IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO.

2, FEBRUARY 2021 1299

A Model Predictive Control Method for

Grid-Connected Power Converters
Without AC Voltage Sensors
Leilei Guo , Nan Jin , Member, IEEE, Yanyan Li, and Kui Luo

Abstract—A model predictive control (MPC) strategy for as wind power systems, and photovoltaic power systems, as they
grid-connected power converters is proposed in this article can convert direct current (DC) electric power to alternating
to reduce the current ripples without using ac voltage sen- current (AC) electric power smoothly [1]–[3].
sors. First, a sliding-mode observer is designed to estimate
the grid voltage with an adaptive compensation algorithm. To improve the power quality of grid-connected power con-
Thus, grid-voltage sensors are removed. The novelty of verters, many control strategies are proposed, such as grid-
the proposed grid-voltage observation method is that it voltage-oriented vector control [4], virtual-flux-oriented control
is inherent frequency adaptive without using the accurate [5], direct power control [6], and direct current control [7].
grid angular frequency. By implementing the MPC strategy
Conventionally, in these control systems, five sensors, including
using the estimated grid voltage, the current ripples are
reduced, especially under distorted grid-voltage conditions two ac current sensors, two ac voltage sensors, and one dc
because of the low-pass filter used in the observer. Next, to voltage sensor, are required, which increase the hardware cost.
further reduce the current ripples, a double-voltage vector- Moreover, once these sensors fail, the control system will lose
based MPC strategy is proposed based on the principle stability. As a result, to reduce the hardware cost and improve the
of the modulated MPC. Detailed theoretical analyses are
operation reliability of grid-connected power converters, voltage
also carried out to show its effectiveness for the first time.
Finally, experimental studies are carried out to verify the va- and current sensors need to be minimized.
lidity of the proposed ac voltage sensorless MPC strategy. In recent years, several methods have been proposed to reduce
the current and voltage sensors. In [8]–[11], to reduce the current
Index Terms—AC voltage sensorless, grid-connected
power converter, model predictive control (MPC), sliding- sensors, phase current reconstruction techniques are studied.
mode observer (SMO). However, the reconstruction precision is limited because of the
dead time effects as well as the sampling noise. As accurate ac
current is very important for the control and overload protection
I. INTRODUCTION of grid-connected power converters, it is preferable to sample it
N RECENT years, with the fast development of renewable instead of reconstruction.
I energy technologies, grid-connected power converters have
been widespread used in many renewable energy systems, such
As the ac current sensors cannot be removed, in [12]–[23],
strategies are presented to remove the ac voltage sensors.
Typically, there are two kinds of approaches, i.e., virtual-flux
Manuscript received April 29, 2019; revised September 4, 2019, observation-based methods and grid-voltage observation-based
November 14, 2019, and December 13, 2019; accepted January 20, methods. To calculate the virtual-flux, pure integration operation
2020. Date of publication February 5, 2020; date of current version
October 30, 2020. This work was supported in part by the National is required, which is sensitive to the dc-drift and initial bias in
Nature Science Foundation of China under Grant 51707176, and Grant practical application. In [12] and [13], virtual-flux is estimated
51607159, in part by the Key Scientific Research Project in Universities using first-order low-pass filter (LPF) instead of pure integrator
of Henan Province under Grant 18A470020, and Grant 20A470011,
in part by the Scientific and Technological Project in Henan Province to avoid these problems associated with pure integrator. How-
under Grant 202102210103, in part by the Open Fund of Operation ever, magnitude and phase errors are generated inevitably be-
and Control of Renewable Energy & Storage Systems (China Electric cause of the differences between the pure integrator and the LPF
Power Research Institute) under Grant 1810-00889, in part by the Youth
Talent Support Project of Hanan Province under Grant 2019HYTP021, [14], [15]. In [16], a second-order LPF is designed to estimate
and in part by Science & Technology Innovation Talents in Universities the virtual-flux without magnitude and phase errors. In [17] and
of Henan Province under Grant 18HASTIT025. (Corresponding author: [18], second-order generalized integrators are utilized to observe
Nan Jin.)
L. Guo, N. Jin, and Y. Li are with the School of Electrical and Infor- the virtual-flux. However, all these filter-based virtual-flux ob-
mation Engineering, Zhengzhou University of Light Industry, Zhengzhou servation methods are sensitive to grid frequency variations.
450002, China (e-mail: [email protected]; [email protected]; On the other hand, to control the grid-connected power
[email protected]).
K. Luo is with the State Key Laboratory of Operation and Control of converters without ac voltage sensors, grid-voltage estimation
Renewable Energy & Storage Systems, China Electric Power Research methods are researched in [19]–[23]. In [19], an adaptive neural
Institute, Beijing 100192, China (e-mail: [email protected]). estimator is presented to eliminate the grid-voltage sensors. In
Color versions of one or more of the figures in this article are available
online at https://ieeexplore.ieee.org. [20], a neural-network-based adaptation algorithm is proposed
Digital Object Identifier 10.1109/TIE.2020.2970638 to estimate the grid voltage. However, too many parameters
0278-0046 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1300 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

require to be adjusted online in [19] and [20]. In [21] and

[22], adaptive state observers are proposed for estimation of the
positive- and negative-sequence components of the grid voltage
under unbalanced grid conditions. Nevertheless, observer gains
should be selected carefully to ensure the stability of the ob-
server. In [23], a sliding-mode observer (SMO) is designed to
observe the grid voltage under unbalanced grid conditions, and
both of the positive- and negative-sequence components of the
grid voltage are obtained. However, the accurate grid angular
Fig. 1. Topology and VVs of the grid-connected power converter.
frequency is required. Although phase-locked loop (PLL) or (a) Topology. (b) VVs.
frequency-lock loop (FLL) can be utilized to observe the grid
angular frequency, the observation error caused by the detun-
ing of the controllers may enlarge the grid-voltage estimation is calculated based on the modulated MPC principle as it is
errors. So, further studies should be carried out to design the simple and effective. Moreover, a detailed theoretical analysis
grid-voltage observation methods without using the accurate is also carried out for the first time to show the effectiveness of
grid angular frequency. the proposed double-VV-based MPC strategy in theory, which
Besides, although Yang et al. [23] has studied the model provides a solid theoretical support for double-VV-based mod-
predictive power control strategy without ac voltage sensors, ulated MPC strategy. That is another contribution of this article.
the power ripples are enlarged as the conventional single-voltage Finally, experimental results are given to verify the effectiveness
vector (VV)-based model predictive control (MPC) method is of the proposed MPC strategy without ac voltage sensors.
adopted. As known, MPC has received much attention in the last
few years in the field of power electronics and power drives [24], II. CONVENTIONAL MPC METHODS
[25]. To reduce the power and current ripples, multi-VV-based
MPC strategies have been studied in [26]–[32]. Typically, there The topology and VVs of the grid-connected power converter
are two ways to calculate the duration time of each VV when are shown in Fig. 1. The basic VVs are u0 (000), u1 (100),
multiple VVs are applied per control period. The first way u2 (110), u3 (010), u4 (011), u5 (001), u6 (101), and u7 (111). In the
is based on the deadbeat control principle [26]–[28], and the conventional MPC system, one of the eight VVs is selected by
second one is named modulated MPC [29]–[32], it is assumed minimizing the defined cost function, and then it is applied for a
that the duration time of each VV is inversely proportional to whole control period to control the current of the grid-connected
its cost function value. Although the accurate duration time can power converter.
be calculated using the first way, a large calculation amount The mathematical model of the grid-connected power
is required. As an alternative way, modulated MPC has been converter shown in Fig. 1(a) can be depicted as
studied by many researchers in [29]–[32] in recent years as it dig
is simple but effective. However, although a lot of experimental L = −Rig + u − e (1)
dt
studies have shown the superiority of the modulated MPC, few
where u and ig are the output voltage and current of the grid-
papers have carried out studies to prove its effectiveness in
connected power converter on the stationary αβ reference frame,
theory. As it is not reasonable to assume that the duration time
L is the filter inductance, R is its parasitic resistance, e is the grid
of each VV is inversely proportional to its cost function value,
VV, and u = [uα , uβ ]T , ig = [igα , igβ ]T , e = [eα , eβ ]T .
detailed theoretical analyses are required to provide theoretical
To predict the current, the model shown in (1) is discretized,
supports for the modulated MPC.
and (2) is obtained as follows:
In this article, an MPC strategy for grid-connected power  
converters without ac voltage sensors is proposed. First, to R Ts
ig (k + 1) = 1 − Ts ig (k) + (u(k) − e(k)) (2)
remove the ac voltage sensors, an SMO with a novel frequency- L L
adaptive compensation algorithm is proposed to estimate the grid
where Ts is control period.
voltage. Different to the conventional grid-voltage estimation
To compensate the inherent one step delay, the current
strategies, the accurate grid angular frequency is not required
ig (k + 1) is first predicted based on (2) using the applied VV
in the proposed method. Thus, it is frequency adaptive without
u(k) in the last control period.
PLL and FLL. That is one of the most important contributions
Next, to select an optimal VV in the next control period, all
in this article. Then, a grid-voltage sensorless MPC system
the eight VVs depicted in Fig. 1(b) are substituted into (3) to
is established by implementing the predictive current control
predict the current at (k+2)th instant
using the estimated grid voltage. Because of the LPF used  
in the observer, the influences of the background harmonics R Ts
ig (k + 2) = 1 − Ts ig (k + 1)+ (u(k + 1)−e(k + 1)).
of the grid voltage are reduced. Thus, the current ripples are L L
reduced significantly, especially under distorted grid-voltage (3)
conditions. Additionally, to further reduce the current ripples, Then, all of the eight predicted currents are substituted into
a double-VV-based MPC strategy without ac voltage sensors the cost function given in (4) to evaluate their control effects.
is also presented in this article. The duration time of each VV The VV that can minimize the cost function g is selected as the

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: MPC METHOD FOR GRID-CONNECTED POWER CONVERTERS WITHOUT AC VOLTAGE SENSORS 1301

optimal one, which will be applied in the next control period

g = |igref (k + 2) − ig (k + 2)| (4)
where igref (k + 2) is the reference current, and igref =
[igαref , igβref ]T .
In the MPC system of grid-connected power converters, the
grid-voltage-oriented control is usually used to achieve power
decoupling control. The angle of the grid voltage is often ob-
tained by a PLL. And the grid voltage is often oriented on the Fig. 2. Conventional compensation algorithm.
d-axis. The active current igdref and the reactive current igqref
are often given as the reference currents.
Then, the reference current igref (k + 2) can be calculated can be obtained as follows:
based on (5) and (6)
 e = ksgn(Δig ) − [ksgn(Δig )]noise (10)
igαref (k + 1) = igdref cos θg − igqref sin θg
(5) where [ksgn(Δig )]noise stands for the sliding-mode noise con-
igβref (k + 1) = igdref sin θg + igqref cos θg tained in the sliding-mode term ksgn(Δig ).
 It can be seen from (10) that the grid voltages can be calcu-
igαref (k + 2) = 3igαref (k + 1) − 3igαref (k)+igαref (k − 1)
lated now. However, as known, there are a lot of sliding-mode
igβref (k + 2) = 3igβref (k + 1) − 3igβref (k)+igβref (k − 1) chattering in the term ksgn(Δig ). Thus, to get the fundamental
(6) frequency signal contained in ksgn(Δig ), LPF is required to
where θg is the angle of the grid voltages. remove the high frequency sliding-mode chattering, and an
From (2)–(6), it is easy to understand that to achieve the additional compensation algorithm is also required.
predictive current control, the grid voltage is often required,
which can be obtained by using two ac voltage sensors. B. Conventional Compensation Algorithm
Once the voltage sensors fail, the MPC system will lose its To get the grid voltages from the sliding-mode signal
stability. ksgn(Δig ), an LPF is used first to remove the high-frequency
To enhance the reliability of grid-connected power converters noise, shown as
and reduce the hardware cost, a new grid voltage observation
method is proposed in this article to remove the ac voltage eLPF = GLPF ksgn(Δig ) (11)
sensors. The details are depicted in the following. where eLPF = [eLPFα , eLPFβ ]T , and GLPF is shown as
ωc
III. GRID-VOLTAGE OBSERVATION GLPF = (12)
s + ωc
A. Design of the Grid-Voltage Observer
where ωc is the cutoff frequency of the LPF.
To get the grid voltage without ac voltage sensors, an SMO Although the high-frequency noise is eliminated by the
is designed in this article, which is shown as LPF, the fundamental frequency signal is also affected by
dige the LPF. Therefore, to get the actual grid voltage, additional
L = −Rige + u − ksgn(Δig ) (7) compensation is required.
dt
If the actual frequency of the grid voltage is known, the phase
where sgn () is the sign function, ige is the estimated current, and amplitude deviations caused by the LPF shown in (12) can
k is the sliding-mode gain, ige = [igeα , igeβ ]T , and Δig = be calculated and compensated easily, as illustrated in Fig. 2,
ige − ig . where E stands for the amplitude of the grid voltage, and ωg is
According to (1) and (7), (8) can be deduced as follows: the actual gird angular frequency.
dΔig From Fig. 2 it can be known that to get the actual grid voltage,
L = −RΔig − ksgn(Δig )+e. (8)
dt the actual gird angular frequency should be estimated first.
To ensure the stability of the SMO shown in (7), Although it can be obtained by PLL or FLL, the estimation error,
Δi∗g (dΔig /dt) should be less than zero. Thus, (9) can be which may be enlarged due to detuning of the proportional–
deduced as follows: integral controller, will reduce the observation accuracy of the
 grid voltage significantly.
−RΔig − k + e < 0 Δig > 0
. (9)
−RΔig + k + e > 0 Δig < 0 C. Proposed Frequency Adaptive
Compensation Algorithm
From (9), it can be concluded that to ensure the stability of
the SMO, k should be larger than max(|e|). To get the actual grid voltage without using the actual gird
When the SMO converges, it means Δig = 0 and angular frequency, a new compensation algorithm is proposed
dΔig /dt = 0. Thus, from (8) it is easy to understand that (10) in this article, which is depicted in Fig. 3.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1302 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

Fig. 4. Bode diagrams of the transfer function G.

Fig. 3. Improved compensation algorithm.

Finally, by substituting the estimated grid voltage in (17) into

The main principle of the proposed method shown in Fig. 3 is (2)–(4), the predictive current control without ac voltage sensors
that same LPFs produce same phase and amplitude deviations is achieved. Because of the LPF used in (11), the harmonics
for same frequency signals. Thus, another LPF is used here to contained in the grid voltage can be suppressed. Thus, the
estimate the phase and amplitude deviations caused by the first current ripples can be reduced, which is similar to other virtual
LPF. By using this LPF, another filtered grid voltage can be flux-based control strategies [5].
obtained, shown as
D. Parameter Designing
eLPF2 = GLPF eLPF (13)
It is known that there are two key parameters for the proposed
where eLP F 2 = [eLP F 2α , eLP F 2β ]T . SMO, i.e., the sliding-mode gain k and the cutoff frequency ωc
From (11) it is known that grid voltage eLPF has no high- of the LPF. Here, the design methods of them are given.
frequency noise because of the LPF. Thus, its phase θg1 and First, from (8) it is deduced that
amplitude E1 can be directly calculated as
⎧  LsΔig + RΔig + ksgn(Δig ) = e. (18)
⎨ E1 = e 2
LPFα + eLPFβ
2

(14) Meanwhile, it is known that
⎩θg1 = tan−1 eLPFβ .
eLPFα Δig
sgn(Δig ) = . (19)
In the same way, the phase θg2 and amplitude E2 of the filtered |Δig |
grid voltage eLPF2 can also be calculated as Based on (18) and (19), it can be deduced that
⎧ 
⎨ E2 = e 2 ee 1 1
LPFα2 + eLPFβ2
2

G= =k (20)
(15) e L |Δig | s + R
+ k
⎩θg2 = tan−1 eLPFβ2 . L L|Δig |
eLPFα2
where ee stands for the estimated grid voltage.
Next, based on (14) and (15), the phase and amplitude devia- Set |Δig | = 0.2 A, L = 0.02 H, R = 0.05 Ω, the bode diagrams
tions caused by the LPF given in (12) can be calculated, which of the transfer function G are shown in Fig. 4 with the sliding-
is shown as mode gain k set as 20, 50, 100, 150, and 200, respectively.

ΔE = E 1 It can be seen from Fig. 4 that when k is small, the high-
E2 (16) frequency noise can be well suppressed, while the dynamic
Δθg = θg1 − θg2 response speed is slow. To improve the dynamic response speed,
where ΔE denotes the amplitude deviation, and Δθg stands for k should be larger, while the high-frequency noise may also be
the phase deviation. enlarged. Thus, a tradeoff between dynamic and steady state
Then, as the deviations caused by the LPF have been calcu- performance should be made. In this article, considering that
lated, the actual grid voltage can be estimated by compensating max(|e|) is 86.6 V, k is chosen as 100.
these deviations, which is shown as Besides, the cutoff frequency of the LPF used in the proposed
 method also should be selected carefully as it has a large influ-
eα = E1 ΔE cos(θg1 + Δθg ) ence on the dynamic and steady-state performances. If a small
. (17) cutoff frequency is chosen, the low-order harmonic voltages
eβ = E1 ΔE sin(θg1 + Δθg )
can be well suppressed. However, the dynamic response speed
As can be seen above, the actual grid angular frequency is will be reduced. On the contrary, if a large cutoff frequency is
not required in this new compensation strategy. Nevertheless, selected, although the dynamic response speed will be enhanced,
the proposed compensation strategy is still frequency adaptive the low-order harmonic voltages cannot be well suppressed. So,
by directly calculating the deviations caused by the LPF based in this article, the cutoff frequency ωc is set as 100 πrad/s for a
on (14)–(16). tradeoff between dynamic and steady state performances.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: MPC METHOD FOR GRID-CONNECTED POWER CONVERTERS WITHOUT AC VOLTAGE SENSORS 1303

E. Parameter Sensitivity Analysis TABLE I

LOCATION OF THE REFERENCE VV
As the proposed SMO is parameter dependent, a parameter
sensitivity analysis is carried out here.
If there is a parameter mismatch, (7) can be rewritten as
dige
(L + ΔL) = −(R + ΔR)ige + u − ksgn(Δig ) (21)
dt
where ΔL and ΔR stand for the parameter deviation. TABLE II
SELECTION PRINCIPLE OF CANDIDATE VVS
According to (1) and (21), the grid-voltage estimation error
Δe can be deduced as
Δe = jω0 ΔLig + ΔRig . (22)
Transforming (22) to the synchronous rotating coordinate, it
can be deduced that
     
Δed −igq igd
= ΔLω0 + ΔR (23) where ti,uj + ti,uk = Ts , ti,uj , and ti,uk are the duration times
Δeq igd igq of uj and uk , respectively.
where Δed and Δeq are the grid-voltage magnitude observation To obtain the synthesized VV, ti,uj and ti,uk should be calcu-
errors, igd and igq are the grid currents on the synchronous lated first. There are mainly two kinds of approaches to calculate
rotating coordinate. the duration times. The first one is based on the deadbeat control
Furthermore, from (23), the grid-voltage phase observation principle, which is time-consuming. The second one is named
error Δθ0 can also be deduced, which is shown as modulated MPC, which is assumed that the duration time of
    each VV is inversely proportional to its cost function value.
eq + Δeq eq
Δθ0 = tan−1 − tan−1 . (24) As this kind of methods is simple but effective, it has been
ed + Δed ed studied in many papers in recent years to control three-phase
When the grid voltage is oriented on the d-axis, it can be active rectifiers [29], shunt active filters [30], ac/dc converters
known that ed = E and eq = 0, where E stands for the magni- [31], and three-phase inverters [32]. Thus, in this article, this
tude of the grid voltage. Then, based on (24), (25) is deduced kind of duration time calculating method is utilized. The main
  implementation steps are explained below.
ΔLω0 igd + ΔRigq
Δθ0 = tan−1 . (25) First, to reduce the calculating amount, the reference VV
ed uref (k + 1) is calculated first according to the deadbeat control
From (23) and (25), it is known that the proposed SMO is sen- principle, depicted as follows:
sitive to the parameter mismatches, which is similar with [23]. L
Besides, parameter mismatches will also affect the MPC, which uref (k + 1) = [igref (k + 2) − ig (k + 1)]
Ts
has been analyzed in [33]. To further improve the robustness,
parameters should be identified online [34]. + Rig (k + 1) + e(k + 1) (27)
where e(k + 1) is obtained by the proposed grid-voltage observer,
IV. VV DOUBLE-VV-BASED MPC STRATEGY and uref = [urefα , urefβ ]T .
A. Implementation Method Then, the VV plane is divided into six sectors equally, as
shown in Fig. 1(b). Based on (27), the location of the reference
Although the proposed grid-voltage sensorless MPC strategy
VV can be calculated, which is shown in Table I, where θu is
can reduce the current ripples by suppressing the high-frequency
calculated based on (28).
harmonics contained in the grid voltage using LPF, the current
When the location of the reference VV is got, 3 instead of
ripples are still quite large as only one VV among the eight VVs
12 synthesized VVs can be selected as candidate VVs. Then,
are selected and applied per control period.
only these three candidate VVs are required to be utilized to
In this article, to further reduce the current ripples, a simple
implement the MPC. As a result, the calculating amount is
but effective double-VV-based MPC strategy is proposed.
reduced significantly. Table II is given to show the candidate
Based on the 8 basic VVs, 12 synthesized VVs usi (i =
VV selection principle.
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12) are defined, as shown in
Fig. 1(b). They are synthesized by (u0 , u1 ), (u7 , u2 ), (u0 , u3 ),  
(u7 , u4 ), (u0 , u5 ), (u7 , u6 ), (u1 , u2 ), (u2 , u3 ), (u3 , u4 ), (u4 , u5 ), −1 urefβ
θu = tan . (28)
(u5 , u6 ), and (u6 , u1 ), respectively. urefα
The synthesized VV can be expressed as (26) according to
Second, when candidate VVs are chosen, the duration times
the volt–second balance principle
of each VV that composing the synthesized VV are calculated
ti,uj ti,uk based on the modulated MPC principle. For example, if the
usi = uj + uk (26)
Ts Ts reference VV is located at sector I, us1 , us2 , and us7 are

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1304 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

Fig. 6. Schematic diagram for analyzing the effectiveness of the

modulated MPC strategy.

As shown in Fig. 6, us7 is synthesized by u1 and u2 . Thus, it

can be expressed as
t7,u1 t7,u2
us7 = u1 + u2 . (33)
Fig. 5. Implementation steps of the proposed double-VV-based MPC Ts Ts
strategy.
Considering that t7,u1 + t7,u2 = Ts , (34) can be deduced as

selected as candidate VVs. Then, duration times of each VV that t7,u2

us7 = u1 + (u2 − u1 ). (34)
composing us1 , us2 , and us7 are calculated based on (29)–(31), Ts
respectively [29]–[32] According to the geometrical relationship depicted in Fig. 6,
 it can be obtained that |AD| = t7,u2 /Ts∗ |(u2 − u1)| and |BD| =
t1,u0 = G0 + G1 T s
G1
t7,u1 /Ts∗ |u2 − u1|. As t7,u1 /t7,u2 is equal to G2 /G1 according
(29)
t1,u1 = to (31), it can be deduced that |BD|/|AD| = G2 /G1 . Besides,
G0 + G1 T s
G0

 according to (32) it is easy to understand that G1 = |AC| and

t2,u7 = G7 + G2 T s
G2
G2 = |BC|. Thus, |BD|/|AD| = |BC|/|AC|.
(30) Assume that there is a point F located on the segment AB
t2,u2 = G7 + G2 T s
G7
and |BC| = |BF|. As |BC| + |AC|>|AB| and |AB| = |BF| + |AF|,
 |AC|>|AF| is obtained. Then, it can be deduced that
t7,u1 = G1 + G2 T s
G2
(31) |BC| |BD| |BF|
t7,u2 = G1 + G2 T s
G1
= < . (35)
|AC| |AD| |AF|
where Gi (i = 0, 1, 2, 3, 4, 5, 6, 7) is calculated based on (32) As |BD|/|AD| is smaller than |BF|/|AF|, the point F must lie
on the segment AD instead of BD. As |BC| = |BF|, ࢬBCF =
Gi = |uref (k + 1) − ui | . (32)
ࢬBFC is true. Meanwhile, as ࢬCDB is an exterior angle of the
When the candidate synthesized VVs are calculated according triangle CDF, ࢬCDB>ࢬBFC is true. Thus, ࢬCDB>ࢬBCF. As
to (26)–(32), they are substituted into (32) again to evaluate their ࢬCBD is smaller than π/3, ࢬBCF and ࢬBFC should be larger
control performance, and the one that can minimize the cost than π/3. Thus, it is concluded that ࢬCDB> π/3 >ࢬCBD. Then,
function given in (32) is chosen as the best one, which will be according to the law of sines, |BC|>|CD| is true. As |BC| is the
applied in the next control period. voltage control error of u2 , and |CD| is the voltage control error
For clarity, Fig. 5 is given to illustrate the detailed implemen- of us7 , it can be concluded that the control error is reduced when
tation steps of the proposed MPC strategy. us7 instead of u2 is applied.
In the same way, it can be deduced that |AC|>|CD|, meaning
that the control error of us7 is smaller than that of u1 . Similarly,
B. Effectiveness Analysis it can be verified that the control error of us1 is smaller than that
In recent years, a lot of papers have studied the modulated of u0 and u1 , and the control error of us2 is smaller than that of
MPC strategy. However, few papers have validated its effective- u7 and u2 . Thus, it can be understood that it is effective to apply
ness in theory. Although when the cost function of the VV is the 12 synthesized VVs defined in this article instead of the 8
large, its duration time should be short, it is still not reasonable basic VVs to reduce the control errors. Moreover, it can also
to assume that they are inversely proportional. In this article, be concluded that it is effective to calculate the duration times
to show the effectiveness of the proposed double-VV-based of each VV based on (29)–(32), which is the basic principle
modulated MPC strategy in Section IV-A, a detailed theoretical of the modulated MPC. Thus, the proposed double-VV-based
analysis is carried out. In this analysis, it is assumed that the modulated MPC strategy can be utilized to further improve the
reference VV is located at Sector I. current control performance.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: MPC METHOD FOR GRID-CONNECTED POWER CONVERTERS WITHOUT AC VOLTAGE SENSORS 1305

Fig. 7. Photo of the experimental platform.

TABLE III
SYSTEM PARAMETERS Fig. 8. Grid-voltage observation results of the conventional
compensation algorithm. (a) 50 Hz. (b) 55 Hz.

V. SIMULATION AND EXPERIMENTAL RESULTS

To verify the validity of the proposed strategy, an experimental
platform shown in Fig. 7 is established and experimental studies
are carried out. In this platform, a digital signal processor of
DSP28335 is taken as the main controller. A bidirectional dc
power source APL-II is employed to emulate the dc side. A Fig. 9. Grid-voltage observation results of the proposed frequency
adaptive compensation algorithm. (a) 50 Hz. (b) 55 Hz.
programmable ac power source Ametek MX30 is employed to
emulate the grid. YOKOGAWA DLM4000 series oscilloscope
is used to record the experimental waveforms. The parameters
used in the experiment are shown in Table III.

A. Experimental Results of the Proposed

Grid-Voltage Observer
In the first experiment, the proposed grid-voltage observer is
tested to show its effectiveness and superiority. The conventional Fig. 10. Dynamic grid-voltage observation results of the proposed
compensation algorithm shown in Fig. 2 and the proposed fre- frequency adaptive compensation algorithm.
quency adaptive compensation algorithm depicted in Fig. 3 are
compared. The results are shown in Figs. 8 and 9, where eα and
eβ stand for the actual grid voltages, while eeα and eeβ stand enlarge the current control error. To reduce the estimation error,
for the estimated ones. In these experiments, igdref is 3 A with the accurate gird frequency, which is used in the compensation
igqref = 0. The grid-voltage frequency is set as 50 and 55 Hz, algorithm, should be estimated by PLL or FLL.
respectively. On the contrary, as shown in Fig. 9, the proposed method in
In the conventional compensation algorithm, the grid angular this article can observe the gird voltage accurately even when
frequency ωg is set as 100π rad/s, which is similar to [35]. In the the gird frequency is changed without the knowledge of the
proposed frequency adaptive compensation algorithm, the grid accurate gird frequency, which shows its inherent frequency
angular frequency is not required. As shown in Fig. 8(a), when adaptive characteristic.
the gird frequency is 50 Hz, the grid voltage can be estimated Furthermore, to show the dynamic performance of the pro-
accurately by the conventional method. However, once the gird posed grid-voltage observer, another experiment is carried out
frequency is changed, there will be an obvious grid-voltage when the frequency of the grid voltage is suddenly changed from
estimation error, as shown in Fig. 8(b), which may further 50 to 55 Hz. The results are shown in Fig. 10. It is clear to see

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1306 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

Fig. 11. Grid-voltage observation results of the proposed frequency

adaptive compensation algorithm under distorted conditions.

TABLE IV
DIFFERENCES BETWEEN THE THREE METHODS

that the voltage estimation errors are small during the dynamic
process, and the recovery time is also less than one rated period,
which further verifies the effectiveness of the proposed method
in this article.
Additionally, Fig. 11 is given to further show the estimated
grid voltage using the proposed method under distorted condi-
tions. In this test, a 7th harmonic voltage with its amplitude set Fig. 12. Current control performance under ideal grid-voltage
conditions. (a) MPC-I. (b) MPC-II. (c) MPC-III.
as 8.7 V is injected into the grid voltage.
It is clear to see from Fig. 11 that actual grid voltage is
distorted. However, the harmonic is reduced in the estimated
grid voltage. That is mainly because LPFs are used to observe the
grid voltage, as shown in Fig. 3. As a result, the current ripples
can be reduced when the estimated grid voltage is utilized to
implement the MPC algorithm. That is the second advantage of
the proposed grid-voltage observation method.

B. Experimental Results Under Ideal

Grid-Voltage Conditions Fig. 13. Current THD analysis result.

To further verify the effectiveness of the proposed method,

other experimental studies are also carried out. Here,
for convenience, the conventional single-VV-based MPC As illustrated in Table IV, the main differences of the three
strategy with sampled grid voltage is named MPC-I. The methods lie in two parts. First, the strategy MPC-I uses voltage
single-VV-based MPC strategy without ac voltage sensors sensors to sample the actual grid voltage, while the other two
using the proposed grid-voltage observation method illustrated methods use the proposed SMO to estimate the grid voltage.
in Fig. 3 is named MPC-II. The proposed ac voltage sensorless Second, the strategy MPC-I and MPC-II choose and apply only
double-VV-based MPC strategy using the grid-voltage one optimal VV per control period, while the proposed strategy
estimation method depicted in Fig. 3 is named MPC-III. MPC-III applies two VVs per control period. Section IV-B has
In this test, igdref is stepped up from 3 to 10 A with igqref = 0. verified the effectiveness of the proposed two-VV based MPC
To make a fair comparison, the control period is set as 66.67 μs strategy in theory, which means that the voltage error shown
for all the three methods. The main differences of three methods in (32) can be further reduced compared to the single-VV-based
are summarized in Table IV. The results are illustrated in Fig. 12. MPC strategy. As the voltage error is reduced, the current control
As shown in Fig. 12(a) and (b), similar current control perfor- error and the current ripples can also be reduced, which has been
mance can be achieved by using the observed grid voltage, which analyzed in detail in [36]. Thus, the current THDs shown in
verifies the effectiveness of the proposed grid-voltage estimation Fig. 12(c) are smaller.
method. Moreover, it can be seen from Fig. 12(c) that by using Additionally, the current THD analysis result of the proposed
the proposed double-VV-based MPC strategy, the total harmonic method when igdref = 10 A and igqref = 0 is shown in Fig. 13
distortions (THDs) are reduced, which also shows the validity to further show its low-order harmonic along with the IEEE
of the proposed method as well as the theoretical analysis in Std.1547. It is clear that the current low-order harmonic does
Section IV-A and -B. not exceed the limit.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: MPC METHOD FOR GRID-CONNECTED POWER CONVERTERS WITHOUT AC VOLTAGE SENSORS 1307

Fig. 15. Experimental results of the double-VV-based MPC using the

conventional grid-voltage observation method. (a) 50 Hz. (b) 55 Hz.

Fig. 14. Current control performance under distorted grid-voltage

conditions. (a) MPC-I. (b) MPC-II. (c) MPC-III.

C. Experimental Results Under Distorted

Grid-Voltage Conditions
As known, the grid voltage is often distorted due to the
low-frequency harmonics caused by the nonlinear load, which
will affect the current control performance. In this section, the
control performance of the proposed method under distorted
grid-voltage conditions is tested. Here, igdref is set as 10 A
with igqref = 0. A seventh harmonic voltage is injected into
the three-phase grid voltage, which is the same as Fig. 11. The
experiment conditions are the same for all the three methods to
Fig. 16. Experimental results of the double-VV-based MPC using the
ensure the fairness of the comparison. The results are depicted proposed grid-voltage observation method. (a) 50 Hz. (b) 55 Hz.
in Fig. 14.
It is clear to see from Fig. 14(a) that the currents are distorted
when the conventional method is used. That is mainly because of
the influences of the distorted grid voltage. When the proposed grid-voltage observer on the double-VV-based MPC are further
grid-voltage sensorless control method is used, the current dis- tested. The proposed method shown in Fig. 3 and the conven-
tortion is reduced obviously, as shown in Fig. 14(b). The reason tional one depicted in Fig. 2 are compared. igdref is set as 10 A,
is that the harmonics contained in the grid voltage are reduced and igqref is set as 0. The frequency of the grid voltage is set as
because of the LPF used in the proposed grid voltage observation 50 and 55 Hz, and the results are depicted in Figs. 15 and 16,
method. Furthermore, when double VVs are applied per control respectively.
period, the current THDs can be further reduced, as illustrated Fig. 15 is the results when the conventional compensation al-
in Fig. 14(c), which shows the effectiveness of the proposed gorithm shown in Fig. 2 is used. In this algorithm, the frequency
method. of the grid voltage is assumed as 50 Hz. So, the grid voltage
can be observed precisely, as shown in Fig. 15(a), when the
actual grid-voltage frequency is 50 Hz. However, once the actual
D. Experimental Results Under Variable Frequency
frequency of the grid voltage is changed, there will be a phase
Grid-Voltage Conditions and amplitude estimation error in the estimated grid voltage,
The superiority of the proposed grid-voltage observer has as shown in Fig. 8(b). Thus, the current control error is also
been tested in Section V-A. In this section, the effects of the enlarged. As can be seen from Fig. 15(b), the actual current lags

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1308 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

Fig. 17. Experimental results with dc offset.

Fig. 19. Simulation results when the inductance is increased.

(a) Estimated grid voltage. (b) Reference currents. (c) Current control
error.
Fig. 18. Estimated grid voltages with and without white noise.
(a) Without white noise. (b) With white noise.
that when the white noise is added, the THD of the actual current
is only increased from 1.97% to 2.05%, which is very small.
Moreover, it should be pointed out that such a large white
behind the actual grid voltage when the grid-voltage frequency is noise (±1 A) is almost impossible to appear in reality. When it
55 Hz. The reason is that the estimated grid voltage lags behind is reduced, its influence can be further reduced.
the actual grid voltage in this situation, as depicted in Fig. 8(b).
On the contrary, when the proposed method shown in Fig. 3 is
utilized, the change of the gird voltage frequency has no effects G. Simulation Results Considering
on the double-VV-based predictive current control, as shown in Parameter Mismatches
Fig. 16, which shows the superiority of the proposed method. To show the effects of the parameter mismatches, another
simulation study is carried out. Here, igdref is set as 8 A with
E. Experimental Results Considering DC Offsets igqref = 0. The inductance used in the control system is changed
from 0.02 to 0.03 H at 0.02 s. The results are shown in Fig. 19,
Considering that dc offset is a common and unavoidable
where iαref is the actual reference current that is calculated
practical situation due to the nonlinearity of the sampling chan-
based on the reference current igdref , igqref , and the actual gird
nel, its influence on the proposed SMO is studied here. In this
voltage angle, while iαrefe is the estimated reference current that
experiment, igdref is set as 3 A with igqref = 0. DC offsets named
is calculated based on the reference current igdref , igqref , and the
iαdc and iβdc are added to the current iα and iβ , respectively.
estimated gird voltage angle.
Then, iα and iβ with dc offset are used to estimate the grid
It is clear to see from Fig. 19(a) that a voltage estimation error
voltage using the proposed SMO. Here, iαdc is stepped up from
is appeared when inductance is increased, and the estimated
0 to 1 A at the time t1 , while iβdc is stepped down from 0 to −1
grid voltage is lag behind the actual one. From Fig. 19(b) it
A at the time t2 . The results are shown in Fig. 17.
can be understood that the estimated current reference is lag
It can be seen from Fig. 17 that although there is a small
behind the actual one because of the lag of the estimated grid
grid-voltage observation error because of the dc offsets, it is
voltage. Besides, it can be seen from Fig. 19(c) that although
very small, which can be analyzed theoretically.
the estimated current reference iαrefe is changed because of
the variation of the inductance at 0.02 s, the current control
F. Simulation Results Considering White Noise error between iαrefe and iα is still very small, which shows
To further show the robustness of the proposed grid-voltage the robustness of the MPC algorithm. However, as there is a
observation method against the white noise, a simulation study large error between iαrefe and iαref , the actual current control
is carried out. Here, igdref is set as 8 A with igqref = 0. A white error between iαref and iα is large because of the inductance
noise with its amplitude set as ±1 A is added to the currents iα mismatch. To improve the robustness of the observer against
and iβ . Then, the currents with white noise are used to estimate parameter variation, parameter online compensation strategies
the grid voltage based on the proposed SMO. The observed grid should be designed [34], which will be studied in the future.
voltages are shown in Fig. 18. Additionally, the computation time of the proposed method in
It can be seen that the white noise added in the current has this article is also tested and compared with the deadbeat control
an influence on the grid-voltage observation, which will further principle-based MPC method, which shows that the proposed
affect the actual current. Fortunately, simulation results show method has a reduced calculation amount.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 GUO et al.: MPC METHOD FOR GRID-CONNECTED POWER CONVERTERS WITHOUT AC VOLTAGE SENSORS 1309

VI. CONCLUSION [16] A. M. Bozorgi, M. Sanatkar Chayjani, R. Mohammad Nejad, and M.

Monfared, “Improved grid voltage sensorless control strategy for railway
In this article, a grid-voltage sensorless MPC strategy was power conditioners,” IET Power Electron., vol. 8, no. 12, pp. 2454–2461,
proposed for grid-connected power converters. A new SMO Dec. 2015.
[17] J. A. Suul, A. Luna, P. Rodríguez, and T. Undeland, “Virtual-flux-based
with an inherent frequency adaptive compensation algorithm voltage-sensor-less power control for unbalanced grid conditions,” IEEE
was proposed to estimate the grid voltage. Then, a MPC Trans. Power Electron., vol. 27, no. 9, pp. 4071–4087, Sep. 2012.
strategy was designed using the estimated grid voltage. Thus, [18] J. A. Suul, A. Luna, P. Rodrıguez, and T. Undeland, “Voltage-sensor- less
synchronization to unbalanced grids by frequency- adaptive virtual flux
ac voltage sensors were removed. Because of the LPF used estimation,” IEEE Trans. Ind. Electron., vol. 59, no. 7, pp. 2910–2923,
in the observer, the impacts of harmonics contained in the Jul. 2012.
grid voltage on the current control performance were reduced. [19] A. Rahoui, A. Bechouche, H. Seddiki, and D. O. Abdeslam, “Grid voltages
estimation for three-phase PWM rectifiers control without AC voltage
Moreover, to further reduce the current THDs, a double-VV- sensors,” IEEE Trans. Power Electron., vol. 33, no. 1, pp. 859–875,
based MPC strategy without AC voltage sensors was also pro- Jan. 2018.
posed in this article. The effectiveness of this double-VV-based [20] Y. Abdel-Rady Ibrahim Mohamed, E. F. El-Saadany, and M. M. A.
Salama, “Adaptive grid-voltage sensorless control scheme for inverter-
MPC strategy was verified by theoretical analysis results. Fi- based distributed generation,” IEEE Trans. Energy Convers., vol. 24, no. 3,
nally, experimental studies were carried out, which also show pp. 683–694, Sep. 2009.
the validity and superiority of the proposed strategy in this [21] R. Fantino, C. A. Busada, and J. A. Solsona, “Observer-based grid- voltage
sensorless synchronization and control of a VSI-LCL tied to an unbalanced
article. grid,” IEEE Trans. Ind. Electron., vol. 66, no. 7, pp. 4972–4981, Jul. 2019.
[22] J. Kukkola and M. Hinkkanen, “State observer for grid-voltage sensorless
control of a converter under unbalanced conditions,” IEEE Trans. Ind.
REFERENCES Appl., vol. 54, no. 1, pp. 286–297, Jan./Feb. 2018.
[23] H. Yang, Y. Zhang, J. Liang, J. Gao, P. D. Walker, and N. Zhang, “Sliding-
[1] J. Scoltock, T. Geyer, and U. K. Madawala, “Model predictive direct power
mode observer based voltage-sensorless model predictive power control
control for grid-connected NPC converters,” IEEE Trans. Ind. Electron.,
of PWM rectifier under unbalanced grid conditions,” IEEE Trans. Ind.
vol. 62, no. 9, pp. 5319–5328, Sep. 2015.
Electron., vol. 65, no. 7, pp. 5550–5560, Jul. 2018.
[2] N. Altin, S. Ozdemir, H. Komurcugil, and I. Sefa, “Sliding-mode control
[24] L. Guo, N. Jin, C. Gan, and K. Luo, “Hybrid voltage vector preselection
in natural frame with reduced number of sensors for three-phase grid-tied
based model predictive control for two-level voltage source inverters to
LCL-interfaced inverters,” IEEE Trans. Ind. Electron., vol. 66, no. 4,
reduce the common-mode voltage,” IEEE Trans. Ind. Electron., to be
pp. 2903–2913, Apr. 2019.
published. doi: 10.1109/TIE.2019.2931257.
[3] H. Yang, Y. Zhang, J. Liang, J. Liu, N. Zhang, and P. D. Walker, “Robust
[25] L. Guo, N. Jin, C. Gan, L. Xu, and Q. Wang, “An improved model
deadbeat predictive power control with a discrete-time disturbance ob-
predictive control strategy to reduce common-mode voltage for two-level
server for PWM rectifiers under unbalanced grid conditions,” IEEE Trans.
voltage source inverters considering dead-time effects,” IEEE Trans. Ind.
Power Electron., vol. 34, no. 1, pp. 287–300, Jan. 2019.
Electron., vol. 66, no. 5, pp. 3561–3572, May 2019.
[4] Y. Jia, J. Zhao, and X. Fu, “Direct grid current control of LCL-filtered
[26] Z. Zhang, H. Fang, F. Gao, J. Rodríguez, and R. Kennel, “Multiple-vector
grid-connected inverter mitigating grid voltage disturbance,” IEEE Trans.
model predictive power control for grid-tied wind turbine system with
Power Electron., vol. 29, no. 3, pp. 1532–1541, Mar. 2014.
enhanced steady-state control performance,” IEEE Trans. Ind. Electron.,
[5] J. Kim and S. Kwak, “Model predictive virtual flux control to improve
vol. 64, no. 8, pp. 6287–6298, Aug. 2017.
performance under distorted input voltage conditions,” IEEE Access,
[27] S. Kwak, S. Kim, and J. Park, “Predictive current control methods with re-
vol. 6, pp. 34921–34933, 2018.
duced current errors and ripples for single-phase voltage source inverters,”
[6] Y. Zhang, Z. Li, Y. Zhang, W. Xie, Z. Piao, and C. Hu, “Performance
IEEE Trans. Ind. Informat., vol. 11, no. 5, pp. 1006–1016, Oct. 2015.
improvement of direct power control of PWM rectifier with simple cal-
[28] X. Guo, H. Ren, and J. Li, “Robust model predictive control for compound
culation,” IEEE Trans. Power Electron., vol. 28, no. 7, pp. 3428–3437,
active clamp three-phase soft switching PFC converter under unbalanced
Jul. 2013.
grid condition,” IEEE Trans. Ind. Electron., vol. 65, no. 3, pp. 2156–2166,
[7] V. Ambrozic, R. Fiser, and D. Nedeljkovic, “Direct current control-a new
Mar. 2018.
current regulation principle,” IEEE Trans. Power Electron., vol. 18, no. 1,
[29] L. Tarisciotti, P. Zanchetta, A. Watson, J. C. Clare, M. Degano, and S.
pp. 495–503, Jan. 2003.
Bifaretti, “Modulated model predictive control for a three-phase active
[8] Y. Cho, T. LaBella, and J.-S. Lai, “A three-phase current reconstruction
rectifier,” IEEE Trans. Ind. Appl., vol. 51, no. 2, pp. 1610–1620, Mar.–
strategy with online current offset compensation using a single current
Apr. 2015.
sensor,” IEEE Trans. Ind. Electron., vol. 59, no. 7, pp. 2924–2933,
[30] L. Tarisciotti et al., “Model predictive control for shunt active filters with
Jul. 2012.
fixed switching frequency,” IEEE Trans. Ind. Appl., vol. 53, no. 1, pp. 296–
[9] C. Gan, J. Wu, S. Yang, and Y. Hu, “Phase current reconstruction of
304, Jan./Feb. 2017.
switched reluctance motors from DC-link current under double high-
[31] X. Shi, J. Zhu, L. Li, and D. Lu, “Model predictive based duty-cycle-control
frequency pulses injection,” IEEE Trans. Ind. Electron., vol. 62, no. 5,
with simplified calculation and mutual influence elimination for AC/DC
pp. 3265–3276, May 2015.
converter,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 7, no. 1,
[10] S. Xiao, T. Shi, X. Li, Z. Wang, and C. Xia, “Single-current-sensor
pp. 504–514, Mar. 2019.
control for PMSM driven by quasi-Z-source inverter,” IEEE Trans. Power
[32] Y. Yang, H. Wen, and D. Li, “A fast and fixed switching frequency model
Electron., vol. 34, no. 7, pp. 7013–7024, Jul. 2019.
predictive control with delay compensation for three-phase inverters,”
[11] Y. X. Xu, H. Yan, J. B. Zou, B. C. Wang, and Y. H. Li, “Zero voltage
IEEE Access, vol. 5, pp. 17904–17913, 2017.
vector sampling method for PMSM three-phase current reconstruction
[33] H. A. Young, M. A. Perez, and J. Rodriguez, “Analysis of finite-control-
using single current sensor,” IEEE Trans. Power Electron., vol. 32, no. 5,
set model predictive current control with model parameter mismatch in a
pp. 3797–3807, May 2017.
three-phase inverter,” IEEE Trans. Ind. Electron., vol. 63, no. 5, pp. 3100–
[12] Y. Tao, Q. Wu, L. Wang, and W. Tang, “Voltage sensorless predictive direct
3107, May 2016.
power control of three-phase PWM converters,” IET Power Electron.,
[34] Y. Zhang, J. Jiao, and J. Liu, “Direct power control of PWM rectifiers with
vol. 9, no. 5, pp. 1009–1018, May 2016.
online inductance identification under unbalanced and distorted network
[13] J. Hu and B. Wu, “New integration algorithms for estimating motor flux
conditions,” IEEE Trans. Power Electron., vol. 34, no. 12, pp. 12524–
over a wide speed range,” IEEE Trans. Power Electron., vol. 13, no. 5,
12537, Dec. 2019.
pp. 969–977, Sep. 1998.
[35] H. Li, M. Yin, J. Ai, and W. Le, “Three-vector-based low-complexity
[14] X. Xiao, Y. Zhang, X. Song, T. Yildirim, and F. Zhang, “Virtual flux
model predictive direct power control strategy for PWM rectifier without
direct power control for PWM rectifiers based on an adaptive sliding
voltage sensors,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 7, no. 1,
mode observer,” IEEE Trans. Ind. Appl., vol. 54, no. 5, pp. 5196–5205,
pp. 240–251, Mar. 2019.
Sep./Oct. 2018.
[36] A. Abbaszadeh, D. A. Khaburi, H. Mahmoudi, and J. Rodríguez, “Sim-
[15] J. G. Norniella et al., “Improving the dynamics of virtual-flux-based
plified model predictive control with variable weighting factor for current
control of three-phase active rectifiers,” IEEE Trans. Ind. Electron.,
ripple reduction,” IET Power Electron., vol. 10, no. 10, pp. 1165–1174,
vol. 61, no. 1, pp. 177–187, Jan. 2014.
Aug. 2017.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
 1310 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 68, NO. 2, FEBRUARY 2021

Leilei Guo was born in Henan, China, in 1987. Yanyan Li was born in Henan, China, in 1987.
He received the B.S. and Ph.D. degrees in elec- She received the Ph.D. degree in control sci-
trical engineering from the School of Electrical ence and engineering from the College of Artifi-
Engineering and Automation, Hefei University cial Intelligence, Tianjin Key Laboratory of Intelli-
of Technology, Hefei, China, in 2010 and 2016, gent Robotics, Nankai University, Tianjin, China,
respectively. in 2017.
He is currently a Lecturer in Electrical En- She is currently a Lecturer in Electrical Engi-
gineering with Zhengzhou University of Light neering with Zhengzhou University of Light In-
Industry, Zhengzhou, China. He has published dustry, Zhengzhou, China. Her current research
more than 30 technical papers in journals interests include model predictive control for
and conference proceedings. His current re- power converters, grid impedance identification,
search interests include model predictive control of induction motors, and tolerant control of power electronics systems.
permanent-magnet synchronous motors, and power converters.
Dr. Guo was selected into the Young Talent Promotion Plan of Henan
province, in 2018.

Nan Jin (Member, IEEE) received the B.S.

and M.S. degrees in electrical engineering
from Zhengzhou University of Light Industry,
Zhengzhou, China, in 2003 and 2007, respec-
tively, and the Ph.D. degree in power electronics
and electrical drives from Shanghai Jiao Tong Kui Luo received the B.S. and M.S. degrees
University, Shanghai, China, in 2012. in electrical engineering from Hefei University
He is currently an Associate Professor in of Technology, Anhui, China, in 2010 and 2013,
Electrical Engineering with Zhengzhou Univer- respectively.
sity of Light Industry, Zhengzhou, China. He From 2014 to 2015, he was funded by the
was a Visiting Professor with the Department of Chinese Scholarship Council (CSC) as a Vis-
Electrical Engineering and Computer Science, University of Tennessee, iting Scholar with the Department of Electrical
Knoxville, TN, USA. He has published more than 50 technical papers in Engineering, Technical University of Denmark,
journals and conference proceedings, two books, and holds 12 Chinese Lyngby, Denmark. He is currently an engineer
patents. His research interests include model predictive control method in Renewable Energy Center, China Electric
for power converter, fault diagnosis, and tolerant control of power elec- Power Research Institute, Beijing, China. His
tronics system. research interests include distributed optimal control of distributed gen-
Dr. Jin was the recipient of the 2018 Highlighted Paper Award from eration, real-time, and power hardware-in-the-loop simulation.
IEEE TRANSACTIONS OF POWER ELECTRONICS.

Authorized licensed use limited to: Seoul National Univ of Science & Tech (SNUT). Downloaded on February 09,2021 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.