Alexandria Engineering Journal 115 (2025) 286–296

Contents lists available at ScienceDirect

Alexandria Engineering Journal

journal homepage: www.elsevier.com/locate/aej

Original article

An intelligent optimization algorithm with novel fitness function for

high-performance PMSM FOC
Zhou You a , Yongming Bian a , Yifan Zhang a, Li Chen b,*
a
School of Mechanical Engineering, Tongji University, Shanghai 200092, China
b
School of Automotive Studies, Tongji University, Shanghai 200092, China

A R T I C L E I N F O A B S T R A C T

Keywords: This paper proposes an advanced intelligent optimization algorithm (IO-FOC) with a novel fitness function to
Field Oriented Control (FOC) improve the dynamic response and steady-state performance of field-oriented control (FOC) systems for per-
Permanent magnet synchronous motor (PMSM) manent magnet synchronous motor (PMSM). The extended state observer (ESO) dynamically compensates for
Extended state observer (ESO)
nonlinearities and time-varying parameters, while the IO algorithm optimizes ESO gain and current-loop PI
Intelligent optimization algorithm (IO)
parameters in real time. This dual-layer optimization enhances system robustness, anti-interference capability,
Fitness function
and control precision. The novel fitness function, designed using dq-axis current and parameter identification,
predicts and compensates for disturbances in inductance, resistance, and magnetic flux linkage, ensuring system
stability under complex operating conditions. Experimental validation demonstrates that IO-FOC achieves su-
perior performance in start-up, steady-state, and dynamic conditions compared to conventional methods,
highlighting its potential for high-precision industrial applications.

1. Introduction speed estimation [18], the application of real-time evolutionary algo-

rithms for PMSM parameter optimization [19–21], inverter enhance-
Permanent Magnet Synchronous Motor (PMSM) has experienced ment techniques for torque ripple minimization [22], electromagnetic
widespread adoption in renewable energy applications, propelled by torque compensation strategies for improved steady-state response [23],
advancements in control theory, power electronics, and rare-earth ma- and advanced harmonic injection methods for current harmonic sup-
terial processing [1–3]. The PMSM serves as critical components in pression [24]. However, existing solutions often face limitations in terms
modern industrial applications, including electric vehicles, robotics, and of inadequate disturbance rejection, limited system adaptability, and
renewable energy systems, offering superior power density, high effi- compromised real-time performance. These limitations necessitate the
ciency, and excellent dynamic performance [4–7]. Despite these ad- development of more robust and efficient control strategies.
vantages, PMSMs present significant control challenges due to their This study introduces an advanced Intelligent Optimization Field-
inherent nonlinear characteristics and parameter uncertainties, partic- Oriented Control (IO-FOC) strategy to overcome the above challenges
ularly when operating under varying conditions and external distur- and improve PMSM control performance. Unlike traditional methods,
bances [8–11]. the IO-FOC strategy offers three key innovations:
Field-oriented control (FOC), the predominant control strategy for
PMSM, employs a multi-loop architecture with PI controllers to track 1. Enhanced Extended State Observer (ESO): Implements real-time
current and speed parameters [12] and [13]. Although effective under voltage compensation while mitigating system uncertainties and
nominal conditions, traditional FOC implementations often fail to external disturbances.
maintain optimal performance when subjected to parameter variations, 2. Global IO Algorithm: Performs simultaneous optimization of ESO
load disturbances, and system uncertainties [14–17]. parameters and current loop PI gains to enhance system dynamics
Recent research has addressed these limitations through various and stability.
optimization approaches. Notable developments include: key advance- 3. Novel Fitness Function: Incorporates parameter identification tech-
ments include the implementation of Kalman filtering for precise motor niques to ensure robust operation under varying conditions.

* Corresponding author.
E-mail address: [email protected] (L. Chen).

https://doi.org/10.1016/j.aej.2024.12.029
Received 4 November 2024; Received in revised form 5 December 2024; Accepted 8 December 2024
Available online 19 December 2024
1110-0168/© 2024 The Authors. Published by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria University. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

In comparison to traditional FOC, E-FOC, and PE-FOC methods, IO-

FOC significantly improves disturbance rejection, reduces harmonic
distortion, and ensures better system stability under a wide range of
operating conditions. By addressing both dynamic response and steady-
state performance simultaneously, IO-FOC demonstrates superior con-
trol precision and adaptability. Experimental validation on a 15 kW
PMSM test bench confirms the effectiveness of the IO-FOC strategy,
showcasing its exceptional robustness in high-precision applications
under complex environments. These results highlight the promising
potential of IO-FOC for industrial applications where reliable, adaptive,
and high-performance motor control is required.

2. ESO-based voltage real-time compensation method

Fig. 2. Block diagram of FOC system for PMSM based on ESO compensa-
The traditional FOC system of PMSM exhibits strong coupling, tion method.
nonlinearity, and time-varying parameters, resulting in low control ac-
curacy, weak interference rejection, and poor robustness [14–16]. The (4) pace Vector Pulse Width Modulation (SVPWM): The SVPWM
FOC for PMSM, is to decouple torque and the magnetic flux by con- module converts Vα and Vβ into PWM signals that drive the in-
trolling the current space vector, thereby achieving dynamic perfor- verter’s power switches. The inverter then converts these PWM
mance in AC motors comparable to that of DC motors. The goal of FOC is signals into three-phase voltages, controlling the stator currents
to precisely regulate both the magnitude and direction of the magnetic and achieving closed-loop control of id and iq.
field, ensuring smooth torque, low noise, high efficiency, and fast dy-
namic response. The process of the conventional vector current In this process, the current feedback signal of PMSM is continuously
single-loop control algorithm is as follows: compared with the reference current signal, which is regulated by the PI
controller. The controller generates a compensation signal to drive the
(1) Coordinate Transformation: The three-phase current signals (iA, power devices and keep id and iq close to their set values. Based on this,
iB,iC) are transformed into two-phase stationary coordinate sys- the rotational speed of PMSM must also be controlled, requiring the
tem components using the Clarke transformation. The resulting development of a current-speed dual-loop control, built upon the current
two-phase components (iα,iβ) are then transformed into a rotating single-loop control (see Fig. 1). In Fig. 1, the current-speed dual-loop
coordinate system using the Park transformation to obtain the d- adds a speed loop to the current loop. It compares the real-time speed
axis current (id) for controlling the motor’s magnetic flux and the (wr) with the motor speed setpoint (wref r ) using a speed estimator to
q-axis current (iq) for torque control. generate a q-axis current reference for speed control. Next part in-
(2) PI Controller: The id and iq currents are compared with their troduces dq-axis voltage prediction compensation method based on an
respective reference values to calculate the current error in the ESO, which is integrated into PMSM control system(E-FOC). Fig. 2
dq-axis. The error signals are then input into the PI controller, shows the block diagram of the FOC system of PMSM after integrating
which adjusts the voltage control signals (Vd,Vq) using propor- the compensation method.
tional and integral control to compensate for the current To construct the dq-axis voltage prediction compensation method
deviation. based on the ESO, and account for the time-varying motor system pa-
(3) Inverse Park Transformation: The Vd and Vq outputs from the PI rameters, the inductive perturbation ΔL, resistive perturbation ΔR, and
controller are transformed back into Vα and Vβ in the two-phase permanent magnet flux perturbation Δψ f are introduced into the dq-axis
stationary coordinate system using the inverse Park trans- voltage equation, as shown in Eq. (1).
formation, generating three-phase PWM signals.

Udc

Power
ωrefr PI irefq PI Vq uα device
+ + Rev SVP PM
- ωr - iq Park WM SM
irefd =0 PI Vd uβ
+
- id
θ

iα iA
Park iB
iβ Clark
iC

θ Position and
speed estimator
for PMSM

Fig. 1. FOC current-speed dual-loop control block diagram.

287
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

( ) ( ) ( )
⎧ did Similarly, using Eqs. (1)-(7), the discrete extended state observer for
⎪ V = R + R i + L + L − L + L wr iq
⎨ d
⎪ s Δ d Δ
dt Δ the d-axis current id and the total perturbation hd in the PMSM system is
( ) ( ) ( ) (1) expressed as:
⎩ Vq = Rs + R iq + L + L diq + L + L wr id + wr Ψ f
⎪
⎪
⎧
Δ Δ
dt Δ
rep
⎨ Gd (= id −) id (k)
⎪
⎪
[ rep ]
Rearranging Eq. (1) yields the PMSM dq-axis current equation, id k + 1 = irep
rep
d (k) + hd (k) + bVd (k) Ts − βC Gd (8)
⎪ ( )
which includes the parameter perturbation term, as shown in Eq. (2). ⎪
⎩ hrep k + 1 = hrep (k) − βD Gd
d d
⎧ di ( )
d Vd ΔL Rs + Δ R did Δ L
⎨ dt = L + wr iq + L wr iq − id L Here, βC and βD are the gains of the discrete ESO.
⎪
⎪ −
dt L
( ) (2) The observed values of the d-axis current id and the total disturbance
⎪ di
⎪ q Vq L R s + R diq Δ L wr Ψ f hd are denoted as zd, expressed as:
⎩ = + − wr id − Δ wr id − iq Δ
− −
dt L L L dt L L [ rep ]
i
In designing the ESO, the control variable is set as the dq-axis voltage zd = drep (9)
hd
(Vd, Vq). The other system variables and external disturbances in Eq. (2)
are treated as total perturbations of the PMSM control system, yielding Based on the above analysis, the total system perturbation (hd, hq),
the dq-axis total perturbation (hd, hq): which includes each parameter perturbation and state variable, is
⎧ calculated by the state observer. This total perturbation is introduced
ΔL Rs + Δ R did Δ L
⎨ hd = wr iq + wr iq − id
⎪
⎪ − into Eq. (1), discretized, and used to derive the dq-axis voltage predic-
L L dt L
(3) tion compensation module based on the ESO [25], expressed as:
⎩ hq = − wr id − Δ Lwr id − iq Rs + Δ R − diq Δ L − wr Ψ f
⎪
⎪
( )
L L dt L L ⎧
iref k + 1 − id (k)
⎪
⎪ rep d
According to Eq. (3), the total dq-axis perturbation of the PMSM ⎨ V d (k) = L
⎪
⎪ − hrep
d (k)L
Ts
includes resistance, inductance parameter perturbations, permanent ( ) (10)
⎪
magnet flux perturbation, system states, and sensor noise. In this paper,
⎪
⎪
⎪ rep iref
q k + 1 − iq (k)
rep
⎩ V q (k) = L − hq (k)L
the total perturbation (hd, hq) is treated as the observer’s state variable, Ts
and first-order state observer is designed for the dq-axis using the ESO
The ESO based on Eq. (10) calculates the predicted dq-axis voltage
method. The resulting ESO is expressed in Eq. (4).
compensation value in real time, reducing errors from parameter
⎧
⎨ distortion and external disturbances. The control block diagram of this
fq = z1 − iq
(4) compensation module (using the q-axis as an example) is shown in
⎩ z1 = bVq + z2 − β1 fq z2 = − β2 fq
Fig. 3.
Eq. (4) provides the first-order state observer for the q-axis, with z
3. Development of advanced intelligent optimization algorithm
defined by Eq. (5).
[ rep ]
iq 3.1. Particle swarm optimization implementation
zq = (5)
hrep
q
In the dq-axis voltage prediction compensation method based on ESO
Here, (irep rep
q ,hq ) are the extended state observables of iq and hq, while
(Section 2), βA, βB, βC, and βD are used to regulate observer performance.
β1 and β2 in Eq. (4) are the observer gains. These four parameters define the observer’s position and bandwidth,
Discretizing Eq. (4) results in a discrete ESO for the q-axis current iq and their configuration directly influences the FOC system’s tracking
and the total perturbation hq. performance and noise sensitivity [26]. Therefore, establishing an
⎧ adaptive optimization mechanism is crucial for improving system sta-
⎪ Gq = irep
q (k) − iq (k) bility. Research and experiments [27] indicate that optimizing the q-axis
⎪
⎪
⎨ rep ( ) [ ]
iq k + 1 = irep rep gains βA and βB yields the most significant impact among the four pa-
q (k) + hq + bVq (k) Ts − βA Gq (6)
⎪
⎪
⎪ ( ) rameters. The PSO algorithm for these two bandwidth gains is intro-
⎩ hrep k + 1 = hrep (k) − β G
q q B q duced below (PE-FOC), with the control block diagram shown in Fig. 4.

In Eq. (6), βA and βB represent the gains of the discrete ESO, and they
3.2. Advanced intelligent optimization algorithm
are expressed as:
{
βA = Ts β1 Although PSO algorithm possesses global iteration and stochastic
(7)
βB = Ts β2

where Ts is the sampling time.

Fig. 4. Block diagram of FOC system for PMSM based on PSO key parameter
Fig. 3. Control block diagram of voltage prediction compensation module. optimization.

288
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

search capabilities, its slow convergence speed and low control accuracy { ( )
make it unsuitable for handling the strong nonlinearity and multivari- Pij = hj − lj ∗ r3 + lj
(11)
i = 1⋯n, j = 1⋯d
able coupling of PMSM [28–31].
In FOC system of PMSM, the current loop performance directly in- Here, Pij represents the j-th dimension of the i-th particle, hj and lj are
fluences the motor’s current response and torque output. The PI controller the upper and lower limits of the j-th dimension, respectively. The
parameters (Kp, Ki) in this loop are critical for optimal performance [32]. parameter n denotes the population size, d represents the number of
The PI controller, a crucial component of the vector control system, reg- decision variables (six in this case), and r3 is a random number between
ulates the dq-axis current. Its parameters, Kp and Ki, determine the sys- 0 and 1.
tem’s steady-state error, dynamic response speed, and disturbance In the optimization process, P* represents the particle position cor-
rejection capability. The parameters Kpd and Kid govern the dynamic and responding to the minimum value of the fitness function F. At iteration t,
steady-state characteristics of the d-axis current, while Kpq and Kiq regu- the system error Ek(t) is given by:
late the q-axis current to deliver torque. Optimizing the proportional gain
(Kp) improves the dq-axis current’s rapid response to reference values, Ek (t) = P∗ (t − 1) − P(t − 1) (12)
while optimizing the integral gain (Ki) reduces steady-state errors in the Eq. (12) defines the system error Ek(t) at iteration t. The overall
current loop, achieving higher accuracy in magnetic field control and deviation in the previous iteration is Ek-1(t), and the deviation in the
torque output. Optimizing PI parameters improves the PMSM’s adapt- iteration before that is Ek-2(t). The relationship among them is expressed
ability to parameter variations (e.g., inductance, back EMF) and external in Eq. (13) [35].
disturbances (e.g., load changes, power grid fluctuations). This enhances {
operational smoothness, prolongs the lifespan of the motor and power t = 1Ek (t) = Ek− 1 (t) = Ek− 2 (t)
(13)
electronics, and boosts system robustness. The Extended State Observer t > 1Ek− 1 (t − 1) = Ek− 2 (t)
(ESO) is a dynamic observer designed using the principle of expanded In IO algorithm, Ek-1(t) and Ek(t-1) differ because each iteration has a
state observation. It can estimate the total disturbance of PMSM system in unique optimal value P*. Fig. 6 shows the iteration process of Ek-1(t).
real time. The bandwidth gains βA and βB, as ESO parameters, directly As shown in Fig. 6, for iterations greater than 1, Ek-1(t) represents the
influence the accuracy of disturbance estimation and the system’s dy- deviation of the previous true value from the system’s current optimal
namic performance. Optimizing βA and βB allows the ESO to rapidly value, as expressed in Eq. (14).
detect system disturbance changes, particularly under conditions like
sudden load variations and external disturbances. This ensures accurate Ek− 1 (t) = P∗ (t) − P(t − 1) (14)
disturbance compensation and enhances the PMSM’s dynamic response According to Eq. (14), calculating Ek-1(t) requires storing P(t-1),
performance. Optimizing βA and βB improves the steady-state disturbance which increases the algorithm’s complexity. To simplify this, a vector
estimation, minimizing its impact on the control system. This reduces synthesis method shown in Fig. 6 is used to obtain Ek-1(t), as expressed in
steady-state errors and torque fluctuations in the current loop, thereby Eq. (15).
enhancing control accuracy [33].
To address these issues, we propose an IO algorithm with novel Ek− 1 (t) = Ek (t − 1) + P∗ (t) − P∗ (t − 1) (15)
fitness function to optimize the βA and βB parameters of the ESO, along After calculating the system deviation, the regulation output of the
with the dq-axis PI parameters of the current loop. This enhances the IO algorithm at iteration t is:
dynamic response, reduces steady-state error, and achieves both local
and global optimal control. Fig. 5 presents the block diagram of the △ out(t) = λp [Ek (t) − Ek− 1 (t)]rp + λi Ek (t)ri +
(16)
proposed IO algorithm for FOC system of PMSM. λd [Ek (t) − 2Ek− 1 (t) + Ek− 2 (t)]rd
The IO algorithm generates new control signals based on current and Here, λp, λi, and λd are the adjustment coefficients for proportional,
previous system errors, performing recursive calculations to prevent integral, and differential terms, respectively, which must be determined
error accumulation and differential amplification. It integrates the experimentally. The terms rp, ri, and rd are corresponding random
proportional-integral-derivative module and discretization to produce numbers.
appropriate outputs and update the control system of the PMSM. This To prevent the algorithm from falling into a local optimum, a zero-
section describes the iterative process of IO algorithm. First, Eq. (11) output module is added, as shown in Eq. (17).
defines the population initialization process in the optimization algo- ⎧
rithm [34].
[ ( t) ]
⎪
⎪
⎪ zero(t) = Ek (t) ∗ cos 1 − + r4 λzero ∗ L
⎪
⎪
⎪ T
⎪
⎪ [ ]
⎪
⎪
⎪ ln(T − t + 2)
⎪ λzero =
⎪
⎪
⎪
⎪ ln T
⎪
⎪
⎪
⎪ uσ
⎨ L= 1
|v|β (17)
⎪
⎪
⎪
⎪ ⎡ ⎤
⎪
⎪ ( ) ( )
⎪
⎪ πβ
⎪
⎪ ⎢sin ∗Γ 1+β ⎥
⎪
⎪
⎪ ⎢ 2 ⎥
⎪ σ=⎢ ( ) ⎥
⎪
⎪
⎪ ⎢ 1 + β β− 1 ⎥
⎪
⎪ ⎣β∗Γ ∗2 2 ⎦
⎪
⎩ 2

Here, r4 is an nd matrix of random numbers between 0 and 1, λzero is

the conditioning factor for zero output, t is the current iteration number,
and T is the total number of iterations. L is the Lévy flight function, β is
the constant factor, and u and v are nd random matrices following a
standard normal distribution.
For all particles, after calculating the regulated output △out(t) and
the zero output zero(t), the particle position update formula is given by
Fig. 5. Block diagram of FOC system for PMSM based on the optimization of Eq. (18).
key parameters of IO algorithm.

289
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

effective in handling the complex characteristics of PMSM. Next, we

introduce how to construct the fitness function for IO algorithm based on
dq-axis current in Section 3.2. The fitness function design in this study
adheres to the following scientific principles:

(1) Real-time disturbance correction: The recursive compensation

mechanism, integrating parameter identification with current
prediction, enables real-time correction of dynamic disturbances.
(2) Feedback-driven optimization: The fitness function’s feedback
mechanism refines control parameter adjustments, enhancing
overall system performance. First, by removing the inductive and
resistive perturbation terms from Eq. (1), and combining the
MTPA module with discretization, we obtain:

Fig. 6. Iterative process of Ek-1(t).

⎧ ( )
Ts ud (k)
⎪
⎪ ipro k + 1 = id (k) + wr Ts iq (k) +
⎪
⎪
⎪
d
L
⎧ ( ) ( ) ⎪
⎪ ( )
⎨ P t + 1 = P(t) + η ∗ △ out(t) + 1 − η ∗ zero(t)
⎨ T s uq (k)
ipro k + 1 = iq (k) + − Ts wr id (k)− (19)
(18) ⎪ q L
⎩ η = r5 ∗ cos t
( ) ⎪
⎪
⎪ [ ]
⎪
T ⎩ Ts wr RS iq (k) + Ψ
⎪
⎪
f
Here, η is an n1 matrix, and r5 is an n1 matrix of random numbers L wr
between 0 and 1. Here, ipro pro
d (k + 1) and iq (k + 1) are the dq-axis current values at
Based on the above analysis, a flowchart of IO algorithm is summa- moment k + 1, recursively calculated from the dq-axis current and
rized as shown in Fig. 7. voltage at moment k.
From Eq. (19), it is evident that the predicted d-axis current contains
4. Identification-based novel fitness function an inductance term, while the predicted q-axis current includes resis-
tance, inductance, and flux terms. After prolonged use or external
In Section 3.1, the fitness function of PSO algorithm is based on interference, the initial parameters of the PMSM become inaccurate. To
motor speed. However, this approach is insufficient for handling the eliminate interference and obtain accurate data during PMSM operation,
strong coupling and time-varying characteristics of PMSM [33]. Con- Eq. (19) is rearranged as follows:
structing the fitness function based on dq-axis current has proven more ⎧ pro ( )
⎪
⎪ id k + 1 = id (k) + wr Ts iq (k) + ud (k)Pd
⎪
⎪ ( )
⎪
⎪
⎪ pro Ts uq (k)
⎪
⎪
⎪ i k + 1 = iq (k) + − Ts wr id (k) − wr Pd Pq
⎨ q
⎪ L
Ts (20)
⎪
⎪
⎪ Pd =
⎪
⎪
⎪ L
⎪
⎪
⎪
⎪ R i (k)
⎩ Pq = S q
⎪ + Ψf
wr
Here, Pd represents the inductive perturbation term, and Pq repre-
sents the resistive flux perturbation term.
To obtain accurate parameters, predictive current error compensa-
tion controller is constructed. It corrects the values of Pd and Pq based on
motor operation data, specifically Prev rev
d and Pq . For the d-axis, the con-
structed current error compensation controller is given in Eq. (21) [36].
[ ( ) ( )]2 [ ( ) ]2
Fd = ipro
d k + 1 − id k + 1 + gd Prev d k + 1 − Prev
d (k) (21)

Eq. (21) shows that when the first term in the d-axis current ap-
proaches zero, the actual d-axis current is close to the predicted value.

Fig. 7. Flowchart of IO algorithm. Fig. 8. Construction process for Prev

d .

290
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

Following the same analytical process, an expression for gq is

obtained:
( )2
imv
gq = (29)
Pmv
q

Pmv
q can be calculated from Eq. (20) using the corresponding pa-
rameters from the motor manual. Based on the above analysis, the ex-
pressions for ipro pro
d (k + 1) and iq (k + 1) are derived [37]:
⎧ pro ( )
⎪ i k + 1 = id (k) + wr Ts iq (k) + ud (k)Prev
d (k)
⎨ d (
⎪ )
pro rev
iq k + 1 = iq (k) + uq (k)Pd (k) − Ts wr id (k) (30)
⎪
⎪
⎩ rev
− Pd (k)wr Prev
q (k)

The compensated parameter estimates Prev rev

d and Pq are computed
using the recursive Eqs. (24) and (25). These estimates represent cor-
Fig. 9. Construction process for Prev
q .
rections for d-axis inductive perturbations and q-axis resistive and flux
perturbations, forming the core of the fitness function to achieve high-
When the second term reaches zero, Prev
d is close to the motor’s actual precision compensation. Subsequently, Prev rev
d and Pq are incorporated
operating value. When both conditions are met, Prev d represents the into the current prediction model via Eq. (30) to compute predicted dq-
optimal value during motor operation. The recursive formula for Prev
d is axis current values. The novel fitness function for IO algorithm is ob-
obtained by polarizing Prev
d with Fd: tained using ipro pro
d (k + 1) and iq (k + 1) from Eq. (30):
∂Fd [ ( ) ( )] [ ( ) ]2 [ ( ) ]2
) = 2 ipro k + 1 − id k + 1 ud (k)+ k + 1 − iref
(
∂Prev k+1 d F = ipro + ipro k + 1 − iref (31)
d (22) d d q q
[ ( ) ]
2gd Prev
d k + 1 − Prev
d (k) The fitness function in Eq. (31) is based on dq-axis current and
identifies the inductance, resistance, and flux parameters of the PMSM
Setting Eq. (22) to zero yields an expression for Prev
d when Fd is
through parameter identification. It predicts and compensates for dq-
optimal:
[ pro ( ) ( )] axis current, eliminating the effects of PMSM parameter distortion on
( )
id k + 1 − id k + 1 ud (k) the control system during operation. The construction process of F is
Prev k + 1 = Prev
d (k) − (23)
d
gd shown in Fig. 10.
The new fitness function is integrated with IO algorithm to optimize
Rearranging Eq. (23) provides the expression for Prev
d at moment k: the βA and βB parameters of ESO and the dq-axis PI parameters in FOC
( ) [ pro ] ( )
system of PMSM, enhancing motor performance. The IO-FOC strategy
i (k) − id (k) ud k − 1
Prev
d (k) = Pd
rev
k− 1 − d (24) addresses the limitations of traditional methods in handling non-
gd
linearities and coupling, significantly improving the dynamic response
The construction process for Prev
d is illustrated in Fig. 8. and steady-state performance of PMSM system. The fitness function
As shown in Fig. 8, Prev
d depends solely on the voltage and current of design ensures real-time algorithm responsiveness to key parameter
PMSM, and is independent of motor parameters. Similarly, the expres- perturbations, while significantly enhancing dynamic robustness and
sion for Prev
q at moment k is derived: steady-state accuracy [38], [39].
( ) [ pro ] ( )
iq (k) − iq (k) Prev
d k − 1 wr
Prev
q (k) = Pq
rev
k− 1 + (25) 5. Experimental validation and analysis
gq
5.1. Experimental setup
The construction process of Prevq can be derived using Eq. (25), as
shown in Fig. 9.
To validate the proposed strategy for the FOC of PMSM, an experi-
As shown in Fig. 9, Prevq is related only to PMSM voltage, current,
mental test bench was constructed for a 15 kW PMSM (Fig. 11).
speed, Prev
d , and is independent of motor parameters. In Eq. (21), gd is the
The control performance of FOC, E-FOC, PE-FOC, and IO-FOC
adjustment factor for the d-axis current and inductive perturbation
strategy was compared under start-up, steady-state, and dynamic con-
terms, preventing the overemphasis of one term and neglecting the ef-
ditions on this bench, with both qualitative and quantitative analyses
fects of other factors on the motor. To balance the terms in Fd, Eq. (21) is
provided. The architecture of the 15 kW PMSM test bench (Fig. 11) is
rewritten as:
[ pro ( ) ( )]2 [ rev ( ) ]2
i k + 1 − id k + 1 P k + 1 − Prevd (k)
Fd = d + d (26)
imv mv
Pd

Here, imv is the nominal PMSM current, which can be found in the
motor manual, and Pmv d is calculated from the nominal inductance and
sampling time in the motor manual. Similarly, Eq. (26) is derived by
solving and rearranging it in the same way as Eq. (21):
( ) ( )( )
2
[ pro ] Pmv
Prev (k) = Prev
k − 1 − i (k) − i d (k) ud k − 1 d
(27)
d d d
imv

By comparing Eq. (27) with Eq. (24):

( )2
imv
gd = (28)
Pmv
d Fig. 10. Construction process for F.

291
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

Fig. 11. 15 kW PMSM test bench.

depicted in Fig. 11. It mainly includes a three-phase isolation trans-

former, PMSMs, a DC power supply, an inverter, and a motor control
cabinet.
PMSM serves as the test machine to generate torque, with its inte-
grated optical encoder utilized for speed measurement. The control al-
gorithms are implemented on a DSP28379D platform, operating at a
switching and execution frequency of 8 kHz. The FOC algorithm utilizes
82 kB of memory with a runtime of 55 µs per cycle, the E-FOC algorithm
requires 93 kB with a runtime of 64 µs, the PE-FOC algorithm occupies
100kB and executes in 78 µs, and the IO-FOC algorithm consumes
108 kB with a runtime of 82 µs.The parameters of the PMSM (Figs. 11
and 12) are provided in the Table 1.

5.2. Experimental results under start-up conditions

This section first compares and analyzes the start-up performance of

PMSM under the control of four FOC strategies. Fig. 13 illustrates the
speed variation and d-axis current when PMSM accelerates from Fig. 13. Experimental data of PMSM under start-up conditions: (a) FOC, (b) E-
standstill to 2000 rpm and maintains that speed under the control of the FOC, (c) PE-FOC, (d) IO-FOC.
four strategies.
During the start-up condition (Fig. 13), the IO-FOC strategy shows
faster response and smaller speed fluctuations after reaching 2000 rpm,
highlighting its superior performance in fast response and overshoot
suppression. As shown in Fig. 13, the d-axis current initially shows an
abrupt change during acceleration but gradually trends to zero,
following the MPTA control strategy in the FOC of PMSM. The d-axis
current under IO-FOC control exhibits smaller fluctuations, indicating
smoother and more precise control, reducing motor losses and
improving operational efficiency. Fig. 14 offers an amplified comparison
of the starting speeds for further analysis and comparison of the four
strategies’ performance. Fig. 14 clearly shows that the FOC strategy

Fig. 14. Comparison of PMSM speed amplification under starting conditions.

Fig. 12. Schematic diagram of the 15 kw PMSM test bench architecture.

Table 1 Table 2
Parameters of PMSM. Comparison of starting performance of PMSM.
Parameter Symbol Value Performance indicators FOC E-FOC PE-FOC IO-FOC

Pole Pairs Pn 1 Δn1 (rpm) 153 105 69 44

d Axis Inductance (mH) Ld 0.395 Δt1 (s) 0.132 0.1085 0.1012 0.0882
q Axis Inductance (mH) Lq 0.395
Stator Resistance (Ω) RS 0.0485
Magnet Flux (Wb) ψf 0.1194
Moment of Inertia (kg*m)2 J 0.0027

292
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

exhibits overshoot near the set speed, along with speed oscillations. E-
FOC, utilizing an ESO, compensates for external disturbances, reducing
their influence. PE-FOC improves response speed by optimizing the
observer bandwidth via particle swarm algorithm. Under IO-FOC con-
trol, the rise time of the motor is short, with almost no overshoot at
2000 rpm, showcasing its superior control performance and dynamic
response. This is attributed to IO-FOC’s intelligent optimization module,
which iteratively updates key parameters in real-time, ensuring fast
dynamic response while maintaining stability and accuracy.
Previous analyses were qualitative; to quantitatively assess strategy
performance, Table 2 presents the speed overshoot (Δn1) and stabiliza-
tion time (Δt1) of the PMSM after reaching steady-state under the four
strategies.
Based on the analysis of Fig. 14 and Table 2, the traditional FOC
strategy shows the highest overshoot and stabilization time. This is
because FOC heavily relies on PMSM parameters and models, lacking a
compensation module for external disturbances, resulting in poor
robustness and significant overshoot. E-FOC compensates online
through the dq-axis voltage prediction and compensation module,
reducing system uncertainty and minimizing speed fluctuations. PE-FOC Fig. 16. Comparison of three-phase currents under steady-state conditions: (a)
significantly improves overshoot and stabilization time compared to FOC, (b) E-FOC, (c) PE-FOC, (d) IO-FOC.
FOC and E-FOC. Under IO-FOC control, the PMSM exhibits almost no
overshoot and smooth startup, with the smallest overshoot and stabili- 5.3. Experimental results under steady-state conditions
zation time, indicating superior dynamic response, speed stability, and
motor efficiency. After analyzing start-up performance, we further explore and
The outstanding performance of IO-FOC in start-up conditions is due compare the steady-state performance of PMSM under four control
to the IO algorithm combined with an new fitness function, which op- strategies.
timizes key parameters in real time. This ensures high robustness and In the working condition shown in Fig. 15, the PMSM operates stably
fast system response, making it ideal for high-precision applications in at 2000 rpm and 20Nm under the four strategies. The figure shows the
complex conditions. torque, speed, and dq-axis current for each control strategy. As shown in
Fig. 15, compared to other strategies, the speed fluctuation under IO-
FOC control is smaller and remains stable around 2000 rpm, indi-
cating its excellent steady-state performance in maintaining smooth
motor operation at the set speed. Regarding torque, the fluctuation
under IO-FOC control is significantly smaller than other strategies,
consistent with the trend of minimal and stable q-axis current fluctua-
tions. Additionally, torque under other strategies occasionally shows
sudden changes, whereas under IO-FOC control, it remains stable,
indicating effective suppression of transient errors. To further observe
PMSM performance, Fig. 16 presents the corresponding three-phase
current operation.
The three-phase current waveforms in Fig. 16 all exhibit typical si-
nusoidal waves. Compared to other strategies, the waveforms under IO-
FOC control are smoother with smaller fluctuation amplitudes, indi-
cating superior control performance and better harmonic suppression.
For a quantitative analysis of the three-phase current, a Fourier trans-
form was performed. Fig. 17 shows the total harmonic distortion (THD)
of the three-phase currents: FOC at 5.67 %, E-FOC at 5.31 %, PE-FOC at
5.11 %, and IO-FOC at 3.84 %. These results indicate that the IO-FOC
algorithm has the lowest THD and significantly outperforms the other
three algorithms in terms of harmonic suppression. PE-FOC achieves
better harmonic suppression by offline optimization of ESO parameters
using particle swarm optimization (PSO). However, because PSO opti-
mization cannot update in real time, the harmonic suppression effect has
limitations in complex operating conditions. In contrast, IO-FOC ach-
ieves balanced suppression of multiple harmonics by combining intel-
ligent optimization algorithms with a new fitness function,
demonstrating its real-time optimization of control parameters and
disturbance compensation. The IO-FOC algorithm enhances its real-time
tracking ability for high-frequency current error components by
adjusting the ESO bandwidth, thereby reducing high-order harmonic
amplitudes. The IO-FOC algorithm optimizes PI controller parameters in
real time based on the operating status, reducing low-order harmonic
components caused by controller misalignment during steady-state
Fig. 15. Experimental data of PMSM under steady-state conditions: (a) FOC, operation. IO-FOC employs a new fitness function, offering more
(b) E-FOC, (c) PE-FOC, (d) IO-FOC.

293
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

Fig. 18. Comparison of PMSM speed under dynamic conditions: (a) 2000 rpm,
(b) 1000 rpm, (c) 500 rpm, (d) 100 rpm.

lag. PE-FOC improves the speed drop and shortens recovery time. IO-
FOC exhibits the smallest speed drop and shortest recovery time,
demonstrating excellent anti-interference performance at high speeds.
At other speeds, IO-FOC continues to show the smallest speed drop,
confirming its superior dynamic response. Controlled by the IO-FOC
algorithm, the PMSM quickly identifies system perturbations caused
by load shocks and dynamically adjusts control parameters to achieve
Fig. 17. Comparison of THD analysis of three-phase currents in steady-state rapid compensation. This capability significantly enhances the control
operation: (a) FOC, (b) E-FOC, (c) PE-FOC, (d) IO-FOC.

Table 3
precise optimization objectives for harmonic suppression. Through real- Comparison of starting performance of PMAM.
time identification of dynamic parameters such as inductance, resis-
Performance indicators Speed FOC E-FOC PE-FOC IO-FOC
tance, and magnetic flux, the fitness function captures nonlinear effects (rpm):
caused by parameter changes in the system. 2000
The identification results of these parameters directly affect the 1000
prediction of dq-axis current error, which in turn influences the opti- 500
100
mization of control parameters and enhances the harmonic suppression
Δn (rpm) 10Nm 87 72 35 28
performance of the IO-FOC algorithm. 88 63 25 19
48 42 29 23
5.4. Experimental results under dynamic conditions 47 36 15 12
​ 15Nm 105 81 55 47
102 71 32 28
Sections 5.2 and 5.3 provide comparative analyses of the start-up and 73 61 36 30
steady-state performance of PMSM under different control strategies. 69 54 22 19
This section further investigates the dynamic performance of PMSM Δt 10Nm 0.1282 0.1173 0.1111 0.1088
(s) 0.1658 0.1607 0.1219 0.1185
under the four strategies.
0.1818 0.1633 0.1356 0.1305
Fig. 18 shows the dynamic response of PMSM rotational speed under 0.2502 0.1926 0.1783 0.1549
the four control strategies at different speeds (100 rpm, 500 rpm, ​ 15Nm 0.138 0.123 0.1217 0.1108
1000 rpm, 2000 rpm) and load shocks (10 Nm: from 0 Nm to 10 Nm; 0.205 0.1975 0.133 0.1191
15 Nm: from 10 Nm to 25 Nm). At a set speed of 2000 rpm, FOC and E- 0.224 0.181 0.153 0.1376
0.259 0.203 0.1867 0.1812
FOC show significant speed drops and longer recovery times, indicating

294
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

system’s robustness and adaptability. The intelligent optimization al-

gorithm forms the core of the IO-FOC, driving its excellent dynamic
performance through the following optimizations:

(1) The intelligent optimization algorithm dynamically optimizes the

ESO gain, allowing the observer to promptly detect system state
changes during the initial stage of a perturbation and adapt to
both high-frequency and low-frequency disturbances.
(2) PI parameters critically influence the system’s response speed to
disturbances and its ability to suppress steady-state errors. The Fig. 19. Speed fluctuation values at different step loads from 100 to 3000 rpm:
intelligent optimization algorithm dynamically adjusts the pro- (a) 10 Nm, (b) 15 Nm.
portional and integral coefficients, enhancing the current loop’s
dynamic response post-load impact. disturbance scenarios.

Leveraging the intelligent optimization algorithm, the fitness func- 6. Comprehensive performance analysis
tion captures dynamic errors induced by parameter drift or nonlinear
effects through real-time identification of key motor parameters. Future To further analyze the effects of different strategies on PMSM per-
current values are predicted using Eq. (30), and the fitness function is formance, this paper presents speed fluctuation (Δn) and stabilization
evaluated using Eq. (31) to achieve precise compensation for distur- time (Δt) curves under 10 Nm and 15 Nm step loads, as shown in Figs. 19
bance sources. Additionally, the fitness function integrates current error, and 20. Fig. 19 shows the speed drop (Δn) of the PMSM under 10Nm and
disturbance effects, and dynamic response performance to ensure real- 15Nm step loads, ranging from 100 rpm to 3000 rpm, demonstrating the
time adjustment of the ESO and PI controllers’ optimization direction dynamic response of the four control strategies (FOC, E-FOC, PE-FOC,
following a load impact. This dynamic feedback mechanism enables and IO-FOC) to step loads.
rapid system stabilization while avoiding the response lag associated As seen in Fig. 19, under FOC control, PMSM speed drops the most,
with traditional fixed control objectives. indicating weak disturbance immunity. This is because FOC uses current
The previous analyses were qualitative. For a quantitative evalua- closed-loop control, which struggles to quickly compensate for step
tion, Table 3 summarizes the results of comparative experiments on the loads. E-FOC compensates for step loads using the ESO, theoretically
PMSM’s dynamic performance. In evaluating the performance of PMSM improving system robustness. However, the compensation effect is
control algorithms, selecting recovery time and speed drop overshoot limited in practice and falls short of expectations. PE-FOC keeps Δn
under external shocks, holds significant theoretical and practical value. within a reasonable range by optimizing the gain of the extended-state
These parameters intuitively reflect the system’s dynamic response and disturbance observer online, enhancing dynamic response. IO-FOC
robustness, directly influencing the motor’s practical performance and performs best across the full speed range, with PMSM system showing
the overall system’s reliability. A shorter stabilization time minimizes excellent dynamic response at various speeds and loads.
energy waste during transitions, improving the motor system’s energy Fig. 20 reveals a similar trend to Fig. 19, where IO-FOC again dem-
utilization. Speed overshoot reflects the system’s stability and control onstrates the best dynamic response across the full speed range. IO-FOC
precision. Excessive overshoot can induce mechanical vibrations, dynamically adjusts PI controller parameters and the gain of the ESO in
component impacts, or system fluctuations, particularly in scenarios the FOC system via the intelligent optimization module. This ensures
involving large load inertia. optimal response to step loads across the full speed range, enhancing the
The data reveal that the conventional FOC strategy results in sig- system’s adaptability to various conditions. Its excellent performance
nificant speed fluctuations and prolonged recovery times following load under high speeds and dynamic loads highlights its potential for wide
shocks. For instance, under a 10Nm load shock at 2000 rpm, the FOC application in the new energy sector.
strategy exhibits a speed fluctuation of 87 rpm and a stabilization time
of 0.1282 s. This indicates that the FOC strategy is highly parameter- 7. Conclusion
dependent and lacks an effective dynamic compensation mechanism,
limiting its ability to handle load variations. In contrast, E-FOC in- This paper introduces a intelligent optimization field-oriented con-
troduces an Extended State Observer (ESO) to provide online compen- trol (IO-FOC) strategy for PMSM, tackling key challenges in precision
sation for internal disturbances, reducing speed fluctuation to 72 rpm control and system robustness. The proposed strategy offers substantial
and stabilization time to 0.1173 s at 2000 rpm. However, E-FOC’s improvements in dynamic response and steady-state performance
compensation effect is constrained by its fixed gain setting, limiting its compared to conventional FOC methods. The key contributions of this
adaptability to larger disturbances. PE-FOC optimizes ESO’s key gain work are:
parameter using the PSO algorithm, enhancing system performance
under load shocks. For example, under a 10 Nm load impact at
2000 rpm, speed fluctuation decreases to 35 rpm, and stabilization time
is shortened to 0.1111 s. However, as PSO optimization is a static offline
process, it lacks the dynamic adaptability and real-time optimization
required for complex working conditions. The proposed IO-FOC strategy
enhances the system’s dynamic response by integrating an intelligent
optimization algorithm to dynamically adjust ESO gains and PI
controller parameters. According to Table 3, under a 10Nm load impact
at 2000 rpm, IO-FOC achieves a speed fluctuation of only 28 rpm and a
stabilization time of 0.1088 s. For a 15Nm load impact at 2000 rpm, it
maintains the lowest speed fluctuation (47 rpm). The data demonstrate
that IO-FOC performs optimally, with the smallest overshoot and
shortest recovery time. This is due to real-time parameter optimization
and disturbance compensation, which significantly reduce speed over- Fig. 20. Speed stabilization time at different step loads from 100 to 3000 rpm:
shoot and recovery time, making it ideal for high-precision, high-load (a) 10 Nm, (b) 15 Nm.

295
Z. You et al. Alexandria Engineering Journal 115 (2025) 286–296

(1) A comprehensive control framework integrating ESO-based [13] R.G. Shriwastava, S.S. Kadlag, J.G. Chaudhari, P.R. Sonawane, N. Dhote, A multi-
level neutral-point-clamped inverter driven PMSM high-speed electric drive, Int. J.
voltage compensation with the IO algorithm, effectively miti-
Electr. Hybrid. Veh. 16 (2024).
gating inherent nonlinearity and parameter uncertainties in [14] Z. Novak, Confidence weighted learning entropy for fault-tolerant control of a
PMSM systems. PMSM with a high-resolution hall encoder, IEEE Trans. Ind. Electron. 71 (2024)
(2) An advanced IO algorithm that simultaneously optimizes ESO 5176–5186.
[15] A. Aboelhassan, S. Wang, G. Buticchi, V. Varvolik, M. Galea, S. Bozhko, Modulated
gains and current-loop PI parameters. This global optimization model predictive speed controller for PMSM drives employing voltage-based cost
significantly improves system adaptability compared to tradi- function, IEEE Open J. Ind. Electron. 5 (2024) 122–131.
tional FOC under varying load conditions. [16] L.H. Gao, G.M. Zhang, H.M.O. Yang, L. Mei, A novel method of model predictive
control on permanent magnet synchronous machine with Laguerre functions, Alex.
(3) A novel parameter identification-based fitness function incorpo- Eng. J. 60 (2021) 5485–5494.
rating real-time inductance, resistance, and flux linkage estima- [17] L.H. Sun, Low speed sensorless control method of brushless DC motor based on
tion, ensuring robust performance despite parameter variations. pulse high frequency voltage injection, Alex. Eng. J. 61 (2022) 6457–6463.
[18] S. Mondal, P. Roy, A. Banerjee, U. Mondal, A CKF-based sensor-less FOC integrated
with gh-SVPWM for PMSM drives, Electr. Eng. 106 (2024) 3461–3473.
The IO-FOC strategy sets a new benchmark for high-performance [19] M. Nicola, C.-I. Nicola, Improvement of linear and nonlinear control for PMSM
PMSM control, especially in applications requiring precise motion con- using computational intelligence and reinforcement learning, Mathematics 10
(2022).
trol and robust operation under varying conditions. Future research may [20] Z.J. Jin, X.D. Sun, G. Lei, Y.G. Guo, J.G. Zhu, Sliding mode direct torque control of
explore integrating advanced neural network techniques and real-time SPMSMs based on a hybrid wolf optimization algorithm, IEEE Trans. Ind. Electron.
optimization strategies to further boost system performance. 69 (2022) 4534–4544.
[21] W.A.E.M. Ahmed, M.M. Adel, M. Taha, A.A. Saleh, PSO technique applied to
sensorless field-oriented control PMSM drive with discretized RL-fractional
CRediT authorship contribution statement integral, Alex. Eng. J. 60 (2021) 4029–4040.
[22] H. Mahmoudi, M. Aleenejad, R. Ahmadi, Torque ripple minimization for a
Chen Li: Resources, Funding acquisition. Yifan Zhang: Resources. permanent magnet synchronous motor using a modified quasi-Z-source inverter,
IEEE Trans. Power Electron. 34 (2019) 3819–3830.
Yongming Bian: Resources. Zhou YOU: Writing – original draft, [23] A.H. Abosh, Z.Q. Zhu, Y. Ren, Reduction of torque and flux ripples in space vector
Software. modulation-based direct torque control of asymmetric permanent magnet
synchronous machine, IEEE Trans. Power Electron. 32 (2017) 2976–2986.
[24] Z. Wu, Z. Yang, K. Ding, G. He, Transfer mechanism analysis of injected voltage
Declaration of Competing Interest harmonic and its effect on current harmonic regulation in FOC PMSM, IEEE Trans.
Power Electron. 37 (2022) 820–829.
The authors declare that they have no known competing financial [25] A. Ben Sada, A. Khelloufi, A. Naouri, H.S. Ning, S. Dhelim, Hybrid metaheuristics
for selective inference task offloading under time and energy constraints for real-
interests or personal relationships that could have appeared to influence time IoT sensing systems, Clust. Comput. - J. Netw. Softw. Tools Appl. 27 (2024)
the work reported in this paper. 12965–12981.
[26] Y. Wang, Y. Feng, X. Zhang, J. Liang, A new reaching law for antidisturbance
sliding-mode control of PMSM speed regulation system, IEEE Trans. Power
Acknowledgment Electron. 35 (2020) 4117–4126.
[27] J. Song, Y.-K. Wang, Y. Niu, H.-K. Lam, S. He, H. Liu, Periodic event-triggered
This work supported by the National Key Research and Development terminal sliding mode speed control for networked PMSM system: a GA-optimized
extended state observer approach, IEEE-Asme Trans. Mechatron. 27 (2022)
Program of China (Grant No. 2023YFF0613200).
4153–4164.
[28] J.C. Chen, T.Y. Li, Y. Zhang, T. You, Y.T. Lu, P. Tiwari, N. Kumar, Global-and-local
References attention-based reinforcement learning for cooperative behaviour control of
multiple UAVs, IEEE Trans. Veh. Technol. 73 (2024) 4194–4206.
[1] M. Tian, B. Wang, Y. Yu, Q. Dong, D. Xu, Adaptive active disturbance rejection [29] J.C. Chen, C.L. Du, Y. Zhang, P.C. Han, W. Wei, A clustering-based coverage path
control for uncertain current ripples suppression of PMSM Drives, IEEE Trans. Ind. planning method for autonomous heterogeneous UAVs, IEEE Trans. Intell. Transp.
Electron. 71 (2024) 2320–2331. Syst. 23 (2022) 25546–25556.
[2] Z. Zhang, Sensorless back EMF based control of synchronous PM and reluctance [30] Z. Sun, T. Sato, K. Watanabe, An efficient combination of topology optimization
motor drives-a review, IEEE Trans. Power Electron. 37 (2022) 10290–10305. and parameter optimization for electromagnetic devices, Int. J. Appl. Electro Mech.
[3] M.S. Rafaq, W. Midgley, T. Steffen, A review of the state of the art of torque ripple 71 (2023) S47–S56.
minimization techniques for permanent magnet synchronous motors, IEEE Trans. [31] A. Naouri, H. Ning, N.A. Nouri, A. Khelloufi, A. Ben Sada, S. Naouri, A. Qammar,
Ind. Electron. 20 (2024) 1019–1031. S. Dhelim, Maximizing UAV fog deployment efficiency for critical rescue
[4] Z. Zhang, Q. Sun, Q. Zhang, A computationally efficient model predictive control operations: a multi-objective optimization approach, Future Gener. Comput. Syst. -
method for dual three-phase PMSM of electric vehicle with fixed switching Int. J. Esci. 159 (2024) 255–271.
frequency, IEEE Trans. Ind. Electron. 60 (2024) 1105–1116. [32] Z. Qian, T.C. Huang, Q.J. Wang, W.Z. Deng, Q.X. Chen, Z.H. Sun, Q.B. Ye, Torque
[5] C. Zhang, F. Wang, X. Li, Z. Dong, Y. Zhang, Fault diagnosis method of permanent Ripple Reduction of PMSM Based on Modified DBN-DNN Surrogate Model, IEEE
magnet synchronous motor based on WCNN and few-shot learning, Actuators 13 Trans. Transp. Electrif. 9 (2023) 2820–2829.
(2024). [33] Y. Xiang, X. Pei, Y. Zhang, D. Jiang, Feature-engineering enabled multiobjective
[6] S. Pang, Y. Zhang, Y. Huangfu, X. Li, B. Tan, P. Li, C. Tian, S. Quan, A virtual MPC- evolutionary impedance fitting technique, IEEE Trans. Ind. Electron. 71 (2024)
based artificial neural network controller for PMSM drives in aircraft electric 4450–4462.
propulsion system, IEEE Trans. Ind. Electron. 60 (2024) 3603–3612. [34] Y. Gao, PID-based search algorithm: a novel metaheuristic algorithm based on PID
[7] K. Guo, Y. Guo, S. Fang, C. Li, W. Xue, Design and analysis of a permanent magnet algorithm, Expert Syst. Appl. 232 (2023).
frameless motor, IEEE J. Emerg. Sel. Top. P. 12 (2024) 3124–3134. [35] M.R. Arahal, F. Barrero, M.G. Satue, C. Martin, M. Bermudez, Evolutionary gaps
[8] X. Zhang, S. Fang, H. Zhang, Predictive current error compensation-based strong stator current control of multiphase drives balancing harmonic content, IEEE
robust model predictive control for PMSM drive systems, IEEE Trans. Ind. Electron. Trans. Ind. Electron. 71 (2024) 6886–6893.
(2024) (early access). [36] A. Naouri, N.A. Nouri, A. Khelloufi, A.B. Sada, H.S. Ning, S. Dhelim, Efficient fog
[9] S. Liu, Z. Song, Y. Liu, Y. Chen, C. Liu, Flux-weakening controller design of dual node placement using nature-inspired metaheuristic for IoT applications, Clust.
three-phase PMSM drive system with copper loss minimization, IEEE Trans. Power Comput. - J. Netw. Softw. Tools Appl. 27 (2024) 8225–8241.
Electron. 38 (2023) 2351–2363. [37] M. Megrini, A. Gaga, Y. Mehdaoui, J. Khyat, Design and PIL test of extended
[10] Z. Sun, Y. Deng, J. Wang, T. Yang, Z. Wei, H. Cao, Finite control set model-free Kalman filter for PMSM field oriented control, Results Eng. 24 (2024).
predictive current control of PMSM with two voltage vectors based on ultralocal [38] X. Dong, J. Mao, Y. Yan, C. Zhang, J. Yang, Generalized dynamic predictive control
model, IEEE Trans. Power Electron. 38 (2023) 776–788. for nonlinear systems subject to mismatched disturbances with application to
[11] S. Xu, Q. He, S. Tao, H. Chen, Y. Chai, W. Zheng, Pig face recognition based on PMSM drives, IEEE Trans. Ind. Electron. 71 (2024) 954–964.
trapezoid normalized pixel difference feature and trimmed mean attention [39] K. Lu, X. Li, Y. Zhao, P. Yi, B. Yan, W. Hua, A novel three-vector-based model
mechanism, IEEE Trans. Instrum. Meas. 72 (2023). predictive flux control with low computation complexity for SPMSM, IEEE Trans.
[12] C. Candelo-Zuluaga, J.-R. Riba, A. Garcia, PMSM parameter estimation for Transp. Electrif. 10 (2024) 3956–3965.
sensorless foc based on differential power factor, IEEE Trans. Instrum. Meas. 70
(2021).

296