IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO.

7, JULY 2023 7847

Efficiency Optimization of Wireless Power Transfer

System for Electric Vehicle Based on Improved
Marine Predators Algorithm
Quanyi Yu , Jun Lin , Xilai Ma, Bo Li, Linlin Xu, and Tianhao Wang , Member, IEEE

Abstract— Electric vehicle (EV) is the core part of future technology of static WPT is relatively mature [7]. Generally,
automobile technology, and safe and reliable wireless power EVs WPT requires high frequencies and several kilowatts or
transfer (WPT) technology is the key link to improve the even tens of kilowatts of power to achieve fast charging of EVs
intelligent driving technology of EV. In this paper, the uncertainty
quantification method is proposed to guide the optimization batteries. Ensuring and improving the efficiency of wireless
design of WPT structure, so as to improve the efficiency of WPT. charging systems has become the focus of research. The
First this paper establishes a surrogate model of WPT efficiency charging efficiency of various types of EVs batteries has been
based on the adaptive sparse polynomial chaos expansion, extensively tested [8], [9]. Owing to the high complexity of
and the uncertainty of EVs WPT transmission efficiency is the WPT and the uncertainties in the manufacturing process of
quantified, the computational efficiency is improved by about
8.4 times. Then the surrogate model is combined with the global the components of the system (coil offset, coil cross-sectional
sensitivity analysis method to quantify the impact of different area, and capacitance and resistance in the compensation
variables in WPT on efficiency and screen out the variables circuit), the efficiency of the WPT may be greatly changed and
with greater impact. Finally, this paper uses the improved affected. In the above problems, the uncertainty quantification
marine predators algorithm to optimize the selected WPT system (UQ) methods are of great significance because they can
structure parameters. Considering the uncertainty, the average
efficiency of the optimized WPT system is increased from 73.43% predict the influence of the uncertainties of the relevant
to 94.64%. Compared with other optimization methods, it proves parameters on the performance of the WPT. Quantifying the
that the method in this paper can optimize WPT more efficiently, influence of the relevant parameters on the WPT efficiency
and significantly improve the transmission efficiency. can guide the structural optimization of the WPT and further
Index Terms— Adaptive sparse polynomial chaos expansion, improve the performance of the WPT. In order to solve the
efficiency optimization, electric vehicles, marine predators problem that the optimization design of efficiency of EVs WPT
algorithm, uncertainty quantification, wireless power system involves many variables and leads to high optimization
transfer. cost, this paper proposes to use UQ method to guide the
optimization design of EVs WPT. The main contributions of
I. I NTRODUCTION this paper are as follows: 1. This paper uses sparse polynomial
chaos to quantify the uncertainty of EVs WPT’s efficiency. 2.

T O ALLEVIATE the problem of greenhouse gas emissions

from traditional fuel vehicles, the development of electric
vehicles (EVs) has received strong support [1], [2], [3], [4].
Combined with the global sensitivity analysis methods, select
the variables that have a greater impact on the efficiency of
EVs WPT. 3. Propose an improved marine predators algorithm
Compared with wired power transmission technology, the to optimize the design of WPT system, and significantly
convenience and safety of wireless power transfer (WPT) improve the transmission efficiency.
technology make it one of the most promising technologies The main contents of this paper are as follows. Section II
in the field of EVs [5]. EVs WPT methods can be divided introduces the related work of uncertainty quantification of
into static, stationary, and dynamic [6], among which the EVs WPT efficiency and the optimization of EVs WPT;
Section III introduces the ASPCE based on the least angle
Manuscript received 18 October 2022; revised 2 December 2022;
accepted 12 December 2022. Date of publication 22 December 2022; date regression (LAR) and hyperbolic truncation method, and
of current version 7 July 2023. This work was supported in part by the GSA method: Sobol and Morris one-at-a-time are
the Jilin Scientific and Technological Development Program under Grant also introduced. Section IV introduces the multi-objective
20220101196JC and Grant 20190103055JH. The Associate Editor for this
article was Z. Lv. (Corresponding author: Tianhao Wang.) optimization algorithm: MPA, then the opposition-based
Quanyi Yu, Jun Lin, Bo Li, Linlin Xu, and Tianhao Wang are with learning and tournament selection are used to improve the
the College of Instrument Science and Electrical Engineering, Jilin MPA. Section V introduces the WPT model of EVs based on
University, Changchun 130026, China (e-mail: [email protected];
[email protected]; [email protected]; [email protected]; simulation software. ASPCE is used to quantify the uncertainty
[email protected]). of the EVs WPT model, and the MPA is used to optimize
Xilai Ma is with the Electrical Department, Commerical Vehicle Devel- the efficiency of WPT. The relevant statistical characteristic
opment Institute, First Automobile Workshop (FAW) Jiefang, Changchun
130011, China (e-mail: [email protected]). parameters of efficiency before and after optimization are
Digital Object Identifier 10.1109/TITS.2022.3229958 compared. Section VI summarizes all the work in this paper.
1558-0016 © 2022 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: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7848 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

II. R ELATED W ORK optimization design of EVs WPT. In this paper, the UQ
surrogate model of transmission efficiency of EVs WPT
The Monte Carlo (MC) is one of the most classical system is established, the variables that have a great impact on
uncertainty quantification methods, but considering the efficiency in WPT are screened through the results of global
complex structure of the EVs WPT and the computational cost sensitivity analysis (GSA), and the WPT is optimized by using
of the simulation software, the MC is obviously unsuitable for the improved marine predators algorithm (MPA), so as to
the UQ of WPT. In recent years, many UQ methods based on realize the significant improvement of WPT efficiency.
machine learning theory are widely studied, such as support
vector machine [10], Gaussian process regression (GPR) [11],
Bayesian neural network [12], and so on. Aiming at the III. A DAPTIVE S PARSE P OLYNOMIAL C HAOS E XPANSION
problem of UQ of WPT, Larbi et al. [13] took the efficiency AND G LOBAL S ENSITIVITY A NALYSIS
of WPT as the objective of UQ, calculated and analyzed it by
partial least square method, and finally optimized the design of Torre et al. [29] proposed a PC-based surrogate model
WPT in combination with sensitivity analysis and correlation method to solve the problem of small sample UQ. The model
analysis, effectively improving the efficiency. In addition to can also obtain comparable accuracy with machine learning
machine learning, surrogate models are also widely used for methods; moreover, the construction of PC is simpler and
UQ, and one of the most commonly used methods is the requires fewer training sets. Considering the complexity and
polynomial chaos (PC) [14], [15], [16]. Lagouanelle et al. [17] high simulation cost of the WPT of EVs, this study uses the
established a Kriging surrogate model to quickly predict adaptive sparse polynomial chaos expansion (ASPCE) method
the radiation field value of the human exposure near the to quantify the uncertainty of the WPT efficiency. The PCE is
WPT. Bilicz et al. [18] used the generalized polynomial one of the most popular methods in recent years. Xiu proposed
chaos expansion (gPCE) to establish a surrogate model of the Askey scheme based on the Wiener PC and extended it to
the WPT, and they analyzed the influence of the geometric gPCE with wider applicability. Let the model be Y = y (ξ )
design parameters’ uncertainties of the magnetic coupling and expand it into a PC.
mechanism on the electrical characteristics of the system. The ∞ ∞ 
    i1
 
PCE is simple to construct and has high calculation accuracy. Y = c0 I 0 + ci1 I1 ξi1 + ci1 i2 I2 ξi1 , ξi2 + · · ·
Although it can cause the curse of dimensionality in the face i1 =1 i1 =1 i2 =1
of systems with higher dimensions, many sparse calculation ∞

methods have been proposed to alleviate the above problems = ĉi i (ξ ) (1)
[19], [20]. i=0
Efficiency is one of the most important properties of WPT.
 
To improve the efficiency of WPT, researchers proposed many In (1), In ξi1 , · · · , ξin represents the combination
design and optimization methods of WPT. Kürschner et al. [21] of orthogonal polynomials corresponding to the multi- 
improved the efficiency of WPT by optimizing the geometry of dimensional standard random variable ξi1 , · · · , ξin .
coils and ferrites. Zhang et al. [22] optimized the parameters Truncating (1), let the truncation order be P. The truncated
of series compensation circuit and finally realized the ideal gPCE model is
efficiency. Subsequently, the secondary series and parallel
compensation circuits are optimized to obtain the optimal 
P
efficiency [23]. Wen et al. [24] studied series compensation Y = ĉi i (ξ ) (2)
topology and hybrid series-parallel compensation topology, i=0
optimized the related parameters, and improved the efficiency
The number of terms of the truncated gPCE is Q, which
to 89.8%. Andersen et al. [25] optimized the design of
can be obtained according to the truncation order P and the
transmitting coils for biomedical implants and ultimately
dimension n of the model input variable.
maximized the transmission power of the WPT while meeting
exposure limits. Zhang et al. [26] proposed a new type (P + n)!
of orthogonal coil and optimized its structure. Finally, Q= (3)
P!n!
the efficiency of the coil reached 94.22%. Selecting the
appropriate methods to optimize the coil geometry parameters The key of the gPCE is to solve the PC coefficients,
and compensation circuit parameters of WPT can effectively sample the input variables of each dimension and obtain the
improve the efficiency of the coil. The optimization algorithm corresponding response values, and obtain the coefficients
is widely used [27], which can effectively calculate the optimal through the least squares method to complete the construction
result of the problem. On the other hand, solving the model of the gPCE surrogate model. In the face of models with
by numerical algorithm can reasonably analyze the relevant low variable dimensions, the gPCE has high computational
information of the model [28]. Existing optimization studies accuracy and computational efficiency. However, when the
on EVs WPT usually require a large number of optimization model has many input variables, it encounters the curse of
variables, resulting in low optimization efficiency, and the dimensionality. To address this problem, the gPCE is sparsely
impact of the uncertainty of each variable is not taken into processed and calculated from two aspects: truncation scheme
account. This paper will use the UQ method to guide the and coefficient calculation method.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 YU et al.: EFFICIENCY OPTIMIZATION OF WPT SYSTEM FOR EV BASED ON IMPROVED MPA 7849

A. Hyperbolic Truncation Based on the idea of ANOVA, Sobol rewrites (2) into the
Define the highest order in i as pi and the order of each form of incremental summation of polynomials of different
input variable as lk : orders, and takes variance at both ends of the equation.


n var [Y (ξ )]
pi = lk = l1 + l2 + · · · + ln ⎡ ⎤
(4) 
n 
k=1 ⎢ c0 + cα α (ξα ) ⎥
⎢ α∈I ⎥
Referring to the q norm in hyperbolic truncation, (4) can ⎢  
i=1 i
  ⎥
⎢ + cα α ξi1 , ξi2 + · · · ⎥
be written as ⎢ ⎥
⎢ ⎥
 n 1/q ⎢
= var ⎢ 
1≤i 1 <i 2 ≤n α∈I
1 2i ,i ⎥
   ⎥
⎢+ cα α ξi1 , · · · , ξis + · · · ⎥
pi = (lk )q ≤ P, 0 < q ≤ 1 (5) ⎢ ⎥
⎢ 1≤i1 <···<n α∈Ii1 ,··· ,is ⎥
k=1 ⎢  ⎥
⎣ + c  (ξ , · · · , ξ ) ⎦
In (5), q = 1, Pmax = P, the traditional truncation is a α α 1 n
α∈I1,...,n
special form of the hyperbolic truncation. With the decrease
of q, the hyperbolic truncation can significantly reduce the 
n 
= D= Di + Di j + · · · + D1,2,...,n
polynomials in the model and realize the sparsification of
i=1 1≤i< j ≤n
polynomial chaos surrogate models.
Ii1 ,...,is
 
α ∈ (α1 , α2 , · · · , αn ) : αk = 0
B. Least Angle Regression = (6)
k∈/ (i 1 , · · · , i s ) , ∀k = 1, · · · , n
The least angle regression (LAR) method was first proposed
Sobol global sensitivity index can be expressed as:
by Efron [30]. It combines the advantages of the forward
selection algorithm and forward gradient algorithm and Di1 ,...,is
Si1 ,...,is = , 1 ≤ i 1 < . . . < i s ≤ n; s = 1, . . . , n
only needs fewer iterations to complete all the regression D
calculations. Let the original model be Y = i ci . The (7)
calculation of LAR is as follows:
1) Search for the vector 1 most relevant to the model Si in (7) is the first-order sensitivity index, which represents
output Y. the contribution of a single input to the output response
2) Take the direction of vector 1 as the moving direction. variance. The sum of the first-order sensitivity index of each
Search forward along this direction until the angle input variable and the sensitivity index of the interaction
between vectors Y − 1 ∗ c1 and 1 is the same as between variables is defined as the total sensitivity index:
that of vector 2 , that is, the correlation between vector 
SiT = Si + S j,k,i + · · · + S1,2,··· ,n (8)
1 and vector 2 and observation vector γ is equal,
j <i
and c1 is the coefficient of 1 at this time.
3) Continue to proceed according to the above step and Owing to the convenience of variance calculation of PC: var =
search for the same degree of correlation between vector  2
p−1
ci , it is easy to calculate Di1 ,...,is based on the PC and
3 and γ , and calculate the coefficient of this iteration. i=1
4) Update the coefficients at this time and move the chosen obtain the Sobol global sensitivity index.
vector to the active set.
5) Repeat the above steps.
D. Morris One-at-a-Time
The LAR can reduce the number of expansion terms in the
PC on the basis of the hyperbolic truncation, which effectively MOAT is a GSA method different from Sobol. Although the
saves the computational cost. The ASPCE can be realized by calculation accuracy is lower than Sobol, MOAT is calculated
adaptively selecting the truncation order P and q mode with purely based on samples, so the calculation cost is low, and it
the leave-one-out cross-validation error as the threshold. can screen the importance of input variables.
MOAT first uses Morris sampling to obtain r trajectories
and generate l levels for each one-dimensional variable ξi .
C. Sobol Global Sensitivity Analysis Then the sampled data points are ξi, j , j = 1, · · · , r , and the
In order to efficiently optimize the EVs WPT system to influence degree of the i th input variable can be calculated as:
improve the efficiency, this study integrates two GSA methods
to screen the variables. ASPCE and Sobol are combined to Ii, j
   
quantify the influence of input variables [31]. Morris one- y ξ1, j , ξ2, j , . . . , ξi, j ±, . . . , ξn, j − y ξ1, j , ξ2, j , . . . , ξn, j
at-a-time (MOAT) is used to calculate the MOAT mean and =

standard deviation of each variable. The importance of input (9)
variables in the UQ model of WPT efficiency are screened
according to the calculation results of the above two methods where  = l
2(l−1) = 1
2 + 1
2(n−1) .
so as to provide guidance for the optimization of efficiency of According to the elementary effect of r replication
WPT system. points, the mean value of MOAT of the i th input

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7850 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

can be expressed as: exploration strategy is adopted, and the walk mode is
 Brownian motion. When I ter ≤ 13 Max I t er ,
1
r
 
μi =  Ii, j  (10) −−−→ 
→
− −
→ −→ −−−→
r
j =1 s = R B ⊗ Eli te − R B ⊗ Pr eyi i = 1, . . . , d
−−−→ −−−→ → →
−
Similarly, the MOAT standard deviation is: Pr eyi = Pr eyi + M · R ⊗ − s (15)

  −
→
1 r  2 where R B is a vector containing a random number based
σi =  Ii, j − μi (11)
r on a normal distribution representing Brownian motion,
j =1
M = 0.5, R is the random number of [0,1], I ter is the
MOAT mean can estimate the influence of different input current iteration, Max I t er is the maximum number of
variables on the model output, while MOAT standard deviation iterations, and −→s is the moving step size.
can estimate the nonlinear influence of input variables and 3) The mid-optimization phase
the interaction between different variables. In this paper, the when 13 Max I t er ≤ I ter ≤ 23 Max I t er , the speed of the
calculation results of Sobol and MOAT are comprehensively prey and predator are the same. Therefore, the prey
considered to screen the variables with greater influence in the is responsible for the development, and the predator
UQ model of EVs WPT efficiency, and the screened WPT is is responsible for the exploration. Based on the above
optimized to achieve the goal of improving efficiency. rules, in unit speed ratio, the prey moves in Lévy,
where R is a random-number vector based on a Lévy
IV. O PTIMIZATION D ESIGN M ETHOD distribution representing Lévy motion, whereas the best
A. Marine Predators Algorithm strategy for the predator is Brownian motion.
−−−→  
Aiming at the optimization problem of efficiency of EVs →
− −→ → −−−→
−
s = R L ⊗ Eli te − R L ⊗ Pr eyi i = 1, . . . , d 2
WPT, this study uses intelligent optimization algorithm to −−−→ −−−→ → →
−
calculate. The MPA simulates the Lévy movement and Brown Pr eyi = Pr eyi + M · R ⊗ − s (16)
movement adopted by predators in the sea when searching for −
→
prey. Fish aggregating devices’ (FADs) effect and eddy effect where R L is a random-number vector based on a Lévy
are adopted to address and avoid falling into the local optimum distribution representing Lévy strategy. As the step
[32]. The different calculation stages of the MPA are described size of Lévy is small, this stage is suitable for the
below. development. The latter half adopts Brownian motion.
  
1) Initialization phase →
− −→ → −−−→ −−−→
−
The variables in the search space are initialized. s = R B ⊗ R B ⊗ Eli te − Pr eyi i = d 2, . . . , d
−−−→ −−−→
X 0 = X min + r and (X max − X min ) (12) Pr ey = Eli te + M · C F ⊗ −
i
→s (17)
 
where X max and X min are the upper and lower bounds of   2 I ter
Max I ter
where C F = 1 − Max I t er
is the adaptive
the variables, respectively, and rand is a uniform random I ter

variable in the range of 0–1. According to theory of the parameter of the predator step size.
→
− 4) The last optimization phase
fittest’s survival, X l is the optimal vector in the prey when 23 Max I t er ≤ I ter ≤ Max I t er , the predator moves
matrix. In matrix X i , j is the j th dimension of the faster than the prey, which means that in a low speed
i th prey. The elite matrix and predator matrix can be ratio, it enters the last stage of optimization, and the best
expressed as (13) and (14), respectively, where d is the strategy for the predator is Lévy.
number of populations and n is the dimension.  
⎡ l →
− −
→ −→ −−−→ −−−→
l l ⎤
· · · X 1,n s = R L ⊗ R L ⊗ Eli te − Pr eyi i = 1, . . . , d
X 1,1 X 1,2
⎢ X 2,1 n ⎥ −−−→ −−−→
··· Pr eyi = Eli te + P · C F ⊗ − →
l l
⎢ X 2,2 ⎥ s (18)
⎢ .. .. .. .. ⎥
⎢
Eli te = ⎢ . . . . ⎥ (13)
⎥ The introduction of eddy formation and FADs’ effect into
⎢ . . . . ⎥
⎣ . . .
. .
. .
. ⎦ the optimization process can prevent the iterative calculation
l l ··· l from falling into the local optimum solution. Long jumps can
X d,1 X d,2 X d,n
⎡ ⎤d×n help it jump out of the local optimization. The mathematical
X 1,1 X 1,2 · · · X 1,n expression of the FADs’ effect is
⎢ X 1,1 X 2,2 · · · X 2,n ⎥
⎢ ⎥ −−−→
⎢ .. .. .. .. ⎥ Pr eyi
Pr ey = ⎢
⎢ . . . . ⎥⎥ (14) ⎧−−−→   
⎢ . .. .. .. ⎥ ⎪
⎪ Pr eyi + C F X min + R ⊗ X max − X min ⊗ U
⎣ .. . . . ⎦ ⎪
⎪
⎨ if r ≤ F ADs
X d,1 X d,2 · · · X d,n d×n = −−−→ −−−−→ −−−−→
⎪
⎪ Pr eyi + [F ADs (1 − r ) + r ] Pr eyr1 − Pr eyr2
2) Optimization phase ⎪
⎪
⎩
In the initial stage, when the speed ratio is high or ifr ≥ F ADs
the speed of predator is greater than that of prey, the (19)

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 YU et al.: EFFICIENCY OPTIMIZATION OF WPT SYSTEM FOR EV BASED ON IMPROVED MPA 7851

where FADs is the influence probability, with a value of

0.2; U is a two-dimensional random vector of 0 or 1; r
is a random number of [0,1]; and r1 and r2 are random
indices of the prey matrix. After each iteration, the results
of the individuals before and after the iteration are compared,
and the individuals with high fitness are selected to update
the elite matrix until the end condition is met. Finally,
the optimization calculation of the objective function is
realized.

B. Improved Marine Predators Algorithm

The basic MPA has included many improved strategies Fig. 1. Tournament selection.
in the optimization algorithm, such as elitist strategy, levy
motion, Brownian motion and adaptive convergence factor,
et al. These strategies make the MPA have good optimization of MPA in the later stage:
ability. However, considering that MPA is only a simple −−−−→ −−−−→
random in the initialization stage, and in the later stage of Pr eyr1 , Pr eyr2
the algorithm, the location update is limited by the current = Tournament Selection
location, object range and other factors, there is a risk −−−→
Pr eyi
of falling into local optimization. Therefore, aiming at the ⎧−−−→   
initialization and later position update of MPA, this study ⎪
⎪ Pr ey i +C F X min + R ⊗ X max − X min ⊗U
⎪
⎪
proposes to improve MPA by combining opposition-based ⎨ if r ≤ F ADs
learning strategy and tournament selection to form OTMPA = −−−→ −−−−→ −−−−→
⎪
⎪ Pr eyi + [F ADs (1 − r ) + r ] Pr eyr1 − Pr eyr2
algorithm. ⎪
⎪
⎩
1) Opposition-based learning if r ≥ F ADs
In MPA, the optimization calculation starts with the (22)
solution randomly generated according to the population
number. If the solution of random initialization is close Combining the above opposition-based learning and tourna-
to the optimal solution, it is likely to get the correct ment selection mechanism, this study obtains OTMPA based
optimal solution in the end. If the solution of random on MPA. The pseudo-code of OTMPA is below.
initialization is far from the optimal solution, the time of
optimization calculation will be relatively long, or even
Algorithm 1 The Improved MPA Based on Opposition-
the correct solution cannot be obtained in the end.
Based Learning and Tournament Selection (OTMPA)
X 0 = X max + X min − X 0 (20) Input: Initialize search agents (prey) populations
i = 1 . . . , d with (21), dimension of variables n,
In (20), X 0 represents the opposite position of X 0 .
search ranges [u b , lb ], maximum iteration
The solutions of random initialization and reverse
Max I t er , iterations I ter ;
learning will jointly form a solution group. Through
Output: Fitness of each iteration, top predator fitness,
the calculation results of fitness, the best one is
top predator position;
generally selected as the initialization solution, that is,
1 while termination criteria are not met
the initialization solution obtained through opposition-
2 Calculate the fitness, construct the Elite matrix
based learning. The initialization of MPA can be
3 if I ter ≤ 1/3*Max I t er
expressed:
  4 Update prey based on (15)
finitialization = fitness X 0 , X 0 5 else if 1/3*Max I t er ≤ I ter ≤ 2/3*Max I t er
X 0 = upper half of sort ( f init ) (21) 6 for i = 1: d/2
7 Update prey based on (16)
2) Tournament selection 8 end
Tournament selection is a mechanism similar to compe- 9 for i = d/2: d
tition. In the later stage of MPA, a group of individuals 10 Update prey based on (17)
dr is randomly selected from the population d (father) to 11 end
calculate the fitness value of dr . After comparison, the 12 else if 2/3*Max I t er ≤ I ter ≤ Max I t er
optimal value will enter the next iterative calculation 13 Update prey based on (18)
(son). Tournament selection provides fair competition 14 end
for all individuals and protects the diversity of the 15 Applying FADs effect with tournament selection and
population. The tournament selection is shown in Fig. 1: update based on (22)
Applying the tournament selection to the selection of 16 end
Pr eyr1 and Pr eyr2 in (19) can realize the improvement

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7852 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

TABLE I
MPA C ALCULATION

TABLE II
C HAOTIC M AP

Fig. 2. Two-dimensional view of test function, (a) is F1, (b) is F2, (c) is
F3, (d) is F4, and (e) is F5.

deviation is closer to 0 in most test functions, indicating

that the OTMPA has higher robustness. The efficiency
C. Standard Benchmark Function of WPT is taken as the objective function of OTMPA.
Five standard benchmark functions are used as test functions The range of relevant optimization parameters (e.g., coil
to test the optimization performance of OTMPA. The test geometry parameters, compensation circuit parameters, etc.)
functions include unimodal, multimodal (High dimensional) is calculated to maximize the efficiency of WPT system. The
and multimodal (Fixed-dimensional). The test functions used relevant design parameters of the WPT system are obtained at
are shown in Table I. Compared with the basic MPA, the MPA this time.
initialized by chaotic map (the chaotic map used is Logistic
map, Sine map, Singer map, Iterative map and sinusoidal V. Q UANTIFICATION OF E FFICIENCY U NCERTAINTY
map, which are represented by CMPA1, CMPA2, CMPA3, AND S TRUCTURAL O PTIMIZATION D ESIGN
CMPA4 and CMPA5 in turn, chaotic map is shown in Table II), EXPERIMENTS OF EV WPT
improved whale optimization algorithm (IWOA), sparrow In this paper, the simulation model of WPT for EVs is
search algorithm (SSA) and African vultures optimization established. The EVs and WPT are shown in Fig. 4 (a). The
algorithm (AVOA). main material of the EVs model body is aluminum. The size
The two-dimensional view of the optimized calculation of of the body is 4.5 × 2 × 1.5 m (length × width × height),
different test functions by OTMPA is shown in Fig. 2. which is basically equivalent to the size of most domestic
In order to compare and verify the accuracy and stability cars, and has certain reference value. The model of the coil
of OTMPA, the test functions are calculated for 50 times and shielding layer is shown in Fig. 4 (b). The WPT has
continuously, and the mean and standard deviation of 50 times an output of 3.3kW. The size of the transmitting coil TX
calculated by different algorithms are compared, as shown in is the same as that of the receiving coil RX. The external
Fig. 3. size is 0.6 × 0.6m (A × A), the internal size is 0.3 × 0.3m
According to Fig. 2 and Fig. 3, for the single-mode (a × a), the cross-sectional area of the single-turn coil is
test function F1, the OTMPA has better results than o 2e−6 m2 , the number of turns of the coil is 11, and the material
the basic MPA and CMPAs, while for the multimodal test used is copper. The height to ground of most household cars
functions F2, F3, F4 and F5, the OTMPA avoids falling chassis is 0.15–0.2m. Considering the actual usage scenario,
into the local optimal solution, and compared with the newly the distance between TX and RX is set to 0.2m. The size of
proposed optimization algorithms, the calculated mean value shielding layer is the same as that of coil, the thickness is 1 cm,
of OTMPA is closer to the optimal result, and the standard and the material used is non-destructive soft iron. Using SS

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 YU et al.: EFFICIENCY OPTIMIZATION OF WPT SYSTEM FOR EV BASED ON IMPROVED MPA 7853

Fig. 4. (a) View of electric vehicle with WPT, (b) schematic of WPT with
shielding and some of its parameters.

Fig. 5. Schematic of compensation circuits considered for WPT.

TABLE III
U NCERTAIN PARAMETERS C ONSIDERED IN WPT
A PPLICATION E XAMPLES

Fig. 3. The mean and standard deviation of the results of 50 times of

calculating the test function by different algorithms (1 to 10 are OTMPA,
MPA, CMPA1, CMPA2, CMPA3, CMPA4, CMPA5, IWOA, SSA, AVOA),
(a) is F1, (b) is F2, (c) is F3, (d) is F4, and (e) is F5.

results of Sobol and MOAT sensitivity analysis, combined with

compensation circuit, the schematic of compensation circuit the OTMPA introduced in Section III, the WPT is optimized
is shown in Fig. 5. In the following section, the UQ of WPT to improve the efficiency. In the actual charging of the WPT
efficiency in this model and the optimal design of WPT are of EVs, the different chassis heights of different models, the
studied. coil offset caused by the driver’s parking, and the errors in
Next, considering the influence of shielding effect of EV the production and manufacturing of the WPT greatly impact
body on efficiency in practical application, the efficiency the WPT efficiency, making the battery charging time longer or
of EVs WPT was calculated several times combined with even unable to charge smoothly. Considering the above factors,
simulation software and used as the data of UQ and the following parameters are set as random input variables of
optimization design. The ASPCE introduced in Section II the ASPCE surrogate model for EVs WPT efficiency, as shown
is applied to build the above-mentioned efficiency surrogate in Table III. U represents uniform distribution, and N denotes
model of the EVs WPT for UQ analysis. According to the normal distribution.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7854 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

Fig. 6. Probability density distribution function of efficiency of WPT system Fig. 8. Global sensitivity index of input parameters for efficiency of WPT.
calculated by ASPCE, GPR, and MC.

Fig. 7. Comparison between the average efficiency of WPT calculated Fig. 9. MOAT mean and standard deviation of efficiency-related variables
by 15 MC simulations (black cross) and mean value (red dots) and 99% in WPT.
CI (blue bars) of the predicted of ASPCE.

the calculation cost of the surrogate model. Owing to the

According to Table III, Latin hypercube sampling method randomness of sampling, there are certain differences in the
is used for sampling, and the number of sample points is number of retained polynomials in each calculation. To verify
set to 200. The probability density distribution function of that the ASPCE can effectively predict the efficiency of
efficiency is calculated. To verify the effectiveness of ASPCE, the WPT, MC simulations are carried out 15 times through
the calculation results of GPR and MC are compared. Taking the simulation model, and each simulation is 2000 times. The
into account the cost of simulation calculation, the simulation mean value of efficiency from each MC simulation is
number of the MC method is 2000. The comparison results compared with the mean value and the 99% confidence interval
are shown in Fig. 6. (CI) of efficiency obtained from each ASPCE prediction. The
Fig. 6 shows that the calculation results of ASPCE are results are shown in Fig. 7.
basically consistent with those of GPR and MC, and the Fig. 7 shows that the WPT efficiency surrogate model can
truncation order P is 4. The leave-one-out cross-validation effectively predict the uncertainty of efficiency. To quantify
error is 6.91e−4. The most likely efficiency of WPT predicted the influence of different input variables on the efficiency of
by surrogate model is about 80%, the average efficiency WPT, the global sensitivity index of each random variable is
is 73.43%, and the variance is 1.013×e−2. The average calculated by combining with Sobol. The MOAT mean and
efficiency calculated by the MC method is 72.94%, and the standard deviation of each variable are calculated by MOAT.
variance is 9.436×e−3. According to (3), the number of Sobol total sensitivity index and first-order sensitivity index
polynomials is 1001, and the number of polynomials retained are shown in Fig. 8, The MOAT mean and standard deviation
after sparse calculation is 119. ASPCE significantly reduces of each variable are shown in Fig. 9.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 YU et al.: EFFICIENCY OPTIMIZATION OF WPT SYSTEM FOR EV BASED ON IMPROVED MPA 7855

Fig. 12. Comparison of different algorithms on efficiency optimization results

Fig. 10. Global sensitivity index of variables related to efficiency of WPT. of WPT system.

influence of the cross-sectional area of the transmitting coil

S0 and C2 and R2 in the receiving coil compensation circuit
on the efficiency is higher than that of other variables. And
according to the Fig. 11, the MOAT mean of the above three
input variables are also larger. Therefore, after the screening
of the Sobol and the MOAT mean, the above three variables
are mainly optimized in the following optimization design.
As Sobol global sensitivity index and MOAT method are
mainly used to describe the degree of influence of the
variables, determining whether an increase in the variables will
lead to a consequent increase or decrease in the efficiency
of the WPT is impossible. Therefore, in the following
optimization design, based on the mean values of s0 , C2 , and
R2 in Table III, the optimization calculation is carried out in
the range of [s0 /2, s0∗ 2], [C2 /2, C2∗ 2], and [R2 /2, R2∗ 2]. OTMPA
is used to optimize the efficiency of WPT. To verify the
superiority of OTMPA, its optimization results are compared
Fig. 11. MOAT mean and standard deviation of efficiency-related variables with the basic MPA, CMPAs, IWOA, SSA and AVOA. The
in WPT.
population numbers of the above three algorithms are set to 10,
and the maximum number of iterations is 30. The optimization
According to the calculation results of Figs. 8 and 9, results are shown in Fig. 12.
considering the geometric structure parameters of the WPT According to Fig. 12, All the algorithms finally approach
and the factors other than the compensation circuit, that is, the consistent optimal efficiency. Owing to opposition-based
the offset between the transmitting TX coil and the receiving learning and tournament selection, the solution obtained by
RX coil, the variables that have the greatest influence on the OTMPA in the initialization stage is very close to the
efficiency are the coil offset correlation variables x 0 and z 0 final optimization result, which helps OTMPA approach the
and the coil spacing d0 . Among them, the MOAT mean of x 0 final result (96.74%) faster than other algorithms. The result
and z 0 are significantly greater than that of other variables, strongly proves that OTMPA can optimize the WPT system of
which is in line with the performance of WPT efficiency in electric vehicles. The final optimization parameters obtained
practical use [33]. To guide the optimization of WPT so as by different algorithms are shown in Table IV, Conv represents
to improve efficiency, Sobol is carried out only considering the convergence times.
the relevant parameters of WPT, and the MOAT mean and Comparing the parameter optimization results and efficiency
standard deviation are calculated. The calculation results are of WPT in Table IV, due to the randomness in the calculation
shown in Figs. 10 and 11. process of the optimization algorithms, there are slight
Fig. 10 shows that the total sensitivity index of each variable differences in the calculation results of all algorithms, but
is the same as the first sensitivity index, which proves that these differences do not affect the calculation accuracy we
the influence of each variable in the above variables on the need. Therefore, we believe that the calculation results of
efficiency is independent of each other. Among them, the the algorithms used are basically consistent, moreover, the

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7856 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

TABLE IV
C OMPARISON OF E FFICIENCY O PTIMIZATION R ESULTS
OF D IFFERENT A LGORITHMS

Fig. 14. Comparison between the average efficiency of WPT calculated

by 15 MC simulations (black cross) and mean value (red dots) and 99 % CI
(blue bars) of the predicted of ASPCE.

simulation remains within the 99% CI of the average efficiency

predicted by the ASPCE surrogate model, which again proves
that ASPCE can effectively quantify the uncertainty of the
efficiency of the WPT of EVs. Through the above comparative
analysis, on the one hand, ASPCE adopted in this paper
can effectively establish the efficiency surrogate model of the
WPT of EVs and calculate its relevant statistical characteristic
parameters. It can also provide guidance for the optimization
design of the WPT according to the GSA results of Sobol and
MOAT. On the other hand, OTMPA can realize the optimal
Fig. 13. Comparison results before and after optimization of probability design of EVs WPT and ultimately significantly improve the
density distribution function for efficiency of WPT system. efficiency of WPT of EVs.

VI. C ONCLUSION
OTMPA only needs 6 calculations to achieve the optimal
result, which converges faster than other algorithms. It is In order to improve the efficiency of EV-WPT system,
proved that the OTMPA can improve the calculation speed this paper uses the UQ method to guide the optimal design
on the premise of ensuring the calculation accuracy. With of EV-WPT system, this paper introduces the application of
the increase of s0 and R2 , the input variables with positive ASPCE combined LAR with hyperbolic truncation scheme in
influence, the efficiency of WPT also increases. The opposite is the UQ of efficiency of EVs WPT. Its calculation accuracy is
true for C2 . To verify the improvement effect of the efficiency basically consistent with that of MC, and it obviously saves
of the optimized WPT, based on the parameter settings in calculation cost. Combining Sobol and MOAT to calculate
Table III and the optimized parameters, ASPCE is used to the influence degree of each variable on the efficiency, the
establish the surrogate model to calculate the probability variables that have great influence on the efficiency of WPT
density distribution function of the efficiency of the WPT. are screened out, and they are used to guide the optimization
The results before optimization are compared and shown design of EVs WPT. Based on these variables, this study uses
in Fig. 13. Again, 15 times MC simulations are performed opposition-based learning and tournament selection to improve
through the simulation model, and the 99% CI of the mean MPA to obtain OTMPA and use it to optimize the WPT
value of efficiency obtained from each ASPCE prediction are with the optimization objective of maximizing efficiency, and
compared. The results are shown in Fig. 14. the optimized WPT effectively improves efficiency. From the
The average efficiency of the optimized WPT system comprehensive results, the scheme proposed in this paper can
is 94.64%, which is improved compared with the average provide a feasible solution for the UQ of the efficiency of EVs
efficiency before optimization. The variance of efficiency WPT and the optimization of the WPT structure. There are
after optimization is 6.927×e−4, which is significantly less still some limitations in this paper. When considering practical
than that before optimization. According to the results in engineering applications, the number of variables involved in
Figs. 8 and 9, the variables that have the greatest impact on the EVs WPT transmission efficiency may be more, which may
efficiency are usually coil offset-related variables. Therefore, bring computational cost problems for future UQ. Therefore,
the anti-offset capability of the optimized WPT system is also in the future work, we will study the methods suitable for
improved. In Fig. 14, the average efficiency obtained by MC higher dimensions, so as to further improve the efficiency of

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 YU et al.: EFFICIENCY OPTIMIZATION OF WPT SYSTEM FOR EV BASED ON IMPROVED MPA 7857

EVs WPT system. At the same time, we should consider the [21] D. Kurschner, C. Rathge, and U. Jumar, “Design methodology for high
safety of electromagnetic exposure of high-power EVs WPT efficient inductive power transfer systems with high coil positioning
flexibility,” IEEE Trans. Ind. Electron., vol. 60, no. 1, pp. 372–381,
to human body in combination with the actual application Jan. 2013.
situation. [22] W. Zhang, S.-C. Wong, C. K. Tse, and Q. Chen, “Design for efficiency
optimization and voltage controllability of series-series compensated
R EFERENCES inductive power transfer systems,” IEEE Trans. Power Electron., vol. 29,
no. 1, pp. 191–200, Jan. 2014.
[1] H.-Y. Mak, Y. Rong, and Z.-J. M. Shen, “Infrastructure planning for [23] W. Zhang, S.-C. Wong, C. K. Tse, and Q. Chen, “Analysis and
electric vehicles with battery swapping,” Manage. Sci., vol. 59, no. 7, comparison of secondary series- and parallel-compensated inductive
pp. 1557–1575, 2013. power transfer systems operating for optimal efficiency and load-
[2] O. N. Nezamuddin, C. L. Nicholas, and E. C. D. Santos, “The problem independent voltage-transfer ratio,” IEEE Trans. Power Electron.,
of electric vehicle charging: State-of-the-art and an innovative solution,” vol. 29, no. 6, pp. 2979–2990, Jun. 2014.
IEEE Trans. Intell. Transp. Syst., vol. 23, no. 5, pp. 4663–4673, [24] W. Feng, X. Chu, Q. Li, and W. Gu, “Compensation parameters opti-
May 2022, doi: 10.1109/TITS.2020.3048728. mization of wireless power transfer for electric vehicles,” Electronics,
[3] V.-D. Doan, H. Fujimoto, T. Koseki, T. Yasuda, H. Kishi, and T. Fujita, vol. 9, no. 5, pp. 1–12, 2020.
“Allocation of wireless power transfer system from viewpoint of optimal [25] E. Andersen, C. Casados, B. D. Truong, and S. Roundy, “Optimal
control problem for autonomous driving electric vehicles,” IEEE Trans. transmit coil design for wirelessly powered biomedical implants
Intell. Transp. Syst., vol. 19, no. 10, pp. 3255–3270, Oct. 2018. considering magnetic field safety constraints,” IEEE Trans. Electromagn.
[4] Y. Jin, J. Xu, S. Wu, L. Xu, and D. Yang, “Enabling the wireless charging Compat., vol. 63, no. 5, pp. 1735–1747, Oct. 2021.
via bus network: Route scheduling for electric vehicles,” IEEE Trans. [26] X. Zhang, Y. Zhang, Z. Zhang, and M. Li, “Mode conversion
Intell. Transp. Syst., vol. 22, no. 3, pp. 1827–1839, Mar. 2021. and structure optimization of quadrature coils for electric vehicles
[5] A. Triviño-Cabrera, J. M. González-González, and J. A. J. A. Aguado, wireless power transfer,” IEEE Trans. Energy Convers., vol. 35, no. 2,
Wireless Power Transfer for Electric Vehicles: Foundations and Design pp. 575–590, Jun. 2020.
Approach. Cham, Switzerland: Springer, 2020, p. 175. [27] O. A. Arqub and Z. Abo-Hammour, “Numerical solution of systems
[6] A. Triviño, J. M. González-González, and J. A. Aguado, “Wireless of second-order boundary value problems using continuous genetic
power transfer technologies applied to electric vehicles: A review,” algorithm,” Inf. Sci., vol. 279, pp. 396–415, Sep. 2014.
Energies, vol. 14, no. 6, p. 1547, Mar. 2021. [28] B. Maayah, O. A. Arqub, S. Alnabulsi, and H. Alsulami, “Numerical
[7] V. Etacheri, R. Marom, R. Elazari, G. Salitra, and D. Aurbach, solutions and geometric attractors of a fractional model of the cancer-
“Challenges in the development of advanced Li-ion batteries: A review,” immune based on the Atangana-Baleanu-Caputo derivative and the
Energy Environ. Sci., vol. 4, no. 9, pp. 3243–3262, Aug. 2011. reproducing kernel scheme,” Chin. J. Phys., vol. 80, pp. 463–483,
[8] F. Musavi and W. Eberle, “Overview of wireless power transfer Dec. 2022.
technologies for electric vehicle battery charging,” IET Power Electron., [29] E. Torre, S. Marelli, P. Embrechts, and B. Sudret, “Data-driven
vol. 7, no. 1, pp. 60–66, Jan. 2014. polynomial chaos expansion for machine learning regression,” J.
[9] S. Y. Chu, X. Cui, X. Zan, and A.-T. Avestruz, “Transfer-power Comput. Phys., vol. 388, pp. 601–623, Jul. 2019.
measurement using a non-contact method for fair and accurate metering [30] B. Efron, T. Hastie, I. Johnstone, and R. Tibshirani, “Least angle
of wireless power transfer in electric vehicles,” IEEE Trans. Power regression,” Ann. Statist., vol. 32, no. 2, pp. 407–451, 2004.
Electron., vol. 37, no. 2, pp. 1244–1271, Feb. 2022. [31] B. Sudret, “Global sensitivity analysis using polynomial chaos
[10] R. Trinchero, I. S. Stievano, and F. G. Canavero, “Black-box modeling of expansions,” Rel. Eng. Syst. Safety, vol. 93, no. 7, pp. 964–979,
the maximum currents induced in harnesses during automotive radiated Jul. 2008.
immunity tests,” IEEE Trans. Electromagn. Compat., vol. 62, no. 2, [32] A. Faramarzi, M. Heidarinejad, S. Mirjalili, and A. H. Gandomi,
pp. 627–630, Apr. 2020. “Marine predators algorithm: A nature-inspired Metaheuristic,” Expert
[11] T. Houret, P. Besnier, S. Vauchamp, and P. Pouliguen, “Controlled Syst. Appl., vol. 152, Aug. 2020, Art. no. 113377.
stratification based on Kriging surrogate model: An algorithm for [33] Q. Zhu, M. Su, Y. Sun, W. Tang, and A. P. Hu, “Field orientation
determining extreme quantiles in electromagnetic compatibility risk based on current amplitude and phase angle control for wireless power
analysis,” IEEE Access, vol. 8, pp. 3837–3847, 2020. transfer,” IEEE Trans. Ind. Electron., vol. 65, no. 6, pp. 4758–4770,
[12] J. Caldeira and B. Nord, “Deeply uncertain: Comparing methods of Jun. 2018.
uncertainty quantification in deep learning algorithms,” Mach. Learn.,
Sci. Technol., vol. 2, no. 41, pp. 1–11, 2021.
[13] M. Larbi, R. Trinchero, F. G. Canavero, P. Besnier, and M. Swaminathan,
“Analysis of parameter variability in an integrated wireless power
transfer system via partial least-squares regression,” IEEE Trans.
Compon., Package., Manuf. Technol., vol. 10, no. 11, pp. 1795–1802,
Nov. 2020.
[14] D. Xiu and G. E. Karniadakis, “The Wiener–Askey polynomial chaos for
stochastic differential equations,” SIAM J. Sci. Comput., vol. 24, no. 2,
pp. 619–644, Jan. 2002.
[15] I. S. Stievano, P. Manfredi, and F. G. Canavero, “Stochastic analysis
of multiconductor cables and interconnects,” IEEE Trans. Electromagn.
Compat., vol. 53, no. 2, pp. 501–507, May 2011.
[16] A. Kaintura, T. Dhaene, and D. Spina, “Review of polynomial chaos-
based methods for uncertainty quantification in modern integrated
circuits,” Electronics, vol. 7, no. 3, p. 30, Feb. 2018.
[17] P. Lagouanelle, V.-L. Krauth, and L. Pichon, “Uncertainty quantification
in the assessment of human exposure near wireless power transfer
systems in automotive applications,” in Proc. AEIT Int. Conf. Electr.
Electron. Technol. Automot. (AEIT AUTOMOTIVE), Jul. 2019, pp. 1–5. Quanyi Yu received the B.S. degree from the
[18] S. Bilicz, S. Gyimothy, J. Pavo, P. Horvath, and K. Marak, “Uncertainty College of Communication Engineering, Jilin Uni-
quantification of wireless power transfer systems,” in Proc. IEEE versity, Changchun, Jilin, China, in 2016, and the
Wireless Power Transf. Conf. (WPTC), May 2016, pp. 1–3. M.S. degree from the College of Instrumentation
[19] Z. Zhang, T.-W. Weng, and L. Daniel, “Big-data tensor recovery for and Electrical Engineering, Jilin University, in 2020,
high-dimensional uncertainty quantification of process variations,” IEEE where he is currently pursuing the Ph.D. degree
Trans. Compon., Package., Manuf. Technol., vol. 7, no. 5, pp. 687–697, in measurement technology and instruments. His
May 2017. research interests include the uncertainty quantifi-
[20] G. Blatman and B. Sudret, “Adaptive sparse polynomial chaos expansion cation and electromagnetic compatibility of wireless
based on least angle regression,” J. Comput. Phys., vol. 230, no. 6, power transfer of EVs.
pp. 2345–2367, Mar. 2011.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.
 7858 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 24, NO. 7, JULY 2023

Jun Lin received the B.Sc. and M.Sc. degrees in Linlin Xu received the B.S. degree in electrical
applied geophysics from the Changchun College engineering and automation from the College of
of Geology Applied Geophysics Professional in Instrumentation and Electrical Engineering, Jilin
1982 and 1987, respectively. University, Changchun, Jilin, China, in 2020, where
He is currently a Professor with the College she is currently pursuing the M.S. degree in
of Instrumentation and Electrical Engineering, electrical engineering.
Jilin University, Changhchun, China. He is also Her research interests include electromagnetic
an Academician with the Chinese Academy of safety and electromagnetic compatibility of wireless
Engineering, Beijing, China. He is also the Director charging systems in electric vehicles.
of the National Geophysical Exploration Equipment
Engineering Research Center, Jilin University. His
research interests include the geophysical exploration theory, technologies,
and equipment.

Xilai Ma received the B.S. degree in electronic and

information engineering from Changchun University
of Technology, in 2003, and the Ph.D. degree
in instrument science and technology from Jilin
University Changchun, Jilin, China, in 2008.
From 2008 to 2022, he was a Chief Engineer
in vehicle electronic and electrical testing at
First Automobile Workshop, China. His research
interests include computational electromagnetics,
signal integrity, RF signal testing and analysis, elec-
tromagnetic compatibility simulation, and antenna
analysis.

Tianhao Wang (Member, IEEE) received the

B.S. degree in electrical engineering and the
Bo Li received the B.S. degree in electrical Ph.D. degree in vehicle engineering from Jilin
engineering and automation from the College of University, Changchun, Jilin, China, in 2010 and
Instrumentation and Electrical Engineering, Jilin 2016, respectively.
University, Changchun, Jilin, China, in 2020, From 2016 to 2019, he was a Post-Doctoral
where he is currently pursuing the M.S. degree Researcher with the Department of Science and
in electrical engineering. His research interests Technology of Instrument, Jilin University, where
include uncertainty analysis, wireless power transfer, he is currently an Associate Professor with the
magnetic resonance, and human protection. College of Instrumentation and Electrical Engineer-
ing. His research interests include the uncertainty
quantification of wireless power transfer of EVs and human electromagnetic
exposure safety.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on April 22,2024 at 11:02:24 UTC from IEEE Xplore. Restrictions apply.