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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 1

Model-Free Predictive Current Control of PMSM

Drives Based on Extended State Observer Using
Ultra-Local Model
Yongchang Zhang, Senior Member, IEEE, Jialin Jin and Lanlan Huang

Abstract—Conventional model predictive current control to be tuned. Direct torque control (DTC) is another conven-
(MPCC) is a powerful control strategy for three-phase inverters tional control method for PMSM drives that directly uses the
that has the advantages of simple concept, quick response, stator flux linkage and output torque of the motor as the
easy implementation and good performance. However, MPCC is
sensitive to machine parameter variation, and the performance control target. DTC has many advantages, such as simple
degrades substantially if a mismatch exists between the model structure, quick dynamic response and strong robustness, but
parameters and real machine parameters. Model-free predictive DTC suffers from high torque ripples and variable switching
current control (MFPCC) based on an ultra-local model, which frequency [2], [3].
uses only the input and output of the system without considering Recently, model predictive control (MPC) has attracted
any motor parameters, has been proposed to solve this problem.
Since parameters are not required, the robustness of the control increasing attention due to the advantages of the clear con-
system is improved. However, conventional MFPCC based on cept, high dynamic performance and easy implementation
an ultra-local model uses many control parameters, which [4], [5]. According to the different control variables, MPC
increases the tuning work. Furthermore, the control performance can be divided into model predictive torque control (MPTC)
is not ideal at low sampling frequency. This paper proposes and model predictive current control (MPCC), which have
an improved MFPCC based on the extended state observer
(MFPCC-ESO) of PMSM drives that does not require motor achieved good control performance in PMSM control. MPCC
parameters and needs less tuning work and lower computational is a high-performance control method for PMSM drives [5]–
time while achieving better performance in terms of current [9]. However, MPCC relies heavily on the accuracy of the
harmonics, tracking error and dynamic overshoot. The proposed machine parameters [10]. In reality, the machine parameters
method is compared to conventional MPCC and MFPCC, and may vary with the operating point and environment; for
the effectiveness is confirmed by the simulation and experimental
results. example, temperature changes will alter the motor’s stator
resistance and stator inductance. If the parameter variations
Index Terms—Current control, predictive control, robustness, is not considered, the performance of MPCC deteriorates
ultra-local model, extended state observer
substantially [11]. For example, the steady-state and dynamic
performance will decrease, the current will show a steady-state
I. I NTRODUCTION error and the noise will increase during the operation of the
In recent years, permanent magnet synchronous motor motor [12].
(PMSM) drives have received widespread attention in industry To solve the problem of machine parameter dependence,
applications and electric vehicles because of the merits of various various control methods have been proposed in the
high efficiency and high power density. In PMSM drives, the literature. For example, online parameter identification method
stator current directly affects the dynamic and steady-state has been proposed to obtain accurate machine parameters
performance of the motor drive system [1]. [10], [13], [14]. In [14], an online parameter estimation
The conventional PMSM control method is field-oriented method based on a discrete-time dynamic model for PMSM
control (FOC), which controls the stator current vector of the is proposed. Two affine projection algorithms are designed
PMSM to control the torque of the motor directly. FOC is to estimate the stator inductance, stator resistance, rotor flux
widely used and has good steady-state performance. However, linkage, and load torque. However, online parameter identifi-
FOC requires the proportional-integral controller’s parameters cation requires a considerable number of calculations, which
increases the complexity of the system and the identification
Manuscript received August 8, 2019; revised December 20, 2019; accepted accuracy directly affects the control performance of the system
January 16, 2020. This work was supported by the National Natural Sci- [13].
ence Foundation of China under Grant 51577003. (Corresponding author:
Yongchang Zhang). Disturbance observer-based methods can achieve robust
Yongchang Zhang and Jialin Jin are with Inverter Technologies Engineering control against machine parameter variations by observing and
Research Center of Beijing, North China University of Technology, Beijing, compensating the disturbances caused by parameter changes
100144, China (email: yozhang@ieee.org; 353761790@qq.com).
Lanlan Huang was with North China University of Technology, Beijing, [15]. In [16], a predictive control method based on the
100144, China. She is now with Gui Zhou Aerospace Control Technology Luenberger observer is proposed. In [17], a stator current
Corporation, LTD, Guiyang, 550009, China (email: hll08020123@163.com). and disturbance sliding mode observer based on exponential
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. reaching law is proposed. In the 1980s, Chinese scholar
Digital Object Identifier 10.1109/TIE.2019.××××××× Jingqing Han proposed active disturbance rejection control

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
2 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

(ADRC) and the extended state observer (ESO) to estimate inductance, permanent magnet flux, and electrical rotor angle,
the system state, even when the model is unknown [18]. ESO respectively; and ω is the electrical rotor speed.
observes the total disturbance in the system by expanding it The machine equations of the stator current can be ex-
as a state variable to compensate the model uncertainty and pressed in the standard state-space form as:
parameter variations [18]. However, most of them is based
on mathematical model of PMSM and they still use machine dis 1
= (us − Rs is − jωψr ) (3)
parameters [19]. Furthermore, they are usually complicated in dt Ls
principle and require more tuning work.
Recently, the model-free control method proposed by III. C ONVENTIONAL MFPCC USING U LTRA - LOCAL
Michel Fliess in 2009 [20] has received wide attention in the M ODEL
fields of intelligent transportation, energy system management, The first-order ultra-local model of a single-input single-out
and so on. Some studies of MFPCC based on an ultra-local system is expressed as [24]:
model [20] for a PMSM drive system have been conducted
[21]. However, this method suffers from complex tuning work ẏ = F + αu (4)
and the performance may be deteriorated with relatively low where u and y are control and output variables, respectively;
sampling frequency, as confirmed in this paper. In [22], another α represents a nonphysical scaling factor selected by the
kind of model-free predictive current control based on current designer; and F represents the known and unknown part of
difference detection is applied to PMSM. However, this control the system.
strategy requires double current sampling in one control pe- Using the ultra-local model in (4), the control law of model-
riod, which increases the calculation burden of the system [23]. free control is given as [20]:
Furthermore, this method requires a high sampling frequency  
to achieve satisfactory performance. u = −F̂ + ẏ ∗ + Kp e /α (5)
To achieve robust control against parameter variations for
PMSM drives, improve the steady and dynamic performance where F̂ donates the estimated value of F , y ∗ is the desired
of MFPCC [21] with low sampling frequency and simplify output, Kp is the proportional gain and e = y ∗ − y donates
the complex tuning work, this paper proposes an MFPCC- the tracking error.
ESO method based on an ultra-local model [20] and ESO In [21], the principle of model-free control using ultra local
[18]. This method uses only the information of the present and model is applied to the current control of a PMSM drive and
past current to predict the future current value without using the corresponding controller is called “conventional MFPCC”.
any PMSM parameters. Although the model-free control using The reference of stator voltage in synchronous dq reference is
ultra-local model for a PMSM drive has been proposed in [21], calculated as:
there are noticeable differences between the conventional MF-  ˆ
PCC and the proposed MFPCC with ESO. Firstly, the control u∗ = −Fˆd +di∗d /dt+Kdp (i∗d −id )
d αd
parameters to be tuned or designed in the proposed method ˆ (6)
 ∗ −Fˆq +di∗
q /dt+Kqp (iq −iq )
∗
is much less than that of conventional MFPCC. Secondly, the uq = αq
performance of the proposed method is better than that of where the subscript “d” and “q” mean the d-axis and q-axis
conventional MFPCC in terms of current THD and dynamic component in synchronous reference.
overshoot. Finally, the computational time of the proposed
method is also lower than that of conventional MFPCC. IV. P RINCIPLE OF THE P ROPOSED MFPCC-ESO
A detailed comparison of the conventional MPCC, MFPCC A. ESO Based on Ultra-local Model
and the proposed method is carried out in this paper. The
simulation and experimental results confirm the effectiveness In this paper, according to formulas (3) and (4), the ultra-
of the proposed MFPCC-ESO. local mathematical model of PMSM can be expressed in a
stationary αβ frame using a complex vector as:
dis
II. M ODEL OF THE PMSM D RIVE S YSTEM = F + αus (7)
dt
This paper is based on a surface-mounted PMSM (SPMSM), where F = (−Rs is − jωψr ) /Ls is assumed to be the
whose dq-axis inductances are equal. To avoid synchronous unknown part and α = 1/Ls is the gain of the input value.
transformation and reduce the complexity of the control sys- Based on the ultra-local model of PMSM in (7), a linear
tem, the mathematical model of the PMSM is expressed in a ESO taking is and F as state variables with the feedback of
stationary αβ frame using complex vectors, as in [1]: current error can be constructed as [25]:
dψs 
us = Rs is + (1) err = z1 − is
dt

ψs = Ls is + ψr (2) ż1 = z2 + αus − β1 err (8)

ż2 = −β2 err

where ψr = ψf ejθe is the rotor flux vector; us , is , Rs , ψ s ,
Ls , ψf , θe , and ω = dθe /dt are the stator voltage vector, stator where z1 = iˆs and z2 = F̂ are the estimation of is and F in
current vector, stator resistance, stator flux vector, synchronous (7); β1 and β2 are the error feedback gains of the observer.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
ZHANG et al.: MODEL-FREE PREDICTIVE CURRENT CONTROL 3

B. ESO Parameter Design id iα* u * S

a
0
Te*  dq MFPCC S
b
In the proposed method, parameters β1 and β2 affect system r∗
Te* iq iβ* -ESO u * PWM
 Sc
PMSM
PI
1.5 pf

performance, so it is necessary to design the parameters of β1  Inverter
r i i
and β2 . (8) can be rewritten in matrix form as:
( 
ż = Az + Bus + D(y − ŷ) abc
(9)
ŷ = Cz 

    eQEP
z1 0 1
where z = , ŷ is the observation of y, A = ,
  z2   0 0 Fig. 1. Control diagram of the PMSM drive system with an ESO based on
α   β1 an ultra-local model.
B= ,C= 1 0 ,D= .
0 β2
According to (9), the characteristic equation of the ESO can
be expressed as: ω0 is such selected that z1,2 are within the unit circle of the
z domain.
id   Generally, ifu ω0 is too u smallS (z12 is close to 1), the
*
a
d
|sI − (A − DC)| = s2 + β1 s + β2 (10)
 
observer’s
iq    dynamic
id 基于超局部的
无模型算法 u
dq
performanceu will
PWM
*
q
ω0 is too
beS worse, and if IPMSM b

r  S
large (z12 isi close to zero), the robustness
ASR of the system will be
Inverter
c

where I refers to the identity matrix.  iq andd may lead to system divergence [26]. For PMSM

affected
r iq u u d q
To make the roots of the characteristic equation both fall at drives, the inner current loop usually requires a high bandwidth
dq 
−ω0 , β1 and β2 are obtained as to satisfy the requirements of dynamic response. Hence, in
abc

β1 = 2ω0 (11) this paper, z12 is set to 0.15 to achieve quick response and
the corresponding ω0 is 8500 when  the sampling frequency is
β2 = ω02 (12)
10 kHz. For other sampling frequency 编码器 such as 5 kHz and 20
where ω0 is the bandwidth of the LESO and it determines the kHz, the corresponding ω0 are 4250 and 17000, respectively.
steady-state and dynamic performance of the observer. The value of α is set to 50 in various sampling frequency.
It is not intuitive to directly design proper value of ω0 . In
this paper, the design of ω0 will be carried out in z domain.
The formula (8) can be discretized as: C. Reference Voltage Vector Calculation
 According to (7), uks can be obtained using first-order
err (k)

 = iˆs (k) − is (k)
  discretization as:
iˆs (k + 1) = iˆs (k) + Tsc F̂ (k) + αus (k) − β01 err (k)
ik+1 − iks

F̂ (k + 1) = F̂ (k) − β e (k)

s F̂ (k)
02 rr uks = − (18)
(13) αTsc α
where β01 = Tsc β1 and β02 = Tsc β2 are the gains of observer.
The values of β01 and β02 affect the distribution of the closed- where F is replaced by F̂ because F can not be directly
loop poles of the system, thereby affecting the stability of the obtained.
observer. Therefore, reasonable values of β01 and β02 must The final voltage reference can be obtained from (18) by
be selected to ensure the stability of the system and to ensure assuming ik+1
s reaches its reference value iref
s as:
that the controller achieves good control performance.
According to (13), the transfer function of ESO can be iref − iks F̂ (k)
s
obtained as uref
s = − (19)
αTsc α

iˆs (z) β01 z − β01 + β02 tsc where uref

s is the reference value of output voltage vector and
G (z) = = 2 (14) it is synthesized by SVM to generate the final gating pulses
is (z) (z − 1) + β01 z − β01 + β02 tsc
in this paper.
and its characteristic equation is Considering one-step delay in digital implementation, (19)
can be rewritten as:
z 2 + (β01 − 2) z + 1 − β01 + β02 tsc = 0 (15)
iref
s (k + 2) − ik+1
s F̂ (k + 1)
Considering β01 = 2ω0 tsc and β02 = ω02 tsc , the poles of uref
s = − (20)
αTsc α
G (z) is solved from (15) as
where
z1,2 = 1 − ω0 tsc (16)
k
j (θr +2ωr Tsc )
and ω0 can be calculated from the z domain poles as iref
s (k + 2) = iref
s (k) · e (21)
1 − z12
ω0 =
tsc
(17) ik+1
s and F̂ (k + 1) can be obtained from (13).

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
4 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

TABLE I
S YSTEM AND CONTROL PARAMETERS

DC-bus voltage Udc 540 V 2000

Rated power PN 2.4 kW

(rpm)
Rated voltage UN 380 V
Rated current IN 4.1 A

r
Rated frequency fN 100 Hz
Rated torque TN 10 Nm 0
Pole pairs Np 4
5
Stator resistance Rs 2.34 Ω
Inductance of d-axis Ld 19.36 mH

Iq(A)
Inductance of q-axis Lq 19.36 mH Iq ref
Permanent magnet flux ψf 0.402 Wb Iq
Parameter 1 α 50 1
Parameter 2 ω0 8500
Sampling period Tsc 100 µs 5

Ia(A)
0

V. S IMULATION AND E XPERIMENTAL R ESULTS -5

To confirm the effectiveness of the proposed method, simu- 0 0.2 0.4 0.6 0.8 1
time(s)
lations and experimental tests are implemented with SPMSM
(a)
based on a two-level inverter. The control diagram of the
PMSM drive system based on the proposed method is shown 2000
in Fig. 1. Table I lists the parameters of the control system used

(rpm)
in the digital simulation and experimental tests. The sampling
frequency is set to 10 kHz unless explicitly indicated. The r
value of ω0 is 4250, 8500 and 17000 when the sampling 0
frequency is 5 kHz, 10 kHz and 20 kHz, respectively. 5
Iq(A)

Iq
A. Simulation Results
Iqref
1
In order to verify the effectiveness of the proposed method,
simulation comparisons of conventional model-based MPCC 5
[6], MFPCC using ultra-local model [21] and the proposed
Ia(A)

0
MFPCC-ESO are performed in Matlab/Simulink.
Fig. 2 present the simulation results of the conventional -5
MPCC, MFPCC and the proposed MFPCC-ESO with accurate 0 0.2 0.4 0.6 0.8 1
machine parameters under the same sampling frequency of time(s)
10 kHz. The test condition is that the speed reference steps (b)
from 25% to 50% rated speed at t=0.1 s and then steps to
2000
rated speed of 1500 rpm at t=0.5 s. The first channel shows
(rpm)

the reference and actual speed of PMSM, the second channel

shows the reference and actual current of the q-axis and
r

the last channel presents the stator current of the A-phase.

0
It is seen that conventional MPCC presents the best steady 5
state performance with accurate machine parameters, followed
Iq(A)

by the proposed MFPCC-ESO. The conventional MFPCC Iq

presents higher current ripples than other two methods and Iq ref
there is evident overshoot in the q-axis current for stepped 1

change in the q-axis current reference. 5

By increasing the sampling frequency from 10 kHz to 20
Ia(A)

0
kHz in Fig. 3, the current ripples are significantly reduced in
the three methods, but there are still evident overshoot in the
-5
conventional MFPCC. The proposed MFPCC-ESO presents 0 0.2 0.4 0.6 0.8 1
very similar performance to conventional MPCC in terms of time(s)
current ripples and dynamic response. (c)
For PMSM, the stator inductance and permanent magnet Fig. 2. Simulated dynamic response with a sampling frequency of 10 kHz
flux have a strong influence on the control performance. under accurate parameters for (a) conventional MPCC, (b) MFPCC and (c)
proposed MFPCC-ESO.
Therefore, the simulation is conducted under the condition
that the inductance and flux linkage of the motor change
simultaneously.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
ZHANG et al.: MODEL-FREE PREDICTIVE CURRENT CONTROL 5

2000 2000
(rpm)

(rpm)
r

r
0 0
5 11
Iq ref

Iq(A)
Iq(A)

Iq
Iqref 6
Iq
1 2

10
5
Ia(A)

Ia(A)
0 0

-5
-10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
time(s) time(s)
(a) (a)

2000 2000
(rpm)

(rpm)
r

r
0 0
5 11
Iq
Iq(A)

Iq(A)

Iq Iq ref
6
ref
Iq
1 2

5 10
Ia(A)

Ia(A)

0 0
-5
-10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
time(s) time(s)
(b) (b)
2000 2000
(rpm)

(rpm)
r

0 0
5 11
Iq
Iq(A)

Iq ref
Iq(A)

Iq 6
ref
Iq
1 2

5 10
Ia(A)

Ia(A)

0 0

-5 -10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
time(s) time(s)
(c) (c)
Fig. 3. Simulated dynamic response with a sampling frequency of 20 kHz Fig. 4. Simulated dynamic response with a sampling frequency of 10 kHz
under accurate parameters for (a) conventional MPCC, (b) MFPCC and (c) under mismatched parameters for (a) conventional MPCC, (b) MFPCC and
proposed MFPCC-ESO . (c) proposed MFPCC-ESO.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
6 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

4
MPCC

Phase current harmonic content (%)

3.5 MFPCC
2000
MFPCC-ESO
(rpm)

3
accurate paramaters 2L and 0.5 f
r

2.5
0
11 2
Iqref
Iq(A)

Iq 1.5
6

2 1

10 0.5
Ia(A)

0 0
10 kHz 20 kHz 10 kHz 20 kHz
-10 (a)
0 0.2 0.4 0.6 0.8 1 Fig. 6. Simulated FFT analysis for a phase current with accurate parameters
time(s) and mismatched parameters.
(a)

2000
Fig. 4 shows the simulation results of the three methods
(rpm)

under the same sampling frequency of 10 kHz. The test

condition is the same as that in Fig. 2 and 3, except that the
r

permanent magnet flux and dq-axis inductance in the controller

0
11
are 50% and 200% of the actual values in the machine.
Iq It is seen that, in this scenario, the mismatched parameters
Iq(A)

6 Iq ref clearly have a considerable influence on the MPCC. The q-

axis current cannot cannot track the reference value accurately
2
but presents some evident offset. Furthermore, there is evident
10 oscillations in the q-axis current during the stepped change in
the q-axis current reference. On the contrary, the steady state
Ia(A)

0 tracking error of q-axis current is eliminated in both MFPCC

and the proposed MFPCC-ESO, exhibiting strong robustness
-10
against machine parameter variations. However, there is still
0 0.2 0.4 0.6 0.8 1
time(s) evident overshoot in the q-axis current of MFPCC. The results
(b) confirm that the proposed MFPCC-ESO has the best overall
performance in terms of current ripples, tracking accuracy and
2000
dynamic overshoot.
(rpm)

Similar conclusion can be obtained when increasing the

sampling from 10 kHz to 20 kHz, as shown in Fig. 5. With
r

20 kHz sampling frequency, the q-axis current error of the

0
conventional MPCC is reduced than that in Fig. 4 with 10 kHz
11
Iq sampling frequency, but the current error is not eliminated. The
Iq(A)

6 Iq ref conventional MFPCC can eliminate the current error but there
is still evident overshoot in the q-axis current. Again the best
2 overall performance is obtained in the proposed MFPCC-ESO.
10 Fig. 6 shows the quantitative steady-state performance com-
parison of conventional MPCC, MFPCC and the proposed
Ia(A)

0 MFPCC-ESO in terms of current total harmonic distortion

(THD), which is calculated up to 12.5 kHz harmonics. It
-10
0 0.2 0.4 0.6 0.8 1
is seen that, with accurate machine parameters, the conven-
time(s) tional MPCC presents the lowest current THD at 10 kHz
(c) and 20 kHz sampling frequency, followed by the proposed
Fig. 5. Simulated dynamic response with a sampling frequency of 20 kHz method and conventional MFPCC. However, the difference
under mismatched parameters for (a) conventional MPCC, (b) MFPCC and between MPCC and the proposed method is insignificant. With
(c) proposed MFPCC-ESO. mismatched machine parameters, it is clearly seen that both
MFPCC and the proposed MFPCC-ESO outperform the con-
ventional MPCC and the proposed method exhibits the lowest

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
ZHANG et al.: MODEL-FREE PREDICTIVE CURRENT CONTROL 7

1500 14
(rpm)
1000 MPCC MFPCC MFPCC-ESO
500 ref
12
0
10
iref iq
i q(A)

5 q 10
0

Current THD(%)
0 8
i d(A)

-5 iref id
d
-10
-15 6

ierr ierr
Ierror(A)

10
d q
0 4
-10
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01 2
time(s)
(a)
0
Fig. 7. Simulated response of MFPCC-ESO with a sudden parameter 5 10 20
variations under the condition of 10 kHz sampling frequency Sampling Frequency (kHz)

Fig. 9. Current THD of three methods under accurate parameters.

Magnetic
powder
Oscilloscope brake The rotor speed, q-axis current and its reference are displayed
via an on-board DA converter, and the phase current is directly
Inverter measured by a current probe. All waveforms are displayed by
a digital oscilloscope.
SPMSM Current Probe Fig. 9 presents the current THD for conventional MPCC,
MFPCC and the proposed MFPCC-ESO with accurate ma-
chine parameters at rated speed of 1500 rpm. It can be
seen that the proposed MFPCC-ESO presents very similar
performance to conventional MPCC with accurate machine
parameters. Compared to conventional MFPCC, the current
Fig. 8. Experimental platform of a two-level inverter-fed SPMSM drive.
THDs in the proposed method are reduced by 19.0%, 5.8%
and 11.4% with the sampling frequency of 5 kHz, 10 kHz
and 20 kHz, respectively. The current THD reduction is more
current THD. It should be noted that the current tracking error pronounced in the case of low sampling frequency of 5 kHz.
in MPCC and the oscillation during the dynamic process in Fig. 10 presents the starting response from standstill to
MFPCC do not appear in the proposed method, as shown in rated speed and high speed reversal at rated speed for the
Fig. 4 and 5. Hence, it can be concluded that the proposed three methods with a sampling frequency of 10 kHz under
MFPCC-ESO presents the best overall performance among accurate parameters. It is seen that the proposed method
the three methods in comparison when there are parameter exhibits excellent dynamic performance, which is similar to
mismatches. that of conventional MPCC. For conventional MFPCC, there
Fig. 7 presents the simulated response of MFPCC-ESO with are evident overshoot in the q-axis current during the dynamic
a sudden parameter variations under the condition of 10 kHz process, which does not appear in other two methods. The
0
sampling frequency. The motor parameters change to L = experimental results are in accordance to the simulation results
0 0
0.8L, ψf = 1.5ψf , Rs = 0.5Rs at t=0.005 s. In this figure, in Fig. 2.
the first three channels shows the reference and actual value of Fig. 11 shows the dynamic response to stepped load torque
speed, q-axis current and d-axis current, respectively. The last change with a sampling frequency of 10 kHz under accurate
channel shows the error between the estimated current and the parameters. It is seen that the dynamic response of the three
actual current in (13). It is seen that the current error caused methods are very similar. The q-axis current increases quickly
by machine parameter change is reduced to zero quickly and to balance the sudden load and the motor speed recovers to
the dq-axis currents recovers to their reference value in a short its reference value quickly. However, there are some tracking
time. The simulation results confirm the effectiveness of the error in the conventional MFPCC, which does not appear in
proposed method in suppressing the influence of parameter the proposed method. The results confirm the effectiveness
mismatch. of the proposed method in achieving quick dynamic response
with accurate machine parameters.
B. Experimental Results The results above are obtained with accurate machine pa-
The proposed MFPCC-ESO, conventional MPCC and MF- rameters. When the machine parameters used in the controller
PCC are experimentally tested on a two-level inverter-fed 2.4 differ significantly from the actual machine parameters, the
kW/380 V/100 Hz SPMSM drive platform, as shown in Fig. 8. conventional MPCC even cannot start normally. As shown

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
8 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

r
[1500rpm/div]
r
[1500rpm/div]
iref
q
[10A/div] iq [10A/div]
iref
q
[4A/div]
iq [4A/div]

ia [5A/div]
ia [5A/div]

(a) (a)

[1500rpm/div]
r
r
[1500rpm/div]
iref
q
[10A/div] iref [4A/div]
q

iq [10A/div] iq [4A/div]

ia [5A/div]
ia [5A/div]

(b) (b)

[1500rpm/div]
r
r
[1500rpm/div]
iref
q
[10A/div] iq [10A/div] iref [4A/div]
q

iq [4A/div]
ia [5A/div]
ia [5A/div]

(c) (c)
Fig. 10. Responses of starting from standstill to rated speed and high speed Fig. 11. Dynamic response to stepped load torque change with a sampling
reversal at rated speed with a sampling frequency of 10 kHz under accurate frequency of 10 kHz under accurate parameters for (a) conventional MPCC,
parameters for (a) conventional MPCC, (b) MFPCC and (c) MFPCC-ESO. (b) MFPCC and (c) MFPCC-ESO.

in the simulation results and studied in other literature [10], The PMSM parameters are set as follows in Fig. 12: the
conventional MPCC suffers from the inaccurate machine pa- dq-axis inductance, stator resistance and flux linkage in the
rameter seriously. Considering the page limitations and that controller are 200%, 1000% and 70% of the actual values,
the focus of this paper is to propose an improved MFPCC respectively. In Fig. 13, the parameters are set as follows:
method based on ESO, in the following results only MFPCC the dq-axis inductance, stator resistance and flux linkage in
and MFPCC-ESO are compared in the experiments. the controller are 50%, 10% and 200% of the actual values,
Fig. 12, Fig. 13 and Fig. 14 show the dynamic responses of respectively. In Fig. 14, the parameters are set as follows: the
MFPCC and the proposed MFPCC-ESO at no-load and load dq-axis inductance, stator resistance and flux linkage in the
with mismatched machine parameters under the condition of controller are 200%, 100% and 200% of the actual values,
10 kHz sampling frequency. From top to bottom, the curves respectively.
shown in Fig. 12, Fig. 13 and Fig. 14 are the rotor speed, the The experimental results show that the proposed MFPCC-
reference value of the q-axis current, the actual value of the ESO has better control performance than does MFPCC at a
q-axis current and the A-phase stator current. sampling frequency of 10 kHz. During the dynamic operation

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
ZHANG et al.: MODEL-FREE PREDICTIVE CURRENT CONTROL 9

r [600rpm/div] r [600rpm/div]
iq[2A/div]
iqref [4A/div]
iqref [2A/div]
i
q [4A/div]

i [10A/div] ia [5A/div]
a

(a) (a)

r [600rpm/div] r [600rpm/div]

iqref [2A/div]
iq [4A/div]

iqref [4A/div] iq [2A/div]

ia [10A/div] ia [5A/div]

(b) (b)

r [600rpm/div] r [600rpm/div]
iq [2A/div]
iqref [4A/div]
iqref
iq [2A/div]
[4A/div]

ia [10A/div] ia [5A/div]

(c) (c)

r [600rpm/div] r [600rpm/div]
iqref [4A/div]
iq [4A/div] iqref [2A/div]
i
q [2A/div]

ia [10A/div] ia [5A/div]

(d) (d)
Fig. 12. Experimental results during a dynamic process with mismatched Fig. 13. Experimental results during a dynamic process with mismatched
parameters at 10R, 0.7ψf , 2L under the condition of 10 kHz sampling parameters at 0.1R, 2ψf , 0.5L under the condition of 10 kHz sampling
frequency for (a) MFPCC no-load, (b) MFPCC-ESO no-load, (c) MFPCC frequency for (a) MFPCC no-load, (b) MFPCC-ESO no-load, (c) MFPCC
with load and (d) MFPCC-ESO with load. with load and (d) MFPCC-ESO with load.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
10 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

TABLE II
I MPLEMENTATION COMPLEXITY COMPARISON OF THREE METHODS

conventional MFPCC-
MFPCC
MPCC ESO
Rs , Ld , Lq ,
Motor parameters
ψf and so Null Null
r [600rpm/div] needed
on
iq [2A/div] Parameters to be
Null
Kdp , Kqp ,
α, ω0
designed or tuned αd , αq , nF
iqref [2A/div]
Computational time 36.97 µs 44.76 µs 38.89 µs

of the PMSM drive system, the current ripple of the MFPCC-

ia [5A/div]
ESO is smaller, the motor runs more stably, the overshoot is
smaller and the parameter robustness is stronger.
(a) A detailed complexity comparison of the conventional
MPCC, MFPCC and the proposed method is listed in Table
II. Compared to conventional MPCC, the computational time
[600rpm/div] of the proposed method is only slightly increased, but the
r
machine parameter robustness is greatly enhanced. Compared
iq [2A/div]
to conventional MFPCC using as much as five control pa-
rameters, the proposed method only requires to tune two
iqref [2A/div] parameters. Furthermore, the computational time is reduced
by 13.1%. As the overall control performance of the proposed
method is also better than that of conventional MFPCC, it is
ia [5A/div]
concluded that the proposed method outperforms MFPCC and
may be a powerful alternative to conventional MPCC. In the
(b) future study, the proposed method may be extended to the
speed sensorless operation.

[600rpm/div]
VI. C ONCLUSION
r
iq [2A/div] This paper proposes a model-free control algorithm based
iqref [2A/div] on ESO using an ultra-local model for a PMSM drive system
that does not require motor parameters and presents strong
parameter robustness. The simulation and experimental results
show that the proposed MFPCC-ESO and conventional MPCC
have similar performance when the parameters are accurate.
ia[5A/div]
However, when the parameters are inaccurate, the dynamic
and steady performance of the proposed method is better than
(c) that of the conventional MPCC. In contrast to the MFPCC,
MFPCC-ESO needs to tune less control parameters and the
computational time is also lower. Furthermore, MFPCC-ESO
[600rpm/div] outperforms MFPCC in terms of current ripples, tracking
r
error and dynamic overshoot. The steady state performance
iq [2A/div] improvement is more pronounced with a relatively low sam-
pling frequency of 5 kHz. The simulation and test results
iqref
[2A/div] confirm that the proposed method achieves good parameter
robustness, fast dynamic response and satisfactory steady-state
performance.
ia [5A/div]
ACKNOWLEDGMENT
(d)
The authors would like to thank Mr Hao Jiang for his
Fig. 14. Experimental results during a dynamic process with mismatched
parameters at 1R, 2ψf , 2L under the condition of 10 kHz sampling frequency support when preparing the experimental results.
for (a) MFPCC no-load, (b) MFPCC-ESO no-load, (c) MFPCC with load and
(d) MFPCC-ESO with load.
R EFERENCES
[1] Y. Zhang and J. Zhu, “Direct torque control of permanent magnet syn-
chronous motor with reduced torque ripple and commutation frequency,”
IEEE Trans. Power Electron., vol. 26, no. 1, pp. 235–248, Jan 2011.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.
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/TIE.2020.2970660, IEEE
Transactions on Industrial Electronics
ZHANG et al.: MODEL-FREE PREDICTIVE CURRENT CONTROL 11

[2] L. Zhong, M. F. Rahman, W. Y. Hu, and K. W. Lim, “Analysis of direct [19] S. Chang, P. Chen, Y. Ting, and S. Hung, “Robust current control-based
torque control in permanent magnet synchronous motor drives,” IEEE sliding mode control with simple uncertainties estimation in permanent
Trans. Power Electron., vol. 12, no. 3, pp. 528–536, May 1997. magnet synchronous motor drive systems,” IET Electr. Power Appl.,
[3] D. Casadei, F. Profumo, G. Serra, and A. Tani, “Foc and dtc: two vol. 4, no. 6, pp. 441–450, July 2010.
viable schemes for induction motors torque control,” IEEE Trans. Power [20] M. Fliess and C. Join, “Model-free control and intelligent pid controllers:
Electron., vol. 17, no. 5, pp. 779–787, Sep. 2002. Towards a possible trivialization of nonlinear control?” Ifac Proceedings
[4] Y. Zhang, Y. Bai, H. Yang, and B. Zhang, “Low switching frequency Volumes, vol. 42, no. 10, pp. 1531–1550, 2009.
model predictive control of three-level inverter-fed im drives with speed- [21] Y. Zhou, H. Li, and H. Yao, “Model-free control of surface mounted
sensorless and field-weakening operations,” IEEE Trans. Ind. Electron., pmsm drive system,” in 2016 IEEE International Conference on Indus-
vol. 66, no. 6, pp. 4262–4272, Jun. 2019. trial Technology (ICIT), March 2016, pp. 175–180.
[5] P. Cortes, M. P. Kazmierkowski, R. M. Kennel, D. E. Quevedo, and [22] C. Lin, T. Liu, J. Yu, L. Fu, and C. Hsiao, “Model-free predictive current
J. Rodriguez, “Predictive control in power electronics and drives,” IEEE control for interior permanent-magnet synchronous motor drives based
Trans. Ind. Electron., vol. 55, no. 12, pp. 4312–4324, Dec 2008. on current difference detection technique,” IEEE Trans. Ind. Electron.,
[6] F. Morel, X. Lin-Shi, J. Retif, B. Allard, and C. Buttay, “A comparative vol. 61, no. 2, pp. 667–681, Feb 2014.
study of predictive current control schemes for a permanent-magnet [24] M. Fliess and C. Join, “Model-free control,” Int. J. Control, vol. 86,
synchronous machine drive,” IEEE Trans. Ind. Electron., vol. 56, no. 7, no. 12, pp. 2228–2252, 2013.
pp. 2715–2728, July 2009. [25] Zhiqiang Gao, “Scaling and bandwidth-parameterization based con-
[7] Hyung-Tae Moon, Hyun-Soo Kim, and Myung-Joong Youn, “A discrete- troller tuning,” in Proceedings of the 2003 American Control Conference,
time predictive current control for pmsm,” IEEE Trans. Power Electron., 2003., vol. 6, June 2003, pp. 4989–4996.
vol. 18, no. 1, pp. 464–472, Jan 2003. [26] Q. Xu, M. Sun, Z. Chen, and D. Zhang, “Analysis and design of the
[8] Y. Zhang, D. Xu, J. Liu, S. Gao, and W. Xu, “Performance improvement extended state observer using internal mode control,” in Proceedings of
of model-predictive current control of permanent magnet synchronous the 32nd Chinese Control Conference, July 2013, pp. 5408–5413.
motor drives,” IEEE Trans. Ind. Appl., vol. 53, no. 4, pp. 3683–3695,
July 2017.
[9] Y. Zhang, D. Xu, and L. Huang, “Generalized multiple-vector-based
model predictive control for pmsm drives,” IEEE Trans. Ind. Electron.,
vol. 65, no. 12, pp. 9356–9366, Dec. 2018.
[10] X. Zhang, L. Zhang, and Y. Zhang, “Model predictive current control Yongchang Zhang (M’10-SM’18) received the B.S.
for pmsm drives with parameter robustness improvement,” IEEE Trans. degree from Chongqing University, China, in 2004
Power Electron., vol. 34, no. 2, pp. 1645–1657, Feb 2019. and the Ph.D. degree from Tsinghua University,
[11] H. A. Young, M. A. Perez, and J. Rodriguez, “Analysis of finite-control- China, in 2009, both in electrical engineering.
set model predictive current control with model parameter mismatch in From August 2009 to August 2011, he was a
a three-phase inverter,” IEEE Trans. Ind. Electron., vol. 63, no. 5, pp. Postdoctoral Fellow at the University of Technology
3100–3107, May 2016. Sydney, Australia. He joined North China University
[12] J. C. Moreno, J. M. Espi Huerta, R. G. Gil, and S. A. Gonzalez, of Technology in August 2011 as an associate pro-
“A robust predictive current control for three-phase grid-connected fessor. Currently he is a full professor and the direc-
inverters,” IEEE Trans. Ind. Electron., vol. 56, no. 6, pp. 1993–2004, tor of Inverter Technologies Engineering Research
June 2009. Center of Beijing. He has published more than 100
[13] T. Boileau, N. Leboeuf, B. Nahid-Mobarakeh, and F. Meibody-Tabar, technical papers in the area of motor drives, pulsewidth modulation and
“Online identification of pmsm parameters: Parameter identifiability and AC/DC converters. His current research interest is model predictive control
estimator comparative study,” IEEE Trans. Ind. Appl., vol. 47, no. 4, pp. for power converters and motor drives.
1944–1957, July 2011.
[14] D. Q. Dang, M. S. Rafaq, H. H. Choi, and J. Jung, “Online parameter Jialin Jin was born in 1993. He received the
estimation technique for adaptive control applications of interior pm B.S. degree in electrical engineering from North
synchronous motor drives,” IEEE Trans. Ind. Electron., vol. 63, no. 3, China University of Technology, Beijing, China, in
pp. 1438–1449, March 2016. 2017, where he is currently working toward the
[15] J. Yang, W. Chen, S. Li, L. Guo, and Y. Yan, “Disturbance/uncertainty M.S. degree in control science and engineering. His
estimation and attenuation techniques in pmsm drives—a survey,” IEEE research interests include model predictive control
Trans. Ind. Electron., vol. 64, no. 4, pp. 3273–3285, April 2017. of permanent magnet synchronous motor drives.
[16] C. Xia, M. Wang, Z. Song, and T. Liu, “Robust model predictive
current control of three-phase voltage source pwm rectifier with online
disturbance observation,” IEEE Trans. Ind. Informat., vol. 8, no. 3, pp.
459–471, Aug 2012.
[17] X. Zhang, B. Hou, and Y. Mei, “Deadbeat predictive current control Lanlan Huang was born in 1994. She received the
of permanent-magnet synchronous motors with stator current and dis- B.S. degree in automation engineering and M.s. de-
turbance observer,” IEEE Trans. Power Electron., vol. 32, no. 5, pp. gree in electrical engineering from North China Uni-
3818–3834, May 2017. versity of Technology, Beijing, China, in 2016 and
[18] J. Han, “From pid to active disturbance rejection control,” IEEE Trans. 2019 respectively. In September 2019, She joined
Ind. Electron., vol. 56, no. 3, pp. 900–906, March 2009. Gui Zhou Aerospace Control Technology Corpo-
[23] Y. Chen, T. Liu, C. Hsiao, and C. Lin, “Implementation of adaptive ration, LTD. Her research interests include model
inverse controller for an interior permanent magnet synchronous motor predictive control of permanent magnet synchronous
adjustable speed drive system based on predictive current control,” IET motor drives.
Electr. Power Appl., vol. 9, no. 1, pp. 60–70, 2015.

0278-0046 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University College London. Downloaded on May 26,2020 at 04:04:00 UTC from IEEE Xplore. Restrictions apply.