Model Predictive Control of a Three-to-Five phase

Matrix Converter
SK. Moin Ahmed* Atif Iqbal Haitham Abu-Rub* Patricio Cortes**

Department of Electrical Engineering, College of Engineering, Qatar University, Qatar

*Department of Electrical & Computer Engineering, Texas A&M University at Qatar
**Departmento de Electronica, Universidad Tecnica Federico Santa Maria, Chile
[email protected] [email protected] [email protected] [email protected]

Abstract: The paper proposes model predictive control strategy to inherent discrete nature of the power converter switching
control active and reactive power simultaneously in addition to the states and the digital implementation [12].This paper describe
source and load current using predictive approach for a three-phase the application of the Predictive current control applied to a
input to five-phase output matrix converter. The proposed method multi-phase topology of direct matrix converter in addition to
selects the actuation states of the matrix converter according to the the active and reactive power control. The proposed control
optimization algorithm of the cost function. The proposed cost strategy achieve the objectives of the simultaneous input
function considers the active and reactive input power control to current, output current and active and reactive power control.
facilitate the control of source side power factor. Additionally source The predictive algorithm uses prediction horizon of unity and
and load side currents are controlled to be sinusoidal. The proposed evaluates at every sampling time Ts all the available two
control approach offer good tracking of the reference and actual hundred and forty-three possible switching states and apply the
source and load current and can control the source side power factor one that returns the minimum value for the formulated cost
to any value. The analytical and simulation approach is adopted in the functional in the next sampling instant. The novelty of the
paper. paper lies in the proposal of Model predictive control of a five-
phase output matrix converter.
Keywords: Model Predictive control, Current control, Power control,
multi-phase system, three-phase to five-phase matrix converter. II. DIRECT THREE-TO- FIVE-PHASE MATRIX CONVERTER

I. INTRODUCTION The power circuit topology of a direct matrix converter is

illustrated in Fig. 1 with three-phase input and five-phase
Direct Matrix converter is increasingly used in several output. The source is connected to the array of bidirectional
applications due to some inherent advantages which they power semiconductor switches through a suitable LC filter.
offered when compared to their counterpart such as back-to- The matrix converter directly connects the output phases to
back converter. Complex control strategies are employed to one of the input phases. The current after filtering (source side)
control the matrix converters, starting from classical Venturuni
is designated as is and the current at the converter side without
method [1] to carrier-based PWM [2-4] and space vector
PWM [5,6]. Among several control strategies Predictive filtering is denoted as iin . The source side and converter side
control becoming very popular for power electronic converters voltages are denoted as vs and vin , respectively. The output of
and has seen unprecedented growth in this area of research the matrix converter is connected to the passive R-L load
work in recent times [7,8]. Model Predictive control along with an active back emf component. The switching
techniques are mostly employed in slow processes in chemical functions of the bi-directional power switches are denoted as
and petrochemical due to slow predictions. Nevertheless, the Sij with i ∈ {a ,b ,c} (input phases) and j ∈ {A, B ,C , D , E}
advancement in digital signal processing has led to the
application of this control technique in power electronic (output phases). When the switching function is Sij = 1 , the
converters and drives [9-13]. Matrix converter is far more input phase ‘i' is connected the output phase ‘j’. The switching
complex than conventional AC/DC/AC converters. However, constrain is that the input should not be short-circuited and the
Predictive control technique is also applied to the matrix output side should not be open-circuited (due to inductive
converter topologies recently [9-11].This simple and powerful nature of the load); the following constraint equation holds
method is based upon the minimization of a cost/objective good
function that ascertain the future impact of system input
sequences at each sampling instant. The procedure is then Saj + Sbj + Scj = 1 ∀j ∈ {A, B,C , D , E} (1)
repeated at the next subsequent sampling instant. The
advantage of the Predictive control approach is that the
Hence with the switching constraints imposed by (1), are 243
process/system non-linearities and constraints can be included
valid switching states. The model of the matrix converter can
explicitly in the cost/objective function formulation [7,8].
then be written as;
Predictive Current Control can be viewed as a special case of
Model Predictive Control which takes into account the

978-1-4577-1914-1/11/$26.00 ©2011 IEEE 36

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 05,2023 at 04:50:52 UTC from IEEE Xplore. Restrictions apply.
 v o (k ) = T v s (k )
(2)
i s (k ) = T i o (k )

Where v o (k ) = [v A (k ) v B (k ) vC (k ) vD (k ) vE (k )]t ,

v s (k ) = [va (k ) vb (k ) vc (k )]t , i s (k ) = [ia (k ) ib (k ) ic (k )]t ,

i o (k ) = [i A (k ) iB (k ) iC (k ) iD (k ) iE (k )]t and

& S aA SbA S cA #
$S SbB S cB !!
$ aB Fig. 1. Three-to-five phase matrix converter power circuit topology.
T = $ S aC SbC S cC ! (3)
$ !
$ S aD SbD S cD ! III. PREDICTIVE CONTROL PRINCIPLE
$S SbE S cE !"
% aE The power converters have valid number of switching states,
e.g. eight switching states in case of a two-level voltage source
The dynamic model of the matrix converter can also be written inverter and twenty seven states in the three to three phase
using the filter capacitor voltage, filter inductor current and matrix converter and 243 in the investigated matrix converter
load current as; topology. The predictive control relies on applying suitable
switching state vectors based on the prediction of the behavior
of the converter. Hence, a quality or cost function is
d i s (t ) Rs 1
=− i s (t ) + (v s (t ) − vi (t )) (4) formulated that is minimized by means of predictions of
dt Ls Ls control variables using dynamic model of the controlled
variable. Thus two aspects are important, minimization of the
d vi 1 appropriate cost function and the accurate model of the
= (i (t ) − ii (t )) (5) controlled variables.
dt Cs s
The control objectives are;
1) Obtain precise control of the load current.
d i o (t ) Ro 1 2) Obtain sinusoidal source currents and unity power factor in
=− i o (t ) + v (t ) − E (t ) (6)
dt Lo Lo o the source side.
3) Obtain simultaneous active and reactive power control.
Meeting these control objectives will lead to reduced
Where Rs , Ls ,Cs are the input side filter parameters and
switching losses and improved efficiency of the system under
Ro , Lo , E are the load parameters. control.
The choice of the optimization function or cost function
The Load Model depended upon the control objectives set for the process or
system under investigation. An intelligent choice of the cost
The load is assumed as a five-phase RLE (Resistance, function leads to an optimum results meeting all the set control
inductance and back emf). The discrete time model of the load objectives. The cost function used in [13] is reproduced in
suitable for current prediction is obtained from [8]; equation (9) as;
T , RT ) g (k + 1) = iα*o − iαo + iβ* o − iβo + λ Qs* − Qs (9)
( )
iˆo (k + 1) = s v o (k ) − eˆ(k ) + i o (k )**1 − o s '' (7)
L + Lo (
Where,
Where Ro and Loare the resistance and inductance of the load,
Tsis the sampling interval, i is the load current space vector, v { }
Qs (k + 1) = Im g v s (k + 1)i s (k + 1) =
(10)
is the inverter voltage space vector used as a decision variable iαs (k + 1)v βs (k + 1) − iβ s (k + 1)vαs (k + 1)
and ê is the estimated back emf obtained from [8]; as
L , L ) The control objectives are to track the load current and to
eˆ(k ) = v o (k ) + i o (k − 1) − i o (k )** Ro + o '' (8)
achieve unity power factor at the source side. This is done by
Ts + Ts (
choosing zero reference reactive power. The tuning parameter
λ dictates the emphasis of control. The results obtained using
Where ê(k ) is the estimated value of e(k ) . However, for this scheme shows unity power factor control and good
simulation purpose the amplitude and frequency of back tracking of the load current. However, the ripple in the source
emf is assumed constant. current is high that lead to additional harmonic losses [13].

37

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 05,2023 at 04:50:52 UTC from IEEE Xplore. Restrictions apply.
 Another scheme suggested in [9] to achieve source side as well IV. SIMULATION RESULTS
as load side current control. The cost function chosen in this
Simulation program is developed in Matlab/Simulink platform
case is reproduced in equation (11) as
using‘s-function’ approach. The total cycle time
! !wassetat!!= 350 ", resulting in fs= 2857#$. The
g (k + 1) = i*αo − iαo + iβ* o − iβo modulation index is set to q=0.4 for all the simulations. The
(11) input voltage is set to 311V (peak per phase). The output
+ λ,* i*αs − iαs + iβ* s − iβ s )' current reference is set to 5 A initially and afterward it is
+ ( stepped up to 10 A. with the same input voltage. The load
parameters are set to R=10 •, L=10 mH. The input filter
The cost function thus formulated to achieve the control parameters are chosen as L =300!H , R =0.1 " and C =15!F.
f f f
objectives requires the computation of the source side current The simulation is done for the output frequencies set at 70 Hz
references. This is done using equation (12) which is obtained and 30 Hz. The simulation results are presented in Fig. 2-Fig.
keeping the power balance at the input and output side 3.Output reference and actual currents are plotted
(neglecting the losses in the converter); simultaneously to observe the error between the reference and
actual currents. Input current as well as source current with
P (k + 1) vs (k + 1)
is* (k + 1) = o 2
• 2
(12) source voltage are shown to validate the unity power factor
vs (k + 1) vs (k + 1) criteria. The output phase voltage (VA), adjacent line voltage
(VAB) and non-adjacent line voltage (VAC), are also plotted. A
The equation for output and input power obtained assuming good tracking of the actual and reference current is observed.
loss-less converter as; The achievable dynamic response of the current waveform is
extremely fast with rise time less than 1 ms. The source side
Ps (k + 1) = Po (k + 1) = current is highly improved when compared with the results of
[13]. The initial transient is also not observed in the waveform
, 2 * / )
( ) ( )
* Re1e k + 1 i o k + 1 . ' as that of [9]. The results for unity power factor is presented,
* 0 - ' (13) nevertheless, this can be varied as the reactive power is
R
* 2* * /' controlled. The voltages shows a typical response validating
* + Re1i o (k + 1)i o (k + 1). ' the control objectives.
+ 0 -(
25
10
The current response improves in [9] when compared to [13].
20
However, the initial transient is still not satisfactory in the 0
Output Ref and Actual Current (A)

waveform. This is due to the fact that the optimization function 15

is forcing the source side current to remain sinusoidal. -10
0.193 0.195 0.197
Nevertheless, the control of source side power factor is 10
missing in this cost function. Hence, a new cost function is
5
formulated and presented in this paper which incorporates the
current tracking as well as controlling active and reactive 0
power. The active power control leads to good source side
current tracking and reactive power control offers the -5
flexibility of controlling the source side power factor. This is
-10
highly important in grid connected converter system for wind
generation and other applications requiring power factor -15
0.15 0.2 0.25
control. The formulated cost function is given in equation (14) Time (s)
as; a.
12
5
Source Currents (A) and Phase 'a' Voltage (V/100)

g (k + 1) = i*αo − iαo + iβ* o − iβo + ixo + i yo + 10

0
(14) 8
λ Q*s − Qs + γ Ps* − Ps -5
6 0.19 0.195 0.2

The output current references are kept as sinusoidal, the 4

reactive power reference can be put to zero for unity power at 2

the source side and can assume some positive or negative
value depending upon the power factor requirement. The 0

active power control is achieved simultaneously and to -2

compute the reference active power, equation (13) can be
employed. It is shown in the next section that this cost function -4

leads to an optimum current tracking behavior along with -6

0.15 0.2 0.25
active and reactive power control of the system. Time (s)
b.

38

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 05,2023 at 04:50:52 UTC from IEEE Xplore. Restrictions apply.
 Phase A (V)

500
0
V. CONCLUSION
-500
0.15 0.2 0.25
The paper presents model predictive control approach to
control simultaneously, the source side current (to obtain
Non-Adjascent AC (V) Adjascent AB (V)

500
sinusoidal current), the load side current (actual should follow
0
the reference), and the active and reactive powers. The
-500 investigated topology of the matrix converter has three-phase
0.15 0.2 0.25 input and five-phase output. The control approach is based on
500 the prediction of the system behavior from the discrete load
model minimizing the cost function and selecting the
0 appropriate actuation state. The simulation results shown
-500
validate the achievement of the control objectives. Very good
0.15 0.2 0.25
tracking of load current is observed. The source current is
Time (s) sinusoidal and the input power factor is controllable.
c.
Fig. 2. 70 Hz operation: a. Output Currents. b. Input current , Source voltage ACKNOWLEDGEMENT
and current. c. Output voltages. This publication was made possible by an NPRP grant No. 08-
25
10 369-2-140, from the Qatar National Research Fund. The
20 statements made herein are solely the responsibility of the
0
authors.
Output Ref and Actual Current (A)

15
-10
10
0.192 0.198 REFERENCES

5
[1] A. Alesina, and M. Venturini, “Analysis and design of optimum
amplitude nine-switch direct ac-ac converters”, IEEE Trans. Power
0 Elect. vol. PE-4, no. 1, pp. 101-112, 1989.
[2] B. Wang, and G. Venkataramanan, “A carrier-based PWM algorithm for
-5 indirect matrix converters,” in Proc. IEEE-PESC 2006, pp. 2780–2787,
2006.
-10 [3] Young-Doo Yoon and Seung-Ki Sul: “Carrier-based modulation
technique for matrix converter,” IEEE Trans. Power Elect., vol. 21, no.
-15
0.15 0.2 0.25 6, pp. 1691–1703, November 2006.
Time (s)
[4] Poh C. Loh, R. Rong, F. Blaabjerg and P. Wang: “Digital carrier
a. Modulation and Sampling Issues of Matrix Converter”, IEEE Trans. On
6
Power Elect., vol. 24, no. 7, July 2009.
Source Currents (A) and Phase 'a' Voltage (V/100)

[5] D. Casadei, G.Grandi, G. Serra, and A.Tani, “Space vector control of

4
matrix converters with unity power factor and sinusoidal input/output
waveforms”, Proc. EPE Conf., vol. 7, pp. 170-175, 1993.
2
[6] L. Huber, and D. Borojevic, “Space vector modulated three-phase to
three-phase matrix converter with input power factor correction”, IEEE
0 Trans. Ind. Appl. Vol. 31, no. 6, pp. 1234-1246, Nov./Dec. 1995.
[7] P.W. Wheeler, P.W., Rodriguez, J., Clare, J,C., Empringham, L., and
-2 Weinstein, A: “Matrix Converters: A Technology Review”, IEEE Trans.
On Ind. Elect. vol. 49, no. 2, April, 2002, pp. 276-288.
-4 [8] P. Cortés, M. P. Kazmierkowski, R. M. Kennel,.D. E. Quevedo, and José
Rodríguez, “Predictive control in power electronics and drives”, IEEE
-6
0.15 0.2 0.25
Tran. On Ind. Electronics, vol. 55, no. 12, pp. 4312-4322, Dec. 2008.
Time (s) [9] René Vargas, José Rodríguez, Ulrich Ammann, and Patrick W. Wheeler,
b. “Predictive Current Control of an Induction Machine Fed by a Matrix
Converter With Reactive Power Control’’, IEEE Tran. On Ind.
Phase A (V)

500
Electronics, vol. 55, no. 12, pp. 4362-4371, Dec. 2008. the Predictive
0
DTC Based Matrix Converter under Unbalanced AC Supply’’, IEEE-
-500 IAS2007, pp. 202-207.
0.15 0.2 0.25 [10] C. Rojas, M. Rivera, J. Rodriguez, A. Wilson, J. Espinoza, F. Villarroel,
Adjascent AB (V)

500 P. Wheeler, ‘‘Predictive Control of a Direct Matrix Converter Operating

under an Unbalanced AC Source’’, Proc. IEEE Ind. Elec. Symposium,
0 ISIE, 4-7 July 2010, Bari, Italy, pp. 3159-3164.
[11] H. Abu-Rub, J. Guziñski, Z. Krzeminski, and H. A. Toliyat, ‘‘Predictive
-500
current control of voltage-source inverters,’’ IEEE Trans. Ind. Electron.,
0.15 0.2 0.25 vol. 51, no. 3, pp. 585---593, Jun. 2004.
Non-Adjascent AC (V)

500 [12] Rodriguez, J. Espinoza, M. Rivera, F. Villarroel and C. Rojas,

‘‘Predictive Control of Source and Load Currents in a Direct Matrix
0 Converter’’, Proc. IEEE ICIT, 4-7 March, 2010, Chile, pp. 1826-1831.
[13] J. Rodriguez, J. Pontt, C. A. Silva, P. Correa, P. Lezana, P. Cortes, and
-500 U. Ammann, “Predictive current control of a voltage source inverter,”
0.15 0.2 0.25 IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495–503, Feb. 2007.
Time (s)
c.
Fig. 3. 30 Hz operation of Matrix converter: a. Output Currents. b. Input
current , Source voltage and current. c. Output voltages.

39

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on March 05,2023 at 04:50:52 UTC from IEEE Xplore. Restrictions apply.