A Low-Complexity Three-Vector-Based Model Predictive Torque Control For SPMSM PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

13002 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO.

11, NOVEMBER 2021

A Low-Complexity Three-Vector-Based Model


Predictive Torque Control for SPMSM
Xianglin Li , Member, IEEE, Zhiwei Xue , Lixia Zhang, and Wei Hua , Senior Member, IEEE

Abstract—In conventional model predictive torque control through the discrete mathematical model, and then select the
(MPTC), only one voltage vector (VV) is applied during a whole optimal voltage vector (VV) by minimizing the cost function
control period, thus causing a large torque ripple. To improve the [13], [14]. The FCS-MPC is easy to implement and does
steady-state performance, some two-vector-based control schemes
have been proposed. However, the selection of the optimal VV pair not require a modulator. It has the advantageous features of
is complex as well as has a large computational burden, and the rapid dynamic response, no parameter tuning, multivariable
improvement of performance is still limited by the direction and control, and easy handling of nonlinear constraints [15]–[18].
amplitude of the output VV. This article proposes a low-complexity Compared with the direct torque control, the VV selected by
three-vector-based MPTC for SPMSM drives, which can precisely FCS-MPC is more accurate and effective. Compared with the
determine the appropriate active voltage vectors (AVVs) with the
predicted torque error. Then, a modified switching table is devel- field-oriented control, it can realize a much quicker dynamic
oped to directly select the AVVs, thus greatly reducing the complex- response.
ity and computational burden of the algorithm. To obtain a better Despite the good features mentioned above, it has some
steady-state performance, a duty cycle calculation method based disadvantages. For the application of FCS-MPC algorithm in the
on torque and stator flux difference parameters is newly proposed two-level voltage sourced inverter (VSI)-fed PMSM drives, its
to achieve the deadbeat control of torque and stator flux. And then,
the experimental comparisons with the double-vector-based MPTC control set only consists eight basic VVs, namely six active volt-
are conducted. The results show that the proposed MPTC can age vectors (AVVs) and two zero voltage vectors (ZVVs) [19].
effectively reduce the steady-state torque ripple while maintaining The conventional FCS-MPC applies only one basic VV during
a good dynamic performance as well as almost a fixed switching the entire control period, which leads to relatively high steady-
frequency for all speed ranges. state ripple and variable switching frequency [20]. Therefore,
Index Terms—Model predictive torque control (MPTC), a high sampling frequency is the key to satisfy the steady-state
permanent magnet synchronous motor (PMSM), ripple reduction, control performance. However, with the increase of sampling
switching table, three vectors. frequency, the requirements for hardware are also higher. Hence,
how to improve the steady-state performance of conventional
I. INTRODUCTION FCS-MPC without increasing the sampling frequency is worth
studying [21].
ERMANENT magnet synchronous motor (PMSM) has
P attracted more and more attention in the fields of electric
vehicles and wind power generation due to its merits of high
To overcome these shortcomings, several studies have been
conducted. In [22] and [23], the control performance was im-
proved by increasing the prediction horizon of the FCS-MPC.
power density, high efficiency, and high reliability [1]–[5]. Re- However, with the increase of the prediction horizon, the com-
cently, with the rapid and steady development of microprocessor putation effort of the control algorithm will greatly increase. An-
technology, finite-control-set model predictive control (FCS- other solution is to introduce duty cycle control into FCS-MPC
MPC) has become a feasible and mature control scheme for [24]–[26]. The duty-cycle-based MPC applies one AVV and one
high-performance motor drives [6]–[12]. ZVV during the entire control period, where the duration of the
The FCS-MPC can effectively utilize the discreteness of the AVV is determined according to a certain principle, such as the
inverter to predict the future behavior of the system variables tracking error minimization [26]. Compared with conventional
FCS-MPC, these methods have better steady-state performance.
Manuscript received November 6, 2020; revised February 25, 2021; accepted However, in these methods, although the amplitude of the VV is
April 26, 2021. Date of publication May 11, 2021; date of current version July
30, 2021. This work was supported by the National Natural Science Foundation adjustable, their phase angle is still fixed. When the motor runs
of China under Grants 51877215 and 51777216. Recommended for publication under certain operating conditions, restricting the second VV as
by Associate Editor T. Shi. (Corresponding authors: Zhiwei Xue; Xianglin Li.) the ZVV will limit the improvement of the steady-state perfor-
Xianglin Li is with the College of Electrical Engineering, Qingdao University,
Qingdao 266071, China (e-mail: lxllcc@126.com). mance. Therefore, to obtain better steady-state performance, in
Zhiwei Xue and Lixia Zhang are with the College of New Energy, China [27], the selection of the second VV is extended to the AVV.
University of Petroleum (East China), Qingdao 266580, China (e-mail: Thus, the amplitude and phase angle of the synthesized VV are
zhiweixue0503@163.com; hdzlx@163.com).
Wei Hua is with the School of Electrical Engineering, Southeast University, adjustable, which can improve the steady-state performance at
Nanjing 210096, China (e-mail: huawei1978@seu.edu.cn). high speed. However, double-vector-based MPC is computation-
Color versions of one or more figures in this article are available at https: ally expensive in determining the optimal VV pair, since there
//doi.org/10.1109/TPEL.2021.3079147.
Digital Object Identifier 10.1109/TPEL.2021.3079147 are a large number of possible combinations of VVs.

0885-8993 © 2021 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: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
LI et al.: LOW-COMPLEXITY THREE-VECTOR-BASED MODEL PREDICTIVE TORQUE CONTROL FOR SPMSM 13003

Therefore, reducing the complexity of FCS-MPC has become


an important research topic in recent years. Most schemes focus
on how to simplify the VV selection process to reduce the
number of calculations. For example, in [28], the number of
candidate VVs is reduced from six to three by considering the
sign of torque deviation, thus avoiding the enumeration-based
prediction and significantly reducing the control complexity. In Fig. 1. Three-phase two-level voltage source inverter fed PMSM system.
addition, in [29], the deadbeat control principle is utilized to
optimize the selection process of the optimal VV, which avoids
the enumeration of all feasible VVs. Although the computational
complexity is reduced, since only one VV is applied during
the entire control period, the steady-state performance is still
unsatisfactory. Moreover, in [30] and [31], the deadbeat control
principle is utilized to determine the sector where the reference
VV is located, and then an AVV followed by either a zero or an
active VV are applied during the entire control period. In this
case, the steady-state performance of the control system can be
improved, however, it is necessary to calculate the tangent in-
Fig. 2. Output voltage vectors in the α−β frame.
verse of the VV spatial angle, and an additional coordinate trans-
formation from the rotating reference frame to the stationary
reference frame is also required. In fact, neither the one-vector
nor the two-vector-based MPC can eliminate the steady-state MPTC scheme is demonstrated by experimental verification in
error theoretically. Therefore, to further improve the steady-state Section VI. Finally, the conclusion is drawn in Section VII.
performance and obtain error-free control, three-vector-based
control schemes are proposed. In [32], a three-vector-based MPC II. MACHINE EQUATION OF SPMSM
for doubly fed induction generator in wind energy application
In the proposed drive system, the voltage to the SPMSM is
was proposed. In [33], a symmetrical three-vector-based MPC
supplied by a three-phase two-level VSI as shown in Fig. 1. This
scheme is also introduced to further improve the steady-state
topology can provide eight basic VVs, including six AVVs and
performance of PMSM. Although the three-vector-based MPC
two ZVVs, which can be represented in the stationary α–β frame
can greatly improve the steady-state performance of the PMSM
as shown in Fig. 2.
system, intensive derivation is inevitable.
The stator voltage equation of SPMSM in the synchronous
The main innovation of this article is proposing a
rotating d-q reference frame can be expressed as
low-complexity three-vector-based model predictive torque
control (MPTC) scheme for SPMSM drives to effectively 
ud = Rs id + dψ
dt − ωe ψq
d

reduce the number of prediction vectors as well as improve dψq (1)


uq = Rs iq + dt + ωe ψd
the control performance. First, a modified switching table
which only considers the sign of torque deviation is constructed where ud,q , ψ d,q , and id,q are the d–q axes components of stator
as well as a simple but effective approach is proposed to voltage, stator flux, and stator current, respectively. Rs is the
accurately select the appropriate AVVs, which eliminates the stator resistance and ω e is the rotor speed. In (1), the d–q axes
enumeration process of cost function minimization and greatly components of the stator flux are given by
improves the computational efficiency. Then, considering the 
simultaneous suppression of torque and stator flux ripples, a ψd = L d id + ψ f
(2)
novel duty cycle calculation method based on torque and stator ψ q = Lq iq
flux difference parameters is proposed to achieve the deadbeat where Ld,q are the d–q axes components of stator inductance
control of torque and stator flux. Furthermore, to obtain a small and ψ f is flux linkage of permanent magnets. Finally, the elec-
steady-state ripple and reduce current harmonics, the selected tromagnetic torque equation can be written as
two adjacent AVVs with a ZVV are symmetrically applied to
the inverter. Finally, the experimental testing is conducted to 3
Te = pψf iq (3)
validate the superiority of the proposed scheme by comparing 2
with the double-vector-based MPTC. where Te is the electromagnetic torque and p is the number of
The article is organized as follows. The system configuration pole pairs.
and modeling of the SPMSM are presented in Section II. Then,
Section III will be devoted to the torque ripple analysis. In
Section IV, the principle of the proposed MPTC is discussed III. TORQUE RIPPLE ANALYSIS
in detail. The theoretical comparison of the proposed scheme In this section, the torque ripple of the SPMSM will be
with prior similar schemes is conducted in Section V. Conse- analyzed from the perspective of the torque angle and stator
quently, the performance of SPMSM controlled by the proposed flux space vector. For SPMSM, the electromagnetic torque can

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
13004 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO. 11, NOVEMBER 2021

Fig. 3. Effect of the voltage vector on the stator flux change.

be expressed as Fig. 4. Control block diagram of the proposed three-vector-based MPTC.

3 p |ψs |
Te = (ψf sin δ) (4)
2 Lq
adjusted, the torque ripple of the MPTC can be suppressed.
where |ψ s | is the stator flux magnitude and δ is the torque angle. Therefore, in double-vector-based MPTC, an AVV followed by
Then, by differentiating (4), the torque variation can be derived either an active or a zero VV are applied during a whole control
as period. The torque ripple can be reduced and the steady-state
3 p performance can be improved. However, the error between the
ΔTe = ψf (sin δ × Δψs + |ψs | Δδ cos δ) (5)
2 Lq expected torque and the actual torque cannot be completely
eliminated due to the limitations of the amplitude and direction
where Δψ s and Δδ are the variations of stator flux amplitude
of the available VVs. Therefore, in order to further reduce the
and torque angle within a control period, respectively. From (5),
torque ripple of double-vector-based MPTC at a relatively low
it can be seen that the change of torque depends on the change
sampling frequency, the proposed MPTC scheme uses three
of stator flux amplitude |ψ s | and torque angle δ.
vectors instead of two vectors in one control period, which will
In the case of ignoring the stator resistance voltage drop, the
→ → be a cost-effective solution.
relationship between the VV u and the stator flux vector ψ can
s s
be given by IV. PROPOSED MPTC SCHEME
→ →
Δψ = u Δt (6) The control block diagram of the proposed three-vector-based
s s
MPTC is illustrated in Fig. 4, which mainly includes four parts:
where Δt is the applied time of VV. And the change of torque stator flux estimation, torque and stator flux prediction, VV
angle can be expressed as selection, and duration calculation. The specific implementation
of the proposed MPTC will be elaborated as follows.
Δδ = Δθ(V T ) − ωe Δt (7)
where Δθ(VT) is the change of stator flux angle when a VV is A. Predictive Models
applied.
From (6), it can be found that the variation direction of the From (1) and (2), the current differential terms of SPMSM
stator flux vector is always consistent with the direction of can be expressed as
the applied VV, which is evidenced in Fig. 3. Assuming that  di
1
the stator flux vector is located in the first sector, the torque dt
d
= Ld (ud − Rs id + ωe Lq iq )
diq 1 (8)
variation in one control period will be analyzed as follows. As dt = Lq (uq − Rs iq − ωe Ld id − ωe ψf ) .
shown in Fig. 3, ψ s (k) and ψ f (k) represent the stator and rotor
flux vectors at the kth period, respectively. And ψ s (k+1) and To predict the stator current after one sampling period, using
ψ f (k+1) are the stator and rotor flux vectors at the end of one the forward Euler approximation, the stator current differential
control period when the VV V2 is applied. As can be seen from terms in (8) can be described in the discrete-time manner:
Fig. 3, the stator flux variation Δψ s , stator flux angle variation
⎧  
Δθ(VT) , and torque angle variation Δδ caused by the VV V2 are ⎨ id (k + 1) = 1 − Rs Ts id (k) + ωe Ts iq (k) + Ts ud (k)
approximately proportional to the control period Ts . Therefore,  L d  Ld

according to (5)–(7), the torque variation ΔTe is approximately ⎩ iq (k + 1) = 1 − Rs Ts iq (k) − ωe Ts id (k) + Ts uq (k) .
Lq Lq
proportional to the control period Ts . As a result, the torque (9)
ripple will increase with the increase of the control period. Accordingly, the stator flux at the (k+1)th period in the d–q
In conventional MPTC, the optimal VV is applied during axis can be predicted as
a whole control period, and torque ripple will increase with 
the decrease of sampling frequency. In fact, if the duration ψd (k + 1) = Ld id (k + 1) + ψf
(10)
of the applied optimal VV within one control period can be ψq (k + 1) = Lq iq (k + 1) .

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
LI et al.: LOW-COMPLEXITY THREE-VECTOR-BASED MODEL PREDICTIVE TORQUE CONTROL FOR SPMSM 13005

TABLE I
SWITCHING TABLE FOR THE PROPOSED MPTC

Fig. 6. Torque response when the estimated torque is lower than the reference
torque.

Fig. 7. Torque response when the estimated torque is higher than the reference
torque and the initial torque error is large. (a) Applying ZVV alone. (b) Applying
combination of the selected AVVs and ZVV.
Fig. 5. Schematic diagram of the effect of voltage vector on the torque and
stator flux.

Then, the electromagnetic torque at the (k+1)th period can be


given by
3
Te (k + 1) = pψf iq (k + 1) . (11)
2

B. Voltage Vector Selection


In the double-vector-based MPTC, due to a large number
Fig. 8. Torque response when the estimated torque is higher than the reference
of possible combinations of VVs, the final determining of the torque and the initial torque error is small. (a) Applying ZVV alone. (b) Applying
optimal VV pair brings a large computational burden to digital combination of the selected AVVs and ZVV.
signal processor. Therefore, a new switching table is used in
the proposed MPTC to quickly select the AVVs. As shown in
Table I, the proposed VV selection table is simple because the
is, the AVVs V5 (001) and V6 (101) can generate negative slope
selection of AVVs only involves the torque error and does not
torque, while the ZVV can also generate negative slope torque.
consider the error of stator flux vector. As illustrated in Fig. 5,
Assuming that the initial torque error is large and the ZVV cannot
assuming that the stator flux vector is located in the Sector I,
compensate the torque error alone in one control period as shown
V2 (110) and V3 (010) are the two candidate AVVs selected from
in Fig. 7(a). Thus, the AVVs should be added to rapidly reduce
the switching table. It can be seen that the selected two AVVs
the torque error, which is described in Fig. 7(b).
have the same effect on torque slope, but have opposite effect
However, the abovementioned VV selection method is not
on stator flux slope.
suitable for some conditions. When the estimated torque Te (k)
For example, assuming that the stator flux vector is located in
is higher than the reference torque Te ∗ and the initial torque
the first sector, and the estimated torque Te (k) is lower than the
error is small enough that it can be compensated by ZVV alone
reference torque Te ∗ as shown in Fig. 6. According to Table I,
as shown in Fig. 8(a). In this case, if the two AVVs that can
the VVs V2 (110) and V3 (010) that can increase the torque will
produce negative torque slope are selected from the switching
be selected as the AVVs, and then the ZVV will be applied to
table and applied during the control period, the torque error will
reduce the torque ripple. The reduction of the stator flux ripple is
further increase as shown in Fig. 8(b).
achieved by adjusting the duty cycles of the two AVVs because
From (1) to (3), the derivative of the torque with respect to
they have opposite effects on the stator flux slope.
time t can be obtained as
Then, considering the case that the estimated torque Te (k) is

higher than the reference torque Te ∗ , and the AVVs that can dTe 1 3 3
= −Rs Te (k) + pψf uq − pωe ψf ψd . (12)
reduce the torque will be selected from the switching table. That dt Lq 2 2

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
13006 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO. 11, NOVEMBER 2021

Fig. 11. Switching sequence of each phase.

Fig. 9. Torque variation with different loads and rotor speeds under ZVV
application.
C. Duty Cycle Determination
After the AVVs are obtained from Table I, the durations of the
three VVs in one control period should be reasonably allocated to
reduce the torque ripple and stator flux fluctuation. In this article,
a new duty cycle calculation method based on torque and stator
flux difference parameters is proposed, which can coincidentally
use the predicted torque error of the ZVV obtained from the
VV selection to calculate the duty cycle, thus reducing the
computational complexity and computation burden. The torque
and stator flux difference parameters mi and ni are, respectively,
Fig. 10. Torque response of the proposed MPTC scheme when the predicted
torque is lower than the reference torque.
defined as

mi = Te∗ − Tei (k + 1)
(14)
ni = ψs∗ − ψsi (k + 1)
It can be seen that the derivative of the torque is composed
where the subscript i corresponds to the ith VV applied on the
of three terms. The first term is always negative. The second
SPMSM.
term is positive, which reflects the influence of stator voltage
The duty cycles of the AVVs are calculated based on the
on torque, and the last term is negative, which is proportional to
deadbeat control of the torque and stator flux. Namely, Te ∗ =
rotor speed. When the ZVV is applied, the torque differentiation
Te (k+1), ψ s ∗ = ψ s (k+1). It yields
can be derived as


dTe 1 3 d1 m1 + d2 m2 + (1 − d1 − d2 ) m0 = 0
= −Rs Te (k) − pωe ψf ψd . (15)
dt Lq 2
(13) d1 n1 + d2 n2 + (1 − d1 − d2 ) n0 = 0

Hence, it can be seen that the torque slope caused by ZVV is where d1 and d2 represent the duty cycles of the first and second
always negative, and as the motor speed and load increase, the AVVs, respectively. Subsequently, the duty cycles of the two
slope of torque drop increases proportionally, which is evidenced AVVs can be obtained as

in Fig. 9. Therefore, in the aforementioned method, the torque d1 = m2 ×n0C −m0 ×n2
ripple will deteriorate when the motor is operating at high speed. m0 ×n1 −m1 ×n0 (16)
d2 = C
To overcome this problem, a simple but effective method is
newly proposed in this article to accurately select the AVVs. where C = m1 (n2 −n0 )+m2 (n2 −n0 )+m0 (n1 −n2 ).
The proposed method first predicts the torque of the next control It should be noted that (16) not only considers the torque
period by applying the ZVV, then replaces the initial torque error ripple but also takes into account the stator flux fluctuation when
with the predicted torque error, and finally selects the appropriate calculating the duty cycles of the AVVs, thus the torque ripple
AVVs from the switching Table I. If the predicted torque is higher and the stator flux fluctuation of the motor can be simultaneously
than the reference torque, the AVVs which can reduce the torque suppressed.
will be selected from the switching table to reduce the torque Once the VVs and their durations are determined, the switch-
error. On the other hand, if the predicted torque is lower than ing signals can be generated through the pulsewidth modulation
the reference torque, the AVVs that generate positive torque (PWM). For example, Fig. 11 shows the switching sequence of
slope will be selected from the switching table to compensate each phase, in which the selected VVs are applied symmetrically
for the negative torque slope and the torque error caused by to the inverter. Two AVVs are applied symmetrically on both
the ZVV, which is explained in Fig. 10. Therefore, the VV sides of the PWM period, and a ZVV is applied in the middle
selection method proposed in this section can always be effective of the PWM period, which can effectively reduce the current
in reducing the torque error. harmonics.

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
LI et al.: LOW-COMPLEXITY THREE-VECTOR-BASED MODEL PREDICTIVE TORQUE CONTROL FOR SPMSM 13007

V. THEORETICAL COMPARISON OF THE PROPOSED SCHEME


WITH PRIOR SIMILAR SCHEMES
Actually, some improved MPC schemes based on three VVs
have been presented and studied in [32] and [33]. However,
it should be noted that the proposed scheme in this article
has several notable differences compared with the prior similar
schemes.
First, the control variables of the three control schemes are
different. For the DFIG driver in [32] and the PMSM driver in
[33], power and stator current are selected as the control vari-
ables, respectively, while torque and stator flux are taken as the
control variables in this article. Second, the VV selection method
proposed in this article is different from the prior schemes. In
[32], first, the power error caused by the ZVV at the end of each
control period is calculated, and then coordinate transformation Fig. 12. Experimental setup.
is carried out to determine the sector of power error, and finally
two adjacent AVVs and one ZVV are selected according to the
TABLE II
sector. In [33], intensive derivation is required to determine the RATED PARAMETERS OF THE EXPERIMENTAL SPMSM
reference VV, and then the sector of the VV is obtained by using
the phase angle of the reference VV, and finally two adjacent
AVVs and one ZVV are selected. However, in this article, a
modified switching table which only considers the sign of torque
deviation is constructed to directly select the AVVs, thus greatly
reducing the complexity of the algorithm and computational
burden. Finally, the duty cycle calculation methods of the three
control schemes are different. In [32], the duty cycle of each VV
is determined by deriving the geometric relationship between
the vectors. In [33], the optimal duty cycles of the three VVs are
calculated from a cost function that minimizes the error between
the reference VV and the synthesis VV, but the calculation
equations are cumbersome. In this article, a novel duty cycle that with predicted torque error is termed as TVMPTC-II. For
calculation method based on the deadbeat principle of torque a fair comparison, the sampling frequencies of TVMPTC-I and
and stator flux is proposed, which can make reasonable use of the TVMPTC-II are tuned to achieve the same switching frequency
predicted torque error of ZVV obtained from the VV selection, as DVMPTC.
thus reducing the computation burden. Fig. 13 shows the steady-state responses of stator flux ψ s ,
torque Te , and single-phase current of the three control schemes
at 1000 rpm with rated load. Under the experimental condition
VI. EXPERIMENTAL VERIFICATION
of 10 kHz sampling frequency, the average switching frequency
To verify the effectiveness of the proposed low-complexity of DVMPTC is 3.33 kHz. Thus, the sampling frequencies of the
three-vector-based MPTC, the performance of the proposed proposed TVMPTC-I and TVMPTC-II are configured as 5 kHz.
scheme is compared with the double-vector-based MPTC [31]. As can be seen from Fig. 13, the proposed TVMPTC-II is better
The construction of the experimental setup is shown in Fig. 12, than DVMPTC and TVMPTC-I in terms of current quality and
in which the output of a two-level inverter is connected to an torque ripple. Therefore, the proposed TVMPTC-II has a better
SPMSM, which is loaded by a magnetic powder brake. The steady-state performance than the other two control schemes at
dc-link voltage is measured by using an LEM LV25-P voltage high speed. Meanwhile, Fig. 14 shows the steady-state response
transducer and the phase currents are sensed by using current of each control scheme at a low speed of 100 rpm with the
sensor LEM HAS 50-S. Meanwhile, an incremental encoder rated load. At the 10 kHz sampling frequency and low-speed
with 2500 pulses per revolution is employed to obtain the angular experimental conditions, the average switching frequency of
position of the rotor. The control algorithm is implemented by a DVMPTC is about 3.95 KHz. In this case, the sampling frequen-
dSPACE processor with DS1103 controller board. In the mea- cies of TVMPTC-I and TVMPTC-II are configured as 6 kHz.
surement results, the stator current is directly measured by the As shown in Fig. 14, the three control schemes have similar
current clamp (Pintech PT-2720), and other signals are obtained steady-state performance, since the optimal VV combination
through a digital-to-analog converter connected to dSPACE. The at low speed is mostly composed of an AVV and a ZVV. Be-
parameters of the experimental SPMSM are illustrated in Ta- sides, based on the above experimental results, the steady-state
ble II. For simplicity, the double-vector-based MPTC is termed performance of TVMPTC-I at high speed is worse than that
as DVMPTC, the proposed low-complexity three-vector-based of TVMPTC-II. The main reason is that the VV selection of
MPTC with estimated torque error is termed as TVMPTC-I, and TVMPTC-I at high speed is not particularly reasonable, which

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
13008 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO. 11, NOVEMBER 2021

Fig. 13. Steady-state performance of the three control schemes at 1000 rpm with rated load. (a) DVMPTC. (b) Proposed TVMPTC-I. (c) Proposed TVMPTC-II.

Fig. 14. Steady-state performance of the three control schemes at 100 rpm with rated load. (a) DVMPTC. (b) Proposed TVMPTC-I. (c) Proposed TVMPTC-II.

verifies the theoretical analysis in Section IV. In addition, from Moreover, to compare the steady-state performance of each
the experimental results, it can also be seen that, compared with control scheme more accurately, the standard deviation is intro-
the other two control schemes, the proposed TVMPTC-II can duced to compare the torque and stator flux ripples, and the total
effectively improve the steady-state performance over a wide harmonic distortion (THD) of phase current waveform is intro-
speed range even at the same average switching frequency. duced to evaluate the current quality. To be specific, at 1000 rpm

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
LI et al.: LOW-COMPLEXITY THREE-VECTOR-BASED MODEL PREDICTIVE TORQUE CONTROL FOR SPMSM 13009

TABLE III
PERFORMANCE COMPARISON OF THREE CONTROL SCHEMES

Fig. 15. Dynamic performance of the three schemes under accelerating from standstill to 1000 rpm. (a) DVMPTC. (b) Proposed TVMPTC-I. (c) Proposed
TVMPTC-II.

with rated load, the torque ripples of DVMPTC, TVMPTC-I,


and TVMPTC-II are 0.197, 0.192, and 0.164 Nm, respectively,
and the stator flux ripples are 0.0086, 0.0083, and 0.0078 Wb,
respectively. Compared with DVMPTC and TVMPTC-I, the
torque ripple of TVMPTC-II is reduced by 16.8% and 14.6%,
respectively, and the stator flux ripple is reduced by 10.5%
and 7.2%, respectively. Furthermore, according to the harmonic
spectrum analysis shown in Fig. 13, the stator current THDs
of DVMPTC and TVMPTC-I are 8.97% and 8.25%, respec-
tively, while that of TVMPTC-II is reduced to 5.72%. These
Fig. 16. Average switching frequency of DVMPTC and TVMPTC-II at
experimental results demonstrate that the proposed TVMPTC-II different speeds with the rated load.
has obvious improvements in current quality and steady-state
ripple compared with DVMPTC and TVMPTC-I. Also, the
performance quantitative comparison results of the three con- than DVMPTC and TVMPTC-I. So on the whole, it indicates
trol schemes under different operating conditions at the same that the proposed TVMPTC-II has better steady-state perfor-
average switching frequency are summarized in Table III, which mance while maintaining good dynamic performance.
includes the current THD, torque ripple, stator flux ripple, and The average switching frequency of each control scheme is
turnaround time. also investigated when the sampling frequency is fixed. Fig. 16
Apart from the steady-state performance comparison, the shows the average switching frequencies of the DVMPTC and
dynamic response of each control scheme is also investigated. the proposed TVMPTC-II at different speeds with the rated
Fig. 15 shows the starting responses involving the rotor speed, load, where the sampling frequency of DVMPTC is 10 kHz,
torque, and the stator current under accelerating from standstill and that of the proposed TVMPTC-II is 5 kHz. It can be seen
to 1000 rpm. It can be seen that each scheme has a very that the average switching frequency of DVMPTC in the whole
similar dynamic response, and the rotor speed can be rapidly speed range is variable. However, the proposed TVMPTC-II has
accelerated to the reference value. However, it can also be almost a fixed average switching frequency for all speeds, so the
observed that in terms of current quality and torque ripple, the proposed scheme can achieve better performance in terms of
proposed TVMPTC-II has a little better dynamic performance current harmonics.

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
13010 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO. 11, NOVEMBER 2021

Fig. 17. Turnaround time of the three control schemes.

Then, the calculation burden of the three control schemes is


compared. In the dSPACE1103 control system, turnaround time
as a measurement of computational complexity can be read di-
rectly from the control desk, which includes the communication
time between dSPACE and control desk, A/D conversion time,
data saving time, and code implementation time [18]. Gener-
ally, for different control schemes under the same experimental
conditions, the communication time, A/D conversion time, and
data saving time are basically the same [13]. Therefore, the
main reason why the three control schemes in this article have
different turnaround times is that the code implementation time
is different. For an intuitive comparison, Fig. 17 shows the
histogram of turnaround time of the three control schemes. It can
be seen from the experimental results that the turnaround time of
DVMPTC is 18.5 μs, while that of TVMPTC-I and TVMPTC-II
are about 13.8 μs. Therefore, the experimental results indicate
that compared with DVMPTC, the computational complexity of
the proposed schemes is greatly reduced. This is due to the fact
that in DVMPTC, the determination of the optimal VV pair is
divided into two steps. First, the intensive derivation is required
to determine the reference vector, and the first nonzero vector is
selected according to the sector position of the reference vector.
Then, according to the first nonzero vector, the second VV is
selected based on the enumeration process with state predictions
for several candidate vectors. Overall, the VV selection process
Fig. 18. Steady-state performance of the proposed control scheme under stator
of DVMPTC is complicated, which leads to a high calculation resistance variations. (a) At 100 rpm. (b) At 1000 rpm.
cost. In contrast, the proposed TVMPTC-I and TVMPTC-II
can quickly select the VV only according to the sign of torque
error, which eliminates the enumeration-based process for cost
with the experiments without parameter mismatch, the control
function minimization, and significantly simplifies the computa-
performance of the proposed TVMPTC-II is still satisfactory. It
tion. Moreover, compared with TVMPTC-I, TVMPTC-II does
can be concluded that the proposed TVMPTC-II can maintain
not increase the computational complexity, which is consistent
similar performance before and after the stator resistance varia-
with the above theoretical analysis. Therefore, compared to
tion, which indicates that the robustness of the proposed control
DVMPTC, the proposed scheme has lower calculation com-
scheme against stator resistance variation is pretty good.
plexity and computational burden, and does not require higher
capability processors to implement the algorithm.
Finally, the stator resistance uncertainty test is performed to VII. CONCLUSION
investigate the robustness of the proposed control scheme. When This article proposes a low-complexity three-vector-based
the motor operates for a long time or accelerates or decelerates MPTC scheme for SPMSM drives, which can achieve better
frequently, the temperature of the motor will rise, and the stator steady-state performance while maintaining good dynamic per-
resistance will increase with the temperature rise. Therefore, it formance. The innovation points of the proposed scheme are to
is necessary to investigate the influence of the stator resistance reduce the algorithm complexity and computational burden as
variation on the proposed control scheme. Accordingly, Fig. 18 well as improve the control performance. For the sake of precise
shows the measured results of the proposed TVMPTC-II in the selection of VVs with torque error compensation, the estimated
case of 100% increase in the stator resistance. By comparing torque error is replaced by the torque error between the predicted

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
LI et al.: LOW-COMPLEXITY THREE-VECTOR-BASED MODEL PREDICTIVE TORQUE CONTROL FOR SPMSM 13011

torque Te (k+1) after applying a zero VV and the reference [17] M. Yang, X. Lang, J. Long, and D. Xu, “Flux immunity robust pre-
torque Te ∗ . In addition, a modified switching table which only dictive current control with incremental model and extended state ob-
server for PMSM drive,” IEEE Trans. Power Electron., vol. 32, no. 12,
considers the sign of torque deviation is constructed to directly pp. 9267–9279, Dec. 2017.
select the AVVs, thus greatly reducing the complexity and com- [18] Y. Wang et al., “Deadbeat model-predictive torque control with discrete
putational burden of the algorithm. Then, the optimal duration space-vector modulation for PMSM drives,” IEEE Trans. Ind. Electron.,
vol. 64, no. 5, pp. 3537–3547, May 2017.
for each selected VV is calculated based on the deadbeat control [19] F. Wang et al., “Finite control set model predictive torque control of
of torque and stator flux to improve the steady-state performance. induction machine with a robust adaptive observer,” IEEE Trans. Ind.
Furthermore, the three vectors are symmetrically synthesized Electron., vol. 64, no. 4, pp. 2631–2641, Apr. 2017.
[20] A. Mora, A. Orellana, J. Juliet, and R. Cárdenas, “Model predictive torque
to realize almost fixed switching frequency. Finally, compared control for torque ripple compensation in variable speed PMSMs,” IEEE
with the double-vector-based MPTC, the proposed scheme can Trans. Ind. Electron., vol. 63, no. 7, pp. 4584–4592, Jul. 2016.
improve the control performance in terms of steady-state ripple, [21] W. Chen, S. Zeng, G. Zhang, T. Shi, and C. Xia, “A modified double
vectors model predictive torque control of permanent magnet synchronous
current quality and computational complexity, as indicated by motor,” IEEE Trans. Power Electron., vol. 34, no. 11, pp. 11419–11428,
the experimental results. Nov. 2019.
[22] S. Vazquez, C. Montero, C. Bordons, and L. G. Franquelo, “Model
predictive control of a VSI with long prediction horizon,” in Proc. IEEE
Int. Symp. Ind. Electron., Jun. 2011, pp. 1805–1810.
REFERENCES [23] T. Dorfling, H. du Toit Mouton, T. Geyer, and P. Karamanakos, “Long-
horizon finite-control-set model predictive control with nonrecursive
[1] C. Liu, K. T. Chau, and J. Z. Jiang, “A permanent-magnet hybrid brushless
sphere decoding on an FPGA,” IEEE Trans. Power Electron., vol. 35,
integrated starter-generator for hybrid electric vehicles,” IEEE Trans. Ind.
no. 7, pp. 7520–7531, Jul. 2020.
Electron., vol. 57, no. 12, pp. 4055–4064, Dec. 2010.
[24] Y. Zhang and H. Yang, “Torque ripple reduction of model predictive torque
[2] X. L. Li, F. W. Shen, S. Y. Yu, and Z. W. Xue, “Flux-regulation principle and
control of induction motor drives,” in Proc. IEEE Energy Convers. Congr.
performance analysis of a novel axial partitioned stator hybrid-excitation
Expo., Sep. 2013, pp. 1176–1183.
flux-switching machine using parallel magnetic circuit,” IEEE Trans. Ind.
[25] L. Yan, M. Dou, Z. Hua, H. Zhang, and J. Yang, “Optimal duty cycle model
Electron., vol. 68, no. 8, pp. 6560–6573, Aug. 2021.
predictive current control of high-altitude ventilator induction motor with
[3] X. Li, K. T. Chau, and M. Cheng, “Analysis, design and experimental
extended minimum stator current operation,” IEEE Trans. Power Electron.,
verification of a field-modulated permanent-magnet machine for direct-
vol. 33, no. 8, pp. 7240–7251, Aug. 2018.
drive wind turbines,” IET Elect. Power Appl., vol. 9, no. 2, pp. 150–159,
[26] Y. Zhang and H. Yang, “Model predictive torque control of induction motor
Feb. 2015.
drives with optimal duty cycle control,” IEEE Trans. Power Electron.,
[4] Y. Du, Y. Mao, F. Xiao, X. Zhu, L. Quan, and F. Li, “Partitioned stator
vol. 29, no. 12, pp. 6593–6603, Dec. 2014.
hybrid excited machine with DC-Biased sinusoidal current,” IEEE Trans.
[27] Y. Zhang and H. Yang, “Two-vector-based model predictive torque control
Ind. Electron., to be published.
without weighting factors for induction motor drives,” IEEE Trans. Power
[5] X. Li, X. Wang, and S. Yu, “Design and analysis of a novel transverse-flux
Electron., vol. 31, no. 2, pp. 1381–1390, Feb. 2016.
tubular linear switched reluctance machine for minimizing force ripple,”
[28] M. Habibullah, D. D. C. Lu, D. Xiao, and M. F. Rahman, “A simplified
IEEE Trans. Transp. Electrific., vol. 7, no. 2, pp. 741–753, Jun. 2021.
finite-state predictive direct torque control for induction motor drive,”
[6] M. R. Nikzad, B. Asaei, and S. O. Ahmadi, “Discrete duty-cycle-
IEEE Trans. Ind. Electron., vol. 63, no. 6, pp. 3964–3975, Jun. 2016.
control method for direct torque control of induction motor drives with
[29] W. Xie et al., “Finite-control-set model predictive torque control with a
model predictive solution,” IEEE Trans. Power Electron., vol. 33, no. 3,
deadbeat solution for PMSM drives,” IEEE Trans. Ind. Electron., vol. 62,
pp. 2317–2329, Mar. 2018.
no. 9, pp. 5402–5410, Sep. 2015.
[7] W. Hua, W. Huang, and F. Yu, “Improved model-predictive-flux-control
[30] Y. Zhang, D. Xu, J. Liu, S. Gao, and W. Xu, “Performance improvement
strategy for three-phase four-switch inverter-fed flux-reversal perma-
of model-predictive current control of permanent magnet synchronous
nent magnet machine drives,” IET Elect. Power Appl., vol. 11, no. 5,
motor drives,” IEEE Trans. Ind. Appl., vol. 53, no. 4, pp. 3683–3695,
pp. 717–728, May 2017.
Jul./Aug. 2017.
[8] F. Yu, S. Zhao, Z. Tian, and X. Wu, “Model predictive flux control of
[31] X. Zhang and B. Hou, “Double vectors model predictive torque control
semicontrolled open-winding PMSG with circulating current elimination,”
without weighting factor based on voltage tracking error,” IEEE Trans.
IEEE Trans. Ind. Inform., vol. 17, no. 2, pp. 1438–1448, Feb. 2021.
Power Electron., vol. 33, no. 3, pp. 2368–2380, Mar. 2018.
[9] Y. Luo and C. Liu, “A simplified model predictive control for a dual three
[32] X. Wang and D. Sun, “Three-vector based low-complexity model pre-
phase PMSM motor with reduced harmonic currents,” IEEE Trans. Ind.
dictive direct power control strategy for doubly fed induction generator,”
Electron., vol. 65, no. 11, pp. 9079–9089, Nov. 2018.
IEEE Trans. Power Electron., vol. 32, no. 1, pp. 773–782, Feb. 2016.
[10] Y. Luo and C. Liu, “Multi-vector-based model predictive torque control for
[33] S. Kang, J. Soh, and R. Kim, “Symmetrical three-vector-based model
a six-phase PMSM motor with fixed switching frequency,” IEEE Trans.
predictive control with deadbeat solution for IPMSM in rotating reference
Energy Convers., vol. 34, no. 3, pp. 1369–1379, Sep. 2019.
frame,” IEEE Trans. Ind. Electron., vol. 67, no. 1, pp. 159–168, Jan. 2020.
[11] W. Huang, W. Hua, F. Chen, and J. Zhu, “Enhanced model predictive
torque control of fault-tolerant five-phase permanent magnet synchronous
motor with harmonic restraint and voltage preselection,” IEEE Trans. Ind. Xianglin Li (Member, IEEE) received the B.S. and
Electron., vol. 67, no. 8, pp. 6259–6269, Aug. 2020. M.S. degrees from China University of Petroleum,
[12] X. Li, Z. Xue, X. Yan, L. Zhang, W. Ma, and W. Hua, “Low-complexity Qingdao, China, in 2007 and 2010, respectively, and
multi-vector-based model predictive torque control for PMSM with volt- the Ph.D. degree from Southeast University, Nanjing,
age pre-selection,” IEEE Trans. Power Electron., to be published. China, in 2015, all in electrical engineering.
[13] B. Yu, W. Song, J. Li, B. Li, and M. S. R. Saeed, “Improved finite control He was a Joint-Supervised Ph.D. Student with the
set model predictive current control for five-phase VSIs,” IEEE Trans. Wisconsin Electric Machine and Power Electron-
Power Electron., vol. 36, no. 6, pp. 7038–7048, Jun. 2021. ics Consortium, University of Wisconsin-Madison,
[14] W. Huang, W. Hua, F. Yin, F. Yu, and J. Qi, “Model predictive thrust force Madison, WI, USA, from 2012 to 2013, and a Re-
control of a linear flux-switching permanent magnet machine with voltage search Assistant with the Department of Electri-
vectors selection and synthesis,” IEEE Trans. Ind. Electron., vol. 66, no. 6, cal and Electronic Engineering, University of Hong
pp. 4956–4967, Jun. 2019. Kong, Hong Kong, from July 2014 to January 2015. From 2015 to 2019, he
[15] X. Zhang, L. Zhang, and Y. Zhang, “Model predictive current control was an Associate Professor with China University of Petroleum, Qingdao,
for PMSM drives with parameter robustness improvement,” IEEE Trans. China. Since 2020, he has been a Professor with the College of Electrical
Power Electron., vol. 34, no. 2, pp. 1645–1657, Feb. 2019. Engineering, Qingdao University, Qingdao, China. His research interests include
[16] X. Li and P. Shamsi, “Model predictive current control of switched design, analysis, and control of electrical machines, especially for PM brushless
reluctance motors with inductance auto-calibration,” IEEE Trans. Ind. machines and high temperature superconducting machines.
Electron., vol. 63, no. 6, pp. 3934–3941, Jun. 2016.

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.
13012 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 36, NO. 11, NOVEMBER 2021

Zhiwei Xue received the B.S. degree in electrical en- Wei Hua (Senior Member, IEEE) received the B.Sc.
gineering and automation from Shandong University and Ph.D. degrees in electrical engineering from
of Technology, Zibo, China, in 2018. He is currently Southeast University, Nanjing, China, in 2001 and
working toward the M.S. degree in electrical engi- 2007, respectively.
neering with China University of Petroleum, Qing- From 2004 to 2005, he was with the Depart-
dao, China. ment of Electronics and Electrical Engineering, The
His research interests include the permanent mag- University of Sheffield, Sheffield, U.K., as a Joint-
net machine drive control. Supervised Ph.D. Student. Since 2007, he has been
with Southeast University, where he is currently a
Chief Professor with Southeast University and a Dis-
tinguished Professor of Jiangsu Province. He has
authored or coauthored more than 150 technical papers. He holds 50 patents in
his areas of interest. His research interests include design, analysis, and control
of electrical machines, especially for PM brushless machines and switching
reluctance machines.
Lixia Zhang received the B.S., M.S., and the Ph.D.
degrees from North China Electric Power University,
Baoding, China, in 2000, 2003, and 2008, respec-
tively, all in electrical engineering.
Since 2011, she has been an Associate Professor
with China University of Petroleum, Qingdao, China.
Her research interests include dc ripple suppres-
sion, voltage stability control, and multiphase motor
modeling.

Authorized licensed use limited to: The Islamia University of Bahawalpur. Downloaded on March 18,2023 at 06:44:38 UTC from IEEE Xplore. Restrictions apply.

You might also like