the rotor by means of a frequency converter
consisting of two back-to-back inverters.
The total active power of the DFIG fed into the grid is the sum
of stator and rotor active power.
PT = PS + PR (1)
Taking into account that
PR = −sPS (2)
PT = (1 − s)PS (3)
where PT is the total active power of the DFIG fed into the grid, PS
is the stator active power, and PR is the rotor active power.
In opposition to active power, total reactive power fed into the
grid is not the addition of stator and rotor reactive power because
rotor reactive power cannot flow through the frequency converter.
The grid side inverter of the frequency converter has its own reac-
tive power capability, so total reactive power fed into the grid is
the sum of the stator and the grid side inverter reactive power.
Usually, in the commercial systems, this inverter works with unity
power factor, being total reactive power, in such case, equal to
stator reactive power.

QT = QS (4)

The stator active and reactive power can be expressed as a func-

tion of stator and rotor maximum allowable current [11]:

PS2 + QS2 = (3US IS )2 (5)

 2
2  X 2
US M
PS2 + QS + 3 = 3 US IR (6)
Fig. 1. DFIG wind turbine.
XS XS
700 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706

2.3. Wind farm model

In this paper, the “generator by generator” wind farm model

is considered, which treats each of the wind farm generators sepa-
rately [8]. It consists of N identical DFIG wind turbines whose stators
are connected in parallel to an infinite bus bar supplying a US peak
voltage, being N is the number of DFIG wind turbines in the wind
farm. This model allows considering different wind conditions for
each generator. Consequently, the total active and reactive powers
output of the wind farm equal to the sum of the active and reactive
power generated by each DFIG wind turbine in the wind farm:

n
PWF = PTi (9)
i=1


n

QWF = QTi (10)

i=1

Fig. 4. 1500 kW DFIG capability limits curve.
In the PQ plane, (5) represents a circumference centered at the
origin with radius equal to the stator rated apparent power. Eq.
(6) represents a circumference centered at [−3US2 /XS , 0] and radius
equal to 3US IR XM /XS .
Introducing (3) and (4) into (5) and (6):
 P 2
T
+ QT2 = (3US IS )2
1−s
 P 2  2  X 2
T US2 M
+ QT + 3 = 3 US IR
1−s XS XS
According to (7) and (8), the DFIG capability limits can be
obtained by considering the stator and rotor maximum allowable
currents ISmax and IRmax . Fig. 4 shows the capability limits curves of
1500 kW DFIG, which is obtained by taking into account the max-
imum stator and rotor currents and the steady state stability limit
of the DFIG [11]. The electric parameters of 1500 kW DFIG are given
in Table 1. In this figure, the solid and dashed curves represent the
maximum reactive power that DFIG can generate or absorb cor-
responding to the stator and rotor maximum allowable currents
for terminal voltage US = 1.00 p.u., respectively. The vertical dot-
ted line at [−3US2 /XS , 0] coordinate represents the stability limit of
the DFIG. It means that the generator becomes unstable if reac-
tive power absorption is higher than 3US2 /XS . The shadow part is
the area where the operation of the DFIG can be considered as
feasible. One can observe that when the available active power is
close to its maximum limit, the reactive power operation range
decreases.
The active power that a DFIG should generate is established
through optimum generation curves, which provide the active
power as a function of the generator rotational speed. When the
active power that a DFIG should generate is given, the maximum
reactive power operation range can be obtained.
Table 1
DFIG electric parameters.
where PWF represents the active power output of the wind farm,
QWF represents the reactive power output of the wind farm, PTi rep-
resents the generated active power of each i DFIG and QTi represents
the generated or absorbed reactive power of each i DFIG.
When the reactive power output reference for the wind farm is
obtained, the reactive power reference for each DFIG can be calcu-
lated applying the proportional distribution algorithm:
Q
QTiset = WFeref QTi max (11)
QTi max
(7) where QTiset is the reactive power output set point calculated for
each DFIG, QTimax is the maximum reactive power that each DFIG
can generate or absorb and QWFeref is the reactive power reference
(8) for the wind farm.
2.4. Load flow including wind farm
In this paper, the node integrating wind farm is treated as PQ
nodes in a load flow analysis. In situations where the wind speed
at wind farm is specified and the loads at buses are known, the
real power output of DFIG can be calculated by means of the
power curve. The reactive power of the wind farm is obtained
from the optimization algorithm proposed in this paper. Then a
backward–forward load flow algorithm is utilized to determine the
real and reactive current injection at all the buses. Using these cur-
rents and a backward–forward sweep scheme the branch currents
are found and voltages at all the buses are updated for this iteration.
3. Problem formulation
In this section, wind farm reactive power control and network
reconfiguration joint optimization have been modeled as a multi-
objective, non-differentiable optimization problem. The objective
is to minimize the system real power losses and the deviation of the
bus voltage, subject to operating constraints. In this paper, the con-
stant load models are considered in all time periods. The objective
function is expressed as follows:

Nl
Pi2 + Qi2
min f1 (X̄) = 1 Ri + 2 max|Vi − Vrat | (12)
|Vi |2
i=1

Parameter Value X̄ = [Q̄WF , S̄w ]

RS , stator resistance per phase 0.001692 
XS , stator leakage reactance per phase 0.03692  S w = [S w1 , S w2 , . . . , S wNs ]
XM , mutual reactance 1.4568 
RR , rotor resistance per phase 0.002423  where X̄ is the state variables vector. Q WF is wind farm reac-
XR , rotor leakage reactance per phase 0.03759 
tive power output vector, S w is the switches status vector, which
 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706 701

represents the status of switches specified in terms of on/off status, 4. Particle swarm optimization
taking 0 or 1 as its value. Ri , Pi and Qi are the resistance, real power,
and reactive power of branch i, respectively. Nl is the total number 4.1. The standard particle swarm optimization (PSO)
of branches. Ns is the number of switches. Vi and Vrat are the real
and rated voltage on bus i. 1 and 2 represent weighting factors. The PSO is a population-based optimization method first pro-
1 + 2 = 1. posed by Kennedy and Eberhart [20]. The PSO algorithm is
Owing to the DFIG operational requirements, the minimization initialized with the population of individuals being randomly
of the objective function is subjected to the following constraints: placed in the search space and search for an optimal solution by
updating individual generations. At each iteration, the velocity and
(1) Distribution power flow equations: the position of each particle are updated according to its previous
best position (Pbesti ) and the best position found by informants

Nb
(Gbest). Each particle’s velocity and position are adjusted by the
Pi + PWFi = PDi + Vi Vj (Gij cos ıij + Bij sin ij ) (13) following formula:
j=1
vki (t) = ω · vki (t) + c1 · r1 (Pbestik (t − 1) − xik (t − 1))

Nb
Qi + QWFi = QDi + Vi Vj (Gij sin ıij − Bij cos ij ) (14) + c2 · r2 (Gbest k (t − 1) − xik (t − 1)) (20)
j=1

where Pi and Qi are the substation injected active and reactive xik (t) = xik (t) + vki (t) (21)
power at the ith bus. PWFi and QWFi are the wind farm injected
active and reactive power at the ith bus. PDi and QDi are the where i is the number of the particle in the swarm, k is the number
active and reactive load power at the ith bus. Vi and Vj are the of element in the particle xi (t), and t is the iteration number. vki (t)
amplitude of voltage at the ith and jth bus, respectively. Gij and and xik (t) are the velocity and the position of kth element of the ith
Bij are the conductance and the susceptance between the ith and particle at the tth iteration, respectively. r1 and r2 are the random
jth nodes. ıij and  ij are the phase angle difference between the numbers uniformly distributed between 0 and 1.The constants c1
ith and jth nodes. and c2 are the weighting factors of the stochastic acceleration terms
(2) DFIG active capacity limits: and ω is the positive inertia weight.
The suitable selection of inertia weight ω in (20) provides a bal-
PTimin ≤ PTi ≤ PTi max (15)
ance between global and local explorations [21]. The inertia weight
where PTi , PTimin and PTimax are scheduled, minimum and max- ω can be dynamically set with the following equation:
imum active power output of each i DFIG, respectively.
ωmax − ωmin
(3) DFIG reactive capacity limits: ωt+1 = ωmax − ×t (22)
tmax
 2  P 2
XM Ti US2 where tmax is the maximum number of iteration, and t is the current
QTi ≥ − 3 US IR − −3
XS 1−s XS iteration number. ωmax and ωmin are the upper and lower limits of
 2  P 2
(16) the inertia weight.
XM Ti US2
QTi ≤ 3 US IR − −3
XS 1−s XS 4.2. Binary particle swarm optimization (BPSO)
where QTi is reactive power output of each i DFIG wind turbine.
(4) Node voltage magnitude limits: The BPSO algorithm was introduced by Kennedy and Eberhart
to allow the PSO algorithm to operate in binary problem spaces
Vmin ≤ Vi ≤ Vmax (17) [22]. It uses the concept of velocity as a probability that a bit takes
on one or zero. In the BPSO, (20) for updating the velocity remains
where Vi is the voltage magnitude of node i, Vmin and Vmax are
unchanged, but (21) for updating the position is re-defined by the
low and upper bound of nodal voltage, respectively.
rule:
(5) Distribution line limits:
xik (t) = 1, r < S(vki (t − 1))
|Pijline | < Pijmax
line
(18) (23)
xik (t) = 0, r ≥ S(vki (t − 1))
where |Pijline |
and line
Pijmax
are absolute power flowing over distri-
bution lines and maximum transmission power between nodes where S(vki ) is the sigmoid function for transforming the velocity to
i and j, respectively. the probability as the following expression:
(6) Radial structure of the network. 1
In this paper, the inequality and equality constraints are S(x) = (24)
1 + e−x
included into the objective function by using penalty function
method. Therefore, the objective function in the joint optimiza- 4.3. Hybrid particle swarm optimization with wavelet mutation
tion algorithm is written as (HPSOWM)

Nu

Ne

F(x) = f (x) + k1 Uj (x) + k2 Ej (x) (19) The PSO performs well in the early iterations, but it presents
problems reaching the near optimal solution. The behavior of the
j=1 j=i
PSO presents some problems related with the velocity update. If
where f(x) is the objective function values of optimization prob- a particle’s current position coincides with the global best posi-
lem. Nu and Ne are the number of inequality and equality tion, the particle will only move away from this point if its inertia
constraints, respectively. Uj (x) and Ej (x) are the inequality and weight and previous velocity are different from zero. If their previ-
equality constraints. k1 and k2 are the penalty factors, respec- ous velocities are very close to zero, then all the particles will stop
tively. moving once they catch up with the global best particle, which may
702 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706

lead to premature convergence of the algorithm. This phenomenon

is known as stagnation.
To avoid this problem, a hybrid particle swarm optimization
with wavelet mutation (HPSOWM) is proposed, which incorporates
a genetic algorithm’s evolutionary operations of mutations [23].
In the HPSOWM, a mutation with a dynamic mutating space by
incorporating a wavelet function is proposed. The mutating space
is dynamically varying along the search based on the properties
of the wavelet function. The resulting mutation operation aids the
HPSOWM to perform more efficiently and provides faster conver-
gence. The details of the operation are as follows.
Every particle element of the swarm will have a chance to
mutate what is governed by a probability of mutation pm ∈ [0, 1],
which is defined by the user. For each particle element, a random
number between 0 and 1 will be generated such that if it is less
than or equal to pm , a mutation will take place on that element. For
instance, if xi = [xi1 , xi2 , . . . , xik ] is the selected ith particle, and the
element of particle xik is randomly selected for the mutation, the
resulting particle is given by xi = [xi1 , xi2 , . . . , xik ], i.e.

xik (t) + ı × (parakmax − xik (t)) if ı > 0
x̄ik (t) = (25)
xik (t) + ı × (xik (t) − parakmin ) if ı ≤ 0

where k ∈ 1, 2, . . ., ;  denotes the dimension of the particle and ı

is the mother wavelet.
There are many kinds of wavelets which can be used as a mother
wavelet, such as the Harr wavelet, Meyer wavelet, Coiflet wavelet,
Daubechies wavelet, Morlet wavelet and so on. These wavelets have
different specificities. In this paper, the Morlet wavelet is chosen
as the mother wavelet because the Morlet wavelet offers the best
performance. Its mathematical form is shown as follow:
1 2
  ϕ 
ı(ϕ) = √ e−(ϕ/a) /2 cos ω0 (26)
a a
where ω0 is the central frequency of wavelet.

5. Joint optimization algorithm

5.1. Joint optimization algorithm

Fig. 5. Flowchart of the proposed joint optimization algorithm.

In this paper, a joint optimization algorithm of combining wind

farm reactive power output and network reconfiguration to mini-
generation will lead to a long computing time before reaching an
mize the real power losses of the system and the deviation of the bus
optimal solution. Therefore, in a radial distribution network, when
voltage is proposed. The state variable vector is X̄ = [Q̄WF , S̄w ]. Q̄WF
a tie switch is closed, a loop is formed and a sectionalizing switch
is continuous variable, which represents the wind farm reactive
in the loop should be opened to retain the radial structure of the
power output. S̄w is discrete variable, which represents the status of
system. In this paper, the coding method that recognizes the posi-
switches. In proposed algorithm, HPSOWM is utilized to optimize
tions of the tie switches described in Ref. [24] is utilized. The total
wind farm reactive power output, and BPSO is developed to find
number of open switches is equal to the total number of loops. To
the optimal network structure for each particle updating instance
ensure that no feeder section can be left out of service, a switch not
at the each iteration of wind farm reactive power output optimiza-
included in any loop must be closed.
tion. The stopping criteria of the algorithm is the maximum number
of iteration is reached.
The flow of the joint optimization algorithm is illustrated in 5.3. Wind farm reactive power output optimization using
Fig. 5. HPSOWM

5.2. Distribution network reconfiguration using BPSO
In this paper, we use BPSO-based algorithm for distribution net-
work reconfiguration. The tie and sectionalizing switches status of
all feeders are chosen as a set of control variables. With such a
variable expression, each element of the solution vector represents
one feeder with a switch. The value 0 or 1 of one element in the
solution vector denotes that the status of corresponding switch in
the feeder is open or closed, respectively. It was found that such
a variable expression is often not efficient because the extremely
large number of unfeasible non-radial solutions appearing at each
In this paper, the HPSOWM optimization algorithm described
in the above section is used to optimize reactive power output of
wind farm. In situations where the wind speed at each DFIG wind
turbine is specified, the active power generated by each DFIG wind
turbine can be calculated by means of the power curve. According
to (7) and (8), the maximum reactive power that each DFIG can
generate or absorb can be obtained. The total active power output
and maximum reactive power output of the wind farm are obtained
by (9) and (10).
The steps followed for the implementation of the algorithm are
described as follows:
 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706 703

Step 1: The reactive power output of the wind farm is used as the Table 2
DFIG performance parameters.
control variable. Initialize a population of particles with
random position and velocities within the reactive power Parameter Value
capability limits. Rated capacity 1500 kW
Step 2: Evaluate the objective function values of all particles Cut-in wind speed 4 m/s
according to (19) using the result of distribution load flow. Cut-out wind speed 25 m/s
Step 3: Set Pbest of each particle and its objective value equal to its Rated wind speed 12 m/s
Rated voltage 0.69 kV
current position and objective value, and set Gbest and its
objective value equal to the position and objective value
of the best particle.
Step 4: Select the ith particle. hourly wind speed curves of each wind turbines, obtained from a
Step 5: Update the velocity and position of the ith particle accord- wind speed forecasting for consecutive 24 h of a day.
ing to (20) and (21). The actual active power outputs of each DFIG in each period are
Step 6: Perform wavelet mutation on the selected element of par- shown in Fig. 8, calculated by means of the power curve of each
ticle according to (25) to create new particle. DFIG.
Step 7: Evaluate the objective function value of the new created Considering the DFIG capability limits described in Section 2,
particle, and compare its current objective value with the the maximum reactive power outputs limits of each DFIG wind
objective value of its Pbest. If current value is better, then turbines are shown Fig. 9. In Fig. 9, the undermost curve represents
update Pbest and its objective value with the current posi- the stability limit of the DFIG.
tion and objective value. From Figs. 8 and 9, it can be observed that wind farm made
Step 8: If all individuals are selected, go to the next step, otherwise up of DFIG wind turbine can generate high quantities of reactive
i = i + 1 and returns to Step 4. power when the available active power is far from its maxi-
Step 9: Determine the best particle of current whole population mum. For example, when the average wind speed at wind farm
with the best objective value. If the objective value is bet- is 4.5 m/s, the maximum active and reactive power wind farm can
ter than the objective value of Gbest, then update Gbest and generate are 0.3563 MW and 3.6657 MVAR, respectively. But the
its objective value with the position and objective value maximum reactive power wind farm can generate become very
of the current best particle. low when the available active power is near to its rated power. For
Step 10: If the maximum number of iteration is reached, the search
procedure is stopped, otherwise returns to Step 4.

6. Simulation results

In this paper, the 16 bus distribution network given in Ref. [13]

is used to verify the validity and performance of the proposed joint
optimization algorithm (see Fig. 6). Test system has 13 section-
alizing branches and 3 tie branches, S15, S21 and S26 are three
tie switches. The tie switches and sectionalizing switches are nor-
mally open and closed, respectively. A small wind farm comprising
4 DFIG wind turbines of 1500 kW, with a power installed of 6 MW
is connected at node 12 through a rated 23/0.69 kV transformer.
The performance parameters of 1500 kW DFIG are given in Table 2.
The system loads are 28.7 MW and 17.3 MVAR. Voltage limits are
assumed to be within the range 0.95–1.05 p.u.

6.1. Available active and reactive power in wind farm

Fig. 7. Curve of wind speed.
In this paper, each of wind turbines in wind farm was considered
to have different instantaneous wind speed. Fig. 7 shows the mean

Fig. 6. 16 Node test feeder. Fig. 8. Active power of each DFIG wind turbine.
704 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706

Table 3
Optimization results of different cases.

Case PWF (MW) QWF (MVAR) Tie switch set Objective function value Losses (MW) Minimum nodal voltage (p.u.)

Original system 3.34 0.678 15, 21, 26 0.1860 0.338 0.979

1 3.34 0.678 17, 19, 26 0.1676 0.310 0.983
2 3.34 2.73 15, 21, 26 0.1699 0.347 0.984
3 3.34 2.14 17, 19, 26 0.1555 0.315 0.988
4 3.34 2.73 17, 19, 26 0.1569 0.320 0.988
5 3.34 2.18 17, 19, 26 0.1554 0.315 0.988

example, in the period 10, the average wind speed is 14 m/s, the
active power that wind farm generated is rated power 5.9813 MW,
and the maximum reactive power DFIG can generate reduce to
1.8220 MVAR.

6.2. Joint optimization

In this paper, the parameters of the joint optimization algorithm

are as follows: the number of particles is set as 20 and the maximum
iteration number of the algorithm is set as 100. c1 = c2 = 2, ωmax = 0.9
and ωmin = 0.4.The upper and lower bounds of mutation probability
are set as 0.05 and 0.3, respectively. The Morlet wavelet is chosen
as the mutation wavelet parameter. It is assumed that the network
reconfiguration and reactive power control operate once per hour
and the load is the constant during all time periods.
Choosing the data during period 3 from Figs. 7–9 as the exam-
ple, in this period, the wind speed is 8.1 m/s, the total active power
wind farm generated is 3.3375 MW, and the maximum available
Fig. 10. Convergence performance of PSO for the best solution.
reactive power in wind farm is 2.7253 MVAR. To demonstrate the
performance of the proposed joint optimization algorithm, the
following five cases are studied. Table 3 provides the simulation Table 4
results of these five cases, using MATLAB software carried out on a Maximum active power available in each wind turbine (MW).
P4 1.6 GHz/1 GB RAM computer system. Period Turbine

1 2 3 4 Total
Case 1: Only perform network reconfiguration, power factor of the
DFIG is 0.98. 1 0.1688 0.0938 0.0562 0.0375 0.3563
2 0.5250 0.4688 0.5625 0.4312 1.9875
Case 2: Only perform wind farm reactive power optimization.
8 1.2188 1.2000 1.1438 1.1438 4.7064
Case 3: Perform network reconfiguration first, and then optimize 12 1.5000 1.5000 1.5000 1.4813 5.9813
wind farm reactive power.
Case 4: Perform wind farm reactive power optimization first, and
then carry out network reconfiguration. The computation time used in the proposed joint optimization algo-
Case 5: Perform the proposed joint optimization algorithm. rithm is 232 s. It means that the proposed joint optimization has
proven to be an effective algorithm to find the optimal network
As shown in Table 3, the objective function value in Case 5 is structure and wind farm reactive power output to minimize losses
lowest in the five cases. Compared with original system, real power and improve voltage profiles. The optimization results of Cases 3
losses are reduced about 7%, from 0.338 MW to 0.315 MW, and the and 4 are better than those of Cases 1 and 2. It demonstrates that
minimum node voltage is improved from 0.979 p.u. to 0.988 p.u. performing wind farm reactive power control followed by network
reconfiguration, or vice verse, can do a better job than only perform-
ing wind farm reactive power control or network reconfiguration
alone.
Fig. 10 illustrates the convergence performance of the joint
optimization algorithm for the best solutions. It can be evidently
seen from this figure that the algorithm converged to a good
solution well before the maximum iterations number 100 was
reached.

Table 5
Maximum reactive power available in each wind turbine (MVAR).

Period Turbine

1 2 3 4 Total

1 0.9063 0.9174 0.9205 0.9215 3.6657

2 0.7962 0.8144 0.7835 0.8156 3.2097
8 0.4624 0.4788 0.5250 0.5250 1.9912
Fig. 9. Maximum reactive power limit of each DFIG wind turbine. 12 0.4555 0.4555 0.4555 0.4555 1.8220
 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706 705

Table 6
Results of joint optimization of four chosen periods of time.

Period PWF (MW) QWFmax (MVAR) QWF (MVAR) Tie switch set Objective function value Losses (MW) Minimum nodal voltage (p.u.)

1 0.356 3.66 3.66 17, 19, 26 0.2119 0.465 0.984

2 1.99 3.21 3.21 17, 19, 26 0.1766 0.379 0.988
8 4.71 1.99 1.16 17, 19, 26 0.1409 0.267 0.989
12 5.98 1.82 1.07 17, 19, 26 0.1315 0.236 0.992

Table 7 [2] N. Dizdarevic, M. Majstrovic, S. Zutobradic, Power quality in a distribution net-

Reactive power output of each DFIG (MVAR).
Period Turbine
1 2 3 4
1 0.9049 0.9160 0.9191 0.9201
2 0.7938 0.8119 0.7811 0.8131
8 0.2694 0.2789 0.3058 0.3058
12 0.2675 0.2675 0.2675 0.2675
Tables 4 and 5 show the maximum available active and reactive
power values of each DFIG during four different periods of time,
chosen from Figs. 7 and 8. The first column shows the number of
the time period, and the other columns show the power available at
each DFIG in each period. According to joint optimization algorithm
proposed in this paper, the joint optimization results of four chosen
periods of time are shown in Table 6.
The optimization results show that when wind farm active
power output are 0.356 MW and 1.99 MW, the optimal reactive
power output results equal to the 3.66 MVAR and 3.21 MVAR, which
are the maximum reactive power that wind farm can generate. This
shows that the wind farm needs to generate maximum reactive
power to improve the voltage profile when wind farm active power
output is small. When wind farm active power output increases to
4.71 MW and 5.98 MW, the wind farm optimal reactive power out-
put becomes very low, the results are 1.16 MVAR and 1.07 MVAR.
From Table 6, it can be found that the real power losses are reduced
and the minimum nodal voltage is improved compared with orig-
inal system after joint optimization. After the wind farm optimal
reactive power output is obtained, the reactive power reference
for each DFIG is calculated applying the proportional distribution
algorithm described in (11). The reactive power distribution results
are shown in Table 7.
7. Conclusions
In this paper, a joint optimization algorithm of combining reac-
tive power control of wind farm and network reconfiguration is
proposed. In the proposed joint optimization algorithm, reactive
power output of wind farm and status of switches are utilized
as the control variable for losses minimization and voltage pro-
file improvement. The optimal reactive power output of wind farm
and the optimal network structure are efficiently obtained by taken
into account DFIG reactive capability limits in the simulation. From
the results obtained in the simulations, it can be concluded that
wind farm made up of DFIG can constitute an important continu-
ous reactive power source to support system voltage control. The
simulation results also show that the joint optimization gets better
solution results than using the wind farm reactive power control
optimization algorithm or the network reconfiguration algorithm
alone.
work after wind power plant connection, IEEE Power Systems Conference and
Exposition, NY, USA, 2004, pp. 913–918.
[3] J.M. Rodriguez, J.L. Fernandez, D. Beato, et al., Incidence on power system
dynamics of high penetration of fixed speed and doubly fed wind energy sys-
Total
tems: study of the Spanish case, IEEE Transactions on Power Systems 17 (4)
3.66 (2002) 1089–1095.
3.21 [4] H. Lund, Large-scale integration of wind power into different energy systems,
1.16 Energy 30 (13) (2005) 2402–2412.
1.07 [5] A. Tapia, G. Tapia, J.X. Ostolaza, J.R. Saenz, R. Criado, J.L. Berasategui, Reactive
power control of a wind farm made up with doubly fed induction genera-
tors (I), in: IEEE Porto Power Tech Conference, Porto, Portugal, September,
2001.
[6] D.J. Atkinson, R.A. Lakin, R. Jones, A vector-controlled doubly-fed induction
generator for a variable-speed wind turbine application, Transactions of the
Institute of Measurement & Control 19 (1) (1997) 2–12.
[7] R.S. Peña, J.C. Clare, G.M. Asher, Vector control of a variable speed doubly-
fed induction machine for wind generation systems, EPE Journal 6 (3) (1996)
60–67.
[8] G. Tapia, A. Tapia, J.X. Ostolaza, Two alternative modeling approaches for the
evaluation of wind farm active and reactive power performances, IEEE Trans-
actions on Energy Conversion 21 (4) (2006) 909–920.
[9] R.G. de Almeida, E.D. Castronuovo, J.A.P. Lopes, Optimum generation control in
wind farms when carrying out system operator requests, IEEE Transactions on
Power Systems 21 (2) (2006) 718–726.
[10] A. Tapia, G. Tapia, J.X. Ostolaza, Reactive power control of wind farms for voltage
control applications, Renewable Energy 29 (3) (2004) 377–392.
[11] D. Santos-Martin, S. Arnaltes, J.L.R. Amenedo, Reactive power capability of dou-
bly fed asynchronous generators, Electric Power Systems Research 78 (11)
(2008) 1837–1840.
[12] G. Tapia, A. Tapia, J.X. Ostolaza, Proportional–integral regulator-based approach
to wind farm reactive power management for secondary voltage control, IEEE
Transactions on Energy Conversion 22 (2) (2007) 448–459.
[13] S. Civanlar, J.J. Grainger, H. Yin, S.S.H. Lee, Distribution feeder reconfiguration
for loss reduction, IEEE Transactions on Power Delivery 3 (3) (1988) 1217–
1223.
[14] D. Shirmohammadi, H.W. Hong, Reconfiguration of electric distribution net-
works for resistive line losses reduction, IEEE Transactions on Power Delivery
4 (2) (1989) 1492–1498.
[15] V. Gomes, s. Carneiro, A new reconfiguration algorithm for large distribu-
tion systems, IEEE Transactions on Power Delivery 20 (3) (2005) 1373–
1378.
[16] H.D. Chiang, R.M. Jean-Jumeau, Optimal network reconfiguration distribution
system. Part 1: a new formulation and a solution methodology, IEEE Transac-
tions on Power Delivery 5 (4) (1990) 1902–1909.
[17] Y.J. Jean, J.C. Kim, An efficient simulated annealing algorithm for network
reconfiguration in large-scale distribution systems, IEEE Transactions on Power
Delivery 17 (4) (2002) 1070–1078.
[18] K. Nara, A. Shiose, M. Kitagawoa, T. Ishihara, Implementation of genetic
algorithm for distribution systems loss minimum reconfiguration, IEEE Trans-
actions on Power Systems 7 (3) (1992) 1044–1051.
[19] L.M. Fernández, C.A. García, J.R. Saenz, et al., Aggregated dynamic model for
wind farms with doubly fed induction generator wind turbines, Renewable
Energy 33 (1) (2008) 129–140.
[20] J. Kennedy, R. Eberhart, Particle swarm optimization, IEEE Conf. Neural Net-
works, Perth, Australia, November/December, 1995, pp. 1942–1948.
[21] Y. Shi, R. Eberhart, A modified particle swarm optimizer, IEEE Proc. Computa-
tional Intelligence, Anchorage, AK, USA, May, 1998, pp. 69–73.
[22] J. Kennedy, R. Eberhart, A discrete binary version of the particle swarm algo-
rithm, Proc. Conf. Systems, Man, and Cybernetics, Piscataway, NJ, 1997, pp.
4104–4108.
[23] S.H. Ling, H.H.C. Iu, K.Y. Chan, et al., Hybrid particle swarm optimization with
wavelet mutation and its industrial applications, IEEE Transactions on Systems,
Man and Cybernatics: Part B 38 (3) (2008) 743–764.
[24] J.P. Chiou, C.F. Chang, C.T. Su, Variable scaling hybrid differential evolution for
solving network reconfiguration of distribution systems, IEEE Transactions on
Power Systems 20 (2) (2005) 668–675.

References
[1] R. Piwko, N. Miller, J. Sanchez-Gasca, Integrating large wind farms into weak
power grids with long transmission lines, IEEE/PES Transmission and Distribu-
tion Conference & Exhibition: Asia and Pacific, Dalian, China, 2005, pp. 1–7.
Jingjing Zhao received the M.Sc. degree in software engineering from Chongqing
University, Chongqing, China, in 2004. She is currently pursuing the Ph.D. degree in
the College of Electrical Engineering of Chongqing University. Her current research
interests include soft computing methodologies in power system applications, wind
power generation, distribution network optimal operation with grid-connected dis-
tributed generation.
706 J. Zhao et al. / Electric Power Systems Research 80 (2010) 698–706

Xin Li is a professor at the College of Electrical Engineering of Chongqing Univer-
sity, China. Her research interests include power system operation, power system
dynamics and control.
Jutao Hao was born in 1976. He received the Ph.D. degree in computer science from
Shanghai Jiao Tong University, Shanghai, China, in 2008. He is currently a lecture of
computer science engineering at University of Shanghai for Science and Technol-
ogy. His research interests include soft computing methodologies applied to power
system analysis and planning.
Jiping Lu is a professor at the College of Electrical Engineering of Chongqing Uni-
versity, China. His research interests include power system automation, relay
protection and probability application in power systems.