Electrical Power and Energy Systems 61 (2014) 386–398

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Optimizing current harmonics compensation in three-phase power

systems with an Enhanced Bacterial foraging approach
Sushree Sangita Patnaik ⇑, Anup Kumar Panda
Department of Electrical Engineering, National Institute of Technology, Rourkela, India

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

Article history: A shunt active power filter (APF) comprising of pulse-width modulation (PWM) based voltage–source
Received 17 April 2013 inverter (VSI) is presented in this paper, because it has grabbed tremendous attention as a promising
Received in revised form 21 March 2014 power conditioner. However, it involves huge power loss due to the presence of inductors and semicon-
Accepted 24 March 2014
ductor switching devices, resulting in deterioration of APF performance. So, a Proportional–Integral (PI)
Available online 21 April 2014
controller has been used to minimize this undesirable power loss by regulating the dc-link voltage of
VSI. Conventional linearized tuning of PI controller gains does not yield satisfactory results for a range
Keywords:
of operating conditions due to the complex, non-linear and time-varying nature of power system net-
Active power filter
Current harmonics
works. The goal of this paper is to find out optimized values of PI controller gains by the implementation
Real-time analysis of optimization techniques. Developed by hybridization of Particle swarm optimization (PSO) and Bacte-
Bacterial foraging optimization rial foraging optimization (BFO), an Enhanced BFO technique is presented in this paper so as to overcome
Particle swarm optimization the drawbacks in both PSO and BFO, and accelerate the convergence of optimization problem. Compar-
ative evaluation of PSO, BFO and Enhanced BFO has been carried out with regard to compensation of har-
monics in source current in a three-phase three-wire system. Extensive MATLAB simulations followed by
real-time performance analysis in Opal-RT Lab simulator validate that, the APF employing Enhanced BFO
gives superior load compensation compared to the other alternatives, under a range of supply and sudden
load change conditions. It drastically lowers down the source current total harmonic distortion (THD),
thereby satisfying the IEEE-519 standard recommendations on harmonic limits.
Ó 2014 Elsevier Ltd. All rights reserved.

Introduction traditional passive filters like fixed compensation, large size,

increased risk of harmonic resonance with power system imped-
Current harmonics resulted due to the increased usage of non- ance, and ineffectiveness when the harmonic content in load cur-
linear loads, are the major culprits behind poor efficiency and rent varies randomly [8,9]. Compensation currents generated by
power factor, increased losses in power system, electro-magnetic the shunt APF can simultaneously compensate for current harmon-
interference (EMI) with nearby communication lines, false tripping ics and reactive power (also neutral current and unbalanced load-
of protective relays, failure or misoperation of microprocessors, ing of utility in three-phase four-wire systems) [6,10]. Various
vibration in rotating machines, voltage quality degradation, mal- control schemes have been proposed since the development of
functioning of medical facilities, and overheating of transformers, APFs [11–13]. But with id–iq method, harmonics can be mitigated
motors, neutral conductors, etc. [1,2]. IEEE Recommended Practices under all kinds of supply voltages [14,15].
and Requirements for Harmonic Control in Electric Power Systems During steady state operation, real power supplied by the
(IEEE 519-1992 Standard) [3,4] do an excellent job providing a source is equal to the real power demand of loads plus a small
basis for limiting harmonics. The shunt APF is devised to suppress power to compensate the switching and conduction losses of Insu-
different orders of harmonics simultaneously by injecting current lated gate bipolar transistors (IGBTs) inside APF. However during
harmonics of equal magnitude but in phase opposition with the load variation, real power difference between the two is compen-
load current harmonics at the point of common coupling (PCC) sated by the charging/discharging of dc-link capacitor of VSI. For
between the source and the load as shown in Fig. 1, thereby cancel- proper functioning of APF, dc-link voltage should be maintained
ling out each other [5–8]. This can overcome the shortcomings of constant with the help of a PI controller [16]. APFs with conven-
tional PI controller yield inadequate results under a range of oper-
⇑ Corresponding author. ating conditions [17,18], and are also criticized for being case
E-mail address: [email protected] (S.S. Patnaik). dependent because, when they are applied to same model with

http://dx.doi.org/10.1016/j.ijepes.2014.03.051
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 387

Fig. 1. Compensation principle of shunt APF (a) schematic diagram and (b) waveforms for load current (iL), filter current (ic) and compensated source current (is).

different parameters, the result varies. Hence load compensation compared to the usual PSO and BFO. Results obtained with
capability of APF can be enhanced by optimizing the gains of PI MATLAB/Simulink simulations are compared for APF employing
controller [19–21]. Few solutions to the harmonic mitigation prob- PSO, BFO and Enhanced BFO. It revealed that, Enhanced BFO shows
lem are found in Refs. [22,23], where harmonics are minimized quick convergence to reach the desired solution thereby yielding
through an objective function by optimization. In recent years, superior solution quality. Now-a-days, the testing and validation
many advances have been made to solve optimization problems of power conditioning devices is very essential in the design and
in various fields of application and research by the use of Simulated engineering process. The need for constant improvement of com-
annealing, Tabu search, Genetic algorithm (GA), Ant colony optimi- ponent modeling has led to an increase in the speed of system pro-
zation (ACO), Artificial Bee colony (ABC) optimization, PSO, BFO, totyping. But this has two major drawbacks. (i) There is a big
etc. Some artificial-intelligence based techniques such as Fuzzy hurdle in the design process during the leap over from off-line sim-
logic, Neural network and GA are exploited in Ref. [24] to design ulation to real prototype, as it is prone to many troubles related to
a control scheme for APFs dealing with harmonic and reactive cur- the integration of different modules at a time. (ii) Off-line non-real-
rent compensation. time simulation becomes tediously long for any moderately com-
PSO was introduced by Kennedy and Eberhart, and is being plex system. With advance in real-time simulation techniques,
extensively used due to its simple concept, easy implementation, RT-Lab simulator developed by Opal-RT technologies emerged as
inexpensive computation and well-balanced mechanism to pro- a promising tool. It is an industrial grade, scalable and real-time
mote both local and global explorations [15,19,20,25–27]. Na He platform for simulation, control testing and related applications
et al. demonstrated PSO to be effective and suitable for multi- [15,43,44]. Though several papers based on the implementation
objective optimal design of filters [28]. In spite of so many of evolutionary algorithms are there in literature, this paper
advantages, it suffers from the severe drawback of premature con- describes their performances and figures out the efficiency of these
vergence [29,30]. Since the inception in 2002, BFO has drawn the algorithms in real-time using Opal-RT Lab simulator.
attention of researchers to solve real-world problems such as adap-
tive control [31], harmonic signal estimation [32], optimal power Shunt active power filter
system stabilizers design [33,34], and optimization of real power
loss and voltage stability limit [35]. It is also hybridized with few Several aspects to be considered while designing an APF are
other state-of-the-art evolutionary computation techniques topology, reference compensation current extraction scheme and
[36–39] in order to achieve robust and efficient search perfor- PWM technique used to produce switching signals. Shunt APFs
mances. Over certain real-world optimization problems, BFO is are generally developed with either of the two types of PWM con-
reported to outperform many powerful optimization algorithms verters; current–source inverter (CSI) or voltage–source inverter
like GA and PSO in terms of convergence speed and final accuracy (VSI) [6,45,46]. However, the latter is preferred as it is lighter,
[18,31,32,35,37,38]. An approach based on BFO was proposed by cheaper, and expandable to multilevel and multistep versions for
Mishra that validated optimal control of APF under sudden improved performance in high power ratings with lower switching
switch-on, load change and filter parameters variation with ideal frequencies [6,46]. Fig. 2 depicts the system configuration of shunt
supply voltage [17]. However, there is no explanation towards APF with non-linear load.
the performance of APF when there is a huge unbalance or distor- In Fig. 3, the block diagram for reference current generation
tion in supply voltage along with sudden load variation. In our pre- employing idiq control scheme has been illustrated. The load cur-
vious work [18], a comparative evaluation of conventional, PSO rents iLa, iLb and iLc are tracked, upon which Park’s transformation is
and BFO based APFs was carried out under ideal and unbalanced performed to obtain corresponding d–q axes currents iLd and iLq as
supply without any load variation. Here the performance is inves- given in (1), where x is the rotational speed of synchronously
tigated under ideal, distorted and unbalanced supplies with sud- rotating d–q frame.
den load variation in MATLAB followed by real-time performance 2 3
analysis in Opal-RT Lab. BFO has also been reported to suffer from     rffiffiffi " 1 1
# iLa
iLd iLd1h þ iLdnh 2 cos xt sin xt 1  2  2 6 7
problems due to its fixed step size and uncontrolled particle veloc- ¼ ¼ pffiffi pffiffi 4 iLb 5
iLq iLq1h þ iLqnh 3  sin xt cos xt 0 3  3
ities [34,40–42]. Step size of the bacteria is decided by their veloc- 2 2 iLc
ities. Large step size leads to lesser accuracy though the particles
ð1Þ
arrive at vicinity of optima quickly; whereas smaller step size
slows down the convergence process. According to id–iq control strategy, only the average value of d-
A more powerful search algorithm called as Enhanced BFO is axis component of load current should be drawn from supply. Here
developed here, with the combined advantages of PSO and BFO. iLd1h and iLq1h indicate the fundamental frequency components of
This is found to be faster and provide more accurate results iLd and iLq respectively. The oscillating components iLd and iLq, i.e.,
388 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

Optimization approaches

Particle swarm optimization

Inspired by the social behavior of organisms in a bird flock or

fish school, this optimization technique emerged as a promising
nature-inspired stochastic approach of evolutionary computation
in the year 1995. Each candidate solution is a parameter vector
called as particle. The mechanism of PSO is initialized with a group
of randomly dispersed particles assigned with some arbitrary
velocities. The particles fly through the problem space in search
of global optimum position. The PSO system combines a social-
only model and a cognition-only model [26,47]. The social-only
component suggests the individuals to ignore their own experi-
ences and adjust their behavior according to the intelligence of
neighboring individuals. In contrast, cognition-only component
Fig. 2. Configuration of shunt APF with non-linear load.
treats the individual experience of each particle.
The particles update their positions and velocities in accordance
with (3) and (4), formulated by taking into consideration both the
iLdnh and iLqnh are filtered out using low-pass filter (LPF) consisting social-only and cognition-only components.
of 2nd order Butterworth type filters with cut-off frequency of    
25 Hz. Signal id1h accounts for the losses occurring in VSI and can
v ikþ1 ¼ w  v ik þ c1  r1  xiLbest  xik þ c2  r2  xiGbest  xik ð3Þ
be obtained from dc-link voltage regulator explained in Sec-
tion ‘Problem formulation’. Currents iLdnh and iLqnh along with id1h xikþ1 ¼ xik þ v ikþ1 ð4Þ
 
are utilized to generate reference filter currents icd and icq in d–q In the above expressions,
coordinates, followed by inverse Park’s transformation giving away k and i are indices for number of iterations and particle number;
  
the compensation currents ica , icb and icc in the three wires as xik and v ik are current position and velocity of ith particle at kth
described in (2). iteration;
xikþ1 and v ikþ1 are position and velocity of ith particle at (k + 1)th
2  3 2 32  3
ica sin xt cos xt 1 icd iteration;
   
6  7 6 sin xt  2p cos xt  2p 1 76 i 7
4 icb 5 ¼ 4 5 4 ð2Þ w, c1 and c2 are inertia, cognitive and social constants;
3 3 cq 5

    r1 and r2 are random numbers in the interval [1, 1].
icc sin xt þ 23p cos xt þ 23p 1 
ic0


Acceleration of particles is decided by the values of constants c1
The zero-sequence reference current ic0 in (2) is used just to and c2, whereas w provides a sense of balance between local and
make the transformation matrix a square one. The continuously global search. The exploration of new search space depends upon
tracked actual filter currents (ica, icb, icc) are compared with the ref- the value of w. Eberhart and Shi proposed a value of w that
  
erence filter currents ica ; icb ; icc in a Hysteresis band current con- decreases linearly with successive iterations [48], given as:
troller and consequently switching signals are generated [15].
g
Hysteresis PWM is used here for instantaneous harmonic compen- w ¼ wmax  ðwmax  wmin Þ ð5Þ
sation by the APF on account of simple implementation and quick G
prevail over fast current transitions. The main drawback of Hyster- where g is the index representing the current number of evolution-
esis current control is its high switching frequency that results in ary generation, G is the predefined maximum number of genera-
generation of some higher order harmonics due to the frequent tions, and wmax and wmin are maximal and minimal inertial
switching of semiconductor devices, which can be eliminated weights. Initially, the value of w is taken 0.9 (maximum) in order
using RC high-pass filter. Power losses that take place inside VSI to allow the particles to find the global optimum neighborhood fas-
as a consequence of this high frequency switching are minimized ter. Value of w is set to 0.4 (minimum) upon finding out the optima
by dc-link voltage regulator. so that the search is shifted from exploratory mode to exploitative

Fig. 3. Reference compensation current extraction and generation of switching signals.

 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 389

mode. The search terminates when either the predefined maximum point in the P-dimensional search domain that represents the posi-
number of iterations are executed or a further better optimum solu- tions of bacteria in the swarm, and him is the mth component of the
tion is not available. ith bacterium position hi. The coefficients datt, watt, hrep and wrep are
the measures of quantity and diffusion rate of the attractant signal
Bacterial foraging optimization and the repellant effect magnitude respectively. Now, the resulting
objective function J(h(i, j, k, l)) becomes
This optimization approach is based on the foraging behavior
J ðhði; j; k; lÞÞ ¼ J ðhði; j; k; lÞÞ þ J cc ðhði; j; k; lÞÞ ð8Þ
exhibited by E. coli bacteria inside human intestine. Here the bac-
teria undergo Natural selection and the ones with poor foraging
strategies are eliminated, whereas those with good foraging strat- Reproduction
egies survive. It was introduced by Passino in 2002 and further The fitness value for ith bacterium after travelling Nc chemotac-
established by Mishra in the year 2005 for harmonic estimation tic steps can be evaluated by following (9).
in power system voltage/current waveforms. BFO mimics four
principal mechanisms observed in bacteria viz. chemotaxis, X
N c þ1

J ihealth ¼ J i ðj; k; lÞ ð9Þ

swarming, reproduction and elimination–dispersal. j¼1

Chemotaxis Here Jihealth represents the health of ith bacterium. The least healthy
If h and J(h) represent the position and fitness of a bacterium, we bacteria constituting half of the bacterial population (Sr) are eventu-
want to find the minimum of J(h), h 2 RP ; where we do not have ally eliminated, while each healthier bacterium asexually reproduce
measurements or an analytical description of the gradient rJ(h). by splitting into two, which are then placed in the same location.
Following are the three conditions that arise Ultimately, the population remains constant. If S number of bacteria
constitute the population, then
(i) J(h) < 0; nutrient-rich environment. S
(ii) J(h) = 0; neutral environment. Sr ¼ ð10Þ
2
(iii) J(h) > 0; noxious environment.

Basically, chemotaxis is the movement of E. coli bacteria in Elimination and dispersal

search of nutrient-rich location, away from noxious environment. It is possible that the local environment where bacterial popu-
This is accomplished with the help of the locomotory organelles lation live changes either gradually via consumption of nutrients or
known as Flagella. Chemotactic movement is achieved by either suddenly due to some other influence such as significant heat rise.
of the following two ways, Following this behavior, BFO algorithm makes some bacteria to get
eliminated and dispersed with probability Ped after Nre number of
(i) Swimming (in the same direction as the previous step). reproductive events. This is to ensure that the bacteria do not get
(ii) Tumbling (in an absolutely different direction from the pre- trapped into local optima instead of global optimum.
vious one).
Enhanced Bacterial foraging optimization
Suppose hi(j, k, l) represents the ith bacterium at jth chemotac-
tic, kth reproductive and lth elimination–dispersal step. Then Taking into consideration all the drawbacks and advantages of
movement of the bacterium may be mathematically represented PSO and BFO algorithms discussed in Section ‘Introduction’, an
by (6). Enhanced BFO algorithm is developed here that has the combined
advantages of BFO and PSO, and is also capable of overcoming the
DðiÞ
hi ðj þ 1; k; lÞ ¼ hi ðj; k; lÞ þ CðiÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6Þ limitations in both. The performance of PSO is degraded in prob-
T
D ðiÞ  DðiÞ lems with multiple optima owing to a phenomenon called prema-
ture convergence, where the particles tend to converge and
In the expression, C(i) is the size of unit step taken in a random ultimately get trapped in a local best position as the global best
direction and D(i) indicates a vector in the arbitrary direction remains undiscovered. Enhanced BFO overcomes this drawback
whose elements lie in [1, 1]. through elimination–dispersal of bacteria, thereby ensuring con-
vergence to global optimum. Furthermore, the movement of indi-
Swarming viduals in traditional BFO algorithm is not defined in any specific
This group behavior is seen in several motile species of bacteria, direction. Random search directions delay the convergence to glo-
which helps them to propagate collectively as concentric patterns bal solution. However, unlike BFO, at any particular instant in PSO
of swarms with high bacterial density while moving up in the each particle memorizes its own best solution (local best) as well
nutrient gradient. The cell-to-cell signaling in bacterial swarm as the best solution of entire swarm (global best) owing to the
via attractant and repellant (Jcc(h(i, j, k, l))) may be modeled as memory it possesses. Thereby, velocity and direction of particles
per (7). are obtained as outcome of their social interactions and memory
X
S
storage capability. This characteristic of PSO is incorporated in
J cc ðhði; j; k; lÞÞ ¼ J cc h; hi ðj; k; lÞ Enhanced BFO that improves search efficiency, global optimum
i¼1 solution accuracy and convergence speed, which are the key attri-
" !#
X
S P
X
2 butes of an optimization algorithm. In Enhanced BFO, the chemo-
i
¼ datt  exp watt hm  hm taxis, swarming, reproduction, and elimination–dispersal events
i¼1 m¼1
" !# carried out in BFO realizing cell-to-cell communication, survival
X
S P
X
2 of the fittest, elimination of least healthy bacteria in the
þ hrep  exp wrep hm  him ð7Þ population, and exploration of new search areas is supplemented
i¼1 m¼1
with the ability of PSO to exchange social information and
Here S indicates the total number of bacteria in the population, P is possession of adaptable particle velocity. Hence, this algorithm
the number of variables to be optimized, h = [h1, h2, . . . , hP]T is a yields relatively more optimized result compared to BFO and PSO
390 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

implemented alone. The mechanism of Enhanced BFO and the iter- Step 2: Preliminary assessment of bacterial population.

ative algorithm realizing this mechanism are presented below in (a) For i = 1, 2, . . . , S, current fitness J icurrent of each bacterium
detail. A flowchart of the entire Enhanced BFO technique is in the search space is determined as per the following
depicted in Fig. 4. expression.

J icurrent ¼ Jði; j; k; lÞ
Mechanism of Enhanced Bacterial foraging optimization
In the beginning of search process, a group of bacteria are ran- (b) In the beginning of search process, since the bacterial move-

domly dispersed all throughout the search space. Each bacterium is ment is not yet started, the local best fitness J ilocal and local
assigned with an arbitrary particle velocity. Fitness values of the  
bacteria are calculated by taking into consideration the cell-to-cell best position xiLbest of each bacterium are its current fitness
swarming effect. For the initial population, the local best and glo- value and current position respectively, i.e.
bal best positions are figured out in exactly same way as done in J ilocal ¼ J icurrent
PSO. In order to update the positions of individuals, chemotaxis
is carried out, which utilizes a velocity factor obtained from the
velocity update expression used in PSO. After each chemotactic xiLbest ¼ hði; j; k; lÞ
step, fitness value of each particle is calculated. During the search, (c) The initial global best fitness (Jglobal) of the population is the
reproduction and elimination–dispersal events of BFO are also exe- minimum value of fitness possessed by any of the particle in
cuted. For reproduction, half of the bacterial population with least the population and can be given by,
health are eliminated, while rest half of the population asexually

reproduce by each of them splitting into two. Ultimately, the pop- J global ¼ min J ilocal
ulation size remains constant. To simulate elimination–dispersal
phenomenon, some bacteria are liquidated at random with a very where i = 1, 2, . . . , S.
small probability, while the new replacements are randomly ini- The position corresponding to Jglobal is assigned the global best
 
tialized over the search space. At the end of search process, the position xiGbest .
bacteria reach at the global optimum position.
Step 3: Iterative algorithm.
Iterative algorithm for Enhanced Bacterial foraging optimization (a) Initially the counters for chemotactic loop (j), reproduction
Step 1: Initialization. To begin with, all the parameters related to loop (k), elimination–dispersal loop (l), and counter for swim
proposed algorithm are initialized. Each particle in the group is length (m) are all set to zero.
assigned with a random initial position h(i), and an initial velocity (b) Taking into account the cell-to-cell attractant effect, the cost
(v) which is a random number in the interval [1, 1] with elements function estimated for each of the i = 1, 2, . . . , S bacteria is
n(i); n = 1, 2, . . . , P. calculated as,

Fig. 4. Flowchart of Enhanced BFO algorithm.

 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 391

" !#
X
S P
X
2 (g) If k < Nre, continue with the next reproductive iteration by
i
J cc ðhði; j; k; lÞÞ ¼ datt  exp watt hn  hn setting, k = k + 1.
i¼1 n¼1
" !# The entire iterative process is executed repeatedly until the
X
S P
X
2 specified number of reproductive steps (generations) are
þ hrep  exp wrep hn  hin
executed.
i¼1 n¼1
(h) Elimination–dispersal loop:
With a probability of Ped, the elimination–dispersal event is per-
Jði; j; k; lÞ ¼ Jði; j; k; lÞ þ J cc ðhði; j; k; lÞÞ
formed to ease the exploration of new search areas that may
J last ¼ J ði; j; k; lÞ lead to better optimal solution.
If elimination–dispersal loop counter l < Ned, execute successive
The best cost function value is stored in Jlast until a further
bet-

elimination–dispersal events with the increment of counter l
ter cost is obtained and the best cost of each bacterium J ilocal is after each iteration.
updated as, Terminate the iterative process when the counter l reaches its
J local ði; j; k; lÞ ¼ J last ði; j; k; lÞ maximum specified value, i.e., the number of elimination–dis-
persal events Ned.
(c) Chemotactic loop: Starting with i = 1, the position and cost
function for all the S number of bacteria in the entire popu- Problem formulation
lation are updated using the expressions,
vi Regulation of inverter dc-link voltage
k
hði; j þ 1; k; lÞ ¼ hði; j; k; lÞ þ CðiÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 i T i
vk  vk The dc-side capacitor (Cdc) serves two major purposes, i.e. (i)
maintains a constant dc voltage with small ripples in the steady
Jði; j þ 1; k; lÞ ¼ Jði; j þ 1; k; lÞ þ J cc ðhði; j þ 1; k; lÞÞ state and (ii) serves as an energy storage element to supply real
power difference between load and source during the transient
While m < Ns, period. The dc bus voltage must be higher than peak value of utility
If J(i, j, k, l) < Jlocal, voltage, to force the output current of APF under the command of
Then set Jlocal = J(i, j, k, l). compensating current. This dc-link capacitor acts as an energy
Updating position and cost function we will get, source and maintains energy balance inside the VSI. The compo-
vi nent of supply reference current (id1h) to restore the energy on dc
hði; j þ 1; k; lÞ ¼ hði; j þ 1; k; lÞ þ CðiÞ qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k
T
ðv ik Þ v ik bus is computed based on energy balance. If V dc is the
 reference
value of dc-link voltage, nominal stored energy edc on the dc
Jði; j þ 1; k; lÞ ¼ Jði; j þ 1; k; lÞ þ J cc ðhði; j þ 1; k; lÞÞ
bus of APF is
The current position (xcurrent) of each bacterium can be given  2
by, V dc
edc ¼ C dc ð11Þ
xcurrent ði; j þ 1; k; lÞ ¼ hði; j þ 1; k; lÞ 2

The position corresponding to local best fitness J ilocal is The actual average stored energy (edc) on dc bus is given by (12),
where Vdca is the average value of actual dc-link voltage.
stored in xiLbest .
Increment the counter as, m = m + 1. ðV dca Þ2
End the while loop. edc ¼ C dc ð12Þ
2
To proceed to next bacterium, set i = i + 1 until i = S.
Thus energy loss (Dedc) of dc-link capacitor is
(d) In each chemotactic step, calculate the current global best ( 2 )
V dc  ðV dca Þ2
fitness function value (JGcurrent) and continue the chemotactic Dedc ¼ edc  edc ¼ C dc ð13Þ
2
loop if still j < Nc.

J Gcurrent ¼ minðJ ilocal Þ This energy difference encountered in APF is supplied from ac
mains by regulating the dc-link voltage with the help of a PI
where i = 1, 2, . . . , S. controller.
If JGcurrent < Jglobal, set Jglobal = JGcurrent. The global best position A PI controller offers dual advantages as the Proportional (P)
 i 
xGbest is updated with the position corresponding to Jglobal. action provides fast response and the Integral (I) action provides
zero steady-state error. Block diagram for the process with PI con-
(e) The particles update their new velocities and directions by troller is shown in Fig. 5.
the equation, The output of a PI controller is given by,
    Z t
v ikþ1 ¼ w  v ik þ c1  r1 xiLbest  xicurrent þ c2  r 2 xiGbest  xicurrent uðtÞ ¼ K p  eðtÞ þ K i eðt Þ  dt
0
(f) Reproduction: Z t

The health of each bacterium is calculated using the expression ¼ K p  ½r ðtÞ  cðtÞ þ K i ½rðt Þ  cðtÞ  dt ð14Þ
0
given below and then sorted in ascending order of cost function.
Here, t represents the instantaneous time, e(t) is the system error
X
N c þ1
between the desired output r(t) and actual output c(t), u(t) is the
J ihealth ¼ i
J ðj; k; lÞ
controlled input for non-linear system, Kp is the proportional gain,
j¼1
and Ki is the integral gain. The proportional term considers only
Sr is the number of least healthy bacteria that are discarded out of the current value of error at any time, whereas the integral term
the population and individuals with best health are split into two, considers the sum of instantaneous errors over time, or how far
keeping the population size constant. the actual measured output value has been from the reference since
392 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

Table 1
Values of system parameters used in simulation.

Parameter Notation Value

Supply frequency f 50 Hz
Source impedance (Rs, Ls) (10 mX, 50 lH)
Load-1 parameters (RL1, LL1), (Rdc1, (0.1 X, 3 mH), (25 X,
Ldc1) 25 mH)
Load-2 parameters (RL2, LL2), (0.1 X, 3 mH), (25 X,
(Rdc2, Ldc2) 60 mH)
Fig. 5. Block diagram for PI controller design with optimization technique. dc-link capacitance Cdc 3 mF
Reference dc-link V dc 800 V
  voltage
the start time. Here, the dc-link voltage error DV dc ¼ V dc  V dc is ac-side filter parameters (Rc, Lc) (0.1 X, 1 mH)
minimized using a PI controller and the controller output is
expressed in (15).
Z t
Table 2
id1h ¼ K p  DV dc þ K i DV dc  dt ð15Þ Values of parameters used in Optimization techniques.
0
Parameter Notation Value
Need for optimization Population size S 8
Maximum no. of iterations N 50
In conventional linear PI controller tuning methods based on Dimension of search space P 2
Acceleration constants c1, c2 1.2, 0.12
mathematical modeling, the recommended settings are empirical
Inertia constant wmax, wmin 0.9, 0.4
in nature and are obtained from extensive experimentation. Apart No. of chemotactic steps Nc 5
from this, the power system network presents itself as highly com- Length of swim Ns 3
plex, non-linear and time varying system that involves large number No. of reproduction steps Nre 10
of inequality constraints. Hence, to satisfy the conditions of both No. of elimination–dispersal steps Ned 3
Probability of elimination–dispersal events Ped 0.25
dynamics and stability, optimized values of gains Kp and Ki can always Coefficients of swarming for attractant signal datt, watt 0.01, 0.04
be obtained. The advantages of optimization based controllers over Coefficients of swarming for repellent effect hrep, wrep 0.01, 10
conventional controllers are: (i) no need of accurate mathematical
modeling, (ii) can work with imprecise inputs, (iii) can handle nonlin-
earity, and (iv) more robust than conventional controllers. the above discussed optimization techniques under three different
supply conditions. System configuration of shunt APF along with
Objective function and optimization parameters the three-phase non-linear diode rectifier loads is depicted in
Fig. 6. Values of all the system parameters used for simulation
Here, optimization parameters are the PI controller gains Kp and and parameters used in optimization techniques are clearly indi-
Ki in the range 0 < Kp < 100 and 0 < Ki < 100. Maximum overshoot, cated in Tables 1 and 2 respectively. For ideal supply, completely
rise time, settling time and steady-state error are the constraints balanced and sinusoidal voltage of 230 V RMS is considered. A
that imply optimality of a PI controller. Performance criterion cho- highly distorted supply condition is simulated by incorporating
sen in this paper is integral square error (ISE) that treats both posi- 30% of 3rd harmonic component into the supply voltage. For unbal-
tive and negative errors equally. The objective function to be anced supply, voltage in one of the phases is 200 V RMS, while in
optimized (JISE) is formulated as per (16). other two it is 230 V RMS.
Z t
 
J ISE K p ; K i ¼ ðDV dc Þ2 dt ð16Þ MATLAB simulation results
0

The convergence characteristics of Fig. 7 shows that, Enhanced

Simulation results BFO reaches at minima in least number of generations compared to
PSO and BFO under all kinds of supply voltage conditions. Initially,
Extensive simulation has been carried out in MATLAB/Simulink only Load-1 is put into operation until t = 0.1 s in order to evaluate
and Opal-RT Lab in order to find out the effectiveness of APF with the harmonic compensation capability of APF. Performance under

Fig. 6. VSI-based shunt APF system configuration along with the non-linear loads.
 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 393

Fig. 7. Convergence characteristics of PSO, BFO and Enhanced BFO algorithms under (a) ideal supply, (b) distorted supply and (c) unbalanced supply.

dynamic conditions is observed by sudden switching on of Load-2 Figs. 9a–9c depict simulation waveforms for supply voltage,
at time instant t = 0.1 s. Relative convergence of Vdc with Enhanced load current, compensation current, and source currents for APF
BFO, PSO and BFO based PI controllers has been shown in Fig. 8, employing PSO, BFO and Enhanced BFO under ideal, distorted
which indicates that, Vdc reaches at its reference of 800 V within and unbalanced supply conditions respectively. The nature of
nearly one cycle under ideal and distorted supplies; and within source current before compensation is exactly same as the load
1.5 cycles under unbalanced supply. During load variation at current. It can be clearly observed that, irrespective of the nature
t = 0.1 s, the deviation in Vdc is maximum for PSO and minimum of supply condition, with the implementation of APF, harmonics
for Enhanced BFO signifying its fast return to 800 V and hence in source current have been fully compensated by injecting suit-
quick prevail over harmonics. Besides, ripples in Vdc during steady able compensating filter currents. The nature of compensated
state are observed to be very high for PSO and lowest for Enhanced source currents obtained with PSO, BFO and Enhanced BFO can
BFO. be compared by Fast Fourier Transform (FFT) analysis. The THDs

(a) 900 (b) 900

800 800
Vdc (Volts)

Vdc (Volts)

700 700

600 PSO-based PI 600 PSO-based PI

BFO-based PI BFO-based PI
Enhanced BFO-based PI Enhanced BFO-based PI

500 500
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

(c) 900

800
Vdc (Volts)

700

600 PSO-based PI
BFO-based PI
Enhanced BFO-based PI

500
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec)

Fig. 8. Relative convergence of Vdc with PSO, BFO and Enhanced BFO-based APFs under (a) ideal supply, (b) distorted supply and (c) unbalanced supply.
394 Vs (Volts) S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

is1 (Amp)
300 60
0 0
-300 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

50

is2 (Amp)
iL (Amp)

60
0 0

-50 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

40 60

is3 (Amp)
i c (Amp)

0 0

-40 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

Fig. 9a. Simulation waveforms for supply voltage (Vs), load current (iL), compensation current (ic), and source currents for APF employing PSO (is1), BFO (is2) and Enhanced BFO
(is3) under ideal supply.
Vs (Volts)

is1 (Amp)
300 60
0 0
-300 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

50
iL (Amp)

is2 (Amp)

60
0 0

-50 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

40 60
is3 (Amp)
ic (Amp)

0 0

-40 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

Fig. 9b. Simulation waveforms for supply voltage (Vs), load current (iL), compensation current (ic), and source currents for APF employing PSO (is1), BFO (is2) and Enhanced BFO
(is3) under distorted supply.
Vs (Volts)

300
is1 (Amp)

60
0 0
-300 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

50
is2 (Amp)
iL (Amp)

60
0 0

-50 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

40
is3 (Amp)

60
ic (Amp)

0 0

-40 -60
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (sec) time (sec)

Fig. 9c. Simulation waveforms for supply voltage (Vs), load current (iL), compensation current (ic), and source currents for APF employing PSO (is1), BFO (is2) and Enhanced BFO
(is3) under unbalanced supply.
 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 395

obtained from this analysis have been tabulated in Table 3. The Table 4
compensation of current harmonics to such a large extent irrespec- Components of Opal-RT OP5142 board.

tive of supply voltage is also due to the lower value of source Sl. no. Component name Description
impedance, which contributes to low influence of supply on source 1 S1 FPGA engine manual reset
current and vice-versa. 2 JTAG1 FPGA JTAG interface
3 JTAG2 CPLD JTAG interface
4 JUMP4 JTAG architecture selection
Real-time performance analysis with Opal-RT Lab 5 JTAG3 PCle bridge JTAG interface
6 JTAG4 SerDes JTAG interface
The Simulink model is built in the PC installed with MATLAB, 7 JP1 PCle & synchronization bus and power supply
8 J1/J2/J3 Backplane data, ID and I2C interface
which is integrated with Opal-RT Lab simulator. Once the model
9 JUMP1 Identification EEPROM write protection
is prepared, RT-Lab uses Real-Time Workshop to convert the sepa- 10 JUMP2 FPGA configuration mode selection
rated models into code for compilation as subsystem simulations 11 JUMP3 Flash memory write protection
on each target processor. Data from the model is directed to the 12 J4 Flash memory forced programmation voltage
user via a special subsystem, called the Console, where the signals FPGA – Field-Programmable Gate Array, JTAG – Joint Test Action Group, CPLD –
being generated can be viewed. In Fig. 10, the set-up used for real- Complex Programmable Logic Device, PCIe – Peripheral Component Interconnect
time simulation via Opal-RT Lab is depicted and various compo- Express, SerDes – Serializer/Deserializer, I2C – Inter-Integrated Circuit, EEPROM –
nents of OP5142 have been listed in Table 4. The details regarding Electrically Erasable Programmable Read-Only Memory

each component are described in [15,43,44]. The PCI-Express port

on OP5142 adapter board allows the users to connect the distrib-
source current THDs reveal that, Enhanced BFO offers lesser THDs
uted processors together and operate at faster cycle times than
than PSO and BFO under ideal, distorted and unbalanced voltage
ever before. This real-time link takes advantage of the FPGA power
supplies.
to deliver up to 2.5 Gbits/s full-duplex transfer rates. The OP5142
board is used to translate a Simulink design built using particular
Technical specifications of OP5142 RT-Lab:
library blocks into HDL. The results are observed in a Digital stor-
(i) Digital I/O:
age oscilloscope (DSO).
Number of channels: 256 input/output configurable in 1-
Rigorous analysis of performances of APFs employing PSO, BFO
to 32-bit groups
and the proposed Enhanced BFO is done in RT-Lab by operating
Compatibility: 3.3 V
both the loads simultaneously and the results for ideal, distorted
Power-on state: High impedance
and unbalanced supply voltage conditions have been presented
(ii) Bus:
in Fig. 11. RT-Lab results can now be compared with corresponding
Dimensions (not including connectors): PCI-Express x1
MATLAB results. The waveforms for supply voltage, load current,
Data transfer: 2.5 Gbit/s
compensation filter currents and compensated source currents
(iii) FPGA:
waveforms are exactly similar in both MATLAB and RT-Lab simula-
Device: Xilinx Spartan 3
tions. THD values of source currents signify the major differences
I/O Package: fg676
in their level of distortion. FFT analyses are done to find out the
Embedded RAM available: 216 Kbytes
source current THDs under ideal, distorted and unbalanced sup-
Clock: 100 MHz
plies and are tabulated in Table 5. Comparative evaluation of
Platform options: XC3S5000
Logic slices: 33,280
Table 3 Equivalent logic cells: 74,880
Source current THDs obtained with MATLAB. Available I/O lines: 489
Supply Source current THD (in %)
Technical specifications of the PC used for RT-Lab simulations:
condition Microsoft Windows XP (32-bit version), Xilinx ISE design suite
Without PSO based BFO based Enhanced BFO
APF APF APF based APF
v10.1 with IP update 3, Xilinx System Generator for DSP
v10.1, MATLAB R2007b.
Ideal 30.05 2.12 1.78 1.44
Technical specifications of the computer used for MATLAB sim-
Distorted 30.77 2.27 1.89 1.32
Unbalanced 29.72 2.78 2.14 1.58 ulations: Microsoft Windows 7 (32-bit version), MATLAB
R2009a.

(a) (b)
Fig. 10. Opal-RT Lab (a) set-up and (b) OP5142 system integration diagram showing layout and connectors.
396 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

Fig. 11. RT-Lab results for supply voltage, load current, compensation filter current, dc-link voltage, source current for PSO-based APF, source current for BFO-based APF,
source current for Enhanced BFO-based APF for (a) ideal supply, (b) distorted supply and (c) unbalanced supply respectively.
 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398 397

Table 5 [13] Bhattacharya A, Chakraborty C, Bhattacharya S. Shunt compensation:

Source current THDs obtained with Opal-RT Lab real-time simulator. reviewing traditional methods of reference current generation. IEEE Ind
Electron Mag 2009(September):38–49.
Supply Source current THD (in %) [14] Soares V, Verdelho P, Marques GD. An instantaneous active and reactive
condition current component method for active filters. IEEE Trans Power Electron
Without PSO based BFO based Enhanced BFO
2000;15(4):660–9.
APF APF APF based APF
[15] Patnaik SS, Panda AK. Real-time performance analysis and comparison of
Ideal 32.45 3.45 3.11 2.67 various control schemes for particle swarm optimization-based shunt active
Distorted 33.30 3.59 3.22 2.62 power filters. Electr Power Energy Syst 2013;52:185–97.
Unbalanced 32.07 4.60 3.39 2.91 [16] Jain SK, Agarwal P, Gupta HO. Design simulation and experimental
investigations, on a shunt active power filter for harmonics, and reactive
power compensation. Electr Power Compon Syst 2003;31(7):671–92.
[17] Mishra S, Bhende CN. Bacteria foraging technique-based optimized active
power filter for load compensation. IEEE Trans Power Del 2007;22(1):
457–65.
[18] Patnaik SS, Panda AK. Particle swarm optimization and bacterial foraging
Conclusion optimization techniques for optimal current harmonic mitigation by
employing active power filter. Appl Comput Intell Soft Comput 2012;2012:
Our study proposed the development of a novel Enhanced BFO 1–10.
[19] Chang W-D, Shih S-P. PID controller design of nonlinear systems using an
optimization strategy by hybridization of PSO and BFO to tune the
improved particle swarm optimization approach. Nonlinear Sci Numer Simul
proportional and integral gains of a PI controller. We proposed the 2010;15:3632–9.
implementation of this Enhanced BFO-based PI controller to be [20] Gaing Z-L. A particle swarm optimization approach for optimum design of PID
controller in AVR system. IEEE Trans Energy Conv 2004;19(2):384–91.
used for dc-link voltage regulation in APFs. Rigorous MATLAB and
[21] Lin C-L, Jan H-Y, Shieh N-C. GA-based multiobjective PID control for a linear
RT-Lab simulations are performed by employing PSO, BFO and brushless dc motor. IEEE/ASME Trans Mechatron 2003;8(1):56–65.
Enhanced BFO algorithms to shunt APF, keeping the simulation [22] Uyyuru KR, Mishra MK, Ghosh A. An optimization-based algorithm for shunt
parameters and system configuration same in all the cases. It active filter under distorted supply voltages. IEEE Trans Power Electron
2009;24(5):1223–32.
showed that, Enhanced BFO technique gives excellent Vdc transient [23] Verma V, Singh B. Genetic-algorithm-based design of passive filters for
response as the deviation of dc-link voltage from its reference offshore applications. IEEE Trans Ind Appl 2010;46(4):1295–303.
value could be minimized in the smallest amount of time (approx- [24] Kumar P, Mahajan A. Soft computing techniques for the control of an active
power filter. IEEE Trans Power Del 2009;24(1):452–61.
imately one cycle) irrespective of the supply voltage and sudden [25] Alrashidi MR, El-Hawary ME. A survey of particle swarm optimization
load change conditions. Hence, it provides nearly instantaneous applications in power system operations. Electr Power Compon Syst
compensation over current harmonics. The natures of compen- 2006;34(12):1349–57.
[26] Abido MA. Optimal power flow using particle swarm optimization. Electr
sated source currents obtained with various optimization Power Energy Syst 2002;24:563–71.
approaches are compared to find out the relative harmonic distor- [27] Shayeghi H, Shayanfar HA, Jalilzadeh S, Safari A. A PSO based unified power
tion in them. Less steady state ripples in Vdc transient leads to bril- flow controller for damping of power system oscillations. Energy Convers
Manage 2009;50(10):2583–92.
liant dc-link voltage regulation, resulting in less distorted source
[28] He N, Xu D, Huang L. The application of particle swarm optimization to passive
currents in case of Enhanced BFO. It outperformed all other alter- and hybrid active power filter design. IEEE Trans Ind Electron 2009;56(8):
natives in current harmonics mitigation by yielding the least val- 2841–51.
[29] Chen M-R, Li X, Zhang X, Lu Y-Z. A novel particle swarm optimizer hybridized
ues of source current THDs. Hence, Enhanced BFO algorithm has
with extremal optimization. Appl Soft Comput 2010;10:367–73.
an edge over the classical BFO and PSO algorithms, especially in [30] Monson CK, Seppi KD. Adaptive diversity in PSO. In: Proc 8th annual
context to the convergence behavior of the algorithm very near conference on genetic and evolutionary computation, ACM Press, New York;
to the desired solution. This fact has been supported here both ana- 2006. p. 59–66.
[31] Passino KM. Biomimicry of Bacterial foraging for distributed optimization and
lytically and experimentally using MATLAB and Opal-RT Lab. control. IEEE Control Syst Mag 2002;22(June):52–67.
[32] Mishra S. A hybrid least square-fuzzy bacterial foraging strategy for harmonic
estimation. IEEE Trans Evol Comput 2005;9(February):61–73.
References [33] Das TK, Venayagamoorthy GK. Bio-inspired algorithms for the design of
multiple optimal power system stabilizers: SPPSO and BFA. In: IEEE industry
application society’s 41st annual meeting, Tampa, USA, October 2006. p. 635–
[1] Peng FZ. Harmonic sources and filtering approaches. IEEE Ind Appl Mag
41.
2001(July/August):18–25.
[34] Mishra S, Tripathy M, Nanda J. Multi-machine power system stabilizer
[2] Subjak JS, Mcquilkin JS. Harmonics – causes, effects, measurements, and
design by rule based bacteria foraging. Electr Power Syst Res 2007;77:
analysis: an update. IEEE Trans Ind Appl 1990;26(6):1034–42.
1595–607.
[3] IEEE recommended practices and requirements for harmonic control in electric
[35] Tripathy M, Mishra S. Bacteria foraging-based solution to optimize both real
power systems, 993, IEEE Standard 519-1992.
power loss and voltage stability limit. IEEE Trans Power Syst 2007;22(1):
[4] Blooming TM, Carnovale DJ. Harmonic convergence – IEEE Std. 519-1992 helps
240–8.
define application of harmonic limits. IEEE Ind Appl Mag 2007(January/
[36] Kim DH, Abraham A, Cho JH. A hybrid genetic algorithm and bacterial foraging
February):21–7.
approach for global optimization. Inf Sci 2007;177(18):3918–37.
[5] Singh B, Al-Haddad K, Chandra A. A new control approach to three-phase
[37] Biswas A, Dasgupta S, Das S, Abraham A. Synergy of PSO and bacterial foraging
active power filter for harmonics and reactive power compensation. IEEE Trans
optimization: a comparative study on numerical benchmarks. In: 2nd
Power Syst 1998;13(1):133–8.
International symp hybrid artificial intelligent systems, advances in soft
[6] Singh B, Al-Haddad K, Chandra A. A review of active filters for power quality
computing series, Springer Verlag, Germany, innovations in hybrid intelligent
improvement. IEEE Trans Ind Electron 1999;46(5):960–71.
systems, ASC, vol. 44; 2007. p. 255–63.
[7] Akagi H. New trends in active filters for power conditioning. IEEE Trans Ind
[38] Biswas A, Dasgupta S, Das S, Abraham A. A synergy of differential evolution
Appl 1996;32(6):1312–22.
and bacterial foraging optimization for faster global search. Int J Neural Mass-
[8] Fujita H, Akagi H. A practical approach to harmonic compensation in power
Parallel Comput Inf Syst-Neural Netw World 2007;8.
systems – series connection of passive and active filters. IEEE Trans Ind Appl
[39] Azizipanah-Abarghooee R. A new hybrid bacterial foraging and simplified
1991;27(6):1020–5.
swarm optimization algorithm for practical optimal dynamic load dispatch.
[9] Wu CJ, Chiang JC, Yen SS, Liao CJ, Yang JS, Guo TY. Investigation and mitigation
Electr Power Energy Syst 2013;49:414–29.
of harmonic amplification problems caused by single-tuned filters. IEEE Trans
[40] Tang WJ, Wu QH, Saunders JR. Bacterial foraging algorithm for dynamic
Power Del 1998:800–6.
environment. In: IEEE congress on evolutionary computation, Vancouver,
[10] Chiang SJ, Ai WJ, Lin FJ. Parallel operation of capacity limited three-phase four-
Canada, July 2006. p. 4467–73.
wire active power filters. Proc Inst Electr Eng Electr Power Appl
[41] Das S, Dasgupta S, Biswas A, Abraham A, Konar A. On stability of the
2002;149(5):329–36.
chemotactic dynamics in bacterial foraging optimization algorithm. IEEE Trans
[11] Montero M, Cadaval ER, Gonzalez FB. Comparison of control strategies for
Syst Man Cyber – Pt A 2009;39(3):670–9 [IEEE Press, USA].
shunt active power filters in three-phase four-wire systems. IEEE Trans Power
[42] Biswas A, Das S, Abraham A, Dasgupta S. Analysis of the reproduction operator
Electron 2007;22(1):229–36.
in an artificial bacterial foraging system. Appl Math Comput 2010;215:
[12] Vardar K, Akpınar E, Surgevil T. Evaluation of reference current extraction
3343–55.
methods for DSP implementation in active power filters. Electr Power Syst Res
[43] RT-Lab Professional. <http://www.opal-rt.com/product/rt-lab-professional>.
2009;79(May):1342–52.
398 S.S. Patnaik, A.K. Panda / Electrical Power and Energy Systems 61 (2014) 386–398

[44] Patnaik SS, Panda AK. Three-level H-bridge and three H-bridges-based three- [47] Kennedy J. The particle swarm: social adaptation of knowledge. In: Proc 1997
phase four-wire shunt active power filter topologies for high voltage international conference on evolutionary computation, Indianapolis, IN, USA;
applications. Electr Power Energy Syst 2013;51:298–306. 1997. p. 303–8.
[45] Akagi H. Active harmonic filters. Proc IEEE 2005;93(12):2128–41. [48] Shi Y, Eberhart RC. A modified particle swarm optimizer. In: Proc IEEE
[46] Routimo M, Salo M, Tuusa H. Comparison of voltage–source and current– international conference on evolutionary computation, Anchorage, AK, May
source shunt active power filters. IEEE Trans Power Electron 2007;22(2): 1998. p. 69–73.
636–43.