You are on page 1of 12

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO.

2, MAY 2014 991

A Simplified Finite-Control-Set Model-Predictive


Control for Power Converters
Changliang Xia, Senior Member, IEEE, Tao Liu, Tingna Shi, and Zhanfeng Song

Abstract—Finite-control-set model-predictive control (FCS- inclusion of nonlinear constraints [7]–[20]. FCS-MPC has de-
MPC) requires a large amount of calculation, which is an obstacle veloped rapidly in the past few years, and studies on this topic
for its application. However, compared with the classical linear have been spread out across various fields, for example, renew-
control algorithm, FCS-MPC requires a shorter control loop
cycle time to reach the same control performance. To resolve this
able energy, uninterruptible power supply (UPS) systems, and
contradiction, this paper presents an effective method to simplify electric motor drivers [21]–[27]. However, there exists an ob-
the conventional FCS-MPC. With equivalent transformation and vious challenge for the actual application of FCS-MPC.
specialized sector distribution method, the computation load of As is known, FCS-MPC undergoes no pulse-width modula-
FCS-MPC is greatly reduced while the control performance is tion (PWM) process, and it outputs only one switching state
not affected. The proposed method can be used in various circuit in each control loop cycle, so its switching frequency is un-
topologies and cases with multiple constraints. Experiments on
two-level converter and three-level NPC converter verify the good
fixed. To achieve the same current waveform quality, the con-
performance and application value of the proposed method. trol loop cycle time of FCS-MPC should be much shorter than
that of the conventional linear control methods [28], [29]. As
Index Terms—Finite-control-set model-predictive control
(FCS-MPC), power converter, sector distribution method, simpli-
a result, the time allowed to complete the FCS-MPC algorithm
fied algorithm. becomes so short that it is impractical in reality. For example,
in the case of a three-level neutral point clamped (NPC) con-
verter, the minimum time needed for completing the FCS-MPC
I. INTRODUCTION is approximately 52 s [30]; this is still longer than the control

O PTIMAL control of multiple objectives not only realizes loop cycle time required for achieving satisfying control perfor-
the high-quality and high-efficiency power conversion mances. Moreover, the amount of calculation will also increase
but also ensures the high reliability of the power converters. For with the complexity of circuit topology. The additional multiple
example, in the transformerless photovoltaic (PV) applications, constraints, such as balancing of the neutral point voltage and
the vibration suppression of the common-mode (CM) voltage vibration reduction of the common-mode voltage, will also in-
in power converters helps to reduce the current harmonics and crease the calculation amount of FCS-MPC. In summary, time-
power losses and to improve the system safety [1]–[4]. For consuming computation is an obstacle that limits the application
the medium-voltage power conversion system, the multilevel of FCS-MPC.
power-converter technique is proposed, and the balancing of In recent years, simplification of the FCS-MPC algorithm
the neutral point voltage is considered in order to obtain better has been proposed and discussed [31]–[35]. For example, one
control performance and enhance the reliability of the entire approach uses sector distribution on a source voltage vector
system [5], [6]. to reduce the number of candidate vectors in the prediction
The finite-control-set model-predictive control (FCS-MPC) process [31]. With this approach, the program running time can
can realize the optimal control of multiple objectives, and it has be significantly shortened but the control performance may be
been attracting growing interest due to many of its advantages, affected as the proposed simplified algorithm is not an exact
such as fast dynamic response, inherent decoupling, and easy equivalent to the original algorithm. Another approach, men-
tioned in hybrid hysteresis-SVM algorithms, shed lights also
on reducing the computational time of FCS-MPC. However,
Manuscript received March 02, 2013; revised July 05, 2013 and August 19,
2013; accepted September 27, 2013. Date of publication October 04, 2013;
there is a lack of extensive studies on this issue.
date of current version May 02, 2014. This work was supported in part by In this paper, an effective method for FCS-MPC algorithm
the National Key Basic Research Program of China (973 project) under Grant simplification is proposed to reduce the running time without
2013CB035602, by the National Natural Science Foundation of China under affecting the control performance. This method is divided into
Grant 51107084, and by the Key Technologies Research and Development Pro-
gram of Tianjin under Grant 13ZCZDGX01100. Paper no. TII-13-0125. two steps: the first step is to eliminate the calculation for cur-
C. Xia is with the School of Electrical Engineering and Automation, Tianjin rent prediction; the second step is to reduce the number of cost
University, Tianjin 300072, China, and also with Tianjin Key Laboratory of Ad- function calculations. With the above two steps, the calculation
vanced Technology of Electrical Engineering and Energy, Tianjin Polytechnic
University, Tianjin 300387, China (e-mail: motor@tju.edu.cn). time is shortened while the control performance is not affected.
T. Liu, T. Shi, and Z. Song are with the School of Electrical Engineering The proposed method has some versatility, and it is effective not
and Automation, Tianjin University, Tianjin 300072, China (e-mail: taoliu@tju. only in different circuit topologies but also in cases with mul-
edu.cn; tnshi@tju.edu.cn; zfsong@tju.edu.cn).
Color versions of one or more of the figures in this paper are available online
tiple constraints. Finally, the experiments on a two-level con-
at http://ieeexplore.ieee.org. verter and a three-level NPC converter prove the effectiveness
Digital Object Identifier 10.1109/TII.2013.2284558 of the proposed method.

1551-3203 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
992 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

TABLE I
VOLTAGE VECTORS

Fig. 1. Topological structure of the main circuit of a two-level power converter.

II. OPERATING PRINCIPLE OF CONVENTIONAL FCS-MPC

The computational demand required by the conventional


FCS-MPC for power converters is discussed with the two-level
power converter, shown in Fig. 1, where are the
ac-side source voltages, are the ac-side currents,
are the converter voltages, is the equivalent series
resistance, is the filter inductance; is the dc-link capacitor,
and are the switching states of three converter legs,
respectively.
For simplicity, the vectors are defined as follows:

Fig. 2. Schematic diagram of a conventional FCS-MPC scheme.


(1)

The voltage balance equation based on stationary frame where is the current predicted for time instant
is described as corresponding to the converter voltage acting from time
instant to time instant .
(2) To evaluate the predictions, a cost function is needed and
therefore defined as follows:
where is the vector of ac-side current, is the vector of ac-side
source voltage, and is the vector of converter voltage. (5)
To realize the current control strategy, the forward Euler al-
gorithm (or called standard Euler algorithm) is applied to (2), where is the reference current for time instant .
and the discrete-time expression of (2) is obtained as The realization of FCS-MPC is to find the switching state
, which will minimize the cost function defined in (5),
and then apply it to converter.
(3)
(6)
where is the step size from time instant to time instant ,
; is also the control loop cycle time. where is the combination of the switching states
For the existing error in the forward Euler method, (3) is the during the time interval , that is,
approximate expression of (2). .
Equation (3) says the current at time instant , Finally, for clarity, the following equation summarizes the
, is determined by the source voltage, converter voltage, and mathematic expression of conventional FCS-MPC applied to a
current at time instant , denoted with , , and , power converter:
respectively.
As known, the converter voltage can be any one of the
eight voltage vectors , listed in Table I, due to the (7)
fact that there are total eight switching states generated by the
converter.
The predicted current at time instant can be any one of
the following eight current vectors calculated according to (4): Fig. 2 shows the schematic diagram corresponding to (7).
In conclusion, a conventional FCS-MPC of two-level power
converters, described by (7) and Fig. 2, requires current pre-
(4) diction calculations and calculations of the cost function.
XIA et al.: SIMPLIFIED FINITE-CONTROL-SET MODEL-PREDICTIVE CONTROL FOR POWER CONVERTERS 993

Fig. 4. Schematic diagram of the single predictive FCS-MPC.

Fig. 3. “Required voltage vector” and space distribution of eight voltage vec- (10) and the conventional FCS-MPC (7) are equivalent. By sub-
tors in a two-level power converter.
stituting (4) into (10) and rearranging the equations, the simpli-
fied algorithm (10) can be transformed into
III. SIMPLIFIED FCS-MPC WITH SINGLE PREDICTION METHOD

A. Two-Level Power Converter (11)


As presented above, in the case of two-level power con-
verters, conventional FCS-MPC needs current prediction
calculations. The method proposed here in this paper avoids the where .
current predictions for time instant . Instead, it uses Comparing (11) and (7), one knows that the cost functions
the “required voltage vector” for prediction process. of the “single predictive FCS-MPC” and the conventional
The idea of conventional FCS-MPC is to select voltage vector FCS-MPC are different only by the positive multiplication
which makes the predicted current close to its refer- factor . Such a fact ensures that the performance of “single
ence . Considering the predicted current in (3), one can predictive FCS-MPC” will be the same as that of the conven-
obtain the determination of as follows. Replacing tional FCS-MPC.
with in (3) and making rearrangement of the equation, Again, similarly, Fig. 4 shows the schematic diagram of the
one has “single predictive FCS-MPC” given by (10).
The proposed algorithm reduces the predictions to one,
(8) and this method can also be applied to other circuit topologies.

Equation (8) says the current will be exactly equal B. Three-Level NPC Power Converter
to its reference if the converter voltage acting at time One of main advantages of FCS-MPC is in its ability to in-
instant can be managed to be the same as , calculated clude other constraints, for example, common-mode voltage re-
with (8). duction and neutral point voltage balancing, which will lead to
Fig. 3 shows the space distribution of the “required voltage the realization of the optimal control of multiple objectives. The
vector” along with the eight voltage vectors . general expression of the conventional FCS-MPC is shown as
Then, the current predictive control needs to identify one follows:
voltage vector among which meets the requirement
given below:

(9)
(12)
This means determining the switching state which should be where , , and are the predicted values of current con-
applied to power converter at time instant through identifying straint and additional constraints and , is the pre-
which voltage vector is the one nearest to the “required voltage dictive function of current constraint , and are
vector” . the predictive functions of additional constraints and , re-
Similarly, the following equation summarizes the mathematic spectively, , , and are the weight factors corresponding
description of the FCS-MPC proposed in this paper: to constraints , , and , and means the number of the th
voltage vector .
The general expression of “single predictive FCS-MPC” is
(10) shown as follows:

In this paper, the simplified algorithm described with (10) is


called “single predictive FCS-MPC.” With a simple derivation
process, it is easy to prove that the “single predictive FCS-MPC” (13)
994 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

The conventional FCS-MPC for three-level NPC converters


is described as

(15)
where is the constraint for the balancing of
neutral point voltage, which is necessary in case of three-level
converter circuits and is the weight factor correspondingly.
With the adjusting of , the impact of the neutral point voltage
balancing control can be enhanced or weakened as desired.
With the single prediction method, the “single predictive
FCS-MPC” for three-level NPC converters is obtained as

(16)
Comparing (16) with (15), one can see 27 current predic-
tions are eliminated through one time calculation of the “re-
quired voltage vector” . Compared with two-level con-
verters, the proposed FCS-MPC brings more significant time
savings in three-level NPC converter cases.
Fig. 5. (a) Circuit topology of a three-level NPC converter. (b) Voltage vectors
space distribution of the three-level NPC converter. IV. FURTHER SIMPLIFIED FCS-MPC WITH SECTOR
DISTRIBUTION METHOD

A. Two-Level Power Converter


where is the calculation equation of the “required
voltage vector” and is the weight factor . The “single predictive FCS-MPC” is further simplified by di-
The “single predictive FCS-MPC” is still effective under the viding the space of the eight voltage vectors into sections. As
condition that the additional constraints are considered. shown in Fig. 6(a), a shadowed hexagon region is defined, which
Taking the control of a three-level NPC converter as an ex- is formed with the perpendicular bisectors of the voltage vectors
ample, one has the circuit topology and the voltage vectors dis- . If falls into the shadowed hexagon region, zero
tribution of the three-level NPC converter shown in Fig. 5. In voltage vector or will be closest to . If falls
Fig. 5(a), the switching states of the converter are into sectors I VI rather than the shadowed hexagon region, the
defined as follows. The “1” state means the converter’s phase nonzero voltage vector located in the corresponding sector
leg is connected to positive bus bar (P), whereas the “ 1” state will be closest to .
means the converter’s phase leg is connected to the negative bus Finally, for clarity, steps are summarized below.
bar (N), the “0” state means the converter’s phase leg is con- 1) Calculate using (8);
nected to the neutral point (o) of the dc-link. In Fig. 5(b), to be 2) Identify the sector which falls into (In this paper, it is
concise, the switching states “1,” “0,” “ 1” are replaced with called the “chosen sector”. Approach used in space vector
symbols “ ”, “0,” “ ,” respectively. modulation (SVM) can be adopted for sector determination
In three-level NPC converter cases, the additional constraint [36]).
is the balancing of neutral point voltage. The deviation of the 3) Select either or if falls into the shadowed
neutral point voltage at is given as follows: hexagon region; otherwise, select the nonzero voltage
vector which locates inside “chosen sector”.
With sector distribution method adopted, the proposed
FCS-MPC eliminates not only current predictions but also
cost function evaluations required by the conventional
(14) FCS-MPC. To select the optimal voltage vector, only region
identification is required.
where , , and are the switching state of the Fig. 6(b) is for cases in which additional constraints are con-
corresponding phases at time instant , and is the de- sidered. As the calculation results of the cost function are deter-
viation of the neutral point voltage at time instant , which is mined by both current constraint and additional constraints, the
the difference between the voltages of the two capacitors voltage vectors which have high cost in every one of all con-
and , as shown in Fig. 5(a). straints will be excluded from the candidate voltage vectors be-
. fore the cost function evaluations. Therefore, the total number
XIA et al.: SIMPLIFIED FINITE-CONTROL-SET MODEL-PREDICTIVE CONTROL FOR POWER CONVERTERS 995

under steady state conditions). With the constraint on the


common-mode voltage included, the conventional FCS-MPC
for two-level converters is given as follows:

(18)
where is the common-mode voltage when the voltage
vector is selected. For example, is the common-mode
voltage when the voltage vector is selected.
The whole cost function in (18) takes the format of “current
constraint+invariable constraint.” With the sector distribution
shown in Fig. 6(b), each sector has two zero vectors (000, 111)
and one nonzero vector.
As for the current constraint, the vector which has the lowest
cost belongs to the “chosen sector”; as for the common-mode
voltage constraint, all of the nonzero vectors have the same cost.
Thus, in the final results of the cost function, the vectors outside
the “chosen sector” will impossibly have the lowest cost. The
candidate vectors are the zero vectors and the nonzero vector in
the “chosen sector.” As is known, the costs of two zero vectors
and are equal, and only one of them will be chosen as the
candidate vector. Then, the simplified FCS-MPC can be written
Fig. 6. Sector distribution of voltage vector space for further simplification of as
FCS-MPC algorithm. (a) Only with current constraint. (b) With current con-
straint and common-mode voltage constraint.

TABLE II
CANDIDATE VECTORS FOR EACH SECTOR
(19)
where is the number assigned to the nonzero vector which is
located inside the “chosen sector.”
Table II summarizes the candidate vectors for each sector.
In conclusion, the simplified FCS-MPC is still effective under
the condition that the constraint on the common-mode voltage
of the cost function calculations can be reduced and the running is considered. The simplified FCS-MPC needs calculations
time will be further shortened. of the cost function; while the conventional FCS-MPC requires
The additional constraints can be divided into two types. In eight-time calculations.
this paper, for any voltage vector, the constraints whose results
change with time are referred to as “variable constraint” in this B. Three-Level NPC Power Converter
paper; the constraints whose results do not change with time are
referred to as “invariable constraint.” The sector distribution method is also effective in three-level
For the “invariable constraint,” the corresponding cost of converter cases. Taking the three-level NPC converter as an ex-
each voltage vector is fixed, so the candidate voltage vectors of ample, the additional neutral point voltage constraint is a “vari-
the corresponding sector can be determined in advance. For the able constraint,” and the whole cost function takes the format of
“variable constraint,” the corresponding cost of each voltage “current constraint+variable constraint.” The sector distribution
vector is unfixed, so some candidate voltage vectors of the is similar to that in two-level converter cases, and it is shown in
corresponding sector cannot be determined in advance, and Fig. 7.
they can be determined only when additional calculations are The vector with the lowest cost in current constraint be-
added. longs to the “chosen sector,” but some vectors outside the
Taking common-mode voltage reduction as an example, in “chosen sector” may have lower cost in terms of neutral point
the case of two-level converters, the common-mode voltage cor- voltage constraint, possibly making them the optimal vectors.
responding to is calculated with In Fig. 7, the “chosen sector” is sector III, and the vectors
“ ” belong to this sector. There-
(17) fore, these eight voltage vectors are included into the pool of
candidate vectors. The final cost function results of vectors
It can be seen that the common-mode voltage constraint “ ” may be lower than any vector which belongs
is an “invariable constraint”, because it is only related to the to sector III, so these vectors are also included into the pool of
switching state (the dc-link voltage is almost constant candidate vectors and therefore the shadowed hexagon region is
996 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

TABLE III
CANDIDATE VECTORS FOR EACH SECTOR

Fig. 7. Neutral point voltage influence of all 27 voltage vectors. The vectors
with the same color have the same cost in terms of the neutral point voltage
constraint.

formed. For any vector which does not belong to the shadowed
Fig. 8. Candidate vectors (vectors with dark background) corresponding to
hexagon region, , there is always a vector which belongs sector III when the neutral point voltage constraint is considered.
to the shadowed hexagon region, , making the final cost
function result of less than that of . Therefore, TABLE IV
can be excluded from the candidate vector pool. For example, SMALL VECTORS SELECTED BY EACH SECTOR
for , there is a which gives the same result
in terms of the neutral point voltage constraint. As has
higher cost than in terms of the current constraint. Therefore,
the vector cannot be the optimal vector and is excluded
from the pool of candidate vectors. Similarly, for ,
one has , it can be seen that, in terms of the current
constraint, has higher cost than , while in terms of the
neutral point voltage constraint, the cost of is equal to that
of . Therefore, the vector cannot be the optimal vector
and is excluded from the pool of candidate vectors. At last, all
of the vectors which do not belong to the shadowed hexagon
region are excluded from the pool of candidate vectors. So the When Section III is the “chosen sector,” the small vectors
final candidate vectors corresponding to sector III are those “ ” in the shadowed hexagon region have the same cost
vectors which belong to the shadowed hexagon region. under the current constraint, but different cost under the neutral
The further simplified FCS-MPC for three-level NPC con- point voltage constraint. It is also the same with “ ”
verters with the neutral point voltage constraint considered is and “ .” With simple additional calculations, three small
described as vectors can be excluded from the candidate vectors. Under
the condition that “ ; ;
”, candidate vectors corresponding to
Section III are given in Fig. 8.
When is located in other sectors, the small vectors se-
lected by these sectors are shown in Table IV.
Therefore, the number of voltage vectors to be calculated and
(20) compared with the cost function is reduced to 7, which will
greatly shorten the running time of FCS-MPC.
where means the numbers assigned to the candidate Under the condition that the constraint on the common-mode
vectors which belong to the shadowed hexagon region which is voltage is considered, the cost function takes the format of
determined by the “chosen sector.” “current constraint+variable constraint+invariable constraint.”
The candidate vectors for each sector are shown in Table III. In the case of three-level NPC converters, the common-mode
In addition, the zero vectors have the same cost, so only one voltage constraint is , and can be calculated
of them is included in the candidate vectors. by (17) with , . The conventional
XIA et al.: SIMPLIFIED FINITE-CONTROL-SET MODEL-PREDICTIVE CONTROL FOR POWER CONVERTERS 997

Fig. 10. Candidate vectors (vectors with dark background) corresponding to


sector III when the common-mode voltage constraint and the neutral point
voltage constraint are considered.

Fig. 9. Cost of 27 voltage vectors in terms of the common-mode voltage con-


straint. The vectors with the same color play the same cost.

FCS-MPC algorithm with the constraint on the common-mode


voltage added is described as

Fig. 11. Platform of the two-level converter.


(21)

Analysis similar to what was explained for cases of “cur- where means the numbers assigned to the candi-
rent constraint+neutral point voltage constraint” has been date vectors which belong to the vector set which is determined
performed. Fig. 9 shows the common-mode voltage in- by the “chosen sector”.
fluence of each voltage vector. When is located in It is important to note that the conclusion given above
sector III, with carefully analysis of every voltage vector (reducing the candidate vectors to ten), as shown in Fig. 10,
in Fig. 7 and Fig. 9, It tells only and can compete is achieved based on the assumption that the common-mode
with the vectors in the shadowed hexagon region. There- voltage constraint plays the lowest impact on the cost function
fore, the candidate vectors are those voltage vectors linked result. Such an assumption introduces an implicit priority
with the bold lines. (For sector III, the candidate vectors are among the control objectives; and therefore the proposed
“ ”). FCS-MPC in (22) is approximately equivalent to the conven-
In reality, the weight factor of the common-mode voltage con- tional FCS-MPC. To ensure the proposed FCS-MPC exactly
straint is a small number that makes the current constraint equivalent to the conventional FCS-MPC, two more voltage
to be a dominant term of the cost function. For example, com- vectors need to be added back to the candidate vectors. For
pared with the vectors at the shadowed hexagon region, vectors example, in Section III, vectors “ ” should be kept to be
“ ” are far away from , this makes their cost much the candidate vectors (as shown in Fig. 9). consequently, the
higher than those vectors at the shadowed hexagon region in number of candidate vectors is increased to 12.
terms of the current constraint. Therefore, vectors and
can be excluded from the candidate vectors. Finally, the number V. EXAMINATION VIA EXPERIMENT
of candidate vectors is reduced to ten. Fig. 10 shows the ten The experimental platform is developed using a TI floating-
voltage vectors selected. point digital signal processor (DSP) TMS320F28335. In the ex-
With the inclusion of common-mode voltage constraint, the periment, time-delay compensation [29] is applied to both the
further simplified FCS-MPC is described as conventional FCS-MPC and the simplified FCS-MPC.

A. Two-Level Power Converter


The platform of the two-level converter (with the circuit
topology shown in Fig. 1) is shown in Fig. 11, and the system pa-
rameters are as follows: source line voltage 92 V (rms);
dc-link voltage 160 V; filter inductance 5.0 mH;
(22) equivalent series resistance 1.2 ; dc-link capacitor
998 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

Fig. 13. Comparative experimental results of the two algorithms when the
common-mode voltage constraint is added. (a) Current waveforms ( , and
, 2 A/div) and common-mode voltage ( , 50 V/div) obtained with the
conventional FCS-MPC. (b) Current waveforms ( , and , 2 A/div) and
common-mode voltage ( , 50 V/div) obtained with the sector distribution
based FCS-MPC.

Fig. 12. Comparative experimental results of the conventional FCS-MPC and


the sector distribution based FCS-MPC. (a) Waveform of current (2 A/div)
and its THD. (b) Transient response of three-phase currents , and
(2 A/div).

2200 F; control loop cycle time 33 s; equivalent


load resistance .
Fig. 12 shows the comparative experimental results of Fig. 14. Comparison between the running time required by the conventional
the conventional FCS-MPC and the simplified FCS-MPC FCS-MPC and the simplified FCS-MPC in two-level converter cases. (a) Con-
with sector distribution method. Fig. 12(a) shows that the ventional FCS-MPC. (b) Simplified FCS-MPC with single prediction method.
(c) Simplified FCS-MPC with sector distribution method. In the figures, (1) rep-
steady-state results of the two algorithms are similar to each resents the A/D conversion, (2) represents other algorithms, (3) represents the
other in current waveforms and corresponding total harmonic prediction process, (4) represents the calculation and comparison of cost func-
distortion (THD). Fig. 12(b) shows that the transient responses tion, and (5) represents the free time.
of the two algorithms are similar to each other when the current
amplitude steps from 3 to 6 A. The rising time is about 0.2 ms.
With the inclusion of common-mode voltage constraint, The comparison between the running time required by the
the comparative experimental results of the conventional conventional FCS-MPC and the simplified FCS-MPC is shown
FCS-MPC and the simplified FCS-MPC with sector distribu- in Fig. 14 and Fig. 15. The column chart is used in Fig. 16 to fa-
tion method are shown in Fig. 13. It can be seen that, when the cilitate comparison. Compared with the conventional algorithm,
weight factor changes from 0.0 to 0.1, the suppression the simplified algorithms significantly reduce the running time
effects of the common-mode voltage resulted by these two needed for the prediction and cost function calculation. The run-
algorithms are basically the same. As time goes on, the current ning time of the sector distribution based FCS-MPC is much
harmonics increase, and the high frequency components of shorter than that of the single predictive FCS-MPC. Also, it can
the currents resulted by these two algorithms are basically the be seen that the simplified algorithm is still effective in cases
same. that the additional constraints are added.
XIA et al.: SIMPLIFIED FINITE-CONTROL-SET MODEL-PREDICTIVE CONTROL FOR POWER CONVERTERS 999

Fig. 15. For the case that the common-mode voltage constraint is consid-
ered, the comparison between the running time required by the conventional
FCS-MPC and the simplified FCS-MPC in two-level converter cases. (a) Con-
ventional FCS-MPC. (b) Simplified FCS-MPC with single prediction method.
(c) Simplified FCS-MPC with sector distribution method. In the figures, (1)
represents the A/D conversion, (2) represents other algorithms, (3) represents
the prediction process, (4) represents the calculation and comparison of cost
function, and (5) represents the free time.

Fig. 18. Comparative experimental results of the conventional FCS-MPC and


the sector distribution based FCS-MPC for three-level NPC converter. (a) Wave-
form of current (3 A/div). (b) three-level output voltage (40 V/div).
Fig. 16. Comparison between the running time required by the conventional
FCS-MPC and the simplified FCS-MPC in two-level converter cases. Algo-
rithm A: conventional FCS-MPC; Algorithm B: simplified FCS-MPC with
single prediction method; Algorithm C: simplified FCS-MPC with sector For the case that constraint on the balancing of the neutral
distribution method. (a) Without common-mode voltage constraint. (b) With point voltage is considered, the experimental comparison
common-mode voltage constraint. of the conventional FCS-MPC and the sector distribution
based FCS-MPC is shown in Fig. 18. It can be seen that the
steady-state results of the two algorithms are similar to each
other in current waveforms, current THD and three-level output
voltage waveforms.
For cases that multiple additional constraints are considered,
Fig. 19 shows the comparison of the conventional FCS-MPC
and the sector distribution based FCS-MPC with both the neu-
tral point voltage constraint and the common-mode voltage con-
straint considered. When the weight factor rises from 0.0
to 0.1, the common-mode voltage is gradually reduced,
and the fluctuation of neutral point voltage increases cor-
respondingly. This shows that there is a competition effect be-
tween the two additional constraints, and this effect has basi-
cally the same impact on the two algorithms.
Fig. 17. The platform of three-level NPC converter.
The comparison of running time is shown in Figs. 20–22.
With the increased number of voltage vectors, the simplified
B. Three-Level NPC Power Converter algorithms for three-level NPC converter can save more running
time in the prediction and cost function processes. It can be seen
The platform of three-level NPC converter (with the circuit that the simplified algorithm is also effective with the addition
topology shown in Fig. 5(a)) is shown in Fig. 17, and the system of multiple constraints.
parameters are: Source line voltage 92 V (rms); dc-link
voltage 160 V; filter inductance 5.0 mH; equiva-
VI. CONCLUSION
lent series resistance ; dc-link capacitor
5600 F; control loop cycle time 50 s; equivalent load A simplified FCS-MPC algorithm for power converters is
resistance . proposed to reduce the computational load posed by the conven-
1000 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

Fig. 21. For the case that the neutral point voltage constraint and the common-
mode voltage constraint are considered, the comparison between the running
time required by the conventional FCS-MPC and the simplified FCS-MPC in
three-level NPC converter cases. (a) Conventional FCS-MPC. (b) Simplified
FCS-MPC with single prediction method. (c) Simplified FCS-MPC with sector
distribution method. In the figures, (1) represents the A/D conversion, (2) repre-
sents other algorithms, (3) represents the prediction process, (4) represents the
calculation and comparison of cost function, and (5) represents the free time.

Fig. 19. Comparative experimental results of the two algorithms with the
Fig. 22. Comparison between the running time required by the conventional
inclusion of neutral point voltage constraint and common-mode voltage
FCS-MPC and the simplified FCS-MPC in three-level NPC converter cases.
constraint. (a) Current waveforms ( , and ), neutral point voltage
Algorithm A: conventional FCS-MPC; Algorithm B: simplified FCS-MPC with
and common-mode voltage ( , 25 V/div) resulted with the conventional
single prediction method; Algorithm C: simplified FCS-MPC with sector distri-
FCS-MPC. (b) Current waveforms ( , and ), neutral point voltage
bution method. (a) Only with the neutral point voltage constraint; (b) With the
and common-mode voltage ( , 25 V/div) resulted with the sector
neutral point voltage constraint and common-mode voltage constraint.
distribution based FCS-MPC.

process of current prediction, and the FCS-MPC with sector dis-


tribution method can significantly reduce the running time re-
quired by cost function calculation and comparison. The per-
formance of the simplified FCS-MPC can be maintained to be
the same as the performance of the conventional FCS-MPC
through appropriately adjusting the weight factors, which has
been demonstrated with both theoretical derivation and exper-
imental examination. The simplified method is applicable to
both two-level converter circuits and three-level converter cir-
cuits, without or with additional constraint considered. With the
simplified algorithm, the FCS-MPC can be implemented with
low-cost processors and more room can be made available for
Fig. 20. For the case that the neutral point voltage constraint is considered, the other algorithms required by the control.
comparison between the running time required by the conventional FCS-MPC
and the simplified FCS-MPC in three-level NPC converter cases. (a) Conven-
tional FCS-MPC. (b) Simplified FCS-MPC with single prediction method. (c) REFERENCES
Simplified FCS-MPC with sector distribution method. In the figures, (1) rep-
resents the A/D conversion, (2) represents other algorithms, (3) represents the [1] M. Cavalcanti, K. C. D. Oliveria, A. M. de Farias, F. A. S. Neves, G. M.
prediction process, (4) represents the calculation and comparison of cost func- S. Azevedo, and F. C. Camboim, “Modulation techniques to eliminate
tion, and (5) represents the free time. leakage currents in transformerless three-phase photovoltaic systems,”
IEEE Trans. Ind. Electron., vol. 57, no. 4, pp. 1360–1368, Apr. 2010.
[2] G. Buticchi, D. Barater, E. Lorenzani, and G. Franceschini, “Digital
control of actual grid-connected converters for ground leakage current
tional FCS-MPC. Compared with the conventional algorithm, reduction in PV transformerless systems,” IEEE Trans. Ind. Inf., vol.
the FCS-MPC with single prediction method helps to skip the 8, no. 3, pp. 563–572, Aug. 2012.
XIA et al.: SIMPLIFIED FINITE-CONTROL-SET MODEL-PREDICTIVE CONTROL FOR POWER CONVERTERS 1001

[3] C.-C. Hou, C.-C. Shih, P.-T. Cheng, and A. M. Hava, “Common-mode [24] M. Rivera, V. Yaramasu, A. Llor, J. Rodriguez, B. Wu, and M. Fadel,
voltage reduction pulse width modulation techniques for three-phase “Digital predictive current control of a three-phase four-leg inverter,”
grid-connected converters,” IEEE Trans. Power Electron, vol. 28, no. IEEE Trans. Ind. Electron., vol. 60, no. 11, pp. 4903–4912, Nov. 2013.
4, pp. 1971–1979, Apr. 2013. [25] M. Rivera, J. Rodriguez, J. R. Espinoza, T. Friedli, J. W. Kolar, A.
[4] F. Bradaschia, M. C. Cavalcanti, P. E. P. Ferraz, F. A. S. Neves, E. C. Wilson, and C. A. Rojas, “Imposed sinusoidal source and load currents
dos Santos, and J. H. G. M. da Silva, “Modulation for three-phase trans- for an indirect matrix converter,” IEEE Trans. Ind. Electron., vol. 59,
formerless Z-source inverter to reduce leakage currents in photovoltaic no. 9, pp. 3427–3435, Sep. 2012.
systems,” IEEE Trans. Ind. Electron., vol. 58, no. 12, pp. 5385–5395, [26] P. E. Kakosimos and A. G. Kladas, “Implementation of photovoltaic
Dec. 2011. array MPPT through fixed step predictive control technique,” Renew-
[5] R. Portillo, S. Vazquez, J. I. Leon, M. M. Prats, and L. G. Franquelo, able Energy, vol. 36, no. 9, pp. 2508–2514, Sep. 2011.
“Model based adaptive direct power control for three-level NPC con- [27] R. P. Aguilera, P. Lezana, and D. E. Quevedo, “Finite-control-set
verters,” IEEE Trans. Ind. Inf., vol. 9, no. 2, pp. 1148–1157, May 2013. model predictive control with improved steady-state performance,”
[6] C. Xia, H. Shao, Y. Zhang, and X. He, “Adjustable proportional hy- IEEE Trans. Ind. Inf., vol. 9, no. 2, pp. 658–667, May 2013.
brid SVPWM strategy for neutral-point-clamped three-level inverters,” [28] J. Rodríguez, J. Pontt, C. Silva, P. Correa, P. Lezana, P. Cortés, and
IEEE Trans. Ind. Electron., vol. 60, no. 10, pp. 4234–4242, Oct. 2013. U. Ammann, “Predictive current control of a voltage source inverter,”
[7] M. Rivera, J. Rodriguez, J. Espinoza, and H. Abu-Rub, “Instantaneous IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495–503, Feb. 2007.
reactive power minimization and current control for an indirect matrix [29] P. Cortes, J. Rodriguez, C. Silva, and A. Flores, “Delay compensation
converter under a distorted AC-supply,” IEEE Trans. Ind. Inf., vol. 8, in model predictive current control of a three-phase inverter,” IEEE
no. 3, pp. 482–490, Aug. 2012. Trans. Ind. Electron., vol. 59, no. 2, pp. 1323–1325, Feb. 2012.
[8] P. Martin, O. Machado, E. Bueno, F. J. Rodriguez, and F. Meca, [30] R. Vargas, P. Cortes, U. Ammann, J. Rodriguez, and J. Pontt, “Pre-
“FPGA-based implementation of a predictive current controller for dictive control of a three-phase neutral-point-clamped inverter,” IEEE
power converters,” IEEE Trans. Ind. Inf., vol. 9, no. 3, pp. 1312–1321, Trans. Ind. Electron., vol. 54, no. 5, pp. 2697–2705, Oct. 2007.
Aug. 2013. [31] Y. L. Zhang and H. Lin, “Simplified model predictive current control
[9] J. Scoltock, T. Geyer, and U. K. Madawala, “A comparison of model method of voltage-source inverter,” in Proc. 8th ICPE & ECCE, 2011,
predictive control schemes for MV induction motor drives,” IEEE pp. 1726–1733.
Trans. Ind. Inf., vol. 9, no. 2, pp. 909–919, May 2013. [32] P. Cortes, A. Wilson, S. Kouro, J. Rodriguez, and H. Abu-Rub, “Model
[10] M. Preindl and S. Bolognani, “Model predictive direct torque con- predictive control of multilevel cascaded H-bridge inverters,” IEEE
trol with finite control set for PMSM drive systems part 1: Maximum Trans. Ind. Electron, vol. 57, no. 8, pp. 2691–2699, Aug. 2010.
torque per ampere operation,” IEEE Trans. Ind. Inf., vol. 9, no. 4, pp. [33] T. Benslimane, K. Aliouane, and B. Chetate, “Implementation of
1912–1921, Nov. 2013. SVPWM based on hysteresis control strategy applied on autonomous
[11] M. Preindl and S. Bolognani, “Model predictive direct torque control parallel active filter,” in Proc. 7th WSEAS Int. Conf. Autom. Control,
with finite control set for PMSM drive systems part 2: Field weakening Modeling and Simulation, 2005, pp. 309–313.
operation,” IEEE Trans. Ind. Inf., vol. 9, no. 2, pp. 648–657, May 2013. [34] D. Casadei, G. Serra, and A. Tani, “Implementation of a direct torque
[12] C. Xia, M. Wang, Z. Song, and T. Liu, “Robust model predictive control algorithm for induction motors based on discrete space vector
current control of three-phase voltage source PWM rectifier with modulation,” IEEE Trans. Power Electron., vol. 15, no. 4, pp. 769–777,
online disturbance observation,” IEEE Trans. Ind. Inf., vol. 8, no. 3, Jul. 2000.
pp. 459–471, Aug. 2012. [35] B. Singh and D. Goyal, “Improved DSVM-DTC based current sensor-
[13] M. A. Stephens, C. Manzie, and M. C. Good, “Model predictive control less permanent magnet synchronous motor drive,” in Proc. 7th PEDS,
for reference tracking on an industrial machine tool servo drive,” IEEE 2007, pp. 1354–1360.
Trans. Ind. Inf., vol. 9, no. 2, pp. 808–816, May 2013. [36] Z. Yu, “Space-vector PWM With TMS320C24x/F24x using hardware
[14] F. Kennel, D. Görges, and S. Liu, “Energy management for smart grids and software determined switching patterns,” Texas Instruments, Ap-
with electric vehicles based on hierarchical MPC,” IEEE Trans. Ind. plication Rep. SPRA524.
Inf., vol. 9, no. 3, pp. 1528–1537, Aug. 2013.
[15] T. J. Vyncke, S. Thielemans, and J. A. Melkebeek, “Finite-set model-
based predictive control for flying-capacitor converters: Cost function
design and efficient FPGA implementation,” IEEE Trans. Ind. Inf., vol.
9, no. 2, pp. 1113–1121, May 2013.
[16] D. du Toit, H. D. T. Mouton, R. Kennel, and P. Stolze, “Predictive Changliang Xia (M’08–SM’12) was born in Tianjin,
control of series stacked flying-capacitor active rectifiers,” IEEE Trans. China, in 1968. He received the B.S. degree from
Ind. Inf., vol. 9, no. 2, pp. 697–707, May 2013. Tianjin University, Tianjin, China, in 1990, and the
[17] P. Acuna, L. Moran, M. Rivera, J. Dixon, and J. Rodriguez, “Improved M.S. and Ph.D. degrees from Zhejiang University,
active power filter performance for renewable power generation sys- Hangzhou, China, in 1993 and 1995, respectively, all
tems,” IEEE Trans. Power Electron., vol. 29, no. 2, pp. 687–694, Feb. in electrical engineering.
2014. He is currently a Professor with the School of
[18] Z. Yan and J. Wang, “Model predictive control of nonlinear systems Electrical Engineering and Automation, Tianjin Uni-
with unmodeled dynamics based on feedforward and recurrent neural versity, Tianjin, China, and also with the Tianjin Key
networks,” IEEE Trans. Ind. Inf., vol. 8, no. 4, pp. 746–756, Nov. 2012. Laboratory of Advanced Technology of Electrical
[19] V. Yaramasu, M. Rivera, B. Wu, and J. Rodriguez, “Model predictive Engineering and Energy, Tianjin Polytechnic Uni-
current control of two-level four-leg inverters—Part I: Concept, algo- versity. In 2008, he became “Yangtze Fund Scholar” Distinguished Professor
rithm, simulation analysis,” IEEE Trans. Power Electron., vol. 28, no. and is currently supported by National Science Fund for Distinguished Young
7, pp. 3459–3468, Jul. 2013. Scholars. His research interests include electrical machines and their control
[20] M. Rivera, A. Wilson, C. A, Rojas, J. Rodriguez, J. R. Espinoza, P. W. systems, power electronics, and control of wind generators.
Wheeler, and L. Empringham, “A comparative assessment of model
predictive current control and space vector modulation in a direct ma-
trix converter,” IEEE Trans. Ind. Electron., vol. 60, no. 2, pp. 578–588,
Feb. 2013.
[21] J. Rodriguez, M. P. Kazmierkowski, J. R. Espinoza, P. Zanchetta, H.
Abu-Rub, H. A. Young, and C. A. Rojas, “State of the art of finite Tao Liu was born in Hebei, China, in 1984. He
control set model predictive control in power electronics,” IEEE Trans. received the B.S. degree from Tianjin University,
Ind. Inf., vol. 9, no. 2, pp. 1003–1016, May 2013. Tianjin, China, in 2008, where he is currently
[22] P. Cortes, G. Ortiz, J. I. Yuz, J. Rodriguez, S. Vazquez, and L. G. Fran- working toward the Ph.D. degree in electrical engi-
quelo, “Model predictive control of an inverter with output LC filter neering at the School of Electrical Engineering and
for UPS applications,” IEEE Trans. Ind. Electron., vol. 56, no. 6, pp. Automation.
1875–1883, Jun. 2009. His research interests include power electronics
[23] T. Geyer and S. Mastellone, “Model predictive direct torque control and control of wind generators.
of a five-level ANPC converter drive system,” IEEE Trans. Ind. Appl.,
vol. 48, no. 5, pp. 1565–1575, Sep. 2012.
1002 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 10, NO. 2, MAY 2014

Tingna Shi was born in Zhejiang, China, in 1969. Zhanfeng Song was born in Hebei, China, in 1982.
She received the B.S. and M.S. degrees from He received the B.S., M.S., and Ph.D. degrees from
Zhejiang University, Hangzhou, China, in 1991 Tianjin University, Tianjin, China in 2004, 2005, and
and 1996, respectively, and the Ph.D. degree from 2009, respectively, all in electrical engineering.
Tianjin University, Tianjin, China, in 2009. He is currently a Lecturer with the School of
She is currently a Professor with the School of Electrical Engineering and Automation, Tianjin Uni-
Electrical Engineering and Automation, Tianjin versity, Tianjin, China. His research interests include
University, Tianjin, China. Her research interests in- electrical machines and their control systems.
clude electrical machines and their control systems,
power electronics, and electric drives.

You might also like