Received March 10, 2019, accepted March 24, 2019, date of publication April 1, 2019, date of current version

April 15, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2908446

A Cost-Effective and Low-Complexity Predictive

Control for Matrix Converters Under Unbalanced
Grid Voltage Conditions
WENJING XIONG 1,2 , YAO SUN 1,2 , (Member, IEEE), JIANHENG LIN1,2 , MEI SU1,2 ,
HANBING DAN 1,2 , MARCO RIVERA 3 , (Senior Member, IEEE),
AND JOSEP M. GUERRERO 4 , (Fellow, IEEE)
1 Schoolof Automation, Central South University, Changsha 410083, China
2 Hunan Provincial Key Laboratory of Power Electronics Equipment and Grid, Changsha 410083, China
3 Facultyof Engineering, Universidad de Talca, Talca 2390123, Chile
4 Department of Energy Technology, Aalborg University, 9220 Aalborg, Denmark

Corresponding author: Hanbing Dan ([email protected])

This work was supported in part by the National Key R&D Program of China under Grant 2018YFB0606005, in part by the National
Natural Science Foundation of China under Grant 51807206, in part by the Key Technology R&D Program of Hunan Province of China
under Grant 2018SK2140, in part by the Project of Innovation-Driven Plan in Central South University under Grant 2019CX003, and in
part by the FONDECYT Research Project under Grant 1160690.

ABSTRACT Due to no dc-link energy storage element in the matrix converter (MC), its input and output
performances are highly sensitive to the unbalanced grid conditions. In order to mitigate the adverse effects
of the unbalanced grid voltages, this paper proposes a simple and effective control strategy based on the
finite-control set model predictive control (FCS-MPC). In this case, an extended instantaneous power theory
is adapted to generate the source current references. As a result, both sinusoidal source currents and balanced
output currents are ensured. Besides, complicated positive and negative sequence decomposition is avoided.
Furthermore, an extended state observer (ESO) is designed to eliminate the grid voltages sensors, which
not only estimates the grid voltages but also provides the delayed grid voltage information required for the
source current reference calculation. The effectiveness of the proposed method is verified by the simulation
and experimental results.

INDEX TERMS Matrix converter, unbalanced gird voltages, extended state observer, finite-control set model
predictive control.

I. INTRODUCTION connected or short circuit faults occur. To suppress the

Matrix converter (MC) is a direct AC/AC converter, and it adverse effects of unbalanced source voltages on the input
is featured by sinusoidal input and output currents, bidirec- and output performances of MC, many modulation strategies
tional energy flow, controllable input power factor and no have been presented [5]–[19].
dc-link component [1]–[3]. Due to the elimination of bulky In [5]–[9], several space vector modulation (SVM) meth-
and limited lifetime electrolytic capacitors, the reliability and ods are introduced for MC/indirect MC (IMC) to improve the
power density of MC are higher than those of the conventional source current quality and balance the output currents. In [5],
back-to-back converters [4]. Meanwhile, because of these two SVM methods are presented. One is aligning the input
characteristics, the voltage/current on input and output sides current vector with the input voltage vector. Desired balanced
of MC are directly coupled. Then, all the inner variables, such output currents are obtained, but the source current contains a
as input and output currents, are easily influenced by extra series of positive sequence harmonic components. The other
disturbances. can reduce or eliminate the harmonics of source currents by
The utility grid voltages would become unbalanced utilizing the information of positive and negative sequence
when considerable numbers of asymmetric loads are components of source voltages. The modified SVM method
in [7] can achieve balanced output currents, but the third-
The associate editor coordinating the review of this manuscript and order harmonic component in source currents is relatively
approving it for publication was Huiqing Wen. large. In [8], an online SVM method is introduced to optimize

2169-3536
2019 IEEE. Translations and content mining are permitted for academic research only.
VOLUME 7, 2019 Personal use is also permitted, but republication/redistribution requires IEEE permission. 43895
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

the duty cycles of the best switching states. To avoid the

complex sequence component separation, a simple method
for calculating the source current reference is proposed in [9].
Similar to the latter method in [5], a carrier-based modulation
(CBM) strategy for IMC is presented in [10]. Additionally,
based on the framework of SVM, some closed-loop control
strategies are put forward [11]–[13]. To achieve balanced
output currents and a near unity power factor (UPF) opera-
tion, an advanced closed-loop control for source currents is
proposed for IMC [11]. In [12], two resonant controllers are
adopted to regulate input currents and active power. Thus,
the quality of input currents is improved without degrading
the quality of output currents. In [13], a novel instantaneous
effective power control method (IEPC) for MC under unbal-
anced grids is discussed.
Three enhanced double input line voltage synthesis
(DLVS) methods are investigated in [14]–[15], they are aim-
ing to achieve sinusoidal source currents. In [16], three math- FIGURE 1. Topology of the matrix converter.
ematical construction methods are presented according to
different ways of selecting the input current vector. These in Section III, the proposed control scheme is described in
methods can achieve similar output current quality, but dif- detail; the simulation results of the proposed method and
ferent input current quality. Three model predictive control comparisons with other FCS-MPC-based strategies are pro-
(MPC) methods are proposed in [17], namely, active power vided in Section IV; in Section V, the experimental results
oscillation compensation (APOC) strategy, the instantaneous are illustrated; and finally, the main points of this paper are
unity power factor (IUPF) strategy and positive sequence summarized in Section VI.
(PS) strategy. Among these methods, the APOC strategy
II. MC SYSTEM DESCRIPTION
can achieve the best output and source current performance,
A. TOPOLOGY OF MATRIX CONVERTER
but the calculation of source currents reference is complex.
Literature [18] proposed an MPC method with reactive power The topology of MC is illustrated in Fig. 1. It mainly consists
minimization for IMC and literature [19] introduced a pre- of the power grid interface, input LC filter, nine bidirectional
dictive current control method for MC. The quality of output switches, and three-phase balanced loads.
currents is degraded with above two methods, as the ac- Assuming that the nine bidirectional switches are ideal,
term in instantaneous active power cannot be completely each switch Syx (x =a, b, c; y = A, B, C) has two possible
eliminated. states: Syx = 1 when on state and Syx = 0 when off-state. The
Compared with SVM method, DLVS method, states of nine switches can be represented as the following
CBM method, and the mathematical construction method, switching matrix  
SAa SAb SAc
the MPC method [20]–[24] appears as a promising alternative
S =  SBa SBb SBc  (1)
due to its simplicity and flexibility [3]. To reduce the com-
SCa SCb SCc
plexity and improve the input/output performance of afore-
mentioned methods, a cost-effective and low-complexity For safety reasons, the input of MC cannot be short-
control method based on finite-control set MPC (FCS-MPC) circuited, and the output should not be left open due to the
is proposed for MC under unbalanced grid voltages in this inductive loads. Thus, the switches should meet the following
study. The main ideas include 1) an extended instantaneous constraints
power theory for generating source current references is Sya + Syb + Syc = 1, y ∈ {A, B, C} (2)
adopted to reduce the complexity of algorithm. 2) a non-static
error extended state observer (ESO) based on internal model According to the constraints above, there are 33 = 27
principle is proposed to reduce costs. With the new source feasible switching states.
current references, both the sinusoidal source currents and the From Fig. 1, the relationship between the output and input
balanced output currents are obtained. Besides, complicated voltages/currents are established as follows
positive and negative sequence decomposition is avoided.
    
uoA SAa SAb SAc uia
Because of the non-static error ESO, voltage sensors for grid  uoB  =  SBa SBb SBc   uib  (3)
voltages are saved. As the ESO also provides the delayed uoC SCa SCb SCc uic
source voltages information, it does not increase total com-    T  
putational burden of the proposed method. iia SAa SAb SAc ioA
The remainder of this paper is organized as follows:
 iib  =  SBa SBb SBc   ioB  (4)
Section II introduces the topology and the model of MC; iic SCa SCb SCc ioC

43896 VOLUME 7, 2019

W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

where uix and uoy represent the three phase input and output
voltages, respectively, iix and ioy represent the input and
output currents, respectively.

B. MODELING OF MATRIX CONVERTER SYSTEM

The dynamic behavior of a general three-phase balanced load
can be described as follows:
dEio
L = uEo − REio − Ee (5)
dt
where L and R are the load inductance and resistance, respec-
tively; Eio and uEo are the output current and voltage vector;
Ee is the electromotive force (EMF) vector of the load. The FIGURE 2. Block diagram of the proposed control scheme. (superscript ‘p’
denote as the predictive value at (k+2)-th instant).
load may cover RL load, induction motor, permanent magnet
synchronous motor and grid. In this study, only the RL load
is considered, thus Ee = 0.
The continuous-time model of input side is described as of reactive power is non-unique [25]–[26]. For the same given
the following equations: active and reactive power, the obtained currents are different
      under different definitions of reactive power in the case of
d uEi uEi uEs unbalanced voltages.
= A + B (6)
dt Eis Eis Eii In this paper, the definitions of instantaneous active and
reactive power based on the extension pq theory in [25] are
where
    introduced, where the input instantaneous active and reactive
0 1/Cf 0 −1/Cf power of the MC can be expressed as
A= ,B = ,
−1/Lf −Rf /Lf 1/Lf 0
Ps = Re(Ei∗s uEs ) (7)
Rf , Lf , and Cf are the equivalent line resistance, the filter 0
Qs = Re(Ei∗s uEs ) (8)
inductance and capacitance, respectively, uEs and Eis are the
source voltage and current vectors, respectively, uEi and Eii are where the superscript ‘‘∗’’ denotes the conjugate of a vector;
the input voltage and input current vectors. Re(•) is a function which extracts the real part from its argu-
0
ment. uEs denotes a vector whose phase lags uEs by 90 degrees.
III. PROPOSED CONTROL SCHEME As proved in [26], if Qs is a constant, the oscillatory
Usually, the control targets of MC under unbalanced grids components in Ps can be eliminated naturally. That is, the def-
are: 1) output voltages or currents regulation; 2) low-order initions in (7) and (8) lay the foundation for the two control
harmonics reduction of the source currents with less con- targets under unbalanced grids mentioned before.
tents and input reactive power control. To achieve these Denote the unbalanced source voltage vector uEs as follows
targets, a control method based on FCS-MPC is proposed
for MC. Fig. 2 illustrates the overall control block dia- uEs (t) = UP ejωt + UN e−j(ωt−ϕ) (9)
gram, which mainly involves an ESO and an MPC algorithm
where ω is the angular frequency of source voltages, UP and
block. As shown, based on the novel power definition and
UN are the amplitude of positive and negative sequence volt-
the ESO, the source current reference is calculated. With the
age, respectively, while ϕ is the initial phase angle between
information of measured source currents and input voltages,
positive and negative sequence voltage.
the ESO can accurately estimate the source voltages and 0
Then, the delayed source voltage uEs can be expressed as
the delayed source voltages without using the grid voltage
0
sensors, thus reducing the system costs. The MPC is an uEs (t) = −j(UP ejωt − UN e−j(ωt−ϕ) ) (10)
FCS-MPC, in which the sinusoidal source currents based on
the novel power definition and balanced output currents are Combine (7) ∼ (10), the source currents vector is solved as
controlled. In this Section, the proposed control scheme will 0 ∗
be introduced for MC in detail. Eis = j Ps uEs ∗ −0 Qs uEs (11)
Im(Eus∗ uEs )
A. CALCULATION OF SOURCE CURRENT REFERENCE where Im(•) is a function which extracts the imaginary part
Since there are no storage units in MC, the input instanta- from its argument, xE∗ represents the conjugate of xE.
0
neous active power of MC is equal to its output instantaneous The denominator Im(Eus∗ uEs ) = 32 (UP2 − UN2 ) is a constant,
0
active power when the power losses are neglected. According uEs and uEs are sinusoidal. Thus, according to (11) when the
to the first target, the input instantaneous active power of constant active power Ps and reactive power Qs are given,
MC should keep constant. Regarding the second target, a cer- the sinusoidal source currents will be obtained under both
tain reactive power will be regulated. However, the definition balanced and unbalanced source voltages.

VOLUME 7, 2019 43897

 W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

B. ESO FOR SOURCE VOLTAGES ESTIMATION where ωc represents the location of the pole. Let (15) and (16)
From(6), both the source voltages and input voltages have to equal to each other, then k1 , k2 , and k3 can be obtained
be measured to implement the FCS-MPC. In order to reduce accordingly.
the costs, a non-static error ESO is proposed to estimate the 
source voltages.  k1 = 3ωc Lf − R
f

Assume that the source voltages are unbalanced and its k2 = 3ωc2 − ω2 Lf (17)
k3 = 3ωc ω − ωc /ω Lf
3


frequency is constant and known. Then, the model of the input

side can be expressed as
 When ωc > 0, the designed AE is a Hurwitz matrix. Then,
disx the estimation error will be close to zero, and the convergence
 Lf = usx − uix − Rf isx
speed of the error is determined by ωc .

 du dt



sx 0
On one hand, the grid voltages information are estimated
= −ωusx (12)

 dt0 by observer, which eliminates the voltage sensors and then
 du
 sx = ωusx reduces costs effectively. On the other hand, the delayed


dt source voltage is also obtained by the observer. As a result,
where isx and usx (x =a,b,c) are the source current and the extra strategy for obtaining the delayed source voltage is
0
voltage, respectively, and usx is the delayed source voltages. omitted, which reduces the computational burden to a certain
Based on (12), a non-static error ESO is designed as extent.

d îsx  
C. DISCRETIZATION OF THE MC SYSTEM
L = û − u − R î + k i − î

f sx ix f sx 1 sx sx


 dt To predict the future behavior of the system variables, the


d ûsx 0
 
= −ωûsx + k2 isx − îsx (13) discrete-time model of MC should be developed. Assume that
 dt
 0 the control inputs are piecewise constant over the sampling
 d ûsx = ωûsx + k3 isx − îsx

  
period Ts , the model of input LC filter in (6) is discretized as

dt follows,
where k1 , k2 , and k3 are coefficients, which will be deter-  k+1   k  k
0 uEi uEi uÊs
mined later,îsx , ûsx and ûsx are the estimated value of isx , usx = G + H (18)
0 Eik+1 Eiks Eik
and usx , respectively. s i
Subtracting (12) from (13), we have where
   
ĩ
d  sx 
ĩsx G = eATs , H = A−1 (G − I )B (19)
ũsx = AE  ũsx  (14)
dt 0 0 and the subscript ‘k’ and ‘k + 1’ denote as the k-th and
ũsx ũsx
(k + 1)-th sampling time, respectively.
where As the continuous-time models of the load and non-static
− Rf + k1 /Lf


1/Lf 0
 error ESO in (5) and (13) can be written with the similar form
AE =  −k2 0 −ω  , of (18), so the discrete-time models of the load and ESO can
−k3 ω 0 also be obtained by (19).

and D. CONSTRUCTION OF THE COST FUNCTION


ĩsx
 
îsx − isx
 The tasks of the FS-MPC are to track the desired input cur-
 ũsx  =  ûsx − usx  rents and output currents. Based on the discrete-time models
0 0 0 and control objectives, the cost function can be constructed
ũsx ûsx − usx
as follows.
represents the estimation error.
Then, the characteristic equation of the proposed ESO is 1) CONTROL OF SOURCE CURRENTS
expressed as To achieve the sinusoidal source currents, the following cost
function is constructed.
λ (s) = |sI − AE | 2
g1 = Eisref − Eik+2
 
3 k1 +Rf 2 k2 k1 + R f 2 k3 (20)

=s + s + 2
+ω s + ω − ω s
Lf Lf Lf Lf
(15) where Eisref is the desired vector of source current, which can
be obtained from (11) by replacing Ps and Qs with Psref and
To simplify the tuning process, all the poles are placed in Qsref ; kEx k denotes as the Euclidean norm of the vector xE.
the same location, then the desired λ(s) is formulated as To achieve the UPF operation, Psref and Qsref are given by
λ (s) = (s + ωc )3 (16) Psref = Poref /η, Qsref = 0 (21)

43898 VOLUME 7, 2019

W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

where Poref is the active power reference of load, which is

determined by the type of loads, and η is the efficiency of the
converter.
The active power reference Poref of three phase balanced
load can be written as

3 2
 2 Iom R,
 when RL load
Poref = T , when motor load (22)
o.
3

I U
2 om om cosϕ when grids
where Iom is the amplitude of the output current reference,
Uom is the amplitude of the voltage of balanced grid, ϕo is
the desired displacement angle, T and  are the torque and
angular speed reference of a motor, which can be obtained
from the motor control.
As mentioned above, Psref depends on two factors: η
and Poref . Although η is an unknown and time-varying
parameter and Poref is not accurately known, they have a little
effect on the system performance. The imposed source cur-
rents by the instantaneous power can provide active damping
for MC [29], [30]. And the reason for using the quantities
at (k+2)-thsampling time in (20) is to compensate the time
delay [18].

2) CONTROL OF OUTPUT CURRENTS

To achieve the desired output currents, the following cost FIGURE 3. Flowchart of the proposed FCS-MPC method.
function is considered.
2 Step 3: Calculate the output voltage uEko and input current Eiki
g2 = Eioref − Eik+2 (23)

o by (3), (4) and the current switching state S k , and then predict
the voltages and currents at the (k + 1)-th sampling period
where Eioref = Iom cos (ωo t − ϕo ) + jIom sin (ωo t − ϕo ) is the 0
by combining (5), (6) and (19), include Eik+1 s ,u Êk+1
s ,uÊs k + 1,
desired vector of output currents, ωo is the angular frequency
of output current. uEk+1
i , and Eik+1
o .
The resulting cost function should reflect both the quality Step 4: Repeat step 3 for all the 27 valid switching states
of source and output currents. Thus, the total cost function to predict the voltages and currents at the (k + 2)-th sampling
can be constructed as follows period.
Step 5: Evaluate the cost function (25) and apply the
F = g2 + λg1 (24) optimal switching state at the (k + 1)-th sampling period.
Then, repeat all the steps in the next switching time.
To facilitate the adjustment of the weighting factor λ, the
normalized total cost function is used.
IV. SIMULATION RESULTS
g2 g1
F̄ = + λ (25) In this Section, simulations have been carried out on MC sys-
E 2 E 2

i
oref i
sref tem to illustrate and validate the proposed method. Besides,
the comparisons with other FCS-MPC methods for unbal-
The weighting factor handles the relative importance of anced grids are given.
the two control objectives. Different choices of weighting
factor lead to different comprehensive performance. Usually, A. SIMULATIONS OF PROPOSED METHOD
the weighting factor selected by trial and error method [31]. Numerical simulations are carried out in Matlab/Simulink
platform. The related parameters are listed in Table 1.
E. IMPLEMENTATION PROCESS The proposed ESO, in which ωc is set to 1000π rad/s,
The implementation process of the proposed method is shown is tested under unbalanced source voltages (Usa = Usb =60V,
in Fig. 3. Usc = 40V). The waveforms of the estimated voltage ûsa ,
0
Step 1: Measure the source current Eiks , input voltage uEki and delayed voltage ûsa and estimated error ũsa are shown
0
output current Eiko , and estimate the source voltages uÊks and uÊs k. in Fig. 4. As seen, the estimated error is almost zero,
0 0
Step 2: Calculate the reference signals Eisref and Eioref ûsa has the same amplitude with ûsa , and ûsa lags ûsa
by (11) and (23). 90 degrees, which proves the feasibility of the ESO.

VOLUME 7, 2019 43899

 W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

TABLE 1. Parameters of the matrix converter system.

FIGURE 4. Waveforms of the unbalanced source voltages and the

estimated voltages by ESO.

When the amplitude and frequency of reference output

currents are set to 10A and 30Hz, respectively. The simula-
tion results under proposed method are shown in Fig. 5(a).
From top to bottom, they are source current isa and
source voltage usa , source currents isa , isb , isc , and output
currents ioA , ioB , ioC . As seen, the source currents are basi-
cally sinusoidal, the output currents are balanced, and almost
UPF is achieved.

B. SIMULATION COMPARISONS
The comparisons are performed among the following four
FCS-MPC methods: the proposed method, PS strategy [17],
IUPF strategy [17], [18], and APOC method [17].
In IUPF strategy, the source current reference is calculated
by making the conventional instantaneous power a constant.
The balanced output currents can be achieved, but the quality
of input current cannot be assured. In PS strategy, the positive FIGURE 5. Simulation results of three different FCS-MPC methods.
(a) Proposed method. (b) IUPF strategy. (c) PS strategy. (d) APOC strategy.
sequence component of source voltages is adopted to generate
the source current reference. The balanced source currents
could be obtained. However, the quality of output currents results of IUPF strategy is shown in Fig. 5(b), the balanced
is poor as the ac-term in output active power will exists output currents can be achieved but the source currents are
inevitably in this case. In APOC strategy, sinusoidal source distorted seriously. The waveforms of PS strategy are illus-
currents and balanced output currents could be achieved trated in Fig. 5(c). The balanced source and output currents
simultaneously. The proposed method and APOC are equiva- and almost UPF operation are achieved. Seen from Fig. 5(a)
lent in input-output characteristics. The difference between and Fig. 5(d), the APOC strategy and the proposed method
them lies in the methods to calculate the source current are basically the same, the sinusoidal source currents and
reference, which will be reflected in computational burden. balanced output currents are achieved.
The simulation results of above methods are shown The total harmonic distortions (THDs) of source currents
in Fig. 5. The uniform cost function (25) is adopted and the and output currents under the above methods are listed
weighting factor of four methods is set to 1.0. The simulation in Table 2. As demonstrated, the current quality under the

43900 VOLUME 7, 2019

W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

TABLE 2. THDs of source and output simulation waveforms under

different FCS-MPC methods.

FIGURE 7. Unbalanced source voltages.

FIGURE 6. Experimental setup.

proposed method and the APOC strategy are the same and
they have the best current quality, the quality of output cur-
rents under PS strategy is the worst, and the quality of source
currents under IUPF strategy is the worst.
FIGURE 8. Input and output waveforms under case I.

V. EXPERIMENTAL RESULTS
To validate the proposed control scheme, experiments have A. EXPERIMENTS OF PROPOSED METHOD
been carried out on the MC system. The experimental setup First, the feasibility of designed ESO is tested. By FFT
is shown in Fig. 6, which includes transformer, main cir- analysis, the THDs of source voltages under case II are
0
cuit, input LC filter, controller board, drive boards, clamp 3.37%, 3.58% and 4.95%. The voltages usa , ûsa , ûsa and ũsa
circuit, and a three-phase balanced RLload. The main circuit are shown in Fig. 7(b), usa and ûsa are coincident, and the
consists of nine bi-directional switches, which are realized estimated error is small and within ±3V. As a conclusion,
by the IGBT module FF200R12KT3_E. The main controller the designed ESO can track the source voltages well even
mainly includes: a floating-point digital signal processor though there exist some harmonics in the source voltages.
(DSP) TMS320F28335 and a field programmable gate array In case I, when the amplitude and frequency of output
(FPGA) EP2C8J144C8N. The DSP is responsible for imple- current are set to 10A and 30Hz respectively, the measured
menting the proposed method and transmitting the optimal input and output waveforms are shown in Fig. 8. As seen,
switching states to FPGA. The FPGA is used to generate the source currents and output currents are balanced. Addi-
the PWM signals to the drive boards. The related experi- tionally, the UPF operation is achieved.
mental parameters are the same with the parameters as listed In case II, two different reference output currents are tested:
in Table 1. fo = 30Hz, Iom = 10A and fo = 60Hz, Iom = 10A. The
The experiments are carried out in the following cases: power reference and amplitude of output current reference are
Case I : the source voltages are balanced, and Usa = Usb = the same under the two conditions, so the source currents will
Usc = 60V (RMS); be the same. The waveforms of source currents and calculated
Case II: the source voltages are unbalanced, and Usa = power are illustrated in Fig. 9, where the instantaneous power
Usb = 60V, Usc = 40V. is calculated by DSP, and attenuated by 200 times, then output
Case III: the source voltages are seriously unbalanced, and to the oscilloscope through the D/A converter chip AD5725.
Usa = Usb = 60V, Usc = 120V. The source currents are basically sinusoidal but unbalanced,

VOLUME 7, 2019 43901

 W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

FIGURE 9. Input waveforms of the proposed method under case II.

FIGURE 11. Waveforms with different reactive power settings under

case II (fo = 30Hz, Iom = 6A). (a) Qsref = 0Var; (b) Qsref = 400Var.

FIGURE 10. Output waveforms of the proposed method under case II.
(a) fo = 30Hz, Iom = 10A; (b) fo = 60Hz, Iom = 10A.

FIGURE 12. Experimental waveforms of the proposed method under case

Ps and Qs are controlled to be constant values without III (fo = 60Hz, Iom = 11A).
double-frequency ripple power. The phase shift between
source voltage usa and source current isa is almost zero, but
the conclusion that the system operates at UPF cannot be set to 0Var and 400Var respectively. As seen, the calculated
drawn. The reasons is that the phase-shift of each source cur- power are controlled to be constant values under two condi-
rent in Fig. 9 is not 120 degrees. In fact, Qs equals to zero rep- tions, which means the source currents are sinusoidal and the
resents that the sum of reactive powers of three input phases output currents are balanced. In addition, there is an apparent
is zero, rather than that the source current is in phase with the phase shift between usa and isa when Qsref = 400Var. It can
respective source voltage. Fig. 10 shows the waveforms of be concluded that the proposed method has the capability to
output phase-to-phase voltage and output currents, the output control the reactive power under the premise of ensuring the
currents under different conditions are balanced. Besides, quality of source and output currents.
it can be observed from Fig. 10(a) and Fig. 8 that the output Fig. 12 shows the waveforms in case III with fo = 60Hz,
currents are basically the same. It can be concluded that the Iom = 11A. As seen, the source currents are basically sinu-
balanced output currents, constant instantaneous active power soidal and the output currents are balanced, Ps and Qs are
and minimum instantaneous reactive power can be achieved constant, which confirm the effectiveness of the proposed
under unbalanced input voltages regardless of the output method again.
frequency, which verifies the effectiveness of the proposed In addition, a series of experiments are tested to show the
method. effects of efficiency settings on system performance under
Further, different reactive power references are performed case III with fo = 60Hz, Iom = 11A. The measured values of
by the proposed method under case II. In Fig. 11(a) and source currents and output currents under different efficiency
Fig. 11(b), the output currents are set to 6A/30Hz, Qsref are settings are listed in Table 3. It is observed that the differences

43902 VOLUME 7, 2019

W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

TABLE 3. Amplitude and THD Of source currents and output currents

with different efficiency values.

TABLE 4. THDs of source and output experimental waveforms under

different FCS-MPC methods.

of source current amplitude and output current amplitude are

not significant when the efficiency derivation over a large
interval. The THD of output currents are quite close under
different efficiency settings. As a conclusion, the efficiency
setting η has little effect on the quality of source and output
currents.

B. EXPERIMENTAL COMPARISONS
For comparison purpose, the waveforms of IUPF strategy,
PS strategy and APOC strategy under case II with fo =
30Hz, Iom = 10A are shown in Fig. 13(a), Fig. 13(b) and
Fig.13(c), respectively. From top to bottom, they are source
current isa and source voltage usa , source currents isa , isb , isc ,
conventional instantaneous power Ps and Qc (Qc = Im(Ei∗s uEs )),
output phase-to-phase voltage uAB and output currents ioA ,
ioB , ioC . As illustrated in Fig. 13(a), the output currents are
balanced but the source currents are distorted under IUPF
strategy. In addition, Ps and Qc are controlled to be constant
values, which means that the source currents’ quality are
sacrificed to assure the output currents’ quality. In Fig. 13(b),
the balanced source currents are also obtained, but the instan-
taneous active power and reactive power are fluctuant, so the
output current quality is relatively poor. Comparing Fig. 13(c)
with Fig. 9 and Fig. 10 (a), the source currents and output
currents of the APOC strategy and the proposed method are
basically the same. A clear distinction is that Qs is constant
but Qc is fluctuant.
To evaluate the harmonic performance of four FCS-MPC
methods in an intuitive way, the THDs of the source and FIGURE 13. Experimental waveforms of different FCS-MPC methods
under case II (fo = 30Hz, Iom = 10A). (a) IUPF strategy (b) PS strategy
output currents under case II are illustrated in Table 4. (c) APOC strategy.
As demonstrated, the current quality of APOC strategy and
the proposed method are basically the same and superior to The execution time required by the mentioned four
IUPF strategy and PS strategy, which is in accordance with FCS-MPC methods are listed in Table 5. The required exe-
the simulation results in Section IV (Table 2). cution time for IUPF strategy is the least, because its source

VOLUME 7, 2019 43903

 W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

TABLE 5. Execution time of different FCS-MPC methods. [6] F. Blaabjerg, D. Casadei, C. Klumpner, and M. Matteini, ‘‘Comparison of
two current modulation strategies for matrix converters under unbalanced
input voltage conditions,’’ IEEE Trans. Ind. Electron., vol. 49, no. 2,
pp. 289–296, Apr. 2002.
[7] X. Wang, H. Lin, H. She, and B. Feng, ‘‘A research on space vector
modulation strategy for matrix converter under abnormal input-voltage
conditions,’’ IEEE Trans. Ind. Electron., vol. 59, no. 1, pp. 93–104,
Jan. 2012.
[8] J. D. Dasika and M. Saeedifard, ‘‘An online modulation strategy to control
the matrix converter under unbalanced input conditions,’’ IEEE Trans.
current reference is calculated by the source voltages and Power Electron., vol. 30, no. 8, pp. 4423–4436, Aug. 2015.
expected power directly. As the positive sequence component [9] J. Lei, B. Zhou, J. Bian, X. Qin, and J. Wei, ‘‘A simple method for sinu-
soidal input currents of matrix converter under unbalanced input voltages,’’
of source voltages needs to be extracted, the PS strategy IEEE Trans. Power Electron., vol. 31, no. 1, pp. 21–25, Jan. 2016.
takes slightly longer execution time. It worth noting that [10] X. Liu, F. Blaabjerg, P. C. Loh, and P. Wang, ‘‘Carrier-based modulation
IUPF strategy and PS strategy need less computation time, strategy and its implementation for indirect matrix converter under unbal-
anced grid voltage conditions,’’ in Proc. Conf. EPE/PEMC, Novi Sad,
but their input and output performance are poor. The APOC Serbia, Sep. 2012, pp. LS6a.2-1–LS6a.2-7.
strategy has the excellent input and output current quality, but [11] M. Hamouda, H. F. Blanchette, and K. Al-Haddad, ‘‘Unity power factor
it takes the longest execution time. The reason is that both the operation of indirect matrix converter tied to unbalanced grid,’’ IEEE
Trans. Power Electron., vol. 31, no. 2, pp. 1095–1107, Feb. 2016.
positive and negative sequence component of source voltages
[12] J. Lei et al., ‘‘Feedback control strategy to eliminate the input current
are needed. The reference calculation process of the proposed harmonics of matrix converter under unbalanced input voltages,’’ IEEE
method is very simple. Thus, the total execution time for the Trans. Power Electron., vol. 32, no. 1, pp. 878–888, Jan. 2017.
proposed method is short even though an ESO is added. As a [13] I. Sato, J.-I. Itoh, H. Ohguchi, A. Odaka, and H. Mine, ‘‘An improvement
method of matrix converter drives under input voltage disturbances,’’ IEEE
conclusion, the proposed method can achieve good input and Trans. Power Electron., vol. 22, no. 1, pp. 132–138, Jan. 2007.
output current performance, save the system cost and reduce [14] J.-K. Kang, H. Hara, A. M. Hava, E. Yamamoto, E. Watanabe, and
the computational time. T. Kume, ‘‘The matrix converter drive performance under abnormal
input voltage conditions,’’ IEEE Trans. Power Electron., vol. 17, no. 5,
VI. CONCLUSION pp. 721–730, Sep. 2002.
[15] Y. Yan, H. An, T. Shi, and C. Xia, ‘‘Improved double line voltage synthesis
In this paper, a predictive control based on new reactive power of matrix converter for input current enhancement under unbalanced power
definition is presented for matrix converter under unbal- supply,’’ Power Electron., IET, vol. 6, no. 4, pp. 798–808, Apr. 2013.
anced source voltage conditions. Under this control scheme, [16] X. Li, M. Su, Y. Sun, H. Dan, and W. Xiong, ‘‘Modulation strategies based
the ac-term in active power could be eliminated, together on mathematical construction method for matrix converter under unbal-
anced input voltages,’’ IET Power Electron., vol. 6, no. 3, pp. 434–445,
with balanced output currents and sinusoidal source currents. Mar. 2013.
Moreover, the grid voltage sensors are no longer needed [17] C. Rojas et al., ‘‘Predictive control of a direct matrix converter operating
and the phase-locked loop and positive/negative sequence under an unbalanced AC source,’’ in Proc. IEEE ISIE, Bari, Italy, Jul. 2010,
pp. 3159–3164.
decomposition are not required, resulting in the reduction of [18] M. Rivera, J. Rodríguez, J. R. Espinoza, and H. Abu-Rub, ‘‘Instantaneous
system costs and computational burdens. In fact, it is a general reactive power minimization and current control for an indirect matrix
control strategy for MC under both balanced and unbalanced converter under a distorted AC supply,’’ IEEE Trans. Ind. Informat., vol. 8,
no. 3, pp. 482–490, Aug. 2012.
source voltage conditions. These characteristics make this
[19] T.-L. Nguyen and H.-H. Lee, ‘‘A predictive current control for coordi-
strategy more suitable for practical applications. nate control of current and power of matrix converter under unbalanced
Furthermore, the proposed method can be extended to input voltages,’’ in Proc. ICMWT, Kuala Lumpur, Malaysia, Jun. 2017,
other converters (e.g. three-phase voltage source rectifier pp. 528–538.
[20] S. Muller, U. Ammann, and S. Rees, ‘‘New time-discrete modulation
and three-phase current source rectifier) to improve the scheme for matrix converters,’’ IEEE Trans. Ind. Electron., vol. 52, no. 6,
input or output performance and reduce the system costs pp. 1607–1615, Dec. 2005.
under unbalanced input conditions. [21] M. Siami, D. A. Khaburi, M. Rivera, and J. Rodríguez, ‘‘An experimental
evaluation of predictive current control and predictive torque control for
REFERENCES a PMSM fed by a matrix converter,’’ IEEE Trans. Ind. Electron., vol. 64,
no. 11, pp. 8459–8471, Nov. 2017.
[1] P. W. Wheeler, J. Rodriguez, J. C. Clare, L. Empringham, and A. Weinstein,
[22] R. Vargas, U. Ammann, J. Rodriguez, and J. Pontt, ‘‘Predictive strategy to
‘‘Matrix converters: A technology review,’’ IEEE Trans. Ind. Electron.,
control common-mode voltage in loads fed by matrix converters,’’ IEEE
vol. 49, no. 2, pp. 276–288, Apr. 2002.
Trans. Ind. Electron., vol. 55, no. 12, pp. 4372–4380, Dec. 2008.
[2] D. Casadei, G. Serra, A. Tani, and L. Zarri, ‘‘Matrix converter modulation
strategies: A new general approach based on space-vector representation of [23] R. Vargas, U. Ammann, and J. Rodríguez, ‘‘Predictive approach to increase
the switch state,’’ IEEE Trans. Ind. Electron., vol. 49, no. 2, pp. 370–381, efficiency and reduce switching losses on matrix converters,’’ IEEE Trans.
Apr. 2002. Power Electron., vol. 24, no. 4, pp. 894–902, Apr. 2009.
[3] J. Rodríguez, M. Rivera, J. W. Kolar, and P. W. Wheeler, ‘‘A review of [24] T. Peng et al., ‘‘Open-switch fault diagnosis and fault tolerant for matrix
control and modulation methods for matrix converters,’’ IEEE Trans. Ind. converter with finite control set-model predictive control,’’ IEEE Trans.
Electron., vol. 59, no. 1, pp. 58–70, Jan. 2012. Ind. Electron., vol. 63, no. 9, pp. 5953–5963, Sep. 2016.
[4] P. Zanchetta, P. Wheeler, L. Empringham, and J. Clare, ‘‘Design con- [25] Y. Komatsu and T. Kawabata, ‘‘A control method of active power filter in
trol and implementation of a three-phase utility power supply based on unsymmetrical and distorted voltage system,’’ in Proc. Conf. EPE, Sevilla,
the matrix converter,’’ IET Power Electron., vol. 2, no. 2, pp. 156–162, Spain, Sep. 1995, pp. 904–907.
Mar. 2009. [26] Y. Suh and T. A. Lipo, ‘‘Modeling and analysis of instantaneous active
[5] D. Casadei, G. Serra, and A. Tani, ‘‘Reduction of the input current har- and reactive power for PWM AC/DC converter under generalized unbal-
monic content in matrix converters under input/output unbalance,’’ IEEE anced network,’’ IEEE Trans. Power Del., vol. 21, no. 3, pp. 1530–1540,
Trans. Ind. Electron., vol. 45, no. 3, pp. 401–411, Jun. 1998. Jul. 2006.

43904 VOLUME 7, 2019

W. Xiong et al.: Cost-Effective and Low-Complexity Predictive Control for MCs Under Unbalanced Grid Voltage Conditions

[27] Y. Zhang and C. Qu, ‘‘Model predictive direct power control of PWM rec- HANBING DAN was born in Hubei, China,
tifiers under unbalanced network conditions,’’ IEEE Trans. Ind. Electron., in 1991. He received the B.S. degree in automation
vol. 62, no. 7, pp. 4011–4022, Jul. 2015. and the Ph.D. degree in control science and engi-
[28] Y. Zhang, J. Liu, H. Yang, and J. Gao, ‘‘Direct power control of pulsewidth neering from Central South University, Changsha,
modulated rectifiers without DC voltage oscillations under unbalanced grid China, in 2012, and 2017, respectively.
conditions,’’ IEEE Trans. Ind. Electron., vol. 65, no. 10, pp. 7900–7910, He was a Visiting Researcher with the Faculty
Oct. 2018. of Engineering, University of Nottingham, U.K.,
[29] M. Rivera et al., ‘‘Imposed sinusoidal source and load currents for an
from 2016 to 2017. He is currently an Associate
indirect matrix converter,’’ IEEE Trans. Ind. Electron., vol. 59, no. 9,
Professor with the School of Automation, Cen-
pp. 3427–3435, Sep. 2012.
[30] M. Rivera, C. Rojas, J. Rodriguez, and J. Espinoza, ‘‘Methods of source tral South University, China. His research interests
current reference generation for predictive control in a direct matrix con- include matrix converter, finite control set-model predictive control, motor
verter,’’ IET Power Electron., vol. 6, no. 5, pp. 894–901, 2013. control, fault diagnosis and fault tolerant of power converter, and wireless
[31] P. Cortes et al., ‘‘Guidelines for weighting factors design in model pre- power transfer.
dictive control of power converters and drives,’’ in Proc. IEEE ICIT,
Gippsland, VIC, Australia, Feb. 2009, pp. 1–7. MARCO RIVERA (S’09–M’11–SM’17) received
the B.Sc. degree in electronics engineering and the
M.Sc. degree in electrical engineering from the
Universidad de Concepcion, Chile, in 2007 and
2008, respectively, and the Ph.D. degree from the
WENJING XIONG was born in Hunan, China,
Department of Electronics Engineering, Univer-
in 1991. She received the B.S. degree in automa-
sidad Tecnica Federico Santa Maria, Valparaíso,
tion and the Ph.D. degree in control science
Chile, in 2011, with a scholarship from the Chilean
and engineering from Central South University,
Research Fund CONICYT.
Changsha, China, in 2012 and 2017, respectively,
From 2011 to 2012, he held a postdoctoral posi-
where she is currently a Lecturer with the School
tion and was a part-time Professor of digital signal processors and industrial
of Automation.
electronics with Universidad Tecnica Federico Santa Maria. He is currently
Her research interests include matrix converter
an Associate Professor with the Faculty of Engineering, Universidad de
and ac/dc converter.
Talca, Curicó, Chile. His research interests include matrix converters, predic-
tive and digital controls for high-power drives, four-leg converters, renewable
energies, and development of high-performance control platforms based on
field-programmable gate arrays.
Dr. Rivera is a member of the Advisory Council of the Civil Society for
YAO SUN (M’13) was born in Hunan, China, the Ministry of Energy. He received a scholarship from the Marie Curie Host
in 1981. He received the B.S., M.S., and Ph.D. Fellowships for early stage research training in electrical energy conversion
degrees from the School of Information Sci- and conditioning technology at the University College Cork, Cork, Ireland,
ence and Engineering, Central South University, in 2008. In 2012, he received the Chilean Academy of Science Doctoral
Changsha, China, in 2004, 2007, and 2010, respec- Thesis Award (Premio Tesis de Doctorado Academia Chilena de Ciencias)
tively, where he has been a Professor with the for the Best Ph.D. Thesis published, in 2011, selected from among all
School of Automation. national and international students enrolled in any exact or natural sciences
His research interests include matrix con- program in Chile, and the Outstanding Engineer, in 2015.
verter, and micro-grid and wind energy conversion
systems. JOSEP M. GUERRERO (S’01–M’04–SM’08–
F’15) received the B.S. degree in telecommunica-
tions engineering, the M.S. degree in electronics
engineering, and the Ph.D. degree in power elec-
JIANHENG LIN was born in Fujian, China, tronics from the Technical University of Catalonia,
in 1994. He received the B.S. degree in elec- Barcelona, in 1997, 2000, and 2003, respectively.
tronic engineering from Jimei University, Xiamen, Since 2011, he has been a Full Professor with
China, in 2016. He is currently pursuing the Ph.D. the Department of Energy Technology, Aalborg
degree in control science and engineering with University, Denmark, where he is also responsible
Central South University, Changsha. for the Microgrid Research Program. Since 2014,
His research interests include linear time peri- he has been the Chair Professor with Shandong University. Since 2015,
odic system analysis, and matrix converter. he has been a Distinguished Guest Professor with Hunan University. Since
2016, he has been a Visiting Professor Fellow with Aston University, U.K.,
and a Guest Professor with the Nanjing University of Posts and Telecom-
munications. He has published more than 500 journal papers in the fields
of microgrids and renewable energy systems, which are cited more than
30 000 times. His research interests include different microgrid aspects,
including power electronics, distributed energy-storage systems, hierarchical
MEI SU was born in Hunan, China, in 1967. She and cooperative control, energy management systems, smart metering and
received the B.S. degree in automation, and the the Internet of Things for AC/DC microgrid clusters, and islanded minigrids,
M.S. and Ph.D. degrees in electric engineering recently specially focused on maritime microgrids for electrical ships, ves-
from the School of Information Science and Engi- sels, ferries, and seaports. He received the Best Paper Award of the IEEE
neering, Central South University, in 1989, 1992, TRANSACTIONS ON ENERGY CONVERSION for the period 2014–2015 and the Best
and 2005, respectively, where she has been a Pro- Paper Prize of the IEEE-PES, in 2015. He also received the Best Paper Award
fessor, since 2005. She is currently an Associate of the Journal of Power Electronics, in 2016. During five consecutive years,
Dean of the School of Automation, Central South from 2014 to 2018, he was named Highly Cited Researcher by the Clarivate
University. Analytics (formerly Thomson Reuters). In 2015, he was elevated as an IEEE
Her research interests include matrix converter, Fellow for his contributions on distributed power systems and microgrids.
adjustable-speed drives, and wind energy conversion systems. She is cur- He is also an Associate Editor for a number of IEEE TRANSACTIONS.
rently an Associate Editor of the IEEE TRANSACTIONS ON POWER ELECTRONICS.
VOLUME 7, 2019 43905