IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO.

5, MAY 2019 2775

Asymptotically Stable Predictive Control of

Grid-Connected Converter Based on Discrete
Space Vector Modulation
Kazi Saiful Alam , Student Member, IEEE, Md. Parvez Akter , Student Member, IEEE,
Dan Xiao , Member, IEEE, Daming Zhang , Member, IEEE,
and Muhammed Fazlur Rahman , Fellow, IEEE

Abstract—This paper presents an alternative implemen- unbalanced grid voltage condition. Besides, the proposed tech-
tation of the finite control set model predictive control (FCS- nique can reduce the effect of random sampling and thus im-
MPC) approach for a 3-Ø, two-level grid-connected power proves the quality of the input current. The proposed approach
converter. The presented approach combines the concept of
discrete space vector modulation into the FCS-MPC struc- also highlights the enhanced performance of the controller con-
ture to improve the input current quality of the power con- cerning the interference rejection capability. A cascade-free pre-
verter. Further, with the help of Lyapunov’s stability the- dictive control is proposed in [8] where the control problem is
ory, it is shown that the predictive control law used in the formulated considering dynamic references for directly control-
present analysis ensures the stability of the overall system. ling the real and reactive power of the converter. Besides, the
However, the additional virtual voltage vectors can cause
an unacceptable computational burden to the controller. To dc-voltage control is also highlighted. A direct power control
alleviate this issue, the proposed control approach uses a with the predictive approach is presented in [9] which utilizes
preselection technique that reduces the number of possi- two different cost functions and allows us to operate the con-
ble voltage vectors for the optimization process from 38 verter with a reduced switching frequency. A modified predic-
to 10, thereby allowing the proposed technique to be im- tive control is proposed in [10] concerning the stability of the
plemented in a standard control platform. Finally, measure-
ment results are presented from an experimental prototype grid-connected converter. Besides, the computational effort can
which demonstrates the efficacy of the proposed algorithm be reduced by 18% in comparison with the conventional MPC.
concerning the improvement of the input current quality. A look-up table-based control approach is highlighted in [11]
Index Terms—Direct method of Lyapunov, Lyapunov
featuring new optimal vector to overcome the drawbacks of non-
function, nonlinear stability, virtual voltage vectors. linear direct power control. In [12], a duty cycle-based control
is presented that uses an online optimization of duty cycles in
I. INTRODUCTION different operating conditions. The proposed control approach
N RECENT times, the finite control set model predictive offers improved transient and steady-state feature in compari-
I control (FCS-MPC) is considered as one of the main com-
petitors to replace the conventional control scheme for grid-
son to the other methods. Some other notable contributions are
highlighted in [13]–[17].
connected converters [1]–[2]. The FCS-MPC has several ad- Despite having some useful features, there remain several
vantages, such as easy implementation and understandable con- open problems for FCS-MPC, being one of the most important,
cepts, finite control actions and the inclusion of nonlinearities the stability issue [18]. Stability issues of the FCS-MPC have
in the cost function [1]–[5]. been investigated so far by using Lyapunov’s stability theory
Several FCS-MPC for grid-connected converters has been re- [10], which is a useful technique to investigate the stability is-
ported until the present [6], [7]. In [7], a deadbeat controller sue of the nonlinear systems [19]–[22]. Another drawback of the
is presented for a grid-connected power converter. The pro- FCS-MPC controlled 3-Ø, two-level grid-connected converter is
posed control approach allows the converter operation under the application of single switching state in each sampling inter-
val makes this method difficult to use in applications that require
small filter size and better input current quality at the same time
Manuscript received June 29, 2018; revised September 10, 2018; ac- [23], [24]. Till date, the complete research on the three-phase
cepted October 1, 2018. Date of publication October 16, 2018; date
of current version May 2, 2019. Paper no. TII-18-1695. (Corresponding voltage source inverter (VSI) involving both the stability is-
author: Kazi Saiful Alam.) sue and performance improvement is lacking. Thus, a control
The authors are with the School of Electrical Engineering and Telecom- method is desirable that not only features the explicit investiga-
munications, University of New South Wales, Sydney, NSW 2052,
Australia (e-mail:,[email protected]; [email protected]; tion of stability issue, but also improves the performance of the
[email protected]; [email protected]; f.rahman@unsw. converter regarding the quality of the input current. Motivated
edu.au). by this fact, this paper provides a predictive control implemen-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. tation for a 3-Ø, two-level grid-connected converter that inte-
Digital Object Identifier 10.1109/TII.2018.2876274 grates the virtual voltage vectors in the FCS-MPC architecture
1551-3203 © 2018 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.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
2776 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO. 5, MAY 2019

Fig. 1. 3-Ø, two level grid-connected converter topology.

TABLE I
SWITCHING STATES AND VOLTAGE VECTORS FOR A TWO-LEVEL VSI Fig. 2. Voltage vectors of a 3-Ø grid-connected converter.

next stage involves the modeling of the converter which will be

used for the prediction process.

A. Rectifier Mode
The dynamic equation of the grid-connected converter as a
rectifier can be expressed as
dis rec
vg = vconv + Rs is rec + Ls . (1)
dt
To simplify the analysis, the following vectors are adopted:
vg = [vsa vsb vsc ]T
to improve the quality of the input current. This paper involves
an FCS-MPC structure with 38 modulated voltage space vec- vconv = [vao vbo vco ]T
tors obtained by the discrete space vector modulation (DSVM)
technique [25]. However, this high number of voltage vectors is = [isa isb isc ]T (2)
obtained from the DSVM technique may place an unacceptable where Rs represents the filter inductor leakage resistance and Ls
computational burden to the controller. To alleviate the com- represents the filter inductance. is rec represents the three-phase
putational burden issue, a preselection scheme is adopted that input currents of the grid in the rectifier mode, vconv represents
can reduce the candidate voltage vectors for the optimization the three-phase converter voltages and vg is the three-phase grid
process from 38 to 10 and allows the proposed method to be voltages.
implemented in a standard control platform.
This paper is structured to present in Section II the prediction B. Inverter Mode
model of the power converter. The development of the control
technique and stability assessment using the direct method of The dynamic equation for the grid-connected converter as an
Lyapunov is highlighted in Section III. Section IV provides the inverter can be expressed as
explanation for DSVM concept and highlights the implemen- dis inv
vconv = vg + Rs is inv + Ls (3)
tation of the proposed control scheme. Experimental results to dt
validate the proposed concept are presented in Section V. Fi- where, is inv represents the three-phase currents injected to the
nally, Section VI concludes the paper. grid. The dynamics of the inverter current can be expressed as:
dis inv 1 1 Rs
II. PREDICTION MODEL = vconv − vg − is inv . (4)
dt Ls Ls Ls
Simplified circuit configuration for 3-Ø, two-level grid- To express the prediction equations in digital form, Euler’s
connected converter topology is shown in Fig. 1, which involves forward discretization method can be applied. With this method,
three-pairs of IGBT-diode switches as switching devices and L the derivative can be written as
filters for grid integration.
dx x(k) − x(k − 1)
IGBT devices of each leg are switched in a complementary = (5)
manner. Thus, there can be eight switching states for the power dt Ts
converter as shown in Table I. The voltage space vector corre- where, Ts represents the sampling interval. Using the Euler’s
sponding to Table I is presented in Fig. 2 (in α − β frame). The approximation, the dynamics of the inverter current can be

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
ALAM et al.: ASYMPTOTICALLY STABLE PREDICTIVE CONTROL OF GRID-CONNECTED CONVERTER BASED ON DISCRETE SPACE VECTOR 2777

rewritten as [10] voltage vector of the converter at (k + 1)th sampling instant can
1 be written as (11) [10]
is inv [k + 1] = (Ls is inv [k] + Ts vconv [k + 1]
Rs Ts + Ls vconv [k + 1] = vconv [k + 1] + Δ[k + 1] (11)
− Ts vg [k + 1]). (6) where Δ(k + 1) stands for the error associated with the quanti-
Similarly the dynamics of the input current in rectifier mode can zation process and vconv is the continuous voltage input vector.
be written as [10] Δ(k + 1) satisfies the following condition:
1 |Δ(k + 1)| ≤ φ (φ > 0). (12)
is rec [k + 1] = (Ls is rec [k] − Ts vconv [k + 1]
Rs Ts + Ls
Here, φ is the upper limit of the quantization error. The dynamics
+ Ts vg [k + 1]). (7) of the inverter current expressed by (6) and controlled by (10)
will achieve asymptotic convergence for the following compact
III. DEVELOPMENT OF THE CONTROL LAW AND STABILITY set:
ASSESSMENT    
Ts
σ = ierr |ierr ≤ (φ + ) (13)
A. Control Law RTs + Ls
In the predictive control, the discrete behavior of the converter where φ and  are the constants and represents the upper bound
is used and the finite-state switching actions are applied directly of the errors associated with the quantization and estimation
for the efficient operation of the inverter. Using (6), the future process of the grid voltage, respectively. The definitions for
current error in the discrete form can be written as (inverter practical control Lyapunov function yields to the following in-
mode) equalities to be satisfied [10]:
1 ||V (ierror [k + 1]) − V (ierror [k])||
ierror [k + 1] = (Ls is inv [k] + Ts vconv [k + 1]
Rs Ts + Ls
≤ C1 ||ierror [k + 1] − ierror [k]|| ∀ierror [k]  G (14)
− Ts vg [k + 1]) − i∗s inv [k + 1]
(8) ||V (ierror [k + 1]) − V (ierror [k])||

where i∗s inv [k + 1] is the current reference of the inverter at ≥ C2 ||ierror [k + 1] − ierror [k]|| ∀ierror [k]  Γ (15)
(k + 1)th sampling interval. Moreover, for the present analysis, ΔV (x) < −C3 |ierror [k]|j + C4 (16)
the control law must be involved in the accurate regulation of
the three-phase ac input current/current injected to the grid (is ) where C1 –C4 are positive constants and 1 ≤ j. G ⊆ R rep- n

and must confine the currents within its maximum limit resents a positive control invariant set and Γ ⊂ G represents a
compact set. Next step involves defining a Lyapunov function
im in < is (t) < im ax . (9) as
To effectively track the reference current the converter volt- 1
age vector at (k + 1)th sampling interval/making future current V (ierror [k]]) = [ierror [k]]T [ierror [k]] (17)
2
control error converges to zero, the following equation must be where ierror is the current tracking error. The change of the
satisfied [10]: Lyapunov function can be written as
∗ Ls
vconv [k + 1] = − is inv [k] ΔV (ierror [k]]) = V (ierror [k + 1]) − V (ierror [k]). (18)
Ts
Rs Ts + Ls ∗ Using (8), (18) can be expressed as
+ is inv [k + 1] + vg [k]. (10) 
Ts 1 1
ΔV (ierror [k]) = [Ls is inv [k] + Ts v̂conv [k + 1]
For the present analysis, (10) is chosen as the control law 2 Rs Ts + Ls
to control the dynamics of the three-phase ac current. In T
Section III-B, it is mathematically shown that the proposed con- + Ts Δ[k + 1] − Ts vg [k + 1] + Ts v̂g [k]] − i∗s inv [k + 1]
trol law ensures the asymptotic stability of the overall system.

1
B. Stability Assessment × [Ls is inv [k] + Ts v̂conv [k + 1] + Ts Δ[k + 1]
Rs Ts + Ls
From (10), it is apparent that the three-phase ac current (is ) 
1
will be exactly equal to its reference value when the converter − Ts vg [k +1] + Ts v̂g [k]] − i∗s inv [k +1] − ierr [k]T ierr [k].
∗ 2
voltage vector chosen by the controller will be same as vconv [k +
(19)
1]. However, this may not be possible at all times. The reason
for this is the finite control actions/lack of modulation approach Here, vconv [k + 1] is finite set and bounded. The converter
in the FCS-MPC approach. However, even if the exact voltage current in the inverter mode is represented by is inv [k] is also
vector is not selected, the one closest to that can be selected and bounded. The grid voltages vg [k] are bounded too. The quanti-
thus the control actions will still be the most appropriate. The zation error is delimited by the upper bound φ. Thus, the future

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
2778 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO. 5, MAY 2019

current tracking error is ultimately bounded to a compact set, IV. PROPOSED APPROACH
where ierror [k] ∈ Γ ⊂ R2 . Γ is the compact set which is re-
A. DSVM Theory
lated to the converter voltages and inverter current reference
(bounded). Thus, (19) can be written as Although the control law that controls the load current dynam-
ics ensures the stability of the overall system, one of the draw-
1
ΔV (ierror [k]) ≤ − ierror [k]T ierror [k] backs of the FCS-MPC still remains. The FCS-MPC method
2 applies a single switching state during each sampling inter-
 2
1 Ts val, which leads the poor input current quality. However, this
+ (φ + )2 . (20)
2 Ls + Rs Ts issue can be improved by either choosing higher sampling fre-
quency or using larger filter components. The higher sampling
Conditions that satisfy the stability condition are as follows: frequency entails expensive controller solution, while the larger
1 inductor value leads to increased weight and size of the filter
C1 = C2 = 1, C3 = (21) components. In addition, it does not utilize the whole control
2
 2 region of the converter. As an alternate approach, this study
1 Ts
C4 = (φ + )2 . (22) integrates the virtual voltage vectors into the FCS-MPC archi-
2 Rs Ts + Ls tecture, which is derived from the DSVM technique to improve
This implies the convergence of the system energy to zero or the quality of the input current. The DSVM technique was orig-
convergence of the current control error in a compact set as inally used for DTC in IM drives by Casadei et al. [25] to
   improve the performance of DTC. However, the number of cal-
C4 culations increases rapidly with the larger number of virtual
σ = ierr |ierr ≤ (23)
C3 vectors. Thus, the implementation of the DSVM-based MPC
is challenging concerning the computational effort of the con-
C. Cost Function troller. To overcome this issue, the outcome of the Lyapunov’s
stability theory is used to preselect the candidate voltage vectors
In the proposed approach, the control law is defined as (10) to be used for obtaining the optimal vectors. The preselection
which is mathematically proven to be asymptotically stable by approach can reduce the number of calculations to obtain the
using stability analysis considering the Lyapunov’s stability the- optimal voltage vector from 38 to 10 and makes the proposed
ory. The next step involves the cost-function modification with method implementable in real-time systems. The virtual voltage
the proposed technique considering (10). The reference con- vectors are formed by the combinations of real voltage vectors
verter voltage calculation of (10) can be rewritten in the d − q similar to the DSVM approach applied for induction machine
frame as drives [25] as described in (28). The total number of virtual
Ls − Rs Ts voltage vectors is decided by the chosen number of subdivision
Vd [k + 1] = Vsd [k] + Id inv [k] of the sampling interval
Ts
Ls n
∗
+ ωLs Iq inv [k] − Id inv [k] (24) v vir = tj vjreal (28)
Ts
j =1
Ls − Rs Ts
Vq [k + 1] = Vsq [k] + Iq inv [k] t1 + t 2 + · · · + t n = T s (29)
Ts
Ls ∗ vjreal ∈ {V0 , V1 , . . . , V7 } . (30)
− ωLs Id inv [k] − Iq inv [k] (25)
Ts
Here, Ts is the sampling interval. Any virtual voltage vector,
where Id inv and Iq inv are the dq frame representation of the grid (v vir ) is formulated by linearly combining the real voltage vec-
currents (inverter mode). Vsd and Vsq are the d − q frame rep- tors v real of the converter. Each of the real voltage vectors is
resentation of the grid voltages, whereas Vd and Vq correspond applied for a specific time interval depending on the number
to the d − q frame representation of the converter voltages, and of subdivisions of the sampling interval. The virtual voltage
ω represents the angular frequency of the supply voltage. synthesis technique is shown in Fig. 3. Say, for example, if the
Thus, the cost function to track the d − q component of the subdivision is chosen to be two, then a virtual voltage vector
reference converter voltages can be written as that is located between the V5 and V6 is denoted by V56 and is
g[k +1] = |Vd [k +1] − Vd∗ [k + 1]| + |Vq [k + 1] − Vq∗ [k + 1]|. formed by the combination or averaging of V5 and V6 . For V6
(26) the application of switching state would be, Sa = 1, Sb = 0,
Considering the delay compensation for the implementation, the Sc = 1 and for V5 it would be Sa = 0, Sb = 0, Sc = 1. Thus,
cost function is modified as the switching state for V56 would be Sa = 0.5, Sb = 0, Sc = 1
as shown in Fig. 3. As the number of voltage vectors obtained
g[k +1] = |Vd [k + 2] − Vd∗ [k + 2]| + |Vq [k + 2] − Vq∗ [k + 2]| through this process is still finite, it is possible to implement
(27) the proposed method in real time using look-up table, as shown
(24) and (25) are modified accordingly to predict the (k + 2) in Fig. 3. As the alternative, an external modulator can also be
values of the d − q components of the converter voltage vectors. used as shown in Fig. 3.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
ALAM et al.: ASYMPTOTICALLY STABLE PREDICTIVE CONTROL OF GRID-CONNECTED CONVERTER BASED ON DISCRETE SPACE VECTOR 2779

TABLE II
αβ COMPONENT OF THE CONVERTER VOLTAGE VECTORS FOR THE
PRESENT ANALYSIS

Fig. 3. Illustration of virtual vector synthesis using DSVM theory.

each of voltage vector should be applied for Ts /3. Thus, virtual

voltage vector V112 (referred to Fig. 4) implies the combination
of V1 , V1 and V2 , each of them applied for one-third of the
sampling period. The αβ components of the converter voltage
vectors are given in Table II. This is obtained by using abc to
αβ transformation formula as follows:
⎡ ⎤
vsa
Vsα 2 1 −1/2 −1/2 ⎢ ⎥
= √ √ ⎣ vsb ⎦. (31)
Vsβ 3 0 3/2 − 3/2
vsc
For convenience one case study is presented here. If the virtual
voltage vector is V666 then the switching states of Sa , Sb , and Sc
would be 1, 0, and 1, respectively. Then, using the relationship
Fig. 4. Voltage vectors used in the present analysis. between this switching states (Sa = 1, Sb = 0, and Sc = 1) and
the output voltages of the converter, the output voltages of the
converter can be obtained (va , vb , and vc ). Then using (31), the
The number of voltage vectors that can be obtained from αβ component of the V666 can be obtained. The other voltage
this method is related to the number of subdivisions of the vectors of Table II are also obtained using this concept.
sampling interval. However, a higher number of subdivisions The α − β component of the converter voltage vectors can
will lead to the formation of complex switching table in addition be represented in the d − q frame using the following formula:
to the higher computational burden placed on the controller.
Thus, a compromise is defined by the current ripple reduction Vd = Vα cos (θ) + Vβ sin (θ) (32)
and complexity of the look-up table formulation. The optimal
Vq = − Vα cos (θ) + Vβ sin (θ). (33)
solution for this is found to be three as highlighted in [25] for
DTC. For the present analysis, the subdivision is chosen to be
B. Preselection Approach to Reduce the Number of
three (which yields three-equal time divisions) and 38 voltage
vectors (including two-zero voltage vectors) can be obtained as Calculations
shown in Fig. 4. The formulation of the virtual voltage vector is Even though the optimal number of subdivision (three in this
similar to the illustration of Fig. 3. For further clarification, one case) is chosen here, the number of voltage vectors is still higher
example is presented here. As the number of subdivision is three, to be implemented in a standard control platform. However, it is

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
2780 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO. 5, MAY 2019

Fig. 5. Illustration of the preselection approach based on the sector

information.

TABLE III
PRESELECTION TABLE

Fig. 6. Simplified block diagram of the proposed control technique.

V. EXPERIMENTAL RESULTS
allowed to reduce the number of calculations in the prediction
The proposed work presents an alternative implementation
process by using an effective preselection scheme. The prese-
of predictive control approach based on Lyapunov’s stability
lection scheme will not affect the performance of the controller
theory with 38 modulated voltage vectors [8 real voltage vec-
if the possible voltage vectors required to be applied does not
tors (same as the conventional approach) and 30 virtual voltage
change. As the proposed approach calculates the future refer-
vectors]. However, this approach can increase the number of
ence voltage of the converter, by using the sector information
calculations in the prediction process. To alleviate this problem,
on which the future reference voltage is located, the number of
a preselection scheme is used in the proposed algorithm that can
calculations can be effectively reduced. This will not change
reduce the number of calculations from 38 to 10. The overall
the possible voltage vectors which are required to be applied in
control technique for the proposed technique has been described
order to obtain the desired converter current. Thus, instead of
with a simplified control block and is presented in Fig. 6.
evaluating all the voltage vectors obtained from virtual voltage
To validate the theoretical approach, a scaled down prototype
integrated FCS-MPC, the selected number of voltage vectors
has been built. Fig. 7 illustrates the layout of the experimental
can be utilized. This not only reduces the computational burden
setup. The parameters for the experimental prototype are given
placed on the controller but also reduces the complexity of the
in Table IV. The control scheme is implemented in DSpace
algorithm. The prediction procedure of the proposed approach is
DS1104 RD controller board. The execution of the implemen-
initiated by preselecting the candidate voltage vectors for the op-
tation process can be divided into different tasks including A/D
timization process. The calculation of voltage vector reference
conversion, grid synchronization, preselection, optimal vector
allows using the sector information (as shown in Fig. 5) which
selection, and application of optimal vector with the DSVM pro-
can be effectively used to preselect the candidate voltage vec-
cess. The DSVM task is realized with the pulse width modula-
tors to be used in the optimization process, as given in Table III.
tion peripheral available on the used digital platform. Semikron
Say for example, if the desired output voltage corresponds to
IGBT modules are used as the switching device. The SKYPER
Sector-I, the voltage vectors for Sector-I in Table III can be con-
32 R, a Dual IGBT Driver IC, is used as an interface between
sidered for cost-function minimization as these are the possible
IGBT modules and the controller. As a dc-source, MAGNA-
voltage vectors to be applied. In this case, the candidate voltage
POWER programmable dc-power supply is used.
vectors will be, V000 , V200 , V220 , V222 , V120 , V221 , V112 , V111 , V110 ,
and V100 . If the reference voltage vector corresponds to Sector-II,
then another set of candidate voltage vectors can be chosen from A. Steady-State Performance Evaluation
Table III. Once the preselection process is executed the next step For evaluating the steady-state performance of the power
involves the selection of the optimal voltage vector. The selected converter as a rectifier, the dc-bus voltage is regulated at 180-
voltage vectors are then used to evaluate the cost function. The V using a PI controller as shown in Fig. 8. The real power
voltage vector with minimum cost-function value is chosen as control is achieved by regulating the Id (d- component of the
the optimal vector for the implementation. input current). The proposed control can effectively track the

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
ALAM et al.: ASYMPTOTICALLY STABLE PREDICTIVE CONTROL OF GRID-CONNECTED CONVERTER BASED ON DISCRETE SPACE VECTOR 2781

Fig. 9. Illustration of Id tracking to its reference value.

Fig. 7. Experimental prototype for the present analysis.

TABLE IV
PARAMETERS FOR EXPERIMENTAL SETUP

Fig. 10. Illustration of Iq tracking to its reference value.

Fig. 11. Input current with the conventional FCS-MPC.

Fig. 8. DC-link voltage regulated to 180-V.

d component of the reference current as shown in Fig. 9. As the Fig. 12. Input current with the proposed technique.
converter is operating in the rectifier mode, the reactive power is
regulated at zero to obtain unity power factor. The q component fast Fourier transform (FFT) diagram of the input current
of the input current is regulated to zero as shown in Fig. 10. with the conventional approach is shown in Fig. 13, while
One of the key contributions of the proposed control the FFT of the input current with the proposed approach is
approach is to reduce the THD of the input current. Thus, it shown in Fig. 14. The input current THD with different mag-
is necessary to analyze the quality of the input current. The nitude of the input current along with the power level is shown
input current of the converter with the proposed approach is with the help of a three-dimensional line plot (see Fig. 15). The
shown in Fig. 11. The input current with the conventional improvement in the power quality with the proposed approach
MPC is shown in Fig. 12. Compared to conventional MPC, is particularly striking under very light load. Unlike the con-
the proposed method can improve the THD significantly ventional FCS-MPC, the proposed method ensures higher input
(from 5.72% to 2.93%). To provide a better illustration, the current quality which allows the reduction of filter size. For a fair

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
2782 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO. 5, MAY 2019

Fig. 16. Illustration of the unity power factor operation.

Fig. 13. FFT diagram of the input current with the conventional FCS-
MPC.

Fig. 17. Transient response of Id with the proposed approach.

Fig. 14. FFT diagram of the input current with the proposed method. Fig. 18. Transient response of Id with the conventional approach.

B. Transient Performance Evaluation

To further validate the effectiveness of the proposed approach
transient analysis is performed. For a fair comparison between
the control algorithms, the dc-voltage regulation loop is turned
OFF. Then, the Id reference is suddenly changed from 3-A to 10-
A. The controller tracks this change within a very short time as
shown in Fig. 17. Fig. 18 shows the transient response with the
conventional approach that also yields to the similar response.
Thus, the proposed method does not sacrifice the transient re-
sponse compared to the conventional approach. As the converter
Fig. 15. Comparison of THD between the proposed method and the can operate bidirectionally, the inverter mode is also explored.
conventional approach with the variation of input current and power level. The injected current to the grid in the inverter mode is shown in
Fig. 19.
comparison, the average switching frequency is kept same for
both the methods by adjusting the sampling frequency. Besides, C. Performance Evaluation With Grid Voltage Sag and
Other Abnormal Conditions
in the rectifier mode, the reactive power flow must be controlled
to zero and thus unity power factor should be maintained. The The performance of the proposed control technique is inves-
power factor is regulated to unity as shown in Fig. 16. tigated under voltage sag conditions. The grid voltage sag is

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
ALAM et al.: ASYMPTOTICALLY STABLE PREDICTIVE CONTROL OF GRID-CONNECTED CONVERTER BASED ON DISCRETE SPACE VECTOR 2783

TABLE V
COMPARISON WITH OTHER CONTROL ALGORITHMS

Fig. 19. Injected grid current with the grid voltage in the inverter mode grid voltage [26]–[28]. For the present analysis, the robustness
with the proposed control approach. to grid voltage distortion can be achieved by modifying the
current reference generation scheme as discussed in [26]–[28].
However, this analysis is beyond the scope of this paper and
identified as a future work by the authors.

D. Performance Comparison With Other Methods

In Table V, a comparative study among different other ex-
isting control algorithms is highlighted. For a fair comparison,
the average switching frequency for all the control algorithms
is kept same. Compared to the conventional and the modified
model predictive control method presented in [10], the proposed
Fig. 20. Voltage sag in the grid. method can significantly lower the input current THD. As shown
in [29], the linear controller and the FCS-MPC achieves similar
performance at the same average switching frequency. In the
experiment, the DPC-SVM achieves a THD of 5.68% which is
similar to the FCS-MPC. However, one of the advantages of
the proposed method compared to the linear control scheme is
that it does not require any tuning task with a simpler imple-
mentation process. The inclusion of decoupling networks and
protection against antiwindup is not required. The proposed
method also offers better d and q components decoupling of the
input current. Thus, the proposed method can be considered as
an effective alternative to the linear control schemes.
Fig. 21. Input current under voltage sag condition in the grid.
E. Analysis of Computational Effort
One of the concerns of the FCS-MPC is the execution time
of the algorithm. It is desirable that the proposed algorithm
can be implemented in a standard control platform. Thus, the
analysis of computational burden is necessary. The presented
approach takes 3.42 μs to execute the prediction loop while
the conventional approach takes 3.87 μs. As the computational
burden is still comparable with the conventional FCS-MPC, the
proposed method can be implemented in any standard control
platform.
Fig. 22. DC-bus voltage under voltage sag condition in the grid.
VI. CONCLUSION
shown in Fig. 20. The input current and the dc-link voltage This paper integrates the DSVM theory with the FCS-MPC
during voltage sag is shown in Figs. 21 and 22, respectively. structure to overcome one of the limitations of the conventional
As can be seen from Figs. 21 and 22, the overshoot and rise FCS-MPC. Furthermore, by using Lyapunov’s stability theory, it
time in the input current is acceptable. Also, the dc-bus voltage is mathematically proven that the proposed control law ensures
is tracked back to the reference value in a short time. It is to that the overall system remains asymptotically stable. However,
be mentioned that, the grid voltage can also become distorted as the proposed method employs additional virtual voltage vec-
with lower order harmonics because of the nonlinear loading tors to introduce improvements in terms of quality of the input
effect. In recent times, many other modified forms of predic- current, it may place an unacceptable computational burden in
tive control is investigated to show robustness against distorted the controller. Aiming to solve this problem, the present analysis

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
2784 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 15, NO. 5, MAY 2019

considers a simple preselection scheme based on section infor- [20] I. Sefa, S. Ozdemir, H. Komurcugil, and N. Altin, “Comparative study
mation of the reference voltage vector to reduce the number of on Lyapunov-function-based control schemes for single-phase grid-
connected voltage-source inverter with LCL filter,” IET Renew. Power
calculations and allows the proposed method to be implemented Gener., vol. 11, no. 11, pp. 1473–1482, Sep. 2017.
in a standard control platform. Finally, experimental results are [21] H. Komurcugil, N. Altin, S. Ozdemir, and I. Sefa, “An extended Lyapunov-
shown to support the proposed concept. function-based control strategy for single-phase UPS inverters,” IEEE
Trans. Power Electron., vol. 30, no. 7, pp. 3976–3983, Jul. 2015.
[22] S. M. Wi, J. S. Lee, and M. Kim, “Exponentially stable Lyapunov-function-
REFERENCES based controller for a flyback CCM converter,” IEEE Trans. Ind. Electron.,
vol. 65, no. 2, pp. 1213–1225, Feb. 2018.
[1] F. Donoso, A. Mora, R. Cardenas, A. Angulo, D. Saez, and M. Rivera, [23] I. Gonzalez-Prieto, M. J. Duran, J. J. Aciego, C. Martin, and F. Barrero,
“Finite-set model-predictive control strategies for a 3L-NPC inverter oper- “Model predictive control of six-phase induction motor drives using virtual
ating with fixed switching frequency,” IEEE Trans. Ind. Electron., vol. 65, voltage vectors,” IEEE Trans. Ind. Electron., vol. 65, no. 1, pp. 27–37, Jan.
no. 5, pp. 3954–3965, May 2018. 2018.
[2] C. A. Rojas, M. Aguirre, S. Kouro, T. Geyer, and E. Gutierrez, “Leakage [24] T. Wang, C. Liu, G. Lei, Y. Guo, and J. Zhu, “Model predictive direct
current mitigation in photovoltaic string inverter using predictive con- torque control of permanent magnet synchronous motors with extended
trol with fixed average switching frequency,” IEEE Trans. Ind. Electron., set of voltage space vectors,” IET Elect. Power Appl., vol. 11, no. 8,
vol. 64, no. 12, pp. 9344–9354, Dec. 2017. pp. 1376–1382, Sep. 2017.
[3] S. Vazquez, J. Rodriguez, M. Rivera, L. G. Franquelo, and M. Noram- [25] D. Casadei, G. Serra, and A. Tani, “Implementation of a direct torque
buena, “Model predictive control for power converters and drives: Ad- control algorithm for induction motors based on discrete space vector
vances and trends,” IEEE Trans. Ind. Electron., vol. 64, no. 2, pp. 935–947, modulation” IEEE Trans. Power Electron., vol. 15, no. 4, pp. 769–777,
Feb. 2017. Jul. 2000.
[4] T. Geyer, Model Predictive Control of High Power Converters and Indus- [26] J. R. Fischer, S. A. Gonzalez, I. Carugati, M. A. Herrán, M. G. Judewicz,
trial Drives. New York, NY, USA: Wiley, 2016. and D. O. Carrica, “Robust Predictive control of grid-tied converters based
[5] R. Baidya, R. P. Aguilera, P. Acuna, S. Vazquez, and H. D. T. Mou- on direct power control,” IEEE Trans. Power Electron., vol. 29, no. 10,
ton, “Multistep model predictive control for cascaded H-bridge inverters: pp. 5634–5643, Oct. 2014.
Formulation and analysis,” IEEE Tran. Power Electron., vol. 33, no. 1, [27] T. Liu, C. Xia, and T. Shi, “Robust model predictive current control of
pp. 876–886, Jan. 2018. grid-connected converter without alternating current voltage sensors,” IET
[6] P. Falkowski and A. Sikorski, “Finite control set model predictive control Power Electron., vol. 7, no. 12, pp. 2934–2944, Dec. 2014.
for grid-connected AC-DC converters with LCL filter,” IEEE Trans. Ind. [28] R. Errouissi, S. M. Muyeen, A. Al-Durra, and S. Leng, “Experimental
Electron., vol. 65, no. 4, pp. 2844–2852, Apr. 2018. validation of a robust continuous nonlinear model predictive control based
[7] W. Jiang, W. Ma, J. Wang, L. Wang, and Y. Gao, “Deadbeat control grid-interlinked photovoltaic inverter,” IEEE Trans. Ind. Electron., vol. 63,
based on current predictive calibration for grid-connected converter un- no. 7, pp. 4495–4505, Jul. 2016.
der unbalanced grid voltage,” IEEE Trans. Ind. Electron., vol. 64, no. 7, [29] H. Young and J. Rodriguez, “Comparison of finite-control-set model pre-
pp. 5479–5491, Jul. 2017. dictive control versus a SVM-based linear controller,” in Proc. 15th Eur.
[8] P. Acuna, R. P. Aguilera, A. M. Y. M. Ghias, M. Rivera, C. R. Baier, and Conf. Power Electron. Appl., 2013, pp. 1–8.
V. G. Agelidis, “Cascade-free model predictive control for single-phase
grid-connected power converters,” IEEE Trans. Ind. Electron., vol. 64,
no. 1, pp. 285–294, Jan. 2017.
[9] Y. Sangsefidi, S. Ziaeinejad, and A. Mehrizi-Sani, “Low switching
frequency-based predictive control of a grid-connected voltage-sourced
converter,” IEEE Trans. Energy Conv., vol. 32, no. 2, pp. 686–697, Jun.
2017. Kazi Saiful Alam (S’16) received the B.Sc. Eng.
[10] Md. P. Akter, S. Mekhilef, N. M. L. Tan, and H. Akagi, “Modified model (Hons.) degree in electrical engineering from
predictive control of a bidirectional AC-DC converter based on Lyapunov Bangladesh University of Engineering and Tech-
function for energy storage systems,” IEEE Trans. Ind. Electron., vol. 63, nology, Dhaka, Bangladesh in 2012. He is cur-
pp. 704–715, Feb. 2016. rently working toward the Ph.D. degree at the
[11] K. Kulikowski and A. Sikorski, “New DPC look-up table methods for School of Electrical Engineering and Telecom-
three-level AC/DC converter,” IEEE Trans. Ind. Electron., vol. 63, no. 12, munications, University of New South Wales,
pp. 7930–7938, Dec. 2016. Sydney, NSW, Australia.
[12] Z. Song, Y. Tian, W. Chen, Z. Zou, and Z. Chen, “Predictive duty cy- From 2012 to 2015, he was a Senior Lecturer
cle control of three-phase active-front-end rectifiers,” IEEE Trans. Power with Eastern University, Dhaka, Bangladesh.
Electron., vol. 31, no. 1, pp. 698–710, Jan. 2016. His research interests include solid-state trans-
[13] R. Errouissi, S. M. Muyeen, A. Al-Durra, and S. Leng, “Experimental former, predictive control, and isolated dc/dc converter.
validation of a robust continuous nonlinear model predictive control based
grid-interlinked photovoltaic inverter,” IEEE Trans. Ind. Electron., vol. 63,
no. 7, pp. 4495–4505, Jul. 2016.
[14] S. C. Ferreira, R. B. Gonzatti, R. R. Pereira, C. H. da Silva, L. E. B. da Silva,
and G. Lambert-Torres, “Finite control set model predictive control for
dynamic reactive power compensation with hybrid active power filters,”
IEEE Trans. Ind. Electron., vol. 65, no. 3, pp. 2608–2617, Mar. 2018. Md. Parvez Akter (S’16) was born in Pabna,
[15] B. S. Riar, J. Scoltock, and U. K. Madawala, “Model predictive direct Bangladesh. He received the B.Sc. Eng. degree
slope control for power converters,” IEEE Trans. Power Electron., vol. 32, in electrical and electronics engineering from
no. 3, pp. 2278–2289, Mar. 2017. Chittagong University of Engineering and Tech-
[16] J. Hu and Z. Q. Zhu, “Improved voltage-vector sequences on dead-beat nology, Chittagong, Bangladesh, in 2011 and the
predictive direct power control of reversible three-phase grid-connected M.Eng. Sc. degree in power electronics from the
voltage-source converters,” IEEE Trans. Power Electron., vol. 28, no. 1, University of Malaya, Kuala Lumpur, Malaysia, in
pp. 254–267, Jan. 2013. 2015. He is currently working toward the Ph.D.
[17] M. Gendrin, J. Y. Gauthier, and X. Lin-Shi, “A predictive hybrid pulse- degree at the Faculty of Engineering, Univer-
width-modulation technique for active-front-end rectifiers,” IEEE Trans. sity of New South Wales (UNSE), Sydney, NSW,
Power Electron., vol. 32, no. 7, pp. 5487–5496, Jul. 2017. Australia.
[18] R. P. Aguilera and D. E. Quevedo, “Predictive control of power converters: From 2013 to 2016, he was a Research Assistant with the Power
Designs with guaranteed performance,” IEEE Trans. Ind. Inf., vol. 11, Electronics and Renewable Energy Research Laboratory, Department
no. 1, pp. 53–63, Feb. 2015. of Electrical Engineering, University of Malaya. He is currently a Scientia
[19] H. Beiranvand and E. Rokrok, “Asymptotically stable controller for SSTs Ph.D. Scholar at UNSE. His research interests include power convert-
based on Lyapunov direct stability method,” IET Power Electron., vol. 10, ers and electrical drives, bidirectional power conversion, energy storage,
no. 15, pp. 2065–2075, Dec. 2017. smart grid, renewable energy, and energy efficiency.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.
ALAM et al.: ASYMPTOTICALLY STABLE PREDICTIVE CONTROL OF GRID-CONNECTED CONVERTER BASED ON DISCRETE SPACE VECTOR 2785

Dan Xiao (M’10) received the Bachelor’s and Muhammed Fazlur Rahman (F’14) received
Master’s degrees in electrical engineering from the B.Sc. Eng. degree from the Bangladesh Uni-
Shenyang University of Technology, Shenyang, versity of Engineering and Technology in 1972
China, in 2001 and 2004, respectively, and the and the M.Sc. and Ph.D. degrees from the Uni-
Ph.D. degree in electrical engineering from the versity of Manchester, U.K., in 1975 and 1978
University of New South Wales (UNSW), Syd- respectively, all in electrical engineering.
ney, NSW, Australia. He was a Systems Design Engineer with the
He is currently a Technical Support with the General Electric Projects Co., Rugby, U.K., for
Energy System Research Laboratories, UNSW. two years and with the National University as a
His research interests include sensorless con- Senior Lecturer for eight years. He is currently
trol and online parameter estimation of ac ma- a Professor with the University of New South
chines, model predictive control for power converter and drives, matrix Wales, Sydney, NSW, Australia. His research interests include power
converters, and solid state transformer. electronics, motor drives and design of electrical machines with PM ex-
citation.

Daming Zhang (M’04) received the Bachelor’s

and Master’s degrees from Huazhong Univer-
sity of Science and Technology in 1993 and
1996, respectively, both in electric machine
major, and the Ph.D. degree in power engi-
neering from Nanyang Technological University,
Singapore.
He is currently a Lecturer with the School of
Electrical Engineering and Telecommunication,
University of New South Wales, Sydney, NSW
Australia. His current research interests include
design and operation of ac microgrids, solid-state transformers with em-
phasis on medium-frequency transformers, and condition monitoring in
the scenario of renewable energy generation.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA BAHIA. Downloaded on January 10,2024 at 02:11:40 UTC from IEEE Xplore. Restrictions apply.