Sunteți pe pagina 1din 8

IET Generation, Transmission & Distribution

Research Article

Optimal reactive power and voltage control in


distribution networks with distributed
generators by fuzzy adaptive hybrid particle
swarm optimisation method

ISSN 1751-8687
Received on 25th March 2014
Accepted on 24th December 2014
doi: 10.1049/iet-gtd.2014.1059
www.ietdl.org

Shuheng Chen 1 , Weihao Hu 2, Chi Su 2, Xiaoxu Zhang 2, Zhe Chen 2


1

School of Energy Science and Engineering, University of Electronic Science and Technology of China, Chengdu, CN 611731,
Peoples Republic of China
2
Department of Energy Technology, Aalborg University, Aalborg, DK 9220, Denmark
E-mail: chenshuheng@uestc.edu.cn

Abstract: A new and efficient methodology for optimal reactive power and voltage control of distribution networks with
distributed generators based on fuzzy adaptive hybrid particle swarm optimisation (PSO) is proposed. The objective is
to minimise comprehensive cost, consisting of power loss and operation cost of transformers and capacitors, and
subject to constraints such as minimum and maximum reactive power limits of distributed generators, maximum
deviation of bus voltages and maximum allowable daily switching operation number. PSO is used to solve the
corresponding mixed integer non-linear programming problem and the hybrid PSO (HPSO) method, consisting of three
PSO variants, is presented. In order to mitigate the local convergence problem, fuzzy adaptive inference is used to
improve the searching process and the final fuzzy adaptive inference-based HPSO is proposed. The proposed algorithm
is implemented in VC++ 6.0 program language and the corresponding numerical experiments are finished on the modified
version of the IEEE 33-node distribution system with two newly installed distributed generators and eight newly installed
capacitors banks. The numerical results prove that the proposed method can search a more promising control schedule of
all transformers, all capacitors and all distributed generators with less time consumption, compared with other listed
artificial intelligent methods.

Introduction

Reactive power and voltage control of distribution networks plays a


very important role on power loss reduction and voltage quality
enhancement [1, 2]. In conventional radial distribution networks,
main measures for reactive power and voltage control include
regulating load transformer changers (LTCs) and switching
capacitor banks. Now, many distributed generators are connected
into distribution networks. Distributed generators can also provide
controllable reactive powers into distribution networks as ancillary
support for voltage control while they provide active powers into
distribution networks. By utilising LTCs, capacitor banks and
reactive powers of distributed generators harmoniously, electric
power companies can gain more prots while providing better
electric power quality [36]. Given uncertainties of loads and
distributed generators, all the measures, including LTCs, capacitor
banks and reactive powers of distributed generators, need to be
dispatched dynamically. With the knowledge of the forecast
segmented-time loads on feeders and the forecast segmented-time
active powers of distributed generators, the optimal control scheme
is a long schedule for all schedulable devices in each time interval.
As a result, the corresponding mathematical formulation is a
complex mixed integer non-linear programming (MINLP)
problem. Moreover, as the number of LTCs, capacitor banks and
distributed generators increases, it becomes more difcult to solve
this problem.
In the past 20 years, scientists have made great efforts to study the
dynamic reactive power and voltage control of distribution networks.
Deng et al. [7] propose a heuristic and algorithmic combined
approach to search the optimal schedule. Their approach simplies
the mathematical model, solves the temporal optimisation of each
control device by heuristic rules and then converts the optimisation
model with time-varying loads into the conventional optimisation

1096

model. Liu et al. [8] propose a new method based on interior point
methods. As they treat discrete variables as continuous variables
during the solution process, the MINLP problem is transformed
into a non-linear programming (NLP) problem that can be more
efciently solved by exploiting its highly sparse matrix. It is worth
pointing out that both [7, 8] consider the maximum allowable
daily switching operation number (MADSON) constraints.
However, distributed generators are not considered in their
research and reactive powers of distributed generators are not used
to enhance the operation of distribution systems together with
LTCs and capacitor banks. Madureira and Peas Lopes [9] propose
a new methodology to provide coordinated voltage support for
distribution networks with large integration of distributed
generations (DGs) and microgrids. Tanaka et al. [10] and
Calderaro et al. [11] use reactive powers of distributed generators
to regulate the voltage of distribution networks. Niknam et al. [12]
propose a stochastic multi-objective framework for daily volt/var
control, including hydroturbine, fuel cell, wind turbine and
photovoltaic power plants. Moreover, an evolutionary algorithm
using the modied teaching-learning-algorithm is proposed to
solve the formulated mixed-integer non-linear programming
problem. Although these researchers are nished on distribution
networks with multiple distributed generators, reactive powers of
distributed generators are not utilised together with LTCs and
capacitor banks. Especially, the dispatch schedule is not suitable to
the scenario of the forecast segmented-time loads.
On the other hand, a lot of methods, including deterministic
methods and articial intelligent methods, are used to solve the
corresponding MINLP problems. Yang et al. [13] apply reactive
Tabu searching (RTS) method to solve this problem. In their
scheme, the appropriate size of the Tabu list is learned
automatically through a reactive mechanism, thus the complex
parameter-tuning problem for efcient search is solved. Fang and

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

Guo [14] propose a particle swarm optimisation (PSO)-based


approach to nd a proper combination of online LTCs and
capacitor bank switches in a day. Abu-Mouti and El-Hawary [15]
present a new optimisation approach that employs an articial bee
colony (ABC) algorithm to determine the optimal DG-units size,
power factor and location in order to minimise the real power loss
of total system. In [16], simulated annealing method is used to
minimise power loss by using an automatic switching operation in
large scale distribution systems. In [17], evolutionary algorithm is
used to solve the complex optimisation problem where LTCs,
capacitor banks on feeder lines, capacitor banks on low-voltage
bus in substation are used to enhance the operation of distribution
networks. Among these researches, the factors of affecting the
nal optimal schedule have not been considered completely, so the
feasibility of these methods in distribution networks needs further
verication.
In the past decade, PSO methods have been applied in many elds
in a large scale and many variants have been proposed [1822]. Wu
and Tsai [18] propose an effective approach based on the PSO with
integer coded to determine the switch operation schemes for feeder
reconguration. In [19], a modied PSO is used to support the
minimisation of the operation costs of a virtual power player that
manages the resources in a distribution network and the network
itself. In [20], PSO methods are used to optimally determine the
size and location of multiple DG units in the distribution system
with non-unity power factor considering different load models. In
[21, 22], fuzzy adaptive inferences have been used to enhance
current PSO methods. In addition, other technologies including
chaos sequence and dynamic mutation are also used to enhance
PSO methods [23, 24].
In this paper, the optimal reactive power and voltage control
problem of distribution networks with multiple distributed
generators is investigated with the consideration of the
uncertainties of both loads and distributed generators. The target is
to minimise the comprehensive cost because it is more important
than the separate power loss from the perspective of electric power
companies. The reactive powers of distributed generators are used
to improve the operation of distribution networks together with
LTCs and capacitor banks. A hybrid PSO (HPSO) method
consisting of three PSO variants is presented to search the optimal
solution which is a dispatch schedule of LTCs, capacitor banks
and reactive powers of distributed generators. The fuzzy adaptive
inference is used in order to avoid being trapped in the local
optimum.
The rest of this paper is structured as follows: Section 2 describes
the MINLP mathematical formulation of optimal reactive power and
voltage control of distributed networks. The HPSO method is
presented in Section 3. In Section 4, fuzzy adaptive inference is
used to improve the HPSO method. Numerical experiments are
nished in Section 5. Finally, the conclusions are summarised in
Section 6.

Mathematical formulation

where Ploss is active power loss of distribution network


(kilowatt-hour), KL is the electricity price per kilowatt-hour; AT is
the total operation number of all LTCs, KT is the corresponding
operation cost of moving a LTC one step; AC is the total operation
number of all capacitor bank switches and KC is the average
operation cost of switching a capacitor bank once.
Based on segment number of forecast load curves, the dispatch
schedule V of all LTCs, capacitor banks and reactive powers of
distributed generators is expected to be calculated. Each curve
segment has its own control vector. Here, the number of load
curve segments is supposed to be NL. Then, V is presented as


V = v1 , v2 , . . . , vt , . . . , vNL

(2)

where vt = [st, ct, qt]. st is the position set of all LTCs corresponding
to time segment t, ct is the status set of all capacitor banks
corresponding to time segment t and qt is the reactive power set of
all distributed generators corresponding to time segment t. The
number of all transformers is supposed to be NP and the number
of all capacitor banks is supposed to be NC and the number of all
distributed generators is supposed to be NG. Then
st = [s1t , s2t , . . . , sNP t ]
ct = [c1t , c2t , . . . , cNC t ]
qt = [q1t , q2t , . . . , qNG t ]
The whole control schedule is further presented as


V = s1 , c1 , q1 , s2 , c2 , q2 , . . . , sNL , cNL , qNL

(3)

where the total dimension number of V is NL(NP + NC + NG). At the


same time, PLoss can be presented as
PLoss = F(V ) =

NL


f (vt )

(4)

t=1

At load curve segment t, all branch currents and all node voltages are
calculated by the back/forward sweep power ow method. Then,
active power loss is presented as
f (vt ) =

Nb


(Iit2 Ri ) DTt

(5)

i=1

where Nb is the branch number of the studied distribution networks,


Tt is the length of load curve segment t, Ri is the resistance of
branch i and Iit is the current of branch i corresponding to load
curve segment t.
According to the formulation of V presented in (3), transformer
operation number AT, is presented as
NP


NT 
NL 




snt sn(t1) 

Normally, reducing power loss of distribution networks is chosen to


be the optimisation objective. However, electric power companies
are now paying more attention to reducing the comprehensive cost,
which consists of both power loss and operation cost of primary
devices. Here, the harmonious dispatches of LTCs in the
substations, capacitor banks on feeder lines and reactive powers of
distributed generators are used to optimise distribution networks.
By regulating LTCs, switching capacitor banks and controlling
reactive powers of distributed generators harmoniously,
comprehensive loss is expected to be reduced to the potential
minimum value. Comprehensive cost comprises of power loss cost
and operation cost of LTCs and capacitor bank switches. It is
presented as

where snt is the position of LTC n corresponding to time segment t.


If t = 1, sn, 0 is the position of LTC n corresponding to the last load
curve segment of the previous day. ATn is the total operation number
of LTCs n in the concerned day.
Operation number of capacitor bank switches AC is presented as

F = KL PLoss + KT AT + KC AC

where cmt is the status of capacitor bank switch m corresponding to


load curve segment t. When capacitor bank switch m is on, cmt

(1)

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

AT =

ATn =

n=1

AC =

NC

m=1

ACm

(6)

n=1 t=1

N

NC

L 




=
cmt cm(t1) 
m=1

(7)

t=1

1097

equals to 1; otherwise, when capacitor bank switch m is off, cmt


equals to 0. cm0 is the status of capacitor bank switch m
corresponding to the last load curve segment of the previous day.
At the same time, the objective function presented in (1) is subject
to the following constraints.
For each ATn
ATn ATnMax ,

n [ (1, NT )

(8)

where ATnMax is the maximum allowable daily moving step number


of LTCs n.
For each snt
1

snt
n[(1,NT )

snMax

m [ (1  NC )

qgt
g[(1,NG )

, qgMax

(10)

(11)

where qgMin and qgMax are the minimum reactive power output and
the maximum reactive power output of distributed generator g,
respectively.
For each node
UMin ,

Uti
i[(1N b ),t[(1N L )

, UMax

Basic solution technique

As far as the aforementioned mathematical formulation of reactive


power and voltage control optimisation, which comprises formulas
(1)(12), is concerned, the dimension number of Vs is NL(NP +
NC + NG). As the dimension number increases, it is the curse of
dimensionality. As a result, the conventional enumeration
technique is usually invalid for this problem. In recent years, many
articial intelligent methods are applied to solve this MINLP
problem, such as teaching-learning-algorithm [12], reactive Tabu
search method [13], ABC algorithm [15], simulated annealing
method [16], evolutionary method [17] and PSO method [14, 18
22]. Among them, PSO method has many good performances,
including good robustness and fast convergence. In this paper,
PSO is improved to solve the MINLP problem.
PSO algorithm is an evolutionary method which considers the
behaviour of bird ocking or sh schooling. Its primary idea is
that the potential solution of the studied problem is presented as a
particle and each particle has a tness index, which is calculated
according to the objective function. In the evolutionary process,
each particle has its own speed to regulate its ying direction.
Until now, quite a few variants of PSO algorithm have been
proposed in order to satisfy different requirements, including
continuous PSO, binary PSO, discrete PSO and so on [1822].
As far as the reactive power and voltage control optimisation in
this paper is concerned, positions of LTCs are discrete integer
variables and ON/OFF statuses of capacitor bank switches are
binary variables and reactive power outputs of distributed
generators are continuous variables. Hence, a HPSO algorithm is
presented to solve this problem. Normally, at one bus of

1098

k
k+1
yk+1
id = xid + vid

1,
yk+1
id , 1
k+1
= yid , 1 yk+1
id NiT

NiT ,
yk+1
id . NiT

(14)
(15)

where the function round() is to calculate an integer based on


rounding. rT1 and rT2 are two random functions, which vary in the
interval (01) randomly. cT1 and cT2 are xed learning factors. wT
k
is the inertia coefcient. pkid and gid
are the values of dimension
variable i of the current personal optimal particle and the global
optimal particle, respectively. NiT is the highest position of the
LTC corresponding to dimension variable i.
(2) When current dimension variable is corresponding to the
ON/OFF status of a capacitor bank switch, it is a binary variable.
It varies among 0 or 1. The corresponding PSO algorithm is
presented as
k
k
k
k
k
vk+1
id = wC vid + rC1 cC1 (pid xid ) + rC2 cC2 (gid xid )
k+1
S(vk+1
id ) = 1/(1 + exp ( vid ))
k+1
xid = 1, rand () , S(vk+1
id )
others
xk+1
id = 0,

(12)

where UMin is the lower voltage limit and UMax is the upper voltage
limit.
In addition, the studied distribution network is required to be
radial and branch currents are required to be less than heat limits.

k
k
k
k
k
vk+1
id = round (wT vid + rT1 cT1 (pid xid ) + rT2 cT2 (gid xid )) (13)

xk+1
id

where ACmMax is the MADSON of capacitor bank switch m.


For each qgt
qgMin ,

(1) When current dimension variable is corresponding to the position


of a LTC, it is an integer. It varies from 1 to the highest position of
the corresponding LTC. The corresponding PSO algorithm is
presented as

(9)

where snMax is the highest position of LTC n.


For each ACm
ACm ACmMax ,

distribution networks, two or more capacitor banks are installed


and they are expected to be used averagely in turn. Given these
requirements, the HPSO algorithm is described as follows:

(16)
(17)
(18)

where rC1 , rC2 , cC1 , cC2 and wC are the same as the corresponding
parameters in (13)(15). Function S() is a sigmoid function. It
into a probabilistic value in the random interval
translates vk+1
id
(0 1).
(3) When current dimension variable is corresponding to the reactive
power output of a distributed generator, it is a continuous variable. It
varies in the interval (qiMin qiMax). qiMin and qiMax are the lower
limit and upper limit of reactive power of the distributed generator
corresponding to dimension variable i, respectively. The
corresponding PSO algorithm is presented as
k
k
k
k
k
vk+1
id = wG vid + rG1 cG1 (pid xid ) + rG2 cG2 (gid xid )
k
k+1
zk+1
id = xid + vid

xk+1
id

qiMin ,
= zk+1
,
id
qiMax ,

zk+1
id , qiMin
qiMin zk+1
id qiMax
zk+1
id . qiMax

(19)
(20)
(21)

where rG1 , rG2 , cG1 , cG2 and wG are the same as the corresponding
parameters in (13)(15).
In addition, during each evolution cycle, the data type of each
dimension variable decides the corresponding PSO variant.

4 Further enhancement of HPSO using fuzzy


adaptive inference
The proposed HPSO algorithm in Section 3, comprising formulae
(13)(21), usually suffers from the problem of being trapped in
local optima and its consumed time is very large. In this section,
fuzzy adaptive inference is used to balance the global and local

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

Fig. 1
system

Membership of the normalised comprehensive cost in the fuzzy

Fig. 3 Membership of the inertia weight deviation in the fuzzy system


Table 1 Fuzzy rules for the deviation of the inertia weight
Rule no.

1
2
3
4
5
6
7
8
9

Fig. 2 Membership of the inertia weight in the fuzzy system

Antecedent

Consequent

FNorm

S
S
S
M
M
M
L
L
L

S
M
L
S
M
L
S
M
L

ZE
NE
NE
PE
ZE
NE
PE
ZE
NE

searching ability. In order to gain better performance of the PSO


search process, the inertia weight is expected to be non-linearly
and dynamically changed to achieve better balance between global
and local abilities. According to [21, 22], fuzzy adaptive system
has non-linear and dynamic performance, which is the good
candidate used to tune the inertia weight of PSO in order to avoid
being trapped in local optima.
For a fuzzy system, there are three important steps, namely,
fuzzication, fuzzy logic inference and defuzzication. First, it
maps sets of input variables into fuzzy sets using membership
functions; secondly, according to the predened logic, the output
is assigned based on these fuzzy input sets. In this paper, the input
variables are current normalised total comprehensive cost and
current inertia weight. The output variable is the deviation of the
inertia weight and the result of fuzzy logic calculation is some
fuzzy values and their membership values. Finally, according to its
membership, the true output is calculated. Here, the normalised
total comprehensive cost is used as an input variable in the
interval (0 1) and it is dened as
FNorm =

FNow FMin
FMax FMin

(22)

where FNow is the current total comprehensive cost in each iteration


calculated by (1). FMax is a large value which is greater than or equal
to any feasible comprehensive cost and FMin is a small value which is
less than or equal to any feasible comprehensive cost. The inertia
weight value is limited in the interval (0.30.8) and the deviation
of inertia weight is limited in the interval (0.10.1). The input
fuzzy variables are dened in three fuzzy sets of linguistic values
S (small), M (medium) and L (large) with the associated
membership functions, as shown in Figs. 1 and 2, respectively.
The output variable is presented in three fuzzy sets of linguistic
values NE (negative), ZE (zero) and PE (positive) with associated
membership functions, as shown in Fig. 3.
FNorm, w and w are the true values of the normalised total
comprehensive cost, the inertia weight and the deviation of the
inertia weight, respectively. The memberships of normalised total
comprehensive cost and inertia weight are calculated from their
membership functions, respectively. Input linguistic variables are
connected using the AND operator in the fuzzy rules. Simple
IF/THEN rules are shown in Table 1.
In Table 1, there are nine possible rules for the two input variables
since each input variable has three linguistic values. The Larsen
product has been used as the fuzzy implication operator for the
individual rules. Firing of each rule separately results in clipped
output fuzzy sets, one for each rule. The overall fuzzy output is

Fig. 4 Flowchart of FAHPSO algorithm

being obtained by combining all the clipped fuzzy sets. The


aggregated fuzzy output is converted into a single value using
centroid defuzzication method.

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103

& The Institution of Engineering and Technology 2015

1099

Once the deviation of the inertia weight is calculated based on the


fuzzy system, the inertia weight of the next iteration can be updated
as follows
wk+1 = wk + Dwk

(23)

Moreover, the fuzzy adaptive inference procedure is integrated to the


basic HPSO algorithm in Section 3 and the improved HPSO
algorithm (fuzzy adaptive HPSO FAHPSO) is shown in Fig. 4.
In Fig. 4, fuzzy adaptive inference process is integrated in the
Producing new particles module. By this module, all the particles
nish their evolutions. In the dashed rectangle, the tness indexes
of all particles are calculated and compared with current personal
optimal indices and current global optimal index, respectively. By
comparisons, all the personal optimal particles and the global
optimal particle are updated.

5
5.1

Numerical experiments
Test case network

In order to test the performance of the proposed algorithm, VC++ 6.0


program language is used to develop the corresponding software
system. By the developed software system, the aforementioned
improved PSO method is implemented. The developed software
system runs on a PC computer with Windows 7 operation system.
Here, a modied version of the IEEE 33-node system, which is
shown in Fig. 5, is used to test the proposed algorithm.

Fig. 5 Modied version of IEEE 33-node system

Based on the original IEEE 33-node distribution network, two


new branches are added to this system. One is the branch from
node 32 to node 33 and the other one is the branch from node 16
to node 34. Then, a distributed generator is connected to node 33
and another one is connected to node 34. Their rated active
powers are assumed to be 500 kW and their reactive power
outputs are assumed to be in the range (450450 kVar). At the
same time, four capacitor banks are installed at node 26 and
another four capacitor banks are installed at node 7. The capacity
of each capacitor bank is 0.126 Mvar.
In order to emulate load uncertainties and limit the complexity of
the studied problem, all the loads are divided into three groups.
Group one includes Load 0, Load 3, Load 6, Load 9, Load 12,
Load 15, Load 18, Load 21, Load 24, Load 27 and Load 30.
Group two includes Load 1, Load 4, Load 7, Load 10, Load 13,
Load 16, Load 19, Load 22, Load 25, Load 28 and Load 31.
Group three includes Load 2, Load 5, Load 8, Load 11, Load 14,
Load 17, Load 20, Load 23, Load 26 and Load 29. Each group
has its own load curve and all the loads in the same group have
the same normalised load curve. Three normalised load curves are
shown in Fig. 6, respectively.
At the same time, in order to emulate active power uncertainties of
distributed generators, two different normalised active power curves
are used to represent the uncertainties of DG0 and DG1 as shown in
Fig. 7.
5.2

Test of fuzzy adaptive inference-based HPSO

In this paper, the corresponding cost and operation parameters are


from power supply bureau of China. KL is 1.5, KC is 0.8, KT is
1.0, ATnMax is 25 and ACmMax is 6. At the same time, the typical
PSO parameters are adopted. Particle number is 60, wMax is 0.8
and wMin is 0.3. Both c1 and c2 are 2.0. vmax and vmin are 4.0 and
4.0, respectively. Maximum iteration number is 200.

5.2.1 Test of the rationality of using comprehensive cost as


the optimisation objective: In order to verify the rationality of
using comprehensive cost as the optimisation objective, two
scenarios are chosen to calculate the optimal dispatching schedule.
They are the scenario of minimising power loss (Scenario 1) and
the scenario of minimising comprehensive cost (Scenario 2),
respectively. Under these two scenarios, the proposed method is
used to search the optimal dispatch schedule of LTCs, capacitor

Fig. 6 Three different normalised load curves

1100

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

Fig. 7 Two different normalised curves of DG active powers

Fig. 8 Calculation curves under two different optimisation objective scenarios

banks and reactive power outputs of distributed generators. The


corresponding calculation procedures are shown in Fig. 8.
Four curves are shown in Fig. 8. Two of them are produced under
Scenario 1 and the other two are produced under Scenario 2. These
curves present the evolution processes of power loss and
comprehensive cost, respectively. By the comparisons, it is
veried that the comprehensive cost under Scenario 2 is less than
the comprehensive cost under Scenario 1 in despite of the fact that
the power loss cost under Scenario 2 is greater than the power loss
under Scenario 1. The corresponding indices under these two
scenarios are shown in Table 2.
According to Table 2, when comprehensive cost is chosen as
optimisation objective, operation cost of distribution networks is
reduced further and primary devices are used more averagely. In

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

addition, all the operation number of capacitor banks in Scenario 1


reach their limits, namely, the MADSON constraints.
5.2.2 Performance test of fuzzy adaptive inference: In order
to verify the performance of the fuzzy adaptive inference to improve
the efciency and convergence of the original HPSO method, the
original HPSO method and FAHPSO are both used to search the
optimal schedule of LTCs, capacitor banks and reactive power
outputs of distributed generators. In this experiment, the
optimisation objective is to minimise comprehensive cost. The
corresponding calculation procedures are shown in Fig. 9.
At the same time, the corresponding optimal results are shown in
Table 3.

1101

Table 2 Calculation result comparisons under different optimisation


scenarios
Indices

C1 operation number
C2 operation number
C3 operation number
C4 operation number
C5 operation number
C6 operation number
C7 operation number
C8 operation number
LTC operation number
power loss cost (CNY)
comprehensive cost (CNY)

Scenarios
Scenario 1

Scenario 2

6
6
6
6
6
6
6
6
24
2268.64
2331.04

6
4
4
4
4
4
4
4
21
2272.52
2328.72

Fig. 10 Evolution process curves of different PSO algorithms


Table 4 Calculation result comparison among three different PSO
methods
Indices

Methods

comprehensive cost (CNY)


network loss (CNY)
capacitor operation number
transformer operation number
consumed time, s
evolution number to stable solution

IWAPSO

CMPSO

FAHPSO

2396.83
2396.83
36
24
318
190

2353.44
2302.44
40
19
322
191

2328.72
2272.52
44
21
310
69

the performance of fuzzy adaptive inference. It can be also


observed that both its nal comprehensive cost and power loss are
less than those of other two improved PSO methods.
Fig. 9 Test of the performance of fuzzy adaptive inference

According to Fig. 9 and Table 3, FAHPSO can search better


schedule for the studied distribution network using fewer evolution
cycles, compared with HPSO method. Especially, it can be
observed that FAHPSO has much better convergence than basic
hPSO.
5.2.3 Performance test of FAHPSO compared with other
improved PSO methods: In order to verify further the
performance of fuzzy adaptive inference to HPSO method, other
improved PSO methods based on current emerging technologies
are used to search the optimal schedule together with FAHPSO.
These methods include inertia weight approaching PSO (IWAPSO)
[21] and chaos mutation PSO (CMPSO) [23]. In this experiment,
the optimisation objective is still to minimise comprehensive cost.
Their evolution processes are shown in Fig. 10.
The corresponding optimal results are shown in Table 4.
According to Fig. 10 and Table 4, FAHPSO can gain good
optimal results after just 69 revolution cycles while other two
improved PSO methods need around 190 cycles. This is caused by

5.2.4 Performance test of FAHPSO compared with other


articial intelligent methods: In order to verify further the
performance of the proposed FAHPSO, other articial intelligent
methods, including teaching-learning-algorithm (TLA) [12],
reactive Tabu search RTS [13] and simulating annealing (SA)
[16], are used to search the optimal solution of the aforementioned
optimisation problem together with FAHPSO. In the experiment,
the optimisation objective is still to minimise comprehensive cost.
The corresponding experimental results are shown in Table 5.
In Table 5, it can be found that the operation number of capacitors
and transformers are equal for TLA and FAHPSO. However,
because of the different reactive power outputs of distributed
generators, which are corresponding to the control schemes
calculated by TLA and FAHPSO, respectively, both the
comprehensive cost and the power loss of FAHPSO are less than
TLA. It can be also seen that the nal comprehensive costs of SA
and FAHPSO are almost equal and they are better than RTS and
TLA. This means that both FAHPSO and SA can search more
economic reactive power control schemes for distribution
networks. Further, it can be observed that both the consumed time
and evolution cycle number of FAHPSO are less than SA. It can
Table 5 Experimental results of FAHPSO and other artificial intelligent
methods

Table 3 Calculation result comparison among four PSO methods


Indices
Indices

Methods

Methods
TLA

comprehensive cost (CNY)


network loss (CNY)
total capacitor operation number
total transformer operation number
consumed time, s
evolution number to stable solution

1102

HPSO

FAHPSO

2459.57
2416.37
34
16
320
174

2328.72
2272.52
44
21
310
69

comprehensive cost (CNY)


network loss (CNY)
capacitor operation number
transformer operation number
consumed time, s
evolution number to stable
solution

2352.81
2296.61
44
21
500
98

RTS

SA

FAHPSO

2344.60
2285.60
45
23
430
72

2332.90
2274.10
46
22
420
83

2328.72
2272.52
44
21
300
69

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

presented optimisation problem with relatively little time. Hence, it


is a promising method for reactive power and voltage control of
distribution networks with distributed generators.

Table 6 DG effects on the operation of distribution networks


Scenarios

Indices
Comprehensive
cost (CNY)

scenario
A
scenario
B
scenario
C
scenario
D

Power
loss cost
(CNY)

Operation
number of
transformers

Operation
number of
capacitors

4579.78

4529.58

19

39

3006.35

2953.75

19

42

3658.23

3608.23

18

40

2328.72

2272.52

21

44

This project is sponsored in part by the Fundamental Research Funds


for the Central Universities of China (no. ZYGX2012J160), in part
by the Danish Council for Strategic Research under Grant DSF
09-067255, Development of a Secure, Economic and
Environmentally-friendly Modern Power System (SEEMPS).

8
be explained by the good dynamic performance of fuzzy adaptive
inference on inertia weight. Less consumed time and evolution
cycle number means that FAHPSO can nd a good control
scheme more quickly. As a result, the dispatchers of distribution
networks have more time to analyse, decide and execute the
calculated reactive power control schemes to improve the operation
of distribution networks. Less consumed time and evolution cycle
number also implies that FAHPSO can be applied to a larger
distribution system with hundreds of buses. Hence, FAHPSO can
be regarded as the best method among all the tested intelligent
articial methods.
5.3

Effect of DGs on distribution network operations

In order to verify that the intermittent characteristic of distributed


generators has great effect on the operation of distribution
networks, the proposed FAHPSO method is used further to search
the optimal solutions under four different scenarios. In Scenario A,
no distributed generators are connected into the studied
distribution network. In Scenario B, only DG0 is connected into
the studied distribution network. In Scenario C, only DG1 is
connected into the studied distribution network. In Scenario D,
both DG0 and DG1 are connected into the studied distribution
network. In these scenarios, comprehensive cost minimisation is
selected as the optimisation objective. The corresponding
experimental data are recorded in Table 6.
From Table 6, it can be found that power loss cost and
comprehensive cost are reduced greatly with the effect of
distributed generators. This phenomenon may be explained from
both active power and reactive power of distributed generators.
With the help of distributed generators, both active power and
reactive power absorbed by distribution networks from
transmission networks are reduced. As a result, the corresponding
branch currents are reduced and the total power loss is reduced.

Conclusion

A new and efcient method for optimal reactive power and voltage
control in distribution networks with distributed generators is
proposed in this paper. In this optimisation method, three potential
control measures, including LTCs, capacitor banks and reactive
powers of distributed generators, are used. The corresponding
mathematical formulation can present massive distribution systems
in which there are a large number of control variables, distributed
generators and constraints. At the same time, HPSO algorithm is
used to solve the corresponding mathematical problem. In order to
avoid the premature of being trapped in the local optima, fuzzy
adaptive inference is used to improve the original HPSO method.
This strategy can reduce the time consuming in the process of
searching the optimal schedule. The proposed method is
implemented in VC + + 6.0 program language and the
modication version of IEEE 33-node system is selected to
demonstrate the effectiveness and robustness of the proposed
method. It is also veried that the searching strategy can solve the

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10961103
& The Institution of Engineering and Technology 2015

Acknowledgments

References

1 Carvalho, P.M.S., Correia, P.F., Ferreira, L.A.F.M.: Distributed reactive power


generation control for voltage rise mitigation in distribution networks, IEEE
Trans. Power Syst., 2008, 23, (2), pp. 766772
2 Hien, N.C., Mithulananthan, N., Bansal, R.C.: Location and sizing of distributed
generation units for load-ability enhancement in primary feeder, IEEE Syst. J.,
2013, 7, (4), pp. 797805
3 Deshmukh, S., Natarajan, B., Pahwa, A.: Voltage/VAR control in distribution
networks via reactive power injection through distributed generators, IEEE
Trans. Smart Grid, 2012, 3, (3), pp. 12261234
4 Keane, A., Ochoa, L.F., Vittal, E.: Enhanced utilization of voltage control resource
with distributed generation, IEEE Trans. Power Syst., 2011, 26, (1), pp. 252260
5 Rao, R.S., Ravindra, K., Satish, K.: Power loss minimization in distribution
system using network reconguration in the presence of distribution generation,
IEEE Trans. Power Syst., 2013, 28, (1), pp. 317325
6 Ochoa, L.F., Keane, A., Harrison, G.P.: Minimizing the reactive support for
distributed generation: enhanced passive operation and smart distribution
networks, IEEE Trans. Power Syst., 2011, 26, (4), pp. 21342142
7 Deng, Y., Ren, X., Zhao, C.: A heuristic and algorithmic combined approach for
reactive power optimization with time-varying load demand in distribution
systems, IEEE Trans. Power Syst., 2002, 17, (4), pp. 10681072
8 Liu, M.B., Caizares, C.A., Huang, W.: Reactive power and voltage control in
distribution systems with limited switching operations, IEEE Trans. Power
Syst., 2009, 24, (1), pp. 427436
9 Madureira, A.G., Peas Lopes, J.A.: Coordinated voltage support in distribution
networks with distributed generation and microgrids, IET Renew. Power Gener.,
2009, 3, (4), pp. 439454
10 Tanaka, K., Oshiro, M., Toma, S.: Decentralised control of voltage in distribution
systems by distributed generators, IET Gener. Transm. Distrib., 2010, 4, (11),
pp. 12511260
11 Calderaro, V., Conio, G., Galdi, V.: Optimal decentralized voltage control for
distribution systems with inverter-Based distributed generator, IEEE Trans.
Power Syst., 2014, 29, (1), pp. 230240
12 Niknam, T., Zare, M., Aghaei, J.: Scenario-based multi-objective volt/var control
in distribution networks including renewable energy sources, IEEE Trans. Power
Deliv., 2012, 27, (4), pp. 20042018
13 Yang, Y., Zhang, F., He, C.: A novel method for volt/var control in distribution
systems, Autom. Electr. Power Syst., 2005, 29, (9), pp. 4549 (In Chinese)
14 Fang, X., Guo, Z.: Optimal time-varying reactive power and voltage control in
distribution systems, Autom. Electr. Power Syst., 2005, 29, (9), pp. 4044 (In
Chinese)
15 Abu-Mouti, F.S., El-Hawary, M.E.: Optimal distribution allocation and sizing in
distribution systems via articial bee colony algorithm, IEEE Trans. Power
Deliv., 2011, 26, (4), pp. 20902101
16 Jeon, Y.J., Kim, J.C., Kim, J.O.: An efcient simulated annealing algorithm for
network reconguration in large-scale distribution systems, IEEE Trans. Power
Syst., 2002, 17, (4), pp. 10701078
17 Augugliaro, A., Dusonchet, L., Favuzza, S.: Voltage regulation and power losses
minimization in automated distribution networks by an evolutionary multiobjective
approach, IEEE Trans. Power Syst., 2004, 19, (3), pp. 15161526
18 Wu, W.C., Tsai, M.S.: Application of enhanced integer coded particle swarm
optimization for distribution system feeder reconguration, IEEE Trans. Power
Syst., 2011, 26, (3), pp. 15911599
19 Faria, P., Soares, J., Vale, Z.: Modied particle swarm optimization applied to
integrated demand response and DG resources scheduling, IEEE Trans. Smart
Grid, 2013, 4, (1), pp. 606616
20 El-Zonkoly, A.M.: Optimal placement of muliti-distributed generation units
including different load models using particle swarm optimisation, IET Gener.
Transm. Distrib., 2011, 5, (7), pp. 760771
21 Hu, W., Chen, Z., Bak-Jensen, B.: Fuzzy adaptive particle swarm optimisation for
power loss minimisation in distribution systems using optimal load response, IET
Gener. Transm. Distrib., 2014, 8, (1), pp. 110
22 Bajpai, P., Singh, S.N.: Fuzzy adaptive particle swarm optimization for bidding
strategy in uniform price spot market, IEEE Trans. Power Syst., 2007, 22, (4),
pp. 21522160
23 Jiang, C., Wang, C.: Improved evolutionary programming with dynamic mutation
and metropolis criteria for multi-objective reactive power optimization, IEE Proc.,
Gener. Transm. Distrib., 2005, 152, (2), pp. 291294
24 Caponetto, R., Fortuna, L., Fazzino, S.: Chaotic sequences to improve the
performance of evolutionary algorithms, IEEE Trans. Evol. Comput., 2003, 7,
(3), pp. 289304

1103

S-ar putea să vă placă și