This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

A Model Predictive Current Controller With

Improved Robustness Against Measurement Noise
and Plant Model Variations
Diego Pérez-Estévez, Member, IEEE, Jesús Doval-Gandoy, Member, IEEE

Abstract—This article improves the robustness of a CCS-MPC uses a pulse width modulator (PWM) to generate
finite control set (FCS)-model predictive controller (MPC)
for grid-tied inverters and motor drives applications to plant the firing signals that drive the power converter. The former
parameter variations and noise, without reducing its bandwidth design is more popular among researchers in the field of power
or affecting its excellent transient response to disturbances conversion [3] and it is the type of MPC considered in this
and reference commands. The proposed modification adds article.
an observer to the MPC controller structure, which does not Compared to other applications with slower dynamics [4],
significantly increase the computational burden on the embedded
controller. Traditionally, observers are employed to estimate converter control applications have to meet sub-millisecond
unmeasured variables and cancel the effect of disturbances, execution times, which challenges the implementation of com-
but this article employs the observer to estimate a measured plex MPC formulations and favors the use of simpler control
variable, the converter output current. This solution leverages laws [5]. Current proposals of CCS-MPCs that solve a high-
the benefits of observers from linear controller theory in order order quadratic programming (QP) or linear programming
to remove undesired components in the measured current
and improve the robustness of the controller; hence it is a (LP) optimization problem in real time require a high compu-
valuable solution for practicing power-electronic engineers and tational load [6]. Similarly, a FCS-MPC with a long prediction
researchers in the field of grid-tied inverters and motor drives horizon and several state variables [7] is still difficult to
due to its simplicity compared to some advanced techniques implement due to the extremely high computational load
often required in more complex MPC designs. required to solve an integer optimization problem. Nonethe-
Index Terms—Current controller, L filter, finite control set less, as indicated in [3], long prediction horizons [8]–[10]
(FCS) model predictive controller (MPC), voltage source con- offer performance improvements, despite the fact that their
verter (VSC), inverter, observer.
advantages are often misunderstood due to a poor formulation
of the optimization goal. Recent proposals [9], [11]–[13]
I. I NTRODUCTION have achieved promising results using a novel algorithm [14]
In power electronic applications, model predictive con- borrowed from the digital telecommunication field. Moreover,
trollers (MPCs) are growing in popularity thanks to an increas- the development of solvers for online optimization continue to
ing computational capability in embedded controllers and an be an active field of research that has already provided several
enormous research effort throughout the last two decades [1], solutions, as summarized in [1].
[2]. Compared to traditional error-driven solutions, such as Due to the previous limitations, practical implementations
proportional-integral (PI) and proportional-resonant (PR) con- of FCS-MPCs often use a short prediction horizon. Frequently,
trollers, an MPC can potentially offer better performance be- a one-sample[15]–[20] prediction is selected. The one-sample
cause it incorporates a model of the dynamics and constraints prediction is often modified [19], [20] to compensate for the
in the plant. A model-based controller calculates the best effect of the computational delay. Since FCS-MPCs often use
converter response in terms of an objective or cost function a high sampling frequency compared to PWM-based designs,
while taking into account the nonlinear converter response due a one- or two-sample prediction horizon spans a negligible
to its switching operation. time interval compared to a fundamental grid cycle. Such short
There are two main categories of MPCs, namely, prediction horizon can be significantly affected by the mea-
finite control set (FCS)-MPC and continuous control set surement noise and plant model deviations. Previous literature
(CCS)-MPC. A FCS-MPC directly selects a switching state has studied the operation under distorted grid conditions [21]
and maintains it during a whole sampling period whereas a and the effect of plant model deviations [22] for CCS-MPC.
However, very little progress has been done in studying how
Manuscript received xxxxx; revised xxxxx; accepted xxxxx. Date of current the noise affects the robustness of FCS-MPCs in spite of
version xxxxx. This work was supported by the Spanish Ministry of Science a large research effort [16], [23]–[25] due to the complex
and Innovation and by the European Commission, European Regional Devel-
opment Fund (ERDF) under project PID2019-105612. nonlinear studies often required to assess the robustness and
D. Pérez-Estévez and J. Doval-Gandoy are with the Applied Power Elec- stability of FCS-MPCs.
tronics Technology Group (APET), University of Vigo, Vigo 36310, Spain This article analyzes the limitations of the one-step ahead
(e-mail: [email protected]; [email protected]).
FCS-MPC current controller in terms of robustness to plant
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. model deviations and measurement noise and proposes an
Digital Object Identifier xxxxx observer that can be used in conjunction with the one-step

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

ahead FCS-MPC to improve its robustness with a minimal

L filter
increase in computational load and without affecting the VSC
transient dynamics. Observers are frequently used in the PCC Zg
control of electric drives [26] and grid-tied inverters [27], vVSC,abc
[28]. Depending on the application, an observer can perform vg,abc
different tasks. For example, in electric drive applications an i1,abc
Firing signals sabc
observer can be used to estimate unmeasured variables, such
as the rotor flux orientation and speed, or the temperature of i∗1 FCS
internal elements in a motor [29]. Hence, the observer permits MPC
to obtain information about system variables without requiring vg,abc
additional sensors, which may be difficult or costly to install.
Another common application of observers is the estimation Fig. 1. Grid-tied voltage source converter (VSC) connected at the point of
of disturbances that can affect the performance of the sys- common coupling (PCC) and controlled by a grid current model predictive
controller (MPC) with a finite control set (FCS) operation.
tem. Such disturbance observers [26] permit to compensate
the effect of parameter mismatches and nonlinearities [30],
[31]. Traditionally, disturbance observers estimate an input- performance of the proposal to a conventional FCS-MPC. In
equivalent disturbance that, if subtracted from the plant input, Section V, experimental results that validate the theory are
cancels the effect of disturbances in the plant [32]–[34]. presented. Finally, Section VI concludes this work.
This proposal differentiates from the previous two appli-
cations because it uses an observer to estimate a measured
variable, instead of estimating a disturbance or an unmeasured II. M ODELING OF THE P LANT AND THE C OST F UNCTION
variable. The estimation improves the actual measurement In order to design an MPC, a plant model and a cost function
in the presence of noise and plant model mismatches [35] needs to be defined. In the case of a current controller, the
because it is able to eliminate undesired components in the plant model allows the controller to predict the achievable
measurement and enhance the robustness of the controller. future plant states for the different controller actions. The cost
Moreover, the previously mentioned solutions can be used in function permits the controller to choose the best controller
conjunction with the proposal, i.e., a dual observer design, in action so as to follow a given reference signal.
order to benefit from the advantages of each observer. This article considers a discrete-time second-order plant
The main contribution of this article is an improvement model, which is frequently used in applications such as motor
in the performance of the one-step ahead FCS-MPC in the control and grid-tied inverters with an L filter. A simplified
presence of noise in the current measurements. To the author’s block diagram representation of a grid-tied inverter with an L
knowledge, there are no previous proposals that have studied filter is shown in Fig. 1. The controller measures the grid
the effect of measurement noise in the current distortion for the voltage vg and the grid current i1 and generates the VSC
one-step ahead FCS-MPC. The proposal reduces the current firing-signals. In a grid-tied application, the grid current is
distortion by calculating an estimation of the measured current, usually measured at the point of common coupling (PCC).
without increasing the switching frequency of the voltage The three-phase grid is modeled by a grid impedance Zg and
source converter (VSC) and with a moderate increase in a grid voltage vg . If present, the grid impedance can also
computational load. The proposed modification does not affect model additional filter elements [36]. The control problem is
the transient dynamics of the one-step ahead FCS-MPC and formulated in the grid-aligned αβ frame. If no reference frame
preserves a fast reference tracking and disturbance rejection is denoted in a variable or parameter, the αβ reference frame
responses. is assumed. Such a model can be described by the following
The concept presented in this article can also be imple- first-order differential equation:
mented using other observer structures, such as a Luenberger d i1 (t) Rf 1 1
observer. The proposal selected a Kalman filter due to its = − i1 (t) + vVSC (t) − vg (t). (1)
dt Lf Lf Lf
popularity, simple design process, and good performance.
However, an equivalent performance can be obtained with where the variable vVSC denotes the converter output voltage,
a Luenberger observer that is designed using a direct pole and the parameters Rf and Lf denote the resistive and reactive
placement strategy if the observer poles are placed in the same components, respectively, of the L filter.
locations as the poles of the Kalman filter [35]. It should also In state-space notation, Equation (1) can be rewritten as
be noted that the design or tuning of the observer is not critical d xcont (t)

v (t)

in this application, the proposal uses the same observer design = Acont xcont (t) + Bcont VSC
dt vg (t)
for all tests, although it is also possible to adjust the observer
i1 (t) = Ccont xcont , (2)
design according to the expected noise intensity.
The rest of this paper is organized in the following manner. where xcont is the system state vector, which only contains
Section II describes the plant model and cost function em- one state variable, the grid current in a grid-tied inverter
ployed by the MPC. Section III presents the controller archi- i1 (t), or the motor current in a motor drive application. The
tecture and design process, including the proposed observer. system state matrix Acont is −Rf /Lf ; the input matrix Bcont
Section IV presents a theoretical analysis that compares the is [1/Lf , −1/Lf ]; the output matrix Ccont is equal to one.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

Since an MPC is executed in an embedded controller at a

MPC controller Physical system (plant model)
constant sampling frequency, the continuous-time model (2)
is discretized using a zero-order-hold (ZOH) equivalent and a i∗1 FCS v vd vvsc i1
sampling period of value Ts . Although the ZOH equivalent is z −1 ZOH
MPC
an exact discretization that requires a higher computational vg
load compared to other approximations, such as the Euler vg Comp. VSC mod.
i1 L filter
delay delay
method, this operation does not add up to the computational
load of the controller algorithm because it is performed offline,
during the design process of the controller. The discrete-time
state-space model matrices are:
Fig. 2. Structure of a conventional finite control set (FCS)-model predictive
Adisc = exp {Acont Ts } controller (MPC).

Bdisc = Acont −1 (Acont − 1) Bcont

Cdisc = Ccont . (3) a sampling frequency two orders of magnitude greater than
the switching frequency [3], which significantly increases the
A ZOH accurately describes the FCS operation of the VSC, computational load of the controller. In this article, a simple
which introduces a half sample delay. The resultant discrete- cost function is selected due to its popularity [20], [38], [39]
time state-space model is and low computational load compared to an FCS-MPC with
a long prediction horizon. However, the proposed solution is
 
v (k)
xdisc (k + 1) = Adisc xdisc (k) + Bdisc d not affected by the complexity of the selected cost function
vg (k)
i1 (k) = Cdisc xdisc . (4) and can also be applied to an MPC with a more elaborated
cost function.
The subscripts “cont” and “disc” denote a continuous-time
and a discrete-time model parameter or variable, respectively. III. D ESIGN OF THE C URRENT C ONTROLLER
It should be noted that the resultant discrete-time state-space Compared to a PWM-based controller, an FCS-MPC op-
model is observable, but not controllable because one of the erates directly with actual switching signals (firing signals
inputs, the grid voltage vg , is an uncontrolled variable. and current measurements) instead of their average values, as
Finally, the discrete-time model (4) is augmented so as illustrated in Fig. 2. The switching signals contain a much
to also model a one sample delay in the plant input, which larger frequency content (higher bandwidth) than their average
describes the effect of a computational delay, i.e., vd (k + 1) = value counterparts; therefore, an FCS-MPC operates over a
v(k). The resultant plant model is much larger frequency range than a traditional PWM-based
        design. In fact, an FCS-MPC is regulating the switching noise
xdisc (k + 1) Adisc B1 xdisc (k) 0 B2 v(k) that is generated by the VSC during operation at the same
= +
vd (k + 1) 0 0 vd (k) 1 0 vg (k) time as it controls the fundamental component and the low-
| {z } | {z } | {z } | {z }
xplant (k+1) Aplant xplant (k) Bplant order harmonics. As a result of such high bandwidth, the FCS-


 xdisc (k)
 MPC inherently achieves a fast transient response. However,
i1 = Cdisc 0 a high bandwidth also reduces the robustness of the controller
| {z } vd (k)
Cplant | {z } to noise and plant model deviations. This problem is explained
xplant (k) in the following.
 T Since the controller commands actions at high frequencies,
xplant (k) = i1 vd , (5) it relies on a plant model that describes the plant response
where B1 and B2 are the first and second elements, respec- at high frequencies. However, a plant parameter variation is
tively, of Bdisc . This model relates the output vector v(k) more noticeable at high frequencies than at low frequencies.
selected by the controller to the the grid current i1 (k) in the For example, a one-sample delay introduces a negligible
plant shown in Fig. 2. phase error at the fundamental frequency. But it causes a
polarity inversion at half the sampling frequency. Similarly,
The selected cost function measures the one-step ahead a parameter deviation in an inductive component results in
current error using the following l1 norm: an impedance deviation that increases linearly with frequency.
∗
J(k) = ki1,α (k + 1) − i1,α (k + 1)k+ (6) Moreover, unmodeled effects such as core losses, skin effect,
∗ and measurement noise can become significant at the high
+ki1,β (k + 1) − i1,β (k + 1)k, (7)
frequencies that the MPC operates.
where i∗1,α and i∗1,β are the current references in the αβ In order to reduce the sensitivity of the controller to plant
frame. More complex cost functions have been defined in parameter variations and measurement noise, the proposal
the literature, at the expense of a higher computational load introduces an observer, as shown in Fig. 3. Observers are
and increased design complexity associated to the calculation frequently used in the literature; howerver, to the authors
of the weighting factors [37]. Recent proposals [3] advocate knowledge, an observer has not been previously used in the
towards including in the cost function a penalty on the field of MPC for power converter applications in order to
switching transitions. Such modification frequently results in obtain an estimation of the grid current î1 that improves the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

Proposed current controller Physical system (plant model) A/D i1,abc (t) vg,abc (t)
interrupt
i∗1 FCS v vd vvsc i1 A/D conversion A/D conversion
z −1 ZOH
MPC 3×1 3×1
vg
i1,abc (k) vg,abc (k)
real real
vg Comp. VSC mod. Clarke Clarke
î1 L filter
delay delay
transform. transform.
i1
Observer vg i1 (k) complex
Eqs. (8) and (9)
v(k − 1) vg (k) 4 add.
Observer 4 mul.
complex Eqs. (8) and (9) complex 32 flops
î1 (k) complex
35 add.
Fig. 3. Structure of the proposed current controller, which includes an
i∗1 FCS 14 mul.
observer.
7 comp.
complex MPC
161 flops
v(k) complex
actual measurement i1 . The observer computes every sampling v(k) αβ vector
period an estimation of the instantaneous current value î1 using z −1
complex Firing sig. Total comp. load
all information available, namely, the measured grid current, 3×1
193 flops
the previously commanded switching states, and the measured sabc (k) real
grid voltage.
VSC
The observer selected is the steady-state solution of a
Kalman filter for a linear time-invariant system, due to its
simple design and popularity across numerous research fields. Fig. 4. Execution chronogram of the proposed FCS-MPC plus observer and
computational load associated to each component.
Such Kalman filter for a linear system with a constant variance
noise model consists of the following two equations. The first
one is a prediction equation that estimates the state x̂p (k) from states of a two-level VSC. In order to obtain the current error,
the previous state estimate x̂disc (k − 1) and the last actuation it computes the plant model (5) and the associated current
on the plant v(k − 1). The subscript “p” denotes predicted. error (7) for every possible VSC output vector. The switching
  state with a lower current error is selected. This optimization
v(k − 1)
x̂p (k) = Aplant x̂plant (k − 1) + Bplant . (8) executes five complex additions, two multiplications, and a
vg (k − 1)
comparison operation every loop iteration. Since there are
The second one is a correction equation that modifies the pre- seven different VSC output vectors in the αβ frame, the FCS-
diction x̂p (k) based on the most recent current measurement MPC executes 161 real flops.
i1 (k): In the presented implementation (fs = 16 kHz), a constant
computational load with a value of 3.1 mega-flops per second
x̂plant (k) = x̂p (k) + Ko [i1 (k) − x̂p (k)], (9)
is obtained for the proposal and a value of 2.6 mega-flops per
where Ko is the Kalman gain. The procedure to calculate the second for the conventional FCS-MPC alone. As a reference
Kalman gain is given in the Appendix. for benchmark, these figures are lower than the computational
Fig. 4 shows an execution chronogram of the proposed power of typical microcontrollers, such as the TMS320F335
FCS-MPC plus observer that illustrates the operations exe- from the manufacturer Texas Instruments. Fig. 4 summarizes
cuted by the embedded controller every sampling period. As the computational load results.
shown, the proposed solution does not significantly increases
the computation burden on the embedded controller compared IV. C URRENT D ISTORTION
to a classical FCS-MPC. This section analyzes how the current distortion changes
when the proposed observer is included in the controller.
The performance is studied by simulating the operation of
A. Computational Complexity
a grid-tied inverter during steady-state for different intensities
The computational load of the proposal in number of of measurement noise and computing the current THD. The
floating point operations (flops) is detailed in the following measurement noise is added to the grid current measurements
and the obtained figure is compared to a typical FCS-MPC. before they are sent to the controller. The current THD in a
The proposed controller consists of an observer and a one- grid-tied inverter with an FCS-MPC depends on multiple pa-
step ahead FCS-MPC, as shown in Fig. 3. The observer rameters in addition to the type of controller, such as the filter
consists of two equations, namely, (8) and (9). Such equations parameter values, the switching frequency, and the DC bus
contain four complex additions and four complex products voltage. In order to assess any improvement in performance
in total. Since a complex product requires six real flops between a conventional FCS-MPC and the proposal, the setup
and a complex addition requires two real flops, the total parameters are maintained constant in all the presented tests.
computational load of the observer is 32 real flops. The simulated setup parameters used in this section are the
The FCS-MPC carries out an optimization loop that calcu- same as the parameters of the experimental setup presented in
lates the current error associated to each of the eight switching Section V.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

8 Contrarily, the proposal maintains a consistent performance

independently of the measurement noise intensity and the
Current THD % 7 MPC
Proposal current THD is maintained below a 4%.
6
Fig. 6 shows the current THD for a plant model variation
that consists of a change in the L filter impedance and a one
5 sample delay. The actual filter impedance Lactual is changed
from a value 30% smaller than the nominal value Lf to a
4 value 30% greater than nominal. For the nominal condition
3
Lactual /Lf = 1, the proposal provides a lower THD than a
0 0.025 0.05 0.075 0.1
conventional MPC. When the impedance of the L filter is
Measurement noise standard deviation σ [A] diminished without updating the controller, the conventional
MPC experiments a faster increase in current THD than
(a) the proposal due to its lower robustness to plant parameter
variations. On the other hand, the current THD diminishes
40 in both cases when the filter impedance increases due to
MPC the additional attenuation provided by a filter with a greater
Current THD %

30 Proposal inductance value.

It should be noted that, in a traditional electrical grid, the
20 presence of even harmonics indicates an unbalance between
the positive half of the current or voltage waveform and the
10 negative half. Those even harmonic components cause unequal
positive and negative peak values, which are called waveform
0 asymmetry. Waveform asymmetry has a harmful impact on the
0 0.2 0.4 0.6 0.8 1 loads sensitive to voltage or current peaks. Converters with an
Measurement noise standard deviation σ [A] FCS operation, such as FCS-MPC, generate switching noise
that is spread throughout a wide and continuous frequency
(b)
range. Therefore a direct application of the harmonic limits
Fig. 5. Comparison of the current THD of a conventional FCS-MPC and defined in [40], [41] can be difficult. This problem has
the proposal for different intensities of measurement noise. (a) Measurement
noise standard deviation sweeps from 0 to 0.1 A. (b) Measurement noise been recognized in [3], where the use of an FCS-MPC is
standard deviation sweeps from 0 to 1 A. not recommended for grid-tied inverters due to the difficulty
of meeting the even harmonic limits established for current
waveforms with a periodic distortion.
12.5
MPC
Current THD %

10
Proposal V. E XPERIMENTAL R ESULTS
7.5 This section experimentally validates the previous simula-
5
tion results. The experimental results are carried out in a 5-
kW VSC working as an inverter with a 700-V dc bus vdc
2.5 and connected to a 400-V line-to-line 50-Hz three-phase grid.
In addition to the actual grid, a three-phase grid emulator
0
is employed to simulate adverse grid condition that consists
0.7 1 1.3
of a highly distorted grid voltage an a sudden interruption.
Lactual /Lf
The controller (see Fig. 4) is executed in real-time at a
sampling frequency of 16 kHz, which results in an average
Fig. 6. Comparison of the current THD of a conventional FCS-MPC and the switching frequency of 2 kHz, in an embedded hardware
proposal for a plant model variation that consists of a change in the actual
filter impedance value Lactual from the nominal value Lf and an additional control platform from the German manufacturer dSpace. This
one-sample computational delay. platform is programmed using Simulink programming lan-
guage, Matlab scripts, and C code and it also provides a large
number of analogue input channels, compared to a traditional
Fig. 5 shows a comparison of the current THD of a oscilloscope. By adding the required external voltage and
conventional one-step ahead FCS-MPC and the proposal for current sensors, this platform is able to record in the same time
different intensities of measurement noise and no plant pa- base, i.e. simultaneously, the three-phase grid voltages, the
rameter deviations. In Fig. 5(a), the standard deviation of three-phase grid currents, and signals internal to the controller
the measurement noise is changed from zero (no noise) to a such as the grid-current in the dq frame or the grid current
value of 0.1 A, and in Fig. 5(b), the standard deviation of estimation provided by the observer. Such feature is employed
the measurement noise is changed from zero (no noise) to a to record the experimental results shown in this section.
value of 1 A. As shown, in a conventional FCS-MPC, the The L filter impedance value is indicated in Table I. The
THD rapidly increases as the measurement noise augments. selected filter inductance value and the switching frequency

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

TABLE I
Coupling
vg,abc E XPERIMENTAL S ETUP PARAMETERS
vdc filter
PCC Lg
Base values
Nominal power Po 5 [ kW ]
Grid voltage vg 230 RMS ]
[ Vphase
Grid frequency fg 50 [ Hz ]
Firing i1,abc vPCC L filter
signals i1 Filter inductance Lf 0.32 [p.u.]
abc abc
i∗1,dq± Current αβ αβ
i∗1 VSC
controller
e±jθ vg Sampling frequency fs 16 [ kHz ]
DC bus voltage vdc 700 [ VDC ]
θPLL PLL
Digital controller

(a)
however, there are numerous references that analyze this
problem in detail [46], [47]. In order to synchronize the VSC
with the grid, a synchronous reference frame phase-locked
VSC loop (PLL) with additional filtering to enhance its robustness
to low-order harmonics is used [48]. Figs. 7(a) and 7(b) show
a diagram and a photograph, respectively, of the experimental
setup.
DC source
Fig. 8 shows the experimental waveforms during a 10-A
Digital Coupling reference step in the d axis of the +dq frame for different
controller filter intensities of measurement noise. The experimental results
show that the the FCS-MPC presents a higher current noise
during steady-state operation compared to the proposal. The
difference in performance is larger as the measurement noise
increases, as explained in the previous section. Nevertheless,
the transient response is extremely fast, with a rise time of
2.5 ms and negligible overshoot in both cases.
Fig. 9 shows the experimental waveforms during an
interruption in the grid voltage for different intensities of
Three-phase
measurement noise. For this test, the VSC is connected to
ac source a three-phase ac source that has been programmed to generate
a highly distorted grid voltage before the interruption occurs.
(b) The current reference of the grid-tied inverter is set to zero;
Fig. 7. Experimental setup. (a) Diagram. (b) Photograph. therefore, the measured current is equal to the current error.
As shown, the current noise of the conventional FCS-MPC
is severely affected as the amplitude of the measurement
are commonly employed in low-voltage applications with two- noise increases. Conversely, the proposal maintains a more
level VSCs [42]–[45]. The tests presented in the article show consistent performance for the different intensities of measure-
the performance of the proposed controller for two different ment noise. Similarly to the reference tracking test, both the
grid impedance conditions. For the reference tracking tests proposal and the conventional FCS-MPC show an excellent
the grid-tied inverter is connected to the three-phase grid disturbance rejection capability. No transient events appear in
available in the lab. For the disturbance rejection tests the the current in spite of the large and sudden voltage disturbance.
inverter is connected to a three-phase grid emulator, configured It should be noticed that the shape of the current ripple changes
to generate a highly distorted voltage and an interruption. In when the grid voltage interruption occurs. This effect results
these two configurations, the value of the grid impedance is from the fact that a FCS-MPC varies its switching frequency
assumed to be low compared to the impedance of the L filter. depending on the modulation index value, and when the
It should be noticed that, since the grid impedance is in interruption occurs, there is a large change in the modulation
series with the L filter, a variation in the grid impedance can be index of the VSC.
approximated by a variation in the filter impedance. However, Fig. 10 repeats the previous reference tracking test but the
this is only an approximation because, contrarily to a deviation filter inductance value is a 30% lower than its nominal value.
in the L filter value, an increase in the grid impedance also As expected from the theoretical analysis, cf. Fig. 6, the con-
causes a coupling between the grid current and the voltage ventional MPC experiences a greater degradation compared to
at the PCC, which can affect the grid synchronization mech- the proposal. A plant parameter mismatch causes steady-state
anism. This feature is beyond the scope of this manuscript; error in addition to a greater distortion if no corrective actions,

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

vg,abc vg,abc
300 V/DIV 300 V/DIV

fsw = 2.62 kHz fsw = 1.86 kHz fsw = 2.24 kHz fsw = 1.72 kHz
i1,dq i1,dq
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(a) (b)

vg,abc vg,abc
300 V/DIV 300 V/DIV

fsw = 3.05 kHz fsw = 1.93 kHz fsw = 2.28 kHz fsw = 1.76 kHz
i1,dq i1,dq
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(c) (d)

vg,abc vg,abc
300 V/DIV 300 V/DIV

i1,dq i1,dq
fsw = 2.82 kHz fsw = 1.93 kHz fsw = 2.26 kHz fsw = 1.70 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(e) (f)
Fig. 8. Experimental waveforms during a 10-A reference step in the d axis of the +dq frame for different intensities of measurement noise. Measured grid
current in the dq frame i1,dq , in the abc frame i1,abc , and measured grid voltage vg . (a) FCS-MPC controller with a measurement noise intensity of 0.1 A. (b)
Proposal with a measurement noise intensity of 0.1 A. (c) FCS-MPC controller with a measurement noise intensity of 0.5 A. (d) Proposal with a measurement
noise intensity of 0.5 A. (e) FCS-MPC controller with a measurement noise intensity of 1.0 A. (f) Proposal with a measurement noise intensity of 1.0 A.

such as including a disturbance observer, are incorporated to the observer is able to remove most of the measurement
the controller. The proposed observer is not a disturbance noise from the current measurements, without removing the
observer; therefore, it does not completely eliminate steady- high frequency components that are actually in the current
state errors that may appear due to plant model variations or waveform when the current reference is changed. This feature
noise. Fig. 8 and Fig. 10 show that the steady-state error in the is most noticeable in Fig. 11(c), where it is highlighted using
proposal is lower than in the case of the conventional FCS- dashed red circles.
MPC. The proposal helps to improve the reference-tracking In addition to the voltage and current waveforms, all
steady-state error in the presence of noise. oscilloscope captures detail the switching frequency of the
Fig. 11 shows the difference between the estimated currents VSC. Since the oscilloscope captures show different transient
î1 and the measured currents i1 plus the noise. As shown, events, the average switching frequency is reported before

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

A 40-ms grid voltage A 40-ms grid voltage

vg,abc vg,abc
interruption interruption
250 V/DIV 250 V/DIV

i1,dq fsw = 2.77 kHz fsw = 7.68 kHz fsw = 2.77 kHz i1,dq fsw = 2.36 kHz fsw = 0.15 kHz fsw = 2.36 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
5 A/DIV 5 A/DIV

i∗
1,dq = 0 A i∗
1,dq = 0 A

10 ms/DIV 10 ms/DIV
(a) (b)

A 40-ms grid voltage A 40-ms grid voltage

vg,abc vg,abc
interruption interruption
250 V/DIV 250 V/DIV

i1,dq fsw = 3.10 kHz fsw = 5.75 kHz fsw = 3.10 kHz i1,dq fsw = 2.37 kHz fsw = 0.58 kHz fsw = 2.37 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
5 A/DIV 5 A/DIV

i∗
1,dq = 0 A i∗
1,dq = 0 A

10 ms/DIV 10 ms/DIV
(c) (d)

A 40-ms grid voltage A 40-ms grid voltage

vg,abc vg,abc
interruption interruption
250 V/DIV 250 V/DIV

i1,dq fsw = 3.03 kHz fsw = 5.29 kHz fsw = 3.03 kHz i1,dq fsw = 2.38 kHz fsw = 1.33 kHz fsw = 2.38 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
5 A/DIV 5 A/DIV

i∗
1,dq = 0 A i∗
1,dq = 0 A

10 ms/DIV 10 ms/DIV
(e) (f)
Fig. 9. Experimental waveforms during a 40-ms grid voltage interruption for different intensities of measurement noise. Measured grid current in the dq
frame i1,dq , in the abc frame i1,abc , and measured grid voltage vg . (a) FCS-MPC controller with a measurement noise intensity of 0.1 A. (b) Proposal with a
measurement noise intensity of 0.1 A. (c) FCS-MPC controller with a measurement noise intensity of 0.5 A. (d) Proposal with a measurement noise intensity
of 0.5 A. (e) FCS-MPC controller with a measurement noise intensity of 1.0 A. (f) Proposal with a measurement noise intensity of 1.0 A.

the transient event and after the transient event. In order Ideally, the switching frequency should fall to zero when the
to measure the average switching frequency, an interval of interruption occurs because the current reference is set to zero.
duration equal to 15 power line cycles (PLC) is used, except However, the switching frequency is not zero due to the effect
for the measurement of the switching frequency during the of noise.
grid interruption event (Fig. 9), which only lasts 40 ms; In summary, when the observer is added, the transient
hence a two PLC interval is used instead. The results show dynamics of the controller to reference commands and dis-
that the proposal achieves a reduction in the current noise turbances are not affected, in spite of the additional filtering
compared to a conventional FCS-MPC in the same conditions, provided by the observer during steady-state. The reason
without increasing the switching frequency. This effect is most is that the observer selectively filters out the unmodeled
noticeable during the grid voltage interruption shown in Fig. 9. dynamics and the measurement noise, but it preserves the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

vg,abc vg,abc
300 V/DIV 300 V/DIV

i1,dq fsw = 2.56 kHz fsw = 1.93 kHz i1,dq fsw = 2.22 kHz fsw = 1.74 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(a) (b)

vg,abc vg,abc
300 V/DIV 300 V/DIV

i1,dq fsw = 3.28 kHz fsw = 1.86 kHz i1,dq fsw = 2.24 kHz fsw = 1.79 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(c) (d)

vg,abc vg,abc
300 V/DIV 300 V/DIV

i1,dq fsw = 2.94 kHz fsw = 2.06 kHz i1,dq fsw = 2.45 kHz fsw = 2.04 kHz
5 A/DIV 5 A/DIV

i1,abc i1,abc
10 A/DIV 10 A/DIV

5 ms/DIV 5 ms/DIV
(e) (f)
Fig. 10. Experimental waveforms during a 10-A reference step in the d axis of the +dq frame for different intensities of measurement noise and a plant
model variation that consists of a 30-% reduction in the filter impedance value. Measured grid current in the dq frame i1,dq , in the abc frame i1,abc , and
measured grid voltage vg . (a) FCS-MPC controller with a measurement noise intensity of 0.1 A. (b) Proposal with a measurement noise intensity of 0.1 A.
(c) FCS-MPC controller with a measurement noise intensity of 0.5 A. (d) Proposal with a measurement noise intensity of 0.5 A. (e) FCS-MPC controller
with a measurement noise intensity of 1.0 A. (f) Proposal with a measurement noise intensity of 1.0 A.

frequency components associated to sudden changes in the VI. C ONCLUSION

current reference or the grid voltage. This is possible because
This article has presented an improvement for an FCS-MPC
the observer receives as inputs the VSC switching state v, the
controller that provides a higher robustness to plant parame-
measured voltage at the PCC vg , and the measured current
ter variations and measurement noise, without reducing the
i1 . The presented experimental results validate the theoretical
bandwidth of the controller or affecting its excellent transient
analysis and the previous simulation results.
response to reference steps and disturbances. The proposal
reduces the current distortion and the steady-state error of a an
FCS-MPC controller in the presence of noise. The proposed
modification does not significantly increase the computational

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

î1,abc Algorithm 1 Calculation of the Kalman Gain Ko

10 A/DIV
Inputs: N , Q, Aplant , Cplant , 
1: while kKo (k + 1) − Ko (k)k2 ≥  do
noise
fsw = 2.25 kHz fsw = 1.48 kHz
10 A/DIV 2: Update values from previous iteration
3: P(k) ← P(k + 1)
4: Ko (k) ← Ko (k + 1)
i1,abc + noise
10 A/DIV
5: Project the error covariance ahead
6: Pp ← Aplant P(k)Aplant T + Q
7: Compute the Kalman gain
8: Ko (k + 1) ← Pp Cplant T /(Cplant Pp Cplant T + N )
5 ms/DIV
(a) 9: Update the error covariance
10: P(k + 1) ← (I − Ko (k + 1)Cplant )Pp
î1,abc 11: end while
10 A/DIV

noise
A PPENDIX
10 A/DIV
fsw = 2.25 kHz fsw = 1.51 kHz C ALCULATION OF THE K ALMAN G AIN
This appendix details the procedure to calculate the Kalman
i1,abc + noise
gain Ko required to implement the Kalman filter equa-
10 A/DIV tions [49] presented in Section III. This method has also been
applied to calculate the Kalman gain of a disturbance observer
in [50].
The computation of the Kalman gain is performed offline
5 ms/DIV
to reduce the computational load of the controller. The cal-
(b)
culation process consists of the steps shown in Algorithm I.
Such Algorithm converges, after a few iterations, towards a
î1,abc
constant value in Ko .
10 A/DIV
The input parameters of the algorithm are the measurement
noise N , the process noise Q, the plant model matrices Aplant
noise and Cplant [cf. (5)], and a tolerance , which is used in the stop
fsw = 2.25 kHz fsw = 1.54 kHz condition to detect when convergence is achieved. A value of
10 A/DIV
 = 10−10 ensures a good precision in the coefficients.
The parameters N and Q permit to tune the observer to the
i1,abc + noise
amount of noise in the plant. However, all the tests presented
10 A/DIV
in the article have been carried out using the same observer
gain. A value of one is selected for the measurement noise
N and a value of one percent is used for the process noise
5 ms/DIV parameter Q. These values provide a good performance for the
(c) range of noise intensities that have been tested. For the setup
Fig. 11. Experimental waveforms during a 10-A reference step in the d axis parameters used in the article, the obtained Kalman gain Ko
of the +dq frame for different intensities of measurement noise. Estimated is the following vector:
grid current in the abc frame î1,abc , measurement noise, and measured grid  
current i1,abc plus noise. (a) Proposal with a measurement noise intensity of Ko = 0.1287 0.0087 . (10)
0.1 A. (b) Proposal with a measurement noise intensity of 0.5 A. (c) Proposal
with a measurement noise intensity of 1 A.
R EFERENCES
[1] P. Karamanakos, E. Liegmann, T. Geyer, and R. Kennel, “Model
predictive control of power electronic systems: Methods, results, and
challenges,” IEEE Open Journal of Industry Applications, vol. 1, pp.
burden on the embedded controller nor increases the switching 95–114, 2020.
[2] S. Vazquez, J. Rodriguez, M. Rivera, L. G. Franquelo, and M. No-
frequency of the VSC. This solution leverages the convenience rambuena, “Model predictive control for power converters and drives:
of observers from linear controller theory; hence it is a Advances and trends,” IEEE Trans. Ind. Electron., vol. 64, no. 2, pp.
valuable solution for practicing power-electronic engineers and 935–947, 2017.
[3] P. Karamanakos and T. Geyer, “Guidelines for the design of finite control
researchers due to its simplicity compared to some advanced set model predictive controllers,” IEEE Trans. Power Electron., vol. 35,
techniques often required in more complex MPC designs. no. 7, pp. 7434–7450, 2020.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

[4] S. A. Bonab and A. Emadi, “Mpc-based energy management strategy for [25] V. Yaramasu, M. Rivera, B. Wu, and J. Rodriguez, “Model predictive
an autonomous hybrid electric vehicle,” IEEE Open Journal of Industry current control of two-level four-leg inverters—part i: Concept, algo-
Applications, vol. 1, pp. 171–180, 2020. rithm, and simulation analysis,” IEEE Trans. Power Electron., vol. 28,
[5] D. Dewar, J. Rohten, A. Formentini, and P. Zanchetta, “Decentralised no. 7, pp. 3459–3468, 2013.
optimal controller design of variable frequency three-phase power elec- [26] O. Wallscheid and E. F. B. Ngoumtsa, “Investigation of disturbance
tronic networks accounting for sub-system interactions,” IEEE Open observers for model predictive current control in electric drives,” IEEE
Journal of Industry Applications, vol. 1, pp. 270–282, 2020. Trans. Power Electron., vol. 35, no. 12, pp. 13 563–13 572, 2020.
[6] P. Karamanakos, M. Nahalparvari, and T. Geyer, “Fixed switching [27] J. Kukkola and M. Hinkkanen, “State observer for grid-voltage sensor-
frequency direct model predictive control with continuous and discontin- less control of a converter under unbalanced conditions,” IEEE Trans.
uous modulation for grid-tied converters with LCL filters,” IEEE Trans. Ind. Appl., To be published.
on Control Sys. Tech., pp. 1–16, 2020. [28] D. Pérez-Estévez and J. Doval-Gandoy, “Grid-tied inverter with AC
[7] P. Karamanakos, T. Geyer, and R. P. Aguilera, “Long-horizon direct voltage sensorless synchronization and soft start,” IEEE Trans. Ind.
model predictive control: Modified sphere decoding for transient opera- Appl., vol. 55, no. 5, pp. 4920–4933, Sep./Oct. 2019.
tion,” IEEE Trans. Ind. Appl., vol. 54, no. 6, pp. 6060–6070, Nov./Dec. [29] O. Wallscheid, A. Specht, and J. Böcker, “Observing the permanent-
2018. magnet temperature of synchronous motors based on electrical funda-
[8] P. Karamanakos, T. Geyer, N. Oikonomou, F. D. Kieferndorf, and mental wave model quantities,” IEEE Trans. Ind. Electron., vol. 64,
S. Manias, “Direct model predictive control: A review of strategies that no. 5, pp. 3921–3929, 2017.
achieve long prediction intervals for power electronics,” IEEE Industrial [30] J. Li, X. Huang, F. Niu, C. You, L. Wu, and Y. Fang, “Prediction
Electronics Magazine, vol. 8, no. 1, pp. 32–43, 2014. error analysis of finite-control-set model predictive current control
[9] T. Geyer and D. E. Quevedo, “Performance of multistep finite control for ipmsms,” Energies, vol. 11, no. 8, 2018. [Online]. Available:
set model predictive control for power electronics,” IEEE Trans. Power https://www.mdpi.com/1996-1073/11/8/2051
Electron., vol. 30, no. 3, pp. 1633–1644, 2015. [31] C. Jia, X. Wang, Y. Liang, and K. Zhou, “Robust current controller for
[10] T. Geyer, P. Karamanakos, and R. Kennel, “On the benefit of long- ipmsm drives based on explicit model predictive control with online
horizon direct model predictive control for drives with lc filters,” in disturbance observer,” IEEE Access, vol. 7, pp. 45 898–45 910, 2019.
2014 IEEE Energy Conversion Congress and Exposition (ECCE), 2014, [32] Z. Yin, X. Han, C. Du, J. Liu, and Y. Zhong, “Research on model predic-
pp. 3520–3527. tive current control for induction machine based on immune-optimized
[11] T. Dorfling, H. du Toit Mouton, T. Geyer, and P. Karamanakos, “Long- disturbance observer,” IEEE Journal of Emerging and Selected Topics
horizon finite-control-set model predictive control with nonrecursive in Power Electronics, vol. 6, no. 4, pp. 1699–1710, 2018.
sphere decoding on an FPGA,” IEEE Trans. Power Electron., vol. 35, [33] M. Abdelrahem, C. M. Hackl, R. Kennel, and J. Rodrı́guez, “Efficient
no. 7, pp. 7520–7531, 2020. direct-model predictive control with discrete-time integral action for
[12] T. Geyer and D. E. Quevedo, “Multistep finite control set model PMSGs,” IEEE Transactions on Energy Conversion, vol. 34, no. 2, pp.
predictive control for power electronics,” IEEE Trans. Power Electron., 1063–1072, 2019.
vol. 29, no. 12, pp. 6836–6846, 2014. [34] J. Wang, F. Wang, Z. Zhang, S. Li, and J. Rodrı́guez, “Design and
implementation of disturbance compensation-based enhanced robust
[13] A. Andersson and T. Thiringer, “Assessment of an improved finite
finite control set predictive torque control for induction motor systems,”
control set model predictive current controller for automotive propulsion
IEEE Transactions on Industrial Informatics, vol. 13, no. 5, pp. 2645–
applications,” IEEE Trans. Ind. Electron., vol. 67, no. 1, pp. 91–100,
2656, 2017.
2020.
[35] G. F. Franklin, J. D. Powell, and M. L. Workman, Digital Control of
[14] B. Hassibi and H. Vikalo, “On the sphere-decoding algorithm i. expected
Dynamic Systems, 2nd ed. Addison Wesley Longman, Inc, 1990, p.
complexity,” IEEE Transs on Signal Process., vol. 53, no. 8, pp. 2806–
261.
2818, 2005.
[36] L. Harnefors, A. G. Yepes, A. Vidal, and J. Doval-Gandoy, “Multifre-
[15] Y. Han, C. Gong, L. Yan, H. Wen, Y. Wang, and K. Shen, “Multiobjective
quency current control with distortion-free saturation,” IEEE J. Emerg.
finite control set model predictive control using novel delay compensa-
Sel. Topics Power Electron., vol. 4, no. 1, pp. 37–43, Mar. 2016.
tion technique for PMSM,” IEEE Trans. Power Electron., vol. 35, no. 10,
[37] M. Norambuena, J. Rodriguez, Z. Zhang, F. Wang, C. Garcia, and
pp. 11 193–11 204, 2020.
R. Kennel, “A very simple strategy for high-quality performance of AC
[16] M. Parvez Akter, S. Mekhilef, N. Mei Lin Tan, and H. Akagi, “Modified machines using model predictive control,” IEEE Trans. Power Electron.,
model predictive control of a bidirectional ac–dc converter based on lya- vol. 34, no. 1, pp. 794–800, 2019.
punov function for energy storage systems,” IEEE Trans. Ind. Electron., [38] P. Acuña, L. Morán, M. Rivera, R. Aguilera, R. Burgos, and V. G. Age-
vol. 63, no. 2, pp. 704–715, 2016. lidis, “A single-objective predictive control method for a multivariable
[17] T. Dragičević, “Model predictive control of power converters for robust single-phase three-level npc converter-based active power filter,” IEEE
and fast operation of AC microgrids,” IEEE Trans. Power Electron., Transactions on Industrial Electronics, vol. 62, no. 7, pp. 4598–4607,
vol. 33, no. 7, pp. 6304–6317, 2018. 2015.
[18] Y. Zhang, J. Liu, H. Yang, and S. Fan, “New insights into model [39] J. Rodriguez, J. Pontt, C. A. Silva, P. Correa, P. Lezana, P. Cortes, and
predictive control for three-phase power converters,” IEEE Transactions U. Ammann, “Predictive current control of a voltage source inverter,”
on Industry Applications, vol. 55, no. 2, pp. 1973–1982, 2019. IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 495–503, Feb. 2007.
[19] J. A. Rohten, J. R. Espinoza, J. A. Muñoz, M. A. Pérez, P. E. Melin, [40] IEEE, “Recommended practice and requirements for harmonic control
J. J. Silva, E. E. Espinosa, and M. E. Rivera, “Model predictive control in electric power systems,” IEEE Std 519-2014, pp. 1–29, Jun. 2014.
for power converters in a distorted three-phase power supply,” IEEE [41] “Electromagnetic compatibility (EMC) - part 3-4: Limits - limitation of
Transactions on Industrial Electronics, vol. 63, no. 9, pp. 5838–5848, emission of harmonic currents in low-voltage power supply systems for
2016. equipment with rated current greater than 16 a,” IEC 61000-3-4, pp.
[20] P. Cortes, J. Rodriguez, C. Silva, and A. Flores, “Delay compensation in 1–29, Oct. 1998.
model predictive current control of a three-phase inverter,” IEEE Trans. [42] A. A. Rockhill, M. Liserre, R. Teodorescu, and P. Rodriguez, “Grid-filter
Ind. Electron., vol. 59, no. 2, pp. 1323–1325, 2012. design for a multimegawatt medium-voltage voltage-source inverter,”
[21] C. Xue, L. Ding, Y. Li, and N. R. Zargari, “Improved model predictive IEEE Trans. Ind. Electron., vol. 58, no. 4, pp. 1205–1217, Apr. 2011.
control for high-power current-source rectifiers under normal and dis- [43] M. Zabaleta, E. Burguete, D. Madariaga, I. Zubimendi, M. Zubiaga,
torted grid conditions,” IEEE Trans. Power Electron., vol. 35, no. 5, pp. and I. Larrazabal, “LCL grid filter design of a multimegawatt medium-
4588–4601, 2020. voltage converter for offshore wind turbine using shepwm modulation,”
[22] O. Wallscheid and E. F. B. Ngoumtsa, “Investigation of disturbance IEEE Trans. Power Electron., vol. 31, no. 3, pp. 1993–2001, Mar. 2016.
observers for model predictive current control in electric drives,” IEEE [44] E. Kantar and A. M. Hava, “Optimal design of grid-connected voltage-
Trans. Power Electron., vol. 35, no. 12, pp. 13 563–13 572, 2020. source converters considering cost and operating factors,” IEEE Trans.
[23] X. Zhang, L. Zhang, and Y. Zhang, “Model predictive current control Ind. Electron., vol. 63, no. 9, pp. 5336–5347, Sep. 2016.
for pmsm drives with parameter robustness improvement,” IEEE Trans- [45] K. Park, F. D. Kieferndorf, U. Drofenik, S. Pettersson, and F. Canales,
actions on Power Electronics, vol. 34, no. 2, pp. 1645–1657, 2019. “Weight minimization of LCL filters for high-power converters: Impact
[24] K. S. Alam, M. P. Akter, D. Xiao, D. Zhang, and M. F. Rahman, of PWM method on power loss and power density,” IEEE Trans. Ind.
“Asymptotically stable predictive control of grid-connected converter Appl., vol. 53, no. 3, pp. 2282–2296, May 2017.
based on discrete space vector modulation,” IEEE Trans. on Ind. Inform., [46] M. Boyra and J. L. Thomas, “A review on synchronization methods
vol. 15, no. 5, pp. 2775–2785, 2019. for grid-connected three-phase VSC under unbalanced and distorted

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJIA.2021.3074502, IEEE Open
Journal of Industry Applications

conditions,” in Eur. Conf. on Pow. Electron. and App., Aug. 2011, pp.
1–10.
[47] S. Zhou, X. Zou, D. Zhu, L. Tong, Y. Zhao, Y. Kang, and X. Yuan,
“An improved design of current controller for LCL-type grid-connected
converter to reduce negative effect of PLL in weak grid,” IEEE Trans.
Emerg. Sel. Topics Power Electron., vol. 6, no. 2, pp. 648–663, Jun.
2018.
[48] X. Wang, M. G. Taul, H. Wu, Y. Liao, F. Blaabjerg, and L. Harne-
fors, “Grid-synchronization stability of converter-based resources—an
overview,” IEEE Open Journal of Industry Applications, vol. 1, pp. 115–
134, 2020.
[49] G. C. Goodwin, S. F. Graebe, and M. E. Salgado, Control System Design,
1st ed. Prentice Hall, 2001, p. 355.
[50] D. Pérez-Estévez, J. Doval-Gandoy, A. G. Yepes, O. López, and
F. Baneira, “Generalized multi-frequency current controller for grid-tied
converters with LCL filter,” IEEE Trans. Ind. Appl., vol. 54, no. 5, pp.
4537–4553, Sep. 2018.

Diego Pérez-Estévez (S’15—M’20) received the

M.Sc. and the Ph.D. degrees from the University of
Vigo, Vigo, Spain, in 2014 and 2019, respectively.
Since 2014, he has been with the Applied Power
Electronics Technology Research Group. His re-
search interests include control of grid-connected
converters and distributed power generation systems.

Jesús Doval-Gandoy (M’99) received the M.S. and

Ph.D. degrees in electrical engineering from the
Polytechnic University of Madrid, Madrid, Spain,
and from the University of Vigo, Vigo, Spain, in
1991 and 1999 respectively.
He is a Professor and the head of the Ap-
plied Power Electronics Technology Research Group
(APET), University of Vigo. His research interests
are in the areas of ac power conversion.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/