Received: 26 January 2019 Revised: 18 June 2019 Accepted: 27 July 2019

DOI: 10.1002/2050-7038.12179

RESEARCH ARTICLE

Low switching frequency modulation of a 3 × 3 matrix

converter in UPFC application using differential evolution
method

Salman Ahmad1,2 | Mohammed Al‐Hitmi1 | Atif Iqbal1 | Khaliqur Rahman1 |

Imtiaz Ashraf2

1
Department of Electrical Engineering,
Qatar University, Doha, Qatar
Summary
2
Department of Electrical Engineering,
This paper presents a low switching frequency–based pulse width modulation
Aligarh Muslim University, Aligarh, India technique for a 3 × 3 indirect matrix converter. The nontriplen lower‐order
harmonics are considered for minimization using differential evolution–based
Correspondence
Salman Ahmad, Department of Electrical metaheuristic technique. The mathematical equations are formulated while
Engineering, Aligarh Muslim University, considering minimization of nontriplet lower‐order harmonics along with con-
Aligarh, India.
trol of fundamental voltage. An objective function is constructed by consider-
Email: salmanahmad19@gmail.com
ing all the constraints on the feasible region and then it is minimized by
Funding information using differential evolution technique. It is observed that the solution in the
Qatar University internal grant, Grant/
Award Number: QUCP‐CENG‐ 17\18‐2
entire modulation range exists along with multiple solutions. Various cases
of switching angles are considered, and solution trajectories are reported in
practically feasible range. This type of modulation and control scheme is
extremely useful in high power applications such as unified power flow con-
troller as it minimizes the switching losses in the converter. The theoretical
findings are verified by simulation and hardware results.

KEYWORDS
differential evolution, dSpace, low switching frequency, matrix converter, pulse width modulation,
UPFC

1 | INTRODUCTION

The matrix converter (MC) is a good alternative to the ac‐dc‐ac and conventional voltage source inverters in many appli-
cations because it can provide bidirectional power flow along with very close to sinusoidal input/output waveforms and
controllable power factor at input.1,2 The major advantage lies in the fact that there is no need of DC link reactive com-
ponents as in the case of its counterpart ac‐dc‐ac converters.3 The matrix converters nowadays are used in diverse range
of applications such as effective power supply, maximum power point tracking in wind energy conversion system,
power factor correction, electrical drives, and industrial applications.4-9 After its introduction, the research groups have
worked to improve its performance by introducing new control and modulation schemes.10-18 In comparison with most
commonly used voltage source converters (VSCs) in many industrial applications, the modulation techniques for the

LIST OF SYMBOLS AND ABBREVIATIONS: DE, Differential Evolution; Ns, Number of Switching Angles; UPFC, Unified Power Flow Controller;
VSC, Voltage Source Converter; θ, Switching Angles

Int Trans Electr Energ Syst. 2019;e12179. wileyonlinelibrary.com/journal/etep © 2019 John Wiley & Sons, Ltd. 1 of 23
https://doi.org/10.1002/2050-7038.12179
2 of 23 AHMAD ET AL.

matrix converter are not explored up to that extent. Also, the modulation techniques for the matrix converter are mainly
investigated based on high switching frequency techniques.4-6 In Klumpner et al7 and Huber and Borojevic,8 it is
observed that at low switching frequencies, the VSCs are more efficient than the matrix converter, whereas at higher
frequencies, the reverse is true. The various modulation and control techniques used for the matrix converter are shown
in Figure 1.3-8 At the beginning of the advent of the matrix converter, the transfer function–based method, also known
as Venturini method, is mostly used.10,11 In this method, the desired output voltage is obtained by multiplying the input
voltage and the converter transfer matrix. However, later, the modulation techniques developed for VSCs such as
carrier‐based PWM techniques and space vector–based PWM techniques have also been implemented for the matrix
converter in slightly different ways.12-14 In another method, known as scalar method,14 the active and zero states of
MC switches are generated using instantaneous voltage ratio of a particular input phase voltage.
A breakthrough in the modulation and control techniques of MC is achieved by implementation of the pulse width
modulation (PWM) techniques mostly developed for VSC.15-21 In the PWM technique, the simplest is a carrier‐based
method where a high‐frequency carrier is compared with a reference signal derived from the required output. Another
powerful PWM technique proposed for MC is the space vector modulation. The fast response and easy digital imple-
mentation of space vector PWM (SVPWM) technique make it one of the most competitive solutions with the rest of
the modulation schemes for MC. Other recently proposed modulation schemes for various matrix converter topologies
include direct torque and flux control, predictive control, and sliding mode controls, improving significantly the perfor-
mance of matrix converters in various applications10‐.11
In high power medium voltage applications, the switching losses due to high switching frequency are one of the
important considerations.21 The high switching loss decreases the efficiency of the system and reduces the reliability
of the whole system. Various optimization techniques are proposed in the literature to solve electrical engineering prob-
lems that are diverse in nature.22-24 For VSCs, the harmonics control governing equations are nonlinear transcendental
by nature and therefore will provide unique, multiple, or no solutions in different modulation index or operating points.
In literature, various methods are proposed to obtain solutions for these equations.25-40 These methods are broadly clas-
sified as follows: numerical technique–based iterative methods, algebraic methods, and optimization‐based
metaheuristic methods. The main numerical technique–based method includes Newton‐Raphson25 and sequential
homotopy.26 The algebraic methods such as resultant theory, Groebner bases, and symmetric polynomial–based
methods are used by first converting the trigonometric equations into polynomial equations and then the roots of the
equations are obtained.27-29 The optimization techniques such as genetic algorithm (GA),30 particle swarm optimization
(PSO),31 continuously exploring ant colony optimization (ACO),32 colonial competitive algorithm (CCA),33 artificial bee
algorithm,34 and differential evolution (DE)35 try to minimize an objective function, which results in suppression of
prespecified harmonics along with fundamental voltage magnitude control. The convergence of numerical‐based tech-
niques is highly dependent on initial guess and exact evaluation of Jacobian matrix. The algebraic methods become
complicated when the degree of polynomial increases with the increase in switching angles. Also, optimization tech-
niques sometimes fail to converge and are highly dependent on the parameters selected such as objective function, ini-
tial population, number of generations, number of particles, and accuracy in the solution.36-38
The low switching frequency with harmonics minimization techniques for MC is not significantly explored in the lit-
erature. The low switching frequency techniques for harmonics minimization developed for VSCs and other power con-
verters may be implemented for MC as well. The harmonics minimization equations for MC and VSCs are of similar

FIGURE 1 Classification of modulation schemes for the matrix converter

AHMAD ET AL. 3 of 23

nature, ie, highly nonlinear and transcendental. Therefore, the main challenge in applying harmonics control in MC at
low switching frequency is in obtaining the accurate switching instants in the complete range of operation. Some work
related to harmonics control and input power factor control for single‐phase matrix converters have been reported in
literature.41-46 In Zhao and Wang,40 microprocessor‐based harmonics elimination techniques have been proposed for
the chopper type ac voltage regulator. The paper reported improved power factor improvement in system power factor
and efficiency by minimizing the dominant lower‐order harmonics. An asymmetrical PWM method for harmonics elim-
ination along with power factor control for single‐phase AC chopper is carried out in Addoweesh and Mohamadein.41
In problem formulation, the input power factor has also been included and the numerical technique–based Newton‐
Raphson method has been used in solving the transcendental equations. Similar work has also been reported in
previous works,42-46 where the harmonics minimization problem is first converted into an objective function and then
minimized using various optimization techniques.
In this paper, an optimization technique is used to obtain solutions with random initial population within the
defined constraints for modulation of an indirect three‐phase matrix converter. The experimental results are given to
validate the analytical and simulation results using dSpace and FPGA control board. The computational and hardware
results are in close agreement and thus confirm the validity and effectiveness of the proposed method. The paper is
arranged as follows: Section 2 discusses the fundamentals of matrix converter. In Section 3, the harmonics minimization
equations are formed by using Fourier series analysis. Section 4 describes the differential evolution optimization tech-
nique for obtaining the switching instants of the objective function derived in Section 3. Analytical and computational
results are presented in Section 5. Experimental results from a prototype to validate the analytical and computational
results are given in Section 6. Finally, a conclusion of the work is presented in Section 7.

2 | MATRIX CONVERTER TOPOLOGIES AND CONTROL — AN OVERVIEW

Unlike back‐to‐back ac‐dc‐ac converters, the matrix converter offers a single stage ac‐ac conversion by connecting any
input phase to any output phase directly through bidirectional power switches. Since a matrix converter is normally fed
from a voltage source, there should not be a direct short circuit on the input side. Also, in most of the cases, the con-
verter is connected with an inductive load, and there should not be an open circuit on the output side. The two popular
topologies of a 3 × 3 matrix converter, namely, direct matrix converter and indirect matrix converter, are shown in
Figure 2A,B, respectively.
(
1; switch Sij closed
Sij ¼ i ¼ fA; B; C g and j ¼ fa; b; cg; (1)
0; switch Sij open

SAj þ SBj þ SCj ¼ 1: (2)

By having the conditions defined in (1) and (2), there are 27 possible switching states for a 3 × 3 matrix converter. If
the source neutral is assumed to be a reference, the source and load voltages can be expressed by (3) to (5),3,10.11
2 3 2 3
va ðt Þ vA ðt Þ
v0 ¼ 4 vb ðt Þ 5 and vi ¼ 4 vB ðt Þ 5; (3)
6 7 6 7

vc ðt Þ vC ðt Þ

2 3 2 32 3
va ðt Þ SAa ðt Þ SBa ðt Þ SCa ðt Þ v A ðt Þ
4 vb ðt Þ 5 ¼ 4 SAb ðt Þ SBb ðt Þ SCb ðt Þ 54 vB ðtÞ 5; (4)
6 7 6 76 7

vc ðt Þ SAc ðt Þ SBc ðt Þ SCc ðt Þ vC ðt Þ

vo ¼ T:vi ; (5)

where T is the instantaneous transformation matrix. In a similar way, the input and output currents can be related by
(6) to (8):
4 of 23 AHMAD ET AL.

FIGURE 2 Matrix converter topologies. A, Direct matrix converter; B, indirect matrix converter

2 3 2 3
ia ðt Þ i A ðt Þ
i0 ¼ 4 ib ðt Þ 5 and ii ¼ 4 iB ðt Þ 5; (6)
6 7 6 7

ic ðt Þ i C ðt Þ

2 3 2 3T 2 3
iA SAa SAb SAc ia
4 iB 5 ¼ 4 SBa SBb SBc 5 4 ib 5; (7)
6 7 6 7 6 7

iC SCa SCb SCc ic

ii ¼ T T :i0 ; (8)

where TT is the transpose matrix of instantaneous matrix T. The relation between instantaneous input and output
quantities is given by (1) to (8). The consideration of switching pattern for modulation and control of the converter is
necessary.
AHMAD ET AL. 5 of 23

The bidirectional switches have bidirectional current carrying capabilities as well as both voltage‐blocking capabili-
ties of both the polarity. To achieve this by having proper configurations of IGBT and diodes as it cannot be realized by a
discrete semiconductor switch. The various ways of realizing a bidirectional switch are shown in Figure 3.
The protection schemes such as overcurrent and overvoltage are also required to protect the bidirectional switches of
the matrix converter. The protection is aimed against the short circuit at the input side and open circuit at the output
side. The clamp circuit is employed for protection purposes as shown in Figure 4.
At the input side, the switching frequency component of currents have to be provided a local circulating path
through by an input filter in order to ensure that only the fundamental component is fed to the matrix converter. It also
ensures that a minimum electromagnetic interference (EMI) criterion of the design is satisfied. Also, the filter is
designed in such a way that the performance parameters such as voltage regulation and efficiency are not much
affected. In view of these, the single‐stage LC filter is found to be the most appropriate.
The rectifier and inverter stages of indirect matrix converter operated with space vector modulation technique are
shown in Figure 5. Generally, an indirect optimized switching pattern for control of CSR and VSI stages is used. The
CSR stage has six distinct active current vectors along with three zero current vectors. The VSI stage consists of six active
states and two zero states as shown in Figure 5. The corresponding vectors and determination of their time of applica-
tions are two important aspects of such modulation and control schemes. The proposed control technique in this paper
can be implemented for the unified power flow controller–based applications using matrix converter in previous stud-
ies47‐.48 In the unified power flow controller operation based on matrix converter, the input currents are fed from the
shunt side, and the fundamental component of the current is injected in series with the transmission voltage.
Figure 6 shows a generalized model of equivalent circuit of an UPFC‐MC and a steady‐state equivalent circuit.
The input current expression can be given by (12) to (14):

I i;a ¼ I mi cosðωi t − θi Þ; (12)

I i;a ¼ I mi cosðωi t − θi − 120°Þ; (13)

FIGURE 3 Bidirectional switch realization: A, insulated gate bipolar transistor (IGBT) with diode bridge; B, common emitter; C, common
collector; D, reverse blocking IGBTs

FIGURE 4 Matrix converter topology with filter and protection schemes

6 of 23 AHMAD ET AL.

FIGURE 5 Space vector for indirect matrix converter. A, rectifier stage; B, inverter stage

FIGURE 6 A, UPFC‐MC generalized equivalent circuit model. B, UPFC‐MC steady state equivalent model

I i;a ¼ I mi cosðωi t − θi − 240°Þ: (14)

And the output voltage expression can be given by (15) to (17):

V o;a ¼ V mi cosðωo t − θ°Þ; (15)

V o;b ¼ V mi cosðωo t − θo − 120°Þ; (16)

AHMAD ET AL. 7 of 23

V o;c ¼ V mi cosðωo t − θo − 240°Þ: (17)

The voltage side rectifier (VSR) stage, the mathematical expressions for phase is given by (18) to (20):

dI i;a
Li;a ¼ −Ri;a × I i;a þ V i;t;a − V i;a ; (18)
dt

V i;a ¼ Ai cos δi;a − 30° ; (19)

mi V dc
V i;a ¼ cosðωi t þ θi − 30°Þ: (20)
2

And for the voltage source inverter (VSI) side, the mathematical expressions including the virtual stage dc voltage are
given by (21) to (24):

dI i;a

Li;a ¼ −Ri;a × I i;a þ V i;t;a − Ai;a cos δi;a − 30° ; (21)
dt

dI o;a
V o;t;a ¼ Ao cos δo;A − 30° − Lo;a − Ro;a I o;a ; (22)
dt

dV dc
¼ ∑ V i;k − V o;k ; (23)
dt k¼a; b; c

I dc ¼ I i;dc − I o;dc

I dc ¼ ∑ I i;k × di;k − I o;k × do;k : (24)

k¼a; b; c

In steady state by ignoring the losses and assuming constant Vdc and Idc, the per phase steady‐state expression for VSI
and VSR stages of phase “a” can be written as

 
mi V dc

V i;t;a ¼ cos δi;a − 30° þ jX i × I i;a ; (26)
2

 
mo V dc

V o;t;a ¼ cos δo;a − 30° þ jX o × I o;a : (27)
2

For phase “b” and “c,” the expressions can be obtained by having a phase shifting of 120° and 240° in phase “a,”
respectively. After applying Park transform in the above expressions, the dynamic model equations for rectifier and
inverter stages are written as in (28).
8 of 23 AHMAD ET AL.

0 1
mi V dc
! cosðδ i − 30° ! !
V i;t;d B 2 C 0 −X i I i;d
¼ B þ
B C
C ;
V i;t;q @ mi V dc A Xi 0 I i;q
sinðδi − 30°
2
0 1 (28)
mo V dc
! cosðδ o − 30° ! !
V o;t;d B 2 C 0 −X o I o;d
¼ B þ
B C
C :
V o;t;q @ mo V dc A Xo 0 I o;q
sinðδo − 30°
2

The final controller design for UPFC‐MC should have maximum power transfer and minimum need of power factor
correction. It should have PQ decoupled control features. The PQ decoupled controller equations are given by (29) to
(31).
" #" #
P Vd Vq Id
 
¼ : (29)
Q Vq −V d Iq

And if the reference frame is synchronized with the mains, then

P ¼ V d × I d; (30)

Q ¼ −V d × I d : (31)

On the basis of the above expressions, the reference active and reactive powers are calculated, and it is compared
with the reference values. The parameters are regulated as per the switching sequence and timings to make the errors
in the computed and reference values at the minimum. The complete system configuration in such applications is
shown in Figure 7. Low switching frequency operation can make it feasible for operation at higher power transfer capa-
bilities and thus gives an attractive solution in such application. The active and reactive power variables can be main-
tained at a desired level by having proportionality to a linear combination of errors of these state variables. If active
power and reactive power errors are defined as eP and eQ, which is the difference between reference powers Pref and Qref
with actual powers P and Q respectively and given by (32).

eP ¼ Pref − P;
(32)
eQ ¼ Qref − Q:

The sliding surfaces aims to minimize this error are proportional to magnitude of these errors and become zero after
reaching slide mode. If kP and KQ are gains, then these relationships can be represented by (33).


SP ðeP ; t Þ ¼ k P Pref − P ¼ 0;

(33)
SQ ðeQ ; t Þ ¼ k Q Qref − Q ¼ 0:

The supply side reactive power, Qs, and receiving side reactive power, Qr, and the real power are measured and com-
pared with the reference reactive and real powers. The error signals ε1, ε2, and ε3 are processed through the PI controller
and minimized in order to control ΔMΦ, Δγ, and ΔMf. The M*Φ, M*f, and γ* are computed using the expressions (34) to
(36) and added to the feedback controllers (Figure 8).

Qs ¼ f 1 ðx; t; uÞ; (34)

QR ¼ f 2 ðx; t; uÞ; (35)

P ¼ f 3 ðx; t; uÞ; (36)

AHMAD ET AL. 9 of 23

FIGURE 7 Unified power flow controller using matrix converter

FIGURE 8 Control diagram with matrix converter for UPFC application

10 of 23 AHMAD ET AL.

h i
where vector of component parameters P = [Es, ER, δ, R1, L1, R2, L2, C], input vector u ¼ M *f ; γ * ; M *φ ; φ* , and the state
variables are XT = [iid, i1q, i2d, i2q, v2d, v2q].

3 | LOWER ‐OR DER H ARMONIC S MINIM I ZATION IN INDIRECT MATRIX

CONVERTER

Considering quarter wave and half wave symmetries in output waveform, Ns−1 low‐order harmonics can be minimized
by having Ns switching angles in a quarter cycle while controlling the fundamental voltage magnitude. The general out-
put waveform can be expressed by (37):

∞
vo ðωtÞ ¼ ∑ ½An cosðnωt Þ þ Bn sinðnωt ފ: (37)
n¼1; 2; 3

By considering quarter wave and half wave symmetries, the normalized Fourier coefficients for the nth order har-
monics component can be given by (38):

9
An ¼ 0;
  >
>
sinðn − 1Þθi sinðn þ 1Þθi >
>
Ns >
Bn ¼ ∑ i ¼ 1; 2; 3 − ;>
>
ðn − 1 Þ ðn þ 1 Þ
=
n≠1 (38)
>
>
Ns
 
sinð 2θ Þ
>
>
i i
B1 ¼ 1 − ∑ ð−1Þ θi −
>
>
: >
2
;
i¼1; 2; 3

Where, n is the harmonics order, Ns is the number of switching angles in a quarter period, and i ∈ {1,2,3⋯Ns} is the
index of a particular switching angle. The system of equation represented in (39) includes the control of fundamental
component expressed in terms of modulation index Mi along with minimization of Ns−1 prespecified harmonics. The
third harmonics need not to be minimized as it will be automatically cancelled from three phase line to line balanced
output voltage waveform. The system of equation that needs to be minimized for harmonics minimization is repre-
sented by (39).
Since the system of equation represented by (39) can give a variety of solutions such as multiple, unique, or no solu-
tion in different ranges of modulation, a number of methods can be used to solve these equations.22-30 The solutions are
normally calculated offline and then stored into lookup table and are implemented in various applications for different
operating conditions. In this paper, the solutions are obtained using an evolutionary‐based differential evaluation (DE)
technique, which includes the benefits of GA and PSO techniques.49 DE is a robust technique that tries to minimize an
objective function defined in (40).

½ð2θ1 Þ − sinð2θ1 ފ − ½ð2θ2 Þ − sinð2θ2 ފ þ ⋯ þ ð−1ÞN s þ1 ½ð2θN s Þ − sinð2θN s ފ − ðmi Þ ¼ ξ 1

½ð3sinð4θ1 ÞÞ − 2sinð6θ1 ފ − ½ð3sinð4θ2 ÞÞ − 2sinð6θ2 ފ þ ⋯ þ ð−1ÞN s þ1 ½ð3sinð4θN s ÞÞ − 2sinð6θN s ފ ≤ ξ 5
½ð4sinð6θ1 ÞÞ − 3sinð8θ1 ފ − ½ð4sinð6θ2 ÞÞ − 3sinð8θ2 ފ þ ⋯ þ ð−1ÞN s þ1 ½ð4sinð6θN s ÞÞ − 3sinð8θN s ފ ≤ ξ 7
½ð6sinð10θ1 ÞÞ − 5sinð12θ1 ފ − ½ð6sinð10θ2 ÞÞ − 5sinð12θ2 ފ þ ⋯ þ ð−1ÞN s þ1 ½ð6sinð10θN s ÞÞ − 5sinð12θN s ފ ≤ ξ 11
⋮ (39)
         
nþ1 n−1 nþ1 n−1
ðsinðn − 1Þθ1 Þ − ðsinðn þ 1Þθ1 Þ − ðsinðn − 1Þθ2 Þ − ðsinðn þ 1Þθ2 Þ þ ⋯
2 2 2 2
    
nþ1 n−1
þð−1ÞN s þ1 ðsinðn − 1ÞθN s Þ − ðsinðn þ 1ÞθN s Þ ≤ ξ n
2 2

F ðθ1 ; θ2 ; θ3 ⋯θN s Þ ¼ ξ 21 þ ξ 25 þ ξ 27 þ ⋯ þ ξ 23N s −2 : (40)

The solutions of (39) give nearly zero values of the objective function with the constraints defined in (41) to (44) sub-
ject to
AHMAD ET AL. 11 of 23

Ac :θ ≤ Bc : (41)

The elements of matrix constraints Ac are2 defined as follows:

8
< −1 if i ¼ j
>
Aci; j ¼ þ1 if i ¼ j þ 1 ; (42)
>
0 elsewhere
:

where

0 0 ⋯ 0 0 0
2 3
−1
6
6 1 −1 0 ⋯ 0 0 0 77
6 7
6
6 0 1 −1 ⋯ 0 0 0 77
Ac ¼ 6 ⋮ ⋮ ⋮ ⋯ ⋮ ⋮ ⋮ 7; (43)
6 7
6 7
6
6 0 0 0 ⋯ 1 −1 0 77
0 0 0 ⋯ 0 1
6 7
4 −1 5
0 0 0 ⋯ 0 0 1

0
2 3
2
θ1
3 6
6 0 7
7
6 7
6θ
6 2
7
7
6
6 ⋮ 7
7
⋮ 7:
6 7 6 7
θ¼6
6 ⋮ 7 and Bc ¼ 6
7 6 7 (44)
4 θN s −1 5
6 7 6
6 ⋮ 7
7
6 7
θN s 6 0 7
4 π 5
2

4 | D I F F E R E N T I A L E V O L U T I O N M E T H O D A P P L I E D TO TH R E E‐PHASE
INDIRECT MATRIX CONVERTER

Differential evolution (DE) was first introduced by Storn and Price as an optimization method having advantages of evo-
lutionary and swarm based techniques.49-51 It is a population‐based evolutionary algorithm. It is also a stochastic
method that works on the direct and parallel search in the solution space. Similar to other evolutionary algorithm, it
starts with the initialization of population called vector with random values within the minimum and maximum range
of the solution vectors. All the potential solution vectors are generated covering the entire permissible region goes under
competition and have a unique index of book keeping. The generation of perturbation from two randomly selected vec-
tors available in the solution space is first defined. In mutation operation, another vector is generated with a third ran-
dom vector and the weighted difference vector generated in perturbation process. The new vector obtained is termed as
the trial vector and represented as 0 vector. The objective function will be evaluated at this new vector, and if the gen-
erated vector after crossover have a lesser value, it will be replaced with 0 in the next generation. The trial vector will
compete against the entire population vector.
In minimizing a function f (α1, α2, α3⋯αS) of design variable of dimension D by using DE, the variables are initialized
to create a starting population P of size NP and coded as floating point variables. In the ith population, the jth variable at
starting (G = 0) are generated as a vector within feasible range using (45):

αi; j ð0Þ ¼ αmin þ randð0; 1Þ × ðαmax − αmin Þ;

(45)
i ¼ 0; 1; 2⋯N P − 1 and j ¼ 1; 2; ⋯S;

where S is the number of design variables. The rand(0, 1) function generates uniformly random number in [0,1].
12 of 23 AHMAD ET AL.

4.1 | Mutation

A mutant vector (βi) is generated by a mutation process for each vector αi,j and is given by (46):

βi; j ðG þ 1Þ ¼ αa;G þ F αc;G − αb;G a; b; c ∈ i and i ¼ 1; N P−1 ; (46)

where αa,G, αb,G, and αc,G are randomly chosen from [0,NP−1] such that αa,G ≠ αb, G ≠ αc, and G ≠ i. All the randomly
chosen numbers are different from each other, and none corresponds to the target vector βi. F is a scaling parameter
and the value of F should be chosen carefully as higher value outcomes in larger diversity in the generated population
and lower values will result in faster convergence. Normally, it is a positive real value in the range [0,2].50 In this paper,
the proposed variant in Wang et al51 is adopted, ie, DE/best/1; here, “DE” indicates differential evolution, “best” shows
a best‐chosen base vector αa,G, which is mutated with a single, ie, 1, scaled randomly chosen difference vector F (αc,
G − αb,G).

i. DE/rand/1

βi; j ðG þ 1Þ ¼ αa;G þ F × αc;G − αb;G :

ii. DE/rand/2

βi; j ðG þ 1Þ ¼ αa;G þ F × αc;G − αb;G þ F × αe;G − αd;G :

iii. DE/current‐to‐rand/1

βi; j ðG þ 1Þ ¼ αa;G þ rand × αa;G − αb;G þ F × αd;G − αc;G :

The βi,j(G+1) is the mutation operator, αa,G is the objective vector, and αi,G are individual base point vectors where
b; c; d; e ∈ i and i ¼ ð1; N P−1 Þ.

4.2 | Crossover

The trial vector is created from the target vectors and mutated vector by exchanging some of the components based on
predefined criteria. This process is called crossover operation. There are two types of crossover mostly used in DE and
are binomial and exponential. In this paper, the binomial crossover is used, which is shown in Figure 9.
In a group, each individual βi,j(G+1) is crossed with the mutation vector αi,j(G) to produce the new trial vector γi,j(G
+1). The αi,j(G) should be involved in randomly selected βi,j(G+1) to cofirm the individual evolution process. The Cr fac-
tor is used to determine if βi,j(G+1) or αi,j(G) contributes to a particular γi,j(G+1). Equation (47) gives the relationship in
these three parameters.

(
βi; j ðG þ 1Þ; if ðrandð0; 1ÞÞ ≤ Cr
γ i; j ðG þ 1Þ ¼ : (47)
αi; j ðGÞ; otherwise

Here, rand(0,1) is a randomly distributed uniform number in the jth gene, and Cr represents the crossover rate, and
to control the diversity of the trial vector, its value is normally kept in the range [0, 1]. The crossover operation is per-
formed on each design variable D to have better values as the generation passes.
AHMAD ET AL. 13 of 23

FIGURE 9 Mutation operation

4.3 | Selection

Next, the selection of better vectors from trial and target vectors to propagate in the next generation is carried out. The
fitness value of the trial and corresponding vectors decides its fate to be advanced in the next generation. The objective
function values are evaluated, and in minimization problem, the vector that gives a minimum value is considered for
the next generation. The selection process is described in (48).
8  
< β ðG þ 1Þ; if f β ðG þ 1Þ ≤ f α ðGÞ

i; j i; j i; j
αi; j ðG þ 1Þ ¼ : (48)
: α ðGÞ; else
i; j

4.4 | Generation of trial vector

Sections 4.1 to 4.3 give the formulas for mutation and crossover operations for the trial vectors:

1. DE/rand/1; for

rand ðjÞ ≤ Cr ∪ j ¼ rand nðiÞ;

αi; j ðG þ 1Þ ¼ αa;Gþ1 þ F αc;Gþ1 − αb;Gþ1 :

Otherwise,

αi; j ðG þ 1Þ ¼ αa;Gþ1 :

2. DE/rand/2; for

rand ðjÞ ≤ Cr ∪ j ¼ rand nðiÞ;

14 of 23 AHMAD ET AL.

αi; j ðG þ 1Þ ¼ αa;Gþ1 þ F αc;Gþ1 − αb;Gþ1 þ þ F αe;Gþ1 − αd;Gþ1 :

Otherwise,

αi; j ðG þ 1Þ ¼ αa;Gþ1 :

1. DE/current‐to‐rand/1; for

rand ðjÞ ≤ Cr ∪ j ¼ rand nðiÞ;

αi; j ðG þ 1Þ ¼ αa;Gþ1 þ rand αb;Gþ1 − αa;Gþ1 þ þ F αd;Gþ1 − αc;Gþ1 :

At the start, a set of parameters is first selected randomly. With proper operators and parameters, the initialization is
achieved and later mutation and crossover operations are carried out. To avoid solutions getting to be locally optimal,
the control parameters and mutation operator are taken to be complementary to each other.
Finally, a stopping criterion is set such as maximum generations, the best and worst objective function values, the
difference in objective function values in a predefined generation, to stop the algorithm. In this paper, the solutions
for the switching angles are saved if the algorithm does not reach the maximum iteration points and converges to a tol-
erance value of 10−4.To handle the constrained in the problem, the penalty function method is adopted in this paper.
Let the function be defined as (49):

FIGURE 10 Flow chart DE

AHMAD ET AL. 15 of 23

minimize f ðαÞ
subject to
gk ðαÞ ≤ 0; k ¼ 1; 2; ⋯M (49)
ht ðαÞ ¼ 0; t ¼ 1; 2; ⋯T
i i i
L ≤α ≤U; i ¼ 1; 2; ⋯N

The feasible range of solution in mathematical notation can be represented by (50).

Ω ¼ α ¼ α1 ; α2 ; α3 ; ⋯αN ∈ RN jgk ðαÞ ≤ 0; ht ðαÞ ¼ 0; Li ≤ αi ≤ U i ∀i :

  
(50)

To make the solution within the defined constraint, a penalty function that modifies the objective function to be min-
imized is introduced and is given by (51).
K T

∧ μ μ
f ðαÞ ¼ f ðαÞ þ λ ∑ gk ðαÞ þ ∑ ht ðαÞ ; Where gk ðαÞ ¼ maxf0; gk ðαÞg; ht ðαÞ ¼ jht ðαÞj; (51)
k¼1 t¼1

where λ is a penalty parameter and it is 1 for an exact penalty and 2 for a quadratic penalty. Therefore, the objective
function is heavily penalized for any deviation from the constrained and bounds defined for the problem.
The flowchart with detailed steps is given in Figure 10. Based on the flow chart, a computer program using Matlab
software is written to solve for the switching instants from the harmonics equations defined in (39) by minimizing the
objective function. The defined constraints are incorporated in the penalty function as given in (51). The program is fast
in operation and converges in the entire range of modulation.

FIGURE 11 Solution trajectories (A) Ns = 5, (B) Ns = 7, (C) Ns = 9, (D) Ns = 11

16 of 23 AHMAD ET AL.

5 | COMPUTAT I ON AL A N D S I M UL A T I O N R E S UL T S

The DE parameter used in this paper to minimize lower‐order harmonics is given in Table A1. Four different cases of
switching angles are taken to exemplify the efficacy of the proposed method. The solution trajectories obtained at var-
ious modulation indices are shown in Figure 11. In the presented case of nine switching angles, the nontriplen and low-
est order harmonics are controlled, (e.g. 5th, 7th, 11th, 13th, 17th, 19th, 23rd, and 25th) harmonics are minimized while
controlling the magnitude of the fundamental component. The predominant harmonic component in the output wave-
form is of 29th order, ie, 1450 Hz. This will reduce the filtering requirement of the output waveforms in many appli-
cations. Also, from the solution trajectories, it is clear that the MC can be operated in the whole range of modulation.
The multiple solutions are also found in some of the modulations, which confirm the increased range of operation of
the matrix converter. Figure 12A,B shows the harmonics profile for nine and eleven switching angle cases at a

FIGURE 12 (A) Residual of harmonics function for Ns = 9. (B) Residual of harmonics function for Ns = 11. (C) Harmonics profile at mi =
0.82for Ns = 9. (D) Harmonics profile at mi = 0.87 for Ns = 11

FIGURE 13 Output voltage and FFT spectrum at m = 0.95 for Ns = 9

AHMAD ET AL. 17 of 23

modulation index of 0.82 and 0.87, respectively. Also, the highest inaccuracy in the harmonics function due to the opti-
mization nature of the technique is given in (39) as shown in Figure 12C,D for nine and eleven angles, respectively.
A Simulink model is developed for a 3 × 3 matrix converter. The simulation for high voltage case has been carried
out and the line voltage along with FFT spectrum considering harmonics up to the 49th order is shown in Figure 13.
The results confirm that at the computed switching instants, the low‐order harmonics are considered for control or
have negligible magnitude in the output voltage waveform. Similarly, for other cases, the simulation results can be
obtained.

FIGURE 14 Comparative analysis for

different optimization methods at M = 0.90

FIGURE 15 (A) Block diagram of experimental setup. (B) Prototype photograph developed in laboratory

FIGURE 16 Switching functions (A) gate pulses for the rectifier stage. (B) Space vector for the rectifier stage. (C) Gate pulses for the “a”
upper switch of the inverter stage
18 of 23 AHMAD ET AL.

The numbers of iteration are compared for DE, PSO, ACO, and GA to minimize the same cost function as shown in
Figure 14. The comparison indicates that convergence rate in the proposed technique (DE) is much faster.

FIGURE 17 Condition for getting pulse for output phase “a”

FIGURE 18 Output at mi = 0.9: (A) unfiltered line voltage, phase voltage, and output current. (B) Source voltage and current. (C) Line
voltage harmonics spectrum
AHMAD ET AL. 19 of 23

6 | HARDWARE SETUP A ND EXPERIMENTAL RESULTS

In order to yield PWM signals, the Simulink platform is used in conjunction with dSpace 1006 and FPGA board to gen-
erate gate pulses for the matrix converter switches. Figure 15 shows the block diagram and actual setup of a three‐to‐
three phase matrix converter for testing and validation purposes of the proposed scheme in the laboratory. The detail
of the experimental parameters is given in Table A2.
The whole system is divided into three key units: power, measurement, and control units. The experimental setup for
three‐to‐three phase matrix converter is shown in Figure 15B. For the turning on and off of two switches in the same
leg, a dead band of two microseconds is provided in order to save the switches from damage. For this purpose, an exter-
nal Xilinx FPGA board programed in Verilog high‐definition language (VHDL). The MC is operated for various modu-
lation indexes and set of switching angles. The PWM signals are obtained by developing the space vector model of
regularly sampled PWM waveforms for the rectifier stage at a switching frequency of 1 kHz given in detail.52 This
approach eases the digital implementation of the PWM technique. The switching pulses of inverter stage are obtained
by comparing switching instants with three carriers shifted by an angle of 120° from each other. For a sector, the
switching pulses are shown in Figure 16.
The output from the matrix converter is obtained whenever simultaneously high pulses are applied on upper
switches of inverter stage leg “a” and rectifier stage leg “A.” Similarly, the combination of lower switches of inverter

FIGURE 19 output at Mi = 0.98: (A) unfiltered line voltage, phase voltage, and output current. (B) Source voltage and current. (C) Line
voltage harmonics spectrum
20 of 23 AHMAD ET AL.

stage leg “a” and rectifier stage leg “A” generates output at the converter. For this purpose, AND/OR logic is imple-
mented with inverter stage leg “a” and rectifier stage leg “A.” Hence, the converter output is directed from input phase
B if the similar condition is fulfilled. Same is true with other phases of “B” and “C” and “b” and “c,” respectively.
Among the various input phases available, only one will be obtained at the output of the matrix converter. The logical
operation that incorporates the rectifier and inverter stages is shown in Figure 17. That is only for the output of the
matrix converter phase “a.”
For illustration, the results at modulation indices 0.9 and 0.98 for 11 switching angles at operating frequency of 50 Hz
are shown in Figures 10 and 11, respectively. The switching angles at two different modulations from the continuous
trajectory of Ns = 11 to illustrate the experimental result shown in Figures 18 and 19 are given in Table A3. The wave-
forms of input current and voltages are shown in Figures 18B and 19B, which clearly indicates the input power factor is
very close to unity. The FFT spectra for the two cases given in Table A3 are shown in Figures 18C and 19C, respectively.
It can be observed that the output voltage contains very small magnitude of lower‐order harmonics considered for min-
imization. The corresponding source voltage and current is also shown. Theses waveforms are unfiltered and can be
smoothened by employing suitable filters to minimize the higher‐order harmonics from the final output waveforms.
The proposed method for harmonics control will not only reduce the filters rating and size but also gives considerable
improvement in the source voltage and current waveforms as confirmed from the figures.

7 | CONCLUSION

In this paper, a low switching frequency–based lower‐order harmonics minimization technique for a 3 × 3 matrix con-
verter is proposed. The Fourier series analysis is used to derive nonlinear transcendental equations for minimization of
lower‐order nontriplen harmonics. An objective function is formulated based on the harmonics equation and a robust
evolutionary–based technique, namely, differential evolution, is used to obtain the switching instants for the matrix
converter. By having Ns switching angles Ns−1 harmonics can be minimized along with control of fundamental magni-
tude. The continuous solution trajectories were obtained in the entire range of modulation for various cases of switching
angles. Multiple sets of the solution are also found in some modulation range. A prototype 3 × 3 matrix is developed in
the laboratory. The simulation and experimental results shows that the preassumed lower‐order harmonics for minimi-
zation are of negligible magnitude.

ORCID
Salman Ahmad https://orcid.org/0000-0003-0281-8220
Atif Iqbal https://orcid.org/0000-0002-6932-4367

R EF E RE N C E S
1. Friedli T, Kolar JW, Rodriguez J, Wheeler PW. Comparative evaluation of three‐phase AC‐AC matrix converter and voltage DC‐link back‐
to‐back converter systems. IEEE Trans. Ind. Electron. Dec. 2012;59(12):4487‐4510.
2. Zhang J, Dorrell DG, Li L. “Applications of the direct space vector modulation controlled matrix converter as the unified power flow control-
ler”, 8th Int. Machines and Drives, Glasgow, UK: Conf. Power Electronics; April 2016:1‐6.
3. Abu‐Rub J. G. H, Iqbal A, High performance control of AC drives with MATLAB/Simulink models. Wiley, 2012.
4. Shimada K, Itoh J‐I, Matsukawa M, Kurihara K. A control method of matrix converter for plasma control coil power supply. Elsevier J.
Fusion Eng. Des. 2007;82(5‐14):1513‐1518.
5. Barakati SM, Kazerani M, Aplevich JD. Maximum power tracking control for a wind turbine system including a matrix converter. IEEE
Trans. Energy Convers. Sep. 2009;24(3):705‐713.
6. Siami M, Khaburi DA, Rivera M, Rodríguez J. A computationally efficient lookup table based FCS‐MPC for PMSM drives fed by matrix
converters. IEEE Trans. Ind. Electron. 2017;64(10):7645‐7654.
7. Klumpner C, Nielsen P, Boldea I, Blaabjerg F. A new matrix converter motor (MCM) for industry applications. IEEE Trans. Ind. Electron.
Apr. 2002;49(2):325‐335.
8. Huber L and Borojevic D, “Space vector modulated three‐phase to three‐phase matrix converter with input power factor correction”, IEEE
Trans. Ind. Appl., Vol. 31, no. 6, pp. 1234‐1246, Nov./Dec. 1995.
9. Talaeizadeh V, Kianinezhad R, Seyfossadat SG, Shayanfar HA. Direct torque control of six‐phase induction motors using three‐phase
matrix converter. Elsevier J. Energy Convers. Manag. 2010;51(12):2482‐2491.
AHMAD ET AL. 21 of 23

10. Rodriguiz J, Rivera M, Kolar J.W, and Wheeler P.W, “A review of control and modulation methods for matrix converter”, IEEE Trans. On
Ind. Elect. 2012, Vol. 59, no. 1, pp. 58‐69, Jan.
11. Rodriguez J, Rivera M, Kolar J, Wheeler P. A review of control and modulation methods for matrix converters. IEEE Trans. Ind. Electron.
Jan. 2012;59(1):58‐70.
12. Empringham L, Kolar J, Rodriguez J, Wheeler PW, Clare JC. Technological issues and industrial application of matrix converters: a
review. IEEE Trans. on Ind. Elec. Oct. 2013;60(10):4260‐4271.
13. Gruson F, Le Moigne P, Delarue P, Videt A, Cimetiere X, Arpilliere M. A simple carrier‐based modulation for the SVM of the matrix con-
verter. IEEE Transaction on Industrial Informatics. 2013;9(2):947‐956.
14. Ali M, Iqbal A, Khan MR, Ayyub M, Anees MA. Generalized theory and analysis of scalar modulation techniques for a m × n matrix
converter. IEEE Trans. on Power Electronics. June 2017;32(6):4864‐4877.
15. Helle L, Larsen K, Jorgensen A, Munk‐Nielsen S, Blaabjerg F. Evaluation of modulation schemes for three‐phase to three‐phase matrix
converters. IEEE Trans. Ind. Electron. Feb. 2004;51(1):158‐171.
16. Li X, Sun Y, Zhang J, Su M, Huang S. Modulation methods for indirect matrix converter extending the input reactive power range. IEEE
Trans. Power Electron. 2017;32(6):4852‐4863.
17. Alesina A, Venturini MGB. Analysis and design of optimum amplitude nine‐switch direct AC‐AC converters. IEEE Trans. Power Electron.
Jan. 1989;4(1):101‐112.
18. Casadei D, Serra G, Tani A. The use of matrix converters in direct torque control of induction machines. IEEE Trans. Ind. Electron. Dec.
2001;48(6):1057‐1064.
19. Morel F, Retif J.‐M, Lin‐Shi X, Allard B, and Bevilacqua P, “A predictive control for matrix converter‐fed permanent magnet synchro‐
nous machine,” in Proc. IEEE Power Electron. Spec. Conf., Jun. 2008, pp. 15–21.
20. Ferreira Pinto S, Fernando Silva J. Input filter design for sliding mode‐controlled matrix converters. in Proc. 32nd Annu. IEEE Power Elec-
tron. Spec. Conf. 2001;2:648‐653.
21. Wu B. High‐power converters and AC drives. Wiley‐IEEE Press. 2006.
22. Ahmad S, Iqbal A, Ammari A, Abu‐Rub H. Selected harmonics elimination in multilevel inverter using improved numerical technique.
IEEE 12th International Conference on Compatibility, Power Electronics and Power Engineering CPE‐POWERENG. 2018;1‐6.
23. Ahmad S, Al‐Hitmi M, Iqal A, Ashraf I, Meraj M, Padmanaban S. Low‐order harmonics control in staircase waveform useful in high‐
power application by a novel technique. Int Trans Electr Energ Syst. 2019;29(3):e2769. https://doi.org/10.1002/etep.2769
24. Oshaba AS, Ali ES. Bacteria foraging: a new technique for speed control of DC series motor supplied by photovoltaic system. WSEAS
Transactions on Power Systems. 2014;9:185‐195.
25. Hitmi MA, Ahmad S, Iqbal A, Padmanaban S, Ashraf I. Selective harmonic elimination in a wide modulation range using modified
Newton‐Raphson and pattern generation methods for a multilevel inverter. energies. 2018;11(458):1‐16.
26. Kato T. Sequential homotopy‐based computation of multiple solutions for selected harmonic elimination in PWM inverters. IEEE Trans.
Circuits Syst. Fundam. theory Appl. 1999;46(5):586‐593.
27. Chiasson JN, Tolbert LM, McKenzie KJ, Du Z. Control of a multilevel converter using resultant theory. IEEE Trans. Control Syst. Technol.
2003;11(3):345‐354.
28. Yang KH, Zhang Q, Yuan RY, Yu WS, Yuan JX, Wang J. Selective harmonic elimination with Groebner bases and symmetric polyno-
mials. IEEE Trans. on Power Electron. Apr. 2016;31(4):2742‐2752.
29. Yang K, Yuan Z, Yuan R, Yu W, Yuan J, Wang J. A Groebner bases theory based method for selective harmonic elimination. IEEE Trans.
Power Electron. 2015;8993(c):1‐1.
30. Ozpineci B, Tolbert LM, Chiasson JN. Harmonic optimization of multilevel converters using genetic algorithms. IEEE Trans. Power Elec-
tron. 2005;3(3):92‐95.
31. TarafdarHagh M, Taghizadeh H, Razi K. Harmonic minimization in multilevel inverters using modified species‐based particle swarm
optimization. IEEE Trans. Power Electron. 2009;24(10):2259‐2267.
32. Sundareswaran K, Jayant K, Shanavas TN. Inverter harmonic elimination through a colony of continuously exploring ants. IEEE Trans.
Ind. Electron. 2007;54(5):2558‐2565.
33. Etesami MH, Farokhnia N, Hamid Fathi S. Colonial competitive algorithm development toward harmonic minimization in multilevel
inverters. IEEE Trans. Ind. Informatics. 2015;11(2):459‐466.
34. Kavousi A, Vahidi B, Salehi R, Bakhshizadeh MK, Farokhnia N, Fathi SH. Application of the bee algorithm for selective harmonic elim-
ination strategy in multilevel inverters. IEEE Trans. Power Electron. 2011;27(4):1689‐1696.
35. Salam Z, Majed A, Amjad AM. Design and implementation of 15‐level cascaded multi‐level voltage source inverter with harmonics elim-
ination pulse‐width modulation using differential evolution method. IET Power Electron. 2015;8(9):1740‐1748.
36. Kundu S, Burman AD, Giri SK, Mukherjee S. Banerjee, “Comparative study between different optimisation techniques for finding precise
switching angle for SHE‐PWM of three‐phase seven‐level cascaded H‐bridge inverter”. IET Power Electron. 2018;11(3):600‐609.
37. Dahidah M, Konstantinou G, Agelidis V. A review of multilevel selective harmonic elimination PWM: formulations, solving algorithms,
implementation and applications. IEEE Trans. Power Electron. 2015;30(8):4091‐4106.
22 of 23 AHMAD ET AL.

38. Amjad A. M, and Salam Z, “A review of soft computing methods for harmonics elimination PWM for inverters in renewable energy con-
version systems,” Renewable and Sustainable Energy Reviews, vol. 33, pp. 141‐153, May. 2014.
39. Ahmed M, Sheir A, Orabi M. Real time solution and implementation of selective harmonic elimination of seven‐level multilevel inverter.
IEEE J. Emerg. Sel. Top. Power Electron. 2017;5(4):1700‐1709.
40. Zhao H, Wang S. A four‐quadrant modulation technique to extend modulation index range for multilevel selective harmonic
elimination/compensation using staircase waveforms. IEEE J. Emerg. Sel. Topics Power Electron. Mar. 2017;5(1):233‐243.
41. Addoweesh KE, Mohamadein AL. Microprocessor‐based harmonic elimination in chopper type AC voltage regulators. IEEE Trans. on
Power Electronics. March, 1999;5(2):191‐200.
42. Jang D‐H, Choe G‐H, Ehsani M. Asymmetrical PWM technique with harmonic elimination and power factor control in AC choppers.
IEEE Trans. on Power Electronics. March 1995;10(2):175‐184.
43. Dahidah M,Agelidis V, “SHE‐PWM technique for single‐phase AC‐AC matrix converters”, Power Engineering Conference, 2008. AUPEC
'08. Australasian Universities, 14‐17 Dec. 2008.
44. Dahidah M, Rao M. A hybrid genetic algorithm for selective harmonic elimination PWM AC/AC converter control. Journal of Electrical
Engineering, Springer‐Verlag. March 2006;89(4):285‐291.
45. Deib D, Hill H. Optimal harmonic reduction in ac/ac chopper converters. in Proceedings of 24th annual IEEE power electronics specialist's
conference. 1993;1055‐1060.
46. Khamsen W, Aurasopon A, Sangiamvibool W. Power factor improvement and voltage harmonic reduction in PWM AC chopper using bee
colony optimization. IETE Technical review. 2013;30(3):173‐182.
47. Ooi B‐T, Kazerani M. Unified power flow controller based on matrix converter. PESC Rec. 27th Annu. IEEE Power Electron. Spec. Conf.
1996;1:502‐507.
48. Monteiro J, Member S, Silva JF, Pinto SF, Palma J. Matrix converter‐based unified power‐flow controllers: advanced direct power control
method. IEEE Trans. Power Deliv. 2011;26(1):420‐430.
49. Storn R, Price K. Differential evolution—a simple and efficient heuristic for global optimization over continuous spaces. Journal of Global
Optimization. 1997;2:341‐359.
50. Chakraborty. Advances in differential evolution. Studies in computational intelligence. Berlin Heidelberg: Springer‐Verlag; 2008:1‐10.
51. Wang Y, Cai ZX, Zhang QF. Differential evolution with composite trial vector generation strategies and control parameters. IEEE Trans.
Evol. Comput. 2011;15(1):55‐66.
52. Bowes SR, Lai YS. The relationship between space‐vector modulation and regular‐sampled PWM. IEEE Trans. Ind. Electron. Oct.
1997;44(5):670‐679.

How to cite this article: Ahmad S, Al‐Hitmi M, Iqbal A, Rahman K, Ashraf I. Low switching frequency
modulation of a 3 × 3 matrix converter in UPFC application using differential evolution method. Int Trans Electr
Energ Syst. 2019;e12179. https://doi.org/10.1002/2050‐7038.12179
AHMAD ET AL. 23 of 23

A P P EN D I X A
D A T A F O R SI M U L A T I O N A N D E X P E R I M E N T A L R E S U L T S :

TABLE A1 Algorithm parameters

Design variables (D) 5,7,9,11

Population size (NP−1) 10 × D

Mutation factor ( F ) 0.60
Crossover probability (CR) 0.80
Convergence criterion (ε) 10−4

TABLE A2 Experiment results parameters

Source voltage Vphase = 50 V

Load parameters R = 150 Ω

Switching frequency f s = 550 Hz
Input frequency f in = 50 Hz
Output frequency f out = 50 Hz
Modulation index mi = 0.90 & 0.98

TABLE A3 Switching instants for simulation results

Case Switching angles (radians)

Ns = 11, mi = 0.90 θ1 ¼ 0:369; θ2 ¼ 0:417; θ3 ¼ 0:520; θ4 ¼ 0:592; θ5 ¼ 0:677; θ6 ¼ 0:763;

θ7 ¼ 0:839; θ8 ¼ 0:941; θ9 ¼ 0:984; θ10 ¼ 1:111; θ11 ¼ 1:131
Ns = 11, mi = 0.98 θ1 ¼ 0:221; θ2 ¼ 0:308; θ3 ¼ 0:376; θ4 ¼ 0:489; θ5 ¼ 0:541; θ6 ¼ 0:664;
θ7 ¼ 0:707; θ8 ¼ 0:833; θ9 ¼ 0:869; θ10 ¼ 0:980; θ11 ¼ 1:005