energies

Article
Electric Vehicle Charging Load Prediction Based on Weight
Fusion Spatial–Temporal Graph Convolutional Network
Jun Zhang 1 , Huiluan Cong 1 , Hui Zhou 1 , Zhiqiang Wang 1 , Ziyi Wen 2, * and Xian Zhang 2

1 State Grid Shandong Electric Vehicle Service Company Ltd., Jinan 250117, China;
[email protected] (J.Z.); [email protected] (H.C.); [email protected] (H.Z.);
[email protected] (Z.W.)
2 School of Mechanical Engineering and Automation, Harbin Institute of Technology, Shenzhen 518055, China;
[email protected]
* Correspondence: [email protected]; Tel.: +86-199-7275-4570

Abstract: The rapid increase in electric vehicles (EVs) poses significant impacts on multi-energy
system (MES) operation and energy management. Accurately assessing EV charging demand becomes
crucial for maintaining MES stability, making it an urgent issue to be studied. Therefore, this paper
proposes a novel deep learning-based EV charging load prediction framework to assess the impact
of EVs on the MES. First, to model the EV traffic flow, a modified weight fusion spatial–temporal
graph convolutional network (WSTGCN) is proposed to capture the inherent spatial–temporal
characteristics of traffic flow. Specifically, to enhance the WSTGCN performance, the modified
residual modules and weight fusion mechanism are integrated into the WSTGCN. Then, based on
the predicted traffic flow, an improved queuing theory model is introduced to predict the charging
load. In this improved queuing theory model, special consideration is given to subjective EV user
behaviors, such as refusing to join queues and leaving impatiently, making the queuing model
more realistic. Additionally, it should be noted that the proposed charging load predicting method
relies on traffic flow data rather than historical charging data, which successfully addresses the data
insufficiency problem of newly established charging stations, thereby offering significant practical
value. Experimental results demonstrate that the proposed WSTGCN model exhibits superior
accuracy in predicting traffic flow compared to other benchmark models, and the improved queuing
Citation: Zhang, J.; Cong, H.; Zhou,
H.; Wang, Z.; Wen, Z.; Zhang, X.
theory model further enhances the accuracy of the results.
Electric Vehicle Charging Load
Prediction Based on Weight Fusion Keywords: charging load prediction; deep learning; spatial–temporal network; weight fusion
Spatial–Temporal Graph
Convolutional Network. Energies 2024,
17, 4798. https://doi.org/10.3390/
en17194798 1. Introduction
Academic Editor: Tek Tjing Lie As a promising solution to low-carbon transition, electric vehicle (EV) use is increasing.
Their rapid penetration introduces unignorable uncertainties to the electrical loads, posing
Received: 30 August 2024 challenges to the operation of multi-energy systems (MESs). Addressing these challenges
Revised: 22 September 2024
is crucial for optimizing MES operation and load management.
Accepted: 24 September 2024
The charging load demand can be predicted to effectively quantify the impact of EVs
Published: 25 September 2024
on MESs, providing valuable information for optimizing operations. Most previous studies
have used time series analysis methods to predict EV charging loads [1–3]. However,
time series analysis methods struggle to handle long-term tasks. Recently, researchers
Copyright: © 2024 by the authors.
have begun to apply deep learning technologies to overcome the limitations of time series
Licensee MDPI, Basel, Switzerland. analysis methods. Compared to traditional statistical methods, deep learning methods
This article is an open access article have stronger nonlinear learning capabilities, robustness, and generalization ability, mak-
distributed under the terms and ing them particularly suitable for large-scale data processing. For instance, in [4], a long
conditions of the Creative Commons short-term memory (LSTM) model is used to predict EV charging demand at a charging
Attribution (CC BY) license (https:// station for the next few hours. In [5], both Sequence to Sequence (Seq2seq) and LSTM are
creativecommons.org/licenses/by/ used to predict the EV charging demand, and the study finds that Seq2Seq significantly
4.0/). outperforms other models in multistep prediction. Ref. [6] demonstrates that the LSTM

Energies 2024, 17, 4798. https://doi.org/10.3390/en17194798 https://www.mdpi.com/journal/energies

Energies 2024, 17, 4798 2 of 17

model outperforms the traditional artificial neural network (ANN) in charging load predic-
tion. Ref. [7] compares six different deep neural network models including LSTM, gated
recurrent units (GRUs), and their variants and finds that the multivariate LSTM and multi-
variate GRU are more effective for predicting the EV charging load of fast charging stations
(FCSs). Ref. [8] combines ANN and recurrent neural network (RNN), and proposes a novel
Q-Learning method for EV load predicting. Ref. [9] proposes an attention-based recurrent
convolutional neural network model (LA-RCNN) to predict EV load with consideration of
temporal dependencies in multivariate time series. Ref. [10] combines the niche immunity
lion algorithm (NILA) and convolutional neural network (CNN) for short-term load pre-
diction of FCSs. Additionally, aware that the charging load of FCSs has both temporal and
spatial correlations, some research further studies the spatial–temporal graph convolutional
network (STGCN) and its variants. Ref. [11] proposes a heterogeneous STGCN to model
the complex spatial–temporal correlations. Based on STGCN, a multistep prediction model
is proposed for the load management of charging stations in [12]. Ref. [13] proposes a SED-
former to address the long-term EV load prediction problem. The SEDformer comprises an
attention-based temporal encoder and a channel attention-based spatial encoder, which
can effectively capture the spatial–temporal characteristics of charging load data.
The above-mentioned charging load predicting methods rely on historical data ob-
tained from FCSs. However, for newly established stations, it is difficult to obtain sufficient
and effective historical data. To address the challenges, recent studies explore using traffic
flow information to predict the EV charging load instead of using historical data of the
FCSs directly. For traffic flow modeling, many researchers utilize deep learning models
to capture the complex spatial and temporal characteristics inherent in historical traffic
data. For example, in [14], a spatial–temporal fusion graph neural network (STFGNN) is
developed for traffic flow forecasting, conducting fusion operations of various spatial and
temporal graphs. Ref. [15] proposes a dynamic graph convolutional recurrent network
(DGCRN) to model the dynamic characteristics of the correlations among FCSs in the traffic
network. Also, ASTNet-T and multi-graph STGCN are introduced to predict charging
loads at FCSs directly based on the historical data in [16,17].
Moreover, based on the traffic flow prediction technique, some studies further analyze
the charging process in FCS to predict the charging load. For example, in [18], the CNN
is used to model traffic flow first and then the queuing theory is applied to predict the
EV charging load. Ref. [19] uses wavelet transform and LSTM to predict traffic flow in
road networks first, and then uses queuing theory model to convert the predicted traffic
flow into EV charging demand. In [20], the traffic flow data from real-time closed-circuit
television (CCTV) are used first, and then a Markov chain is applied to determine the
charging load. In [21], a speed-flow model is used to simulate the traffic flow of EVs, and
the Monte Carlo method is used to simulate EV travel and charging behavior, and finally
obtain the charging load. In [22], the traffic flow distribution is modeled by the vehicle
mobility model and Dijkstra algorithm first, and then a charging load determination model
is proposed, considering multiple charging behaviors. However, most previous studies did
not model the spatial–temporal connections inherent in traffic flow when predicting the
traffic flow. Moreover, the impacts of subjective EV user behaviors are always ignored in
charging load modeling, which leads to predictions that are not accurate enough and lack
realism. Therefore, there remains a research gap in developing a high-accuracy charging
load prediction method.
From what has been mentioned above, the main challenges that need to be addressed
include the following: First, it is difficult to obtain sufficient and effective historical data
of newly established FCSs, which makes the traditional historical data-based charging
load predicting methods ineffective. Second, in traffic flow modeling, the spatial–temporal
connection inherent in traffic data needs to be captured to enhance the predicting accuracy.
Third, in charging load prediction, subjective EV user behaviors would significantly influ-
ence modeling accuracy, and they need to be explored to facilitate accurate load prediction.
Therefore, this paper develops a deep learning-based charging load prediction model that
Energies 2024, 17, 4798 3 of 17

models both spatial–temporal connections and EV user behavior. The main contributions
can be concluded as follows:
(1) A traffic flow-based charging load prediction framework is proposed considering
the data insufficiency problem frequently happening in newly established FCSs. Different
from traditional historical data-based methods, the traffic flow-based method does not
rely on the data obtained from the FCSs. Instead, it uses the traffic flow information in
the traffic network and uses queuing theory to quantify the load demand. Therefore, this
method successfully addresses the data insufficiency problem and produces an effective
load prediction method for the newly established FCSs.
(2) A novel weight fusion-based STGCN (WSTGCN) model is proposed to predict the
traffic flow. The spatial–temporal attention mechanism is employed in the proposed WST-
GCN model to capture the spatial–temporal correlations of the traffic network. Compared
with the traditional STGCN, the improved WSTGCN model modifies the residual modules
and integrates the weight fusion mechanism to improve the predicting accuracy. By fusing
the recent, daily, and weekly characteristics of traffic flow, the traffic flow prediction result
is more accurate.
(3) An improved queuing theory is used to model the queuing process of EVs at
FCSs, with special consideration given to the subjective behaviors of EV users. Compared
with traditional queuing theory, the improved queuing theory models EV user behaviors,
including impatience to leave and refusal to join queues, making the queuing model
more comprehensive.
The remainder of this paper is organized as follows. Section 2 introduces the WSTGCN
model for traffic flow prediction. Section 3 illustrates the improved queueing model-based
EV charging demand modeling. Section 4 shows the numerical simulations to demonstrate
the effectiveness of our approach. Section 5 concludes this paper.

2. WSTGCN-Based Traffic Flow Prediction

2.1. WSTGCN Framework
For traffic flow prediction, a novel WSTGCN is developed as an improved variant of
ASTGCN [23]. Compared with ASTGCN, the WSTGCN modifies the residual modules
and integrates the weight fusion mechanism to improve the prediction accuracy. The
framework of the WSTGCN is shown in Figure 1. In the WSTGCN, three different periodic
components, including recent, daily periodic and weekly periodic components, are used to
extract different periodic features of traffic flow data. In each periodic component, multiple
spatial–temporal blocks (ST blocks) are stacked to capture the spatial–temporal features
Energies 2024, 17, x FOR PEER REVIEW 4 of 17
of traffic data. Based on the outputs of these periodic components, the final output of the
WSTGCN model can be obtained by the weight fusion mechanism.

Figure1.1.WSTGCN
Figure WSTGCNmodel.
model.

The final output of the WSTGCN model can be expressed as follows:

ˆ = f ( STr (  r | ψ r ), STd (  d | ψ d ), STw (  w | ψ w ) ) (1)

where ̂ is the predictive output of the WSTGCN, and r, d, and w represent the in-
Energies 2024, 17, 4798 4 of 17

The final output of the WSTGCN model can be expressed as follows:

Ŷ = f (STr (Xr |ψr ), STd (Xd |ψd ), STw (Xw |ψw )) (1)

where Ŷ is the predictive output of the WSTGCN, and Xr , Xd , and Xw represent the inputs
of the recent, daily periodic and weekly periodic components, respectively, as described in
Section 2.2; ψr , ψd , and ψw are the parameters of components; STr (·), STd (·), and STw (·)
denote the ST blocks, as described in Section 2.3; f (·) represents weight fusing of the outputs
of the three components, as described in Section 2.4.

2.2. Periodic Components of WSTGCN

To illustrate the periodic components, the traffic flow data are introduced first. The
traffic network is defined as an undirected graph with N nodes, and each node is equipped
with data sampling facilities. In each traffic node, C traffic features are sampled with
sampling frequency f times per day, that is, each node generates a feature vector of length
C at each sampling time. And the whole traffic network generates N × C × f data per day.
We use xtc,n ∈R to denote the value of the c-th feature of node n at sampling time t, xtc ∈RN
to denote the c-th signal all over the network at time t, xnt ∈RC to denote the values of all
features of node n at sampling time t, and Xt ∈RN ×C to denote the values of all features
of all nodes at sampling time t. Assume that the current time is tp and the size of the
predicting window is Tp . Tr , Td and Tw are the lengths of the inputs of the recent, daily
periodic, and weekly periodic components, and they are all integer multiples of Tp . The
recent, daily periodic, and weekly periodic components are introduced as follows:
(1) The recent component
Intuitively, the past traffic flows significantly influence the near-future traffic flows.
To capture the dependence between near-future data and recent past data, the recent
component is utilized, with its input defined as follows:
n o
Xr = Xt p −Tr +1 , Xt p −Tr +2 , . . . , Xt p −Tr ∈ R N ×C×Tr (2)

where Xr denotes the input of the recent component, including historical traffic flow data
from time tp − Tr to tp .
(2) The daily periodic component
Due to the regular daily routine of people, traffic data may show repeated patterns,
such as daily morning peaks. To model the daily periodic patterns, the daily periodic
component is used. This component captures the dependence between traffic data from the
same time over the past few days and the predicting time, with its input defined as follows:
( )
Xd = X T ,X T ,...,X T ,X Td − Tp ,...,X Td − Tp , Xt p − f +1 , . . . , , Xt p − f +Tp ∈ R N × C × Td (3)
t p − Tdp f +1 t p − Tdp f +2 t p − Tdp f + Tp tp − Tp f +1 tp − Tp f + Tp

where Xd denotes the input of the daily periodic component, including the historical data
from the past few days and the same period as the predicting period.
(3) The weekly periodic component
Traffic patterns on Mondays usually resemble those of previous Mondays, but always
differ significantly from weekends. To capture these weekly periodic features, the weekly
periodic component is designed, with its input defined as follows:

( )
Xw = Xt ,X
− Tw f ·7+1 t − Tw f ·7+2
, . . . , Xt Tw
p − Tp f ·7+ Tp
,X Tw − Tp ,...,X Tw − Tp , Xt p − f ·7+1 , . . . , , Xt p − f ·7+Tp ∈ R N ×C×Tw (4)
p Tp p Tp tp − Tp f ·7+1 tp − Tp f ·7+ Tp

where Xw is the input of the weekly periodic component, including historical data from
the same weekday and period over the past few weeks as the predicting period.
  
 w =  X Tw ,X Tw ,...,X Tw ,X Tw −Tp ,...,X Tw −Tp ,Xtp - f ⋅7 + 1 ,..., ,Xtp - f ⋅7 +Tp  ∈ RN ×C ×Tw (4)
tp - f ⋅7 +1 tp - f ⋅7 + 2 tp - f ⋅7 + Tp f ⋅7 + 1 f ⋅7 + Tp
 
tp - tp -
Tp Tp Tp Tp Tp

where w is the input of the weekly periodic component, including historical data from
Energies 2024, 17, 4798 5 of 17
the same weekday and period over the past few weeks as the predicting period.

2.3. Spatial–Temporal Block of WSTGCN

2.3. Spatial–Temporal Block of WSTGCN
Traffic flow data possess both temporal and spatial attributes. Spatially, traffic con-
Traffic
ditions flow data
at different possessinfluence
locations both temporal and spatial
each other attributes.
dynamically. Spatially,correlations
Temporally, traffic con-
ditions at different locations influence each other dynamically. Temporally,
exist between traffic conditions across different time slices. To capture the temporal and correlations
exist between
spatial traffic conditions
characteristics across spatial–temporal
simultaneously, different time slices. To capture
blocks the are
(ST blocks) temporal and
employed
spatial characteristics simultaneously, spatial–temporal blocks (ST blocks)
in the WSTGCN model. In the WSTGCN, each periodic component has stacks of multiple are employed in
the WSTGCN model. In the WSTGCN, each periodic component has
ST blocks. And each ST block is composed of a spatial–temporal attention mechanismstacks of multiple
ST blocks.
(SAttn And and
+ TAttn) eachaST block is composed
spatial–temporal of a spatial–temporal
convolutional network (CNN attention
+ GCN). mechanism
Figure 2
(SAttn + TAttn) and a spatial–temporal convolutional network (CNN
shows an ST block as an example. Assume that this block is the l-th block in the + GCN). Figure
resent2
shows an ST block as an example. Assume that this block is the l-th block in the resent
component, and the input is r = {X1, X2,…, XT} ∈ N×C ×T . l−1 l−1
l−1 l−1 l−1

component, and the input is Xrl −1 = {X1 , X2 ,. . ., XT } ∈ RN×C ×T .

Figure 2. ST block.
Figure 2. ST block.
(1) Spatial–Temporal Attention Mechanism
(1) Spatial–Temporal Attention Mechanism
The attention mechanism can dynamically adjust weights to prioritize important
The attention mechanism can dynamically adjust weights to prioritize important fea-
features, enhancing focus on important time steps or spatial relationships. To adjust the
tures, enhancing focus on important time steps or spatial relationships. To adjust the
weights of spatial nodes, the spatial attention matrix is used, which can be calculated by
the following:
T
S = VS · σ ((Xrl −1 w1 )w2 (w3 Xrl −1 ) + bS ) (5)

′ exp(Si,j )
Si,j = N
(6)
∑ j=0 exp(Si,j )
where S is the spatial attention matrix and S′ is its normalized counterpart to ensure the
attention weights of a node sum to 1; Si,j represents the correlation between nodes i and j;
w1 , w2 , w3 , VS , and bS are parameters automatically learned by the neural networks; and
sigmoid σ is the activation function.
Similarly, the temporal attention matrix adjusts weights to emphasize more significant
time steps, which can be calculated by the following:
T
T = VT · σ ((Xrl −1 ) u1 )u2 (u3 Xrl −1 ) + bT ) (7)

′ exp(Ti,j )
Ti,j = l −1
(8)
∑ Tj=0 exp(Ti,j )

where T is the temporal attention matrix and T′ is the normalized counterpart; Ti,j rep-
resents the dependent strength between time i and j; and u1 , u2 , u3 , VT , and bT are
Energies 2024, 17, 4798 6 of 17

learnable parameters. Based on the temporal attention matrix, the input can be updated as
l−1 l−1
Xr′l −1 = {X′ 1 , X′ 2 ,. . ., X′ T } = T′ ·{ X1 , X2 ,. . ., XT } ∈ RN×C ×T .
(2) Spatial–Temporal Convolutional Block
Through the spatial–temporal attention mechanism, the network can automatically pay
more attention to valuable spatial and temporal information. Then, the spatial–temporal
convolution module is designed to extract the spatial–temporal features of traffic data. The
spatial–temporal convolution module consists of a graph convolution and a convolution,
as shown in Figure 2.
The graph convolution is used to extract the spatial features of the traffic data. Taking
the traffic flow at time t as an example, the c-th feature data point from the traffic network
is x = xct ∈ RN . According to [23], the graph convolution operation for x can be expressed
as follows:
K −1
gθ ∗G x = ∑ θk Tk (L) ⊙ S′ x


(9)
k =0

where *G denotes the graph convolution operation; gθ is the convolution kernel; θ ∈ RK

is the Chebyshev coefficient vector; L = 2L/λmax − In , in which L is the Laplacian matrix
of the network graph and λmax is the maximum eigenvalue of the Laplacian matrix; Tk (·)
denotes the Chebyshev polynomial, recursively defined as Tk (x) = 2xTk− 1 (x) − Tk− 2 (x),
with T0 (x) = 0, T1 (x) = x. It should be noted that the spatial attention matrix S′ is introduced
in Equation (9) to dynamically adjust the weights of the node. Moreover, Equation (9) only
illustrates the graph convolution operation for single-channel data x = xtc ∈RN , which can
be extended to multiple-channel data. For example, for the recent component, the graph
l−1 l−1
convolution for l-th layer input Xr′l −1 = {X′ 1 , X′ 2 ,. . ., X′ T }∈ RN×C ×T can be expressed
as gθ *G Xr′l −1 .
After the graph convolution operation, the standard 2D convolution operation is used
to capture the temporal features of the traffic data. Also, take the operation on the l-th layer
in the recent component as an example. The temporal convolution operation is shown
as follows:
Xrl = ReLU (ψ ∗ (ReLU ( gθ ∗G X′rl −1 ))) (10)
where * denotes a standard convolution operation; ψ are parameters of the temporal
dimension convolution kernel; ReLU is the activation function.
(3) Modified Residual Module
To connect ST blocks, a modified residual module is introduced into the WSTGCN.
Compared to the residual module in the ASTGCN, the modified residual module integrates
the convolution operation into the residual module, thus making the model training more
efficient and stable. The residual module with convolution operations can be expressed
as follows:
h(Xl −1 ) = Xl −1 + conv(Xl −1 ) (11)
Xl = ReLU (h(Xl −1 ) + ST (Xl −1 , ψl )) (12)
where Xl −1 is the input to the l-th ST block; h(·) is the residual module with a convolution
operation; ST(·) is the operation of the l-th ST block with parameters ψl ; Xl is the output of
the residual module as well as the input of the l + 1-th ST block.

2.4. Weight Fusion Mechanism in WSTGCN

Based on the outputs of periodic components, the weight fusion mechanism is in-
novatively used to obtain the final output of the WSTGCN. Compared with ASTGCN,
which fuses outputs using a fully connected layer, WSTGCN identifies the weight of each
component before the fusing operation, resulting in a more reasonable and accurate final
output. The weight fusion mechanism is shown in Figure 3.
 2.4. Weight Fusion Mechanism in WSTGCN
Based on the outputs of periodic components, the weight fusion mechanism is inno-
vatively used to obtain the final output of the WSTGCN. Compared with ASTGCN, which
fuses outputs using a fully connected layer, WSTGCN identifies the weight of each com-
Energies 2024, 17, 4798 ponent before the fusing operation, resulting in a more reasonable and accurate final7 of 17
out-
put. The weight fusion mechanism is shown in Figure 3.

Figure3.
Figure 3. Weight
Weight fusion
fusionmechanism.
mechanism.

The weights
The weights of
of recent,
recent, daily
daily periodic
periodic and
and weekly
weekly periodic components can be calcu-
lated
lated by
by (13)–(15),
(13)–(15), which
which reflect
reflect the
the influence
influence and
and importance
importance of
of the
the components
components on
on the
the
prediction
predictionresults.
results.
Yr
Wr =Wr = r (13)
(13)
Yr + Yd + Yw
r + d + w
Yd
Wd =  (14)
WdY=r + Y d+ Yw (14)
r + 
dd + w
Yw
Ww =  (15)
WwY=r + Yd w+ Yw (15)
r + d + w
where Yr , Yd , Yw are the outputs of the present, daily periodic and weekly periodic
where r, d, w are the outputs of the present, daily periodic and weekly periodic
components,
components, respectively.
respectively.
Based
Basedon onthe
theobtained weights,
obtained the the
weights, finalfinal
prediction resultresult
prediction after weight fusion isfusion
after weight as follows:
is as
follows:
Ŷ = Ww ⊙ Yr + Ww ⊙ Yd + Ww ⊙ Yw (16)

where ⊙ is the Hadamard product.

3. Electric Vehicle Charging Demand Modeling

After obtaining the predicted traffic flow data by WSTGCN, the electrical load of the
FCS is analyzed by an improved queueing model. In this improved queueing model, the
subjective behaviors of EV users are comprehensively considered, including refusal to
join the queue and impatience to leave the queue. By estimating the EV arrival rate and
analyzing the queuing process, the FCS charging load can be predicted.

3.1. Electric Vehicle Arrival Rate Estimation

As a key factor for charging load prediction, the arrival rate needs to be estimated
first, and it is influenced by the EV state-of-charge (SOC). According to [18], the SOC can
be straightforwardly modeled by the consumption power, which can be illustrated by
the following:
pev Dev
SOC = 1 − (17)
Eev
where Eev is the EV battery capacity; pev is the power consumed during the journey; and
Dev is the travel distance of the EV. Also, the travel distance under a certain SOC can be
obtained by Dev = Eev (1 − SOC )/pev .
However, the travel distance before EVs reach the FCS is difficult to directly deter-
mine, especially when the FCS is newly built. To address this challenge, the daily travel
distance distribution p(x) and driving in progress distribution g(t) [18], which can be more
easily derived from historical data, are used as indirect data sources for estimating travel
Energies 2024, 17, 4798 8 of 17

distance. For example, the daily travel distance probability p(100) = 1% means there is a 1%
probability that an EV will travel 100 km in a day. And, a driving in progress probability
g(5) = 8% indicates an 8% probability that an EV is on the road at 5 a.m. By using mixture
model-based distribution fitting techniques, the probability distribution of daily travel
distance and driving in progress can be easily obtained. Based on this distribution, the
daily travel distance of an EV arriving at the FCS at time t can be expressed as follows:
Z t
T
Dev,t = ( Eev (1 − SOCm ))/( pev g(t)dt) (18)
t0

where SOCm represents the mean SOC of EVs when drivers go for charging.
Based on the daily travel distance, the probability that an EV chooses to charge at the
FCS at time t can be expressed as follows:
Z DT
ev,t−1
Pt = p( x )dx (19)
T
Dev,t

where p(x) is the daily travel distance distribution, which can be obtained by distribution
estimation based on the historical data.
Based on the charging probability Pt , the arrival rate is as follows:

λt = ζPt f t (20)

where ζ is the penetration rate of EVs, and ft is the predicted traffic flow at time t. It should
be noted that since the WSTGCN model is trained on traffic flow data for all vehicle types,
including both traditional internal combustion vehicles and EVs, the predicted results from
the WSTGCN reflect the total flow of all vehicles. To analyze traffic flow specifically for
EVs, the EV penetration rate is utilized. By applying this penetration rate, the traffic flow
of EVs can be estimated, allowing for further analysis of the arrival rate at the FCS.

3.2. Improved Queuing Model Considering Driver Behaviors

After arriving at the FCS, EVs queue for charging. To analyze this process, an improved
queuing model is proposed in this paper. It assumes that there are C servers in the FCS
and up to N cars (N > C) can be in the FCS simultaneously. The number of arriving EVs,
denoted as k, follows a Poisson distribution, which can be expressed as follows:

λk −λ
P(k) = e k = 0, 1, 2, · · · , N (21)
k!
where λ is the average arrival rate.
Also, the charging time of each EV follows an exponential distribution, as shown below:

f (tc ) = µe−µtc (22)

where µ is the average leaving rate, which is equal to the number of EVs leaving the FCS.
The mean of charging time tc is equal to 1/µ.
It should be noted that some EVs may leave the FCS without charging. To make the
analysis more realistic and practical, the refusal to join and impatience to leave behaviors
of users are fully considered in the improved queuing model proposed in [18].
Refusal to join behavior occurs when an EV enters an FCS and finds there is a long
queue, causing the user to hesitate about joining. This hesitation is influenced by the queue
length. The probability of EV choosing to queue is defined as follows:

1
 k<C
αk = e−(k−C)σ C ≤ k < N ,σ ≥ 0 (23)

0 k=N

 The mean of charging time tc is equal to 1/μ.
It should be noted that some EVs may leave the FCS without charging. To make the
analysis more realistic and practical, the refusal to join and impatience to leave behaviors
of users are fully considered in the improved queuing model proposed in [18].
Energies 2024, 17, 4798 Refusal to join behavior occurs when an EV enters an FCS and finds there is a9 long of 17
queue, causing the user to hesitate about joining. This hesitation is influenced by the
queue length. The probability of EV choosing to queue is defined as follows:
where σ is the refusal rate parameter.1As k increases, k <Cthe probability of EV users’ willingness
to join the queue will decrease. αk = e−( k −C )σ C ≤ k < N ,σ ≥ 0 (23)
0
Moreover, impatience to leave behavior k =toNEVs leaving the FCS due to long wait
refers
times
whereinσtheis queue. This impatience
the refusal is primarily
rate parameter. related to the
As k increases, the number of EVs
probability of already in
EV users’
the queue. When there are k EVs in the
willingness to join the queue will decrease.queue, the number of EV users leaving impatiently
can beMoreover,
defined asimpatience
follows: to leave behavior refers to EVs leaving the FCS due to long
wait times in the queue. This ( impatience is primarily related to the number of EVs already
0 k≤C
in the queue. When there are k EVs in the queue, the number of EV users leaving impa-
βk = (24)
δ ln(k − C + 1) C < k ≤ N, δ ≥ 0
tiently can be defined as follows:

βk = 0
where δ is the leaving rate parameter.
also increase.
{ k≤C
As k increases, the number of EVs leaving will
δ ln( k − C + 1) C < k ≤ N , δ ≥ 0 (24)

where δ is the leaving

Considering rateleaving
the two parameter. As k increases,
conditions above, thethe number
average of EVsrate
arriving leaving
and will also
average
increase.
leaving rate can be updated as follows:
Considering the two leavingconditions above, the average arriving rate and average
leaving rate can be updated as follows:
λ
 k<C
λk = λαk = λeλ −( k − C ) σ Ck <≤Ck ≤ N , σ ≥ 0 (25)
k = λe−( k −C )σ
λk = λα
 C ≤ k ≤ N ， σ ≥ 0 (25)
0 k=N
k=N
0

{
(
μ − βk = μ
µ k ≤kC≤ C
µk = µ −μβk k= = μ − δ ln( k − C + 1) C < k ≤ N (26)
(26)
µ − δ ln(k − C + 1) C < k ≤ N
By considering refusal behavior and impatience behavior, the arrival process and
By considering refusal behavior and impatience behavior, the arrival process and
charging duration time of EVs can be modeled more precisely, which leaves a solid foun-
charging duration time of EVs can be modeled more precisely, which leaves a solid founda-
dation for the subsequent modeling of the charging load.
tion for the subsequent modeling of the charging load.
3.3. Markov
3.3. Markov Chain-Based
Chain-Based Charging
Charging Process
ProcessAnalysis
Analysis
Based on
Based on the
the arriving
arriving rate
rate and
and leaving
leaving rate,
rate, the
the Markov
Markov chain
chain is
is used
used to
to model
model the
the
charging state transitions of EVs, which have two states: arriving and leaving.
charging state transitions of EVs, which have two states: arriving and leaving. The state The state
transitiondiagram
transition diagramisisshown
shownininFigure
Figure4.4.

Figure 4. State transition of the improved queueing model.

Figure 4. State transition of the improved queueing model.
According to [18], the FCS balance equation can be derived as follows:

(λk + µk ) Pk = λk−1 Pk−1 + µk+1 Pk+1 (27)

where Pk is the steady-state probability of the Markov chain at the state k.

By solving (25)–(27), the probability of k EVs simultaneously being in the charging
station can be obtained:
 k
λ
P0 0≤k<C


k
 µ k!



Pk = λk e−σ(k−C)(k−C−1)/2 (28)
P0 C ≤ k ≤ N
k −C

 δ
 µk C! ∏ (C + ln( j + 1))



j =1 µ
Energies 2024, 17, 4798 10 of 17

The probability that no EV is in the charging station P0 is given by the following:

  −1
 C ρk N
λk e−σ(k−C)(k−C−1)/2 
P0 =  ∑ + ∑ (29)
 
k=0 k! k=C+1 k −C

δ
µ C! ∏ (C + ln( j + 1))
k

j =1 µ

The number of charging EVs in the FCS can be expressed as follows:

C
λk N
λk e−σ(k−C)(k−C−1)/2
Nch = ∑ k µk k! P0 + C ∑ k −C δ
P0 (30)
k =0 k =C +1 µk C!
∏ (C + ln( j + 1))
j =1 µ

Furthermore, the charging load of the FCS can be calculated by the following:

PFCS = Nch pch (31)

where PFCS is the charging load of the FCS, and pch is the charging power of the charger.

4. Experiments
4.1. Performance Evaluation for the WSTGCN Model
4.1.1. Prediction Accuracy Comparison
To validate the proposed WSTGCN, comparative experiments are conducted with
6 benchmark models. The benchmark models are briefly described as follows:
(1) Historical average model (HA): This uses the average value of the past 12 time
slices to predict the next value.
(2) Autoregressive integrated moving average model (ARIMA) [24]: ARIMA trans-
forms non-stationary series into stationary time series through multiple differencing first,
and then uses stationary time series for predictions.
(3) Long short-term memory network (LSTM) [25]: Thanks to its special gate structure,
LSTM can effectively capture useful information in both the long term and short term, thus
having excellent predictive ability in time series prediction.
(4) Gated recurrent unit (GRU) [26]: GRU is very similar to LSTM, using a gate
mechanism to capture useful information. But GRU only has two gates (i.e., update and
reset), compared with LSTM, which has three gates (forget, input, output). Thus, GRU has
fewer parameters than LSTM, making it converge faster in training.
(5) Spatial–temporal graph convolutional network (STGCN) [27]: STGCN is composed
of multiple spatial–temporal convolution blocks to extract useful spatial and temporal
features. Thus, it can efficiently predict time series with a graph structure, like traffic
flow data.
(6) Attention-based STGCN (ASTGCN) [23]: ASTGCN integrates attention mecha-
nisms into the STGCN; thus, it can capture the spatial–temporal characteristics of historical
data more effectively, making it perform outstandingly in predicting traffic data.
To objectively evaluate the performance of different models in predicting traffic flow
for the next hour, mean absolute error (MAE) and root mean square error (RMSE) are used
as metrics. The predictive performance of each model is shown in Figure 5.
From Figure 5, it is evident that the proposed WSTGCN model achieves the best
prediction accuracy compared to other benchmark models. Also, it can be seen from the
results that temporal models (HA, ARIMA, LSTM, and GRU) all perform poorer than the
spatial–temporal models (STGCN, ASTGCN, and WSTGCN). This may be because tempo-
ral models only consider the temporal features inherent in the traffic data, but fail to capture
the spatial correlation of traffic data at different nodes in the traffic network. Moreover,
among the three spatial–temporal models, the proposed WSTGCN model performs the best.
Compared to the second-best ASTGCN model, the WSTGCN model further reduces the
Energies 2024, 17, 4798 11 of 17

Energies 2024, 17, x FOR PEER REVIEW 11 of 17

MAE and RMSE by 7.5% and 8.6%. This is because the WSTGCN model not only integrates
the attention mechanisms but also implements modified residual modules and employs
a weight fusion mechanism to improve the fusion performance of recent, daily periodic
To objectively
and weekly evaluate
periodic the performance
components. of different
By fully exploiting the models
dynamicinspatial–temporal
predicting trafficcorre-
flow
for the next
lations, hour, meanmodel
the WSTGCN absolute error (MAE)
subsequently and rootpredictive
enhances mean square error in
accuracy (RMSE) are used
the traffic flow
as metrics. task.
prediction The predictive performance of each model is shown in Figure 5.

Figure 5. Predictive performance of different models.

Figure 5. Predictive performance of different models.
4.1.2. Ablation Study
From Figure 5, it is evident that the proposed WSTGCN model achieves the best pre-
To further validate the effectiveness of the modified residual modules and the intro-
diction accuracy compared to other benchmark models. Also, it can be seen from the re-
duced weight fusion in the WSTGCN model, ablation experiments are further conducted.
sults that temporal models (HA, ARIMA, LSTM, and GRU) all perform poorer than the
To validate the modified residual modules, ASTGCN and the residual improved STGCN
spatial–temporal models (STGCN, ASTGCN, and WSTGCN). This may be because tem-
(RSTGCN) are compared. RSTGCN only incorporates the residual improvement module
poral models only consider the temporal features inherent in the traffic data, but fail to
into ASTGCN but does not perform weight fusion. Moreover, RSTGCN and WSTGCN
capture the spatial correlation of traffic data at different nodes in the traffic network. More-
are compared to validate the effectiveness of the weight fusion, since WSTGCN further
over, amongthe
introduces theweight
three spatial–temporal
fusion mechanism models,
based the proposed
on the RSTGCN.WSTGCN modelisperforms
Each model used to
the
predict traffic flow for the next hour, and MAE and RMSE are used as metrics. further re-
best. Compared to the second-best ASTGCN model, the WSTGCN model
ducesFrom
the MAE
Figureand
6, itRMSE
can beby 7.5%
seen thatand
the8.6%.
WSTGCNThis is because
has the bestthe WSTGCN
prediction model not
performance,
only integrates the attention mechanisms but also implements modified
followed by the RSTGCN, and the ASTGCN is the poorest. Comparing RSTGCN and residual modules
AST-
and employs a weight fusion mechanism to improve the fusion performance
GCN, the results indicate that the modified residual module with convolution operations is of recent,
daily periodic
effective and weekly
in enhancing periodic components.
the predicting accuracy of By
thefully exploiting
ASTGCN. the dynamic
Moreover, spatial–
comparing the
temporal correlations, the WSTGCN model subsequently enhances predictive
WSTGCN and RSTGCN, the results demonstrate the effectiveness of the proposed weight accuracy in
Energies 2024, 17, x FOR PEER REVIEW
the traffic flow prediction task. 12 of 17
fusion mechanism. By fully exploiting the interactions among recent, daily periodic, and
weekly periodic components, the accuracy of the results can be further improved.
4.1.2. Ablation Study
To further validate the effectiveness of the modified residual modules and the intro-
duced weight fusion in the WSTGCN model, ablation experiments are further conducted.
To validate the modified residual modules, ASTGCN and the residual improved STGCN
(RSTGCN) are compared. RSTGCN only incorporates the residual improvement module
into ASTGCN but does not perform weight fusion. Moreover, RSTGCN and WSTGCN are
compared to validate the effectiveness of the weight fusion, since WSTGCN further intro-
duces the weight fusion mechanism based on the RSTGCN. Each model is used to predict
traffic flow for the next hour, and MAE and RMSE are used as metrics.
From Figure 6, it can be seen that the WSTGCN has the best prediction performance,
followed by the RSTGCN, and the ASTGCN is the poorest. Comparing RSTGCN and
ASTGCN, the results indicate that the modified residual module with convolution opera-
tions
Figureis6.
Figure 6.effective in enhancing
Performance
Performance of
the
of ASTGCN,
ASTGCN, predicting
RSTGCN,
RSTGCN,
accuracy of the ASTGCN. Moreover, com-
and WSTGCN.
and WSTGCN.
paring the WSTGCN and RSTGCN, the results demonstrate the effectiveness of the pro-
posedPredicting
4.1.3. weight fusion mechanism.
Performance ByDifferent
under fully exploiting the interactions
Predicting Horizons among recent, daily
periodic, and weekly periodic components, the accuracy of the results can be further im-
To study the impact of predicting horizons on the performance of the predicting
proved.
model, comparative experiments are conducted with prediction durations of 15 min, 30
min, and 1 h. The experimental results are shown in Table 1.
From Table 1, it can be seen that as the prediction duration increases, the difficulty of
prediction also increases, leading to a decline in the predictive performance. Among all
predicting horizons, the proposed WSTGCN model has the best predictive performance
Energies 2024, 17, 4798 12 of 17

4.1.3. Predicting Performance under Different Predicting Horizons

To study the impact of predicting horizons on the performance of the predicting model,
comparative experiments are conducted with prediction durations of 15 min, 30 min, and
1 h. The experimental results are shown in Table 1.

Table 1. Performance comparison of different predicting horizons.

Predicting Horizon Metrix HA ARIMA LSTM GRU STGCN ASTGCN RSTGCN WSTGCN
MAE 23.15 18.33 16.89 16.25 15.03 12.45 12.15 11.04
15 min
RMSE 35.96 29.49 26.72 25.91 20.12 18.69 18.24 16.54
MAE 26.54 19.77 19.93 19.47 16.85 14.51 14.16 13.25
30 min
RMSE 38.17 36.91 30.29 29.84 22.37 19.58 19.11 17.63
MAE 29.43 23.58 22.61 21.59 17.76 15.92 15.56 14.72
1h
RMSE 40.39 39.32 34.08 33.26 23.61 20.79 20.36 18.97

From Table 1, it can be seen that as the prediction duration increases, the difficulty of
prediction also increases, leading to a decline in the predictive performance. Among all
predicting horizons, the proposed WSTGCN model has the best predictive performance
compared to other benchmark models, while the HA has the poorest performance. Also, it
can be found that spatial–temporal deep learning models like STGCN, ASTGCN, RSTGCN,
and WSTGCN outperform temporal models like HA, ARIMA, LSTM, and GRU. Addition-
ally, the WSTGCN model performs better than the ASTGCN and RSTGCN, which indicates
the effectiveness of the modified residual modules and the introduced weight fusion.

4.2. Charging Load Model Validation

Based on the traffic flow prediction results from the WSTGCN model, the charging
load of FCS is further analyzed by the proposed improved queuing theory model. In the
experiment, the maximum charging power of the FCS is set to PFCS max = 12 MW, and the

total number of servers is set to C = 25, with each server providing a charging power
of pch = 40 kW. The FCS can accommodate up to N = 30 EVs at a time, and the average
charging time for an EV is 20 min. The probability parameters for refusal to join and
impatience to leave are denoted as σ = 1 and δ = 1. Based on the total vehicle stock [28] and
EV stock data [29], the penetration rate of EVs in the United States in 2020 was 2.2%, and
the probability of an EV needing to charge at any given time is 20%.

4.2.1. Charging Load Prediction Results

The charging load prediction results based on the improved queuing theory model are
shown in Figure 7. It can be observed that the predicted charging load closely approximates
Energies 2024, 17, x FOR PEER REVIEW 13 of 17
the actual values, indicating that the improved queuing model is effective for predicting
the charging load of FCSs.

The prediction
Figure 7. The prediction results of charging load.

4.2.2. Impact of the WSTGCN Model

4.2.2. Impact of the WSTGCN Model
To validate the impact of the proposed WSTGCN model on the accuracy of charging
To validate the impact of the proposed WSTGCN model on the accuracy of charging
load predictions, both the WSTGCN and benchmark ASTGCN models are used for traffic
load predictions, both the WSTGCN and benchmark ASTGCN models are used for traffic
flow prediction. And then, based on the traffic flow prediction results, the improved queu-
ing model is used to predict the charging load of FCS. The charging load prediction results
are shown in Figure 8. To perform further quantitative analysis, the MAE and RMSE are
selected as metrics to compare the prediction accuracy of the two models.
 Figure 7. The prediction results of charging load.

Energies 2024, 17, 4798 4.2.2. Impact of the WSTGCN Model

13 of 17
To validate the impact of the proposed WSTGCN model on the accuracy of charging
load predictions, both the WSTGCN and benchmark ASTGCN models are used for traffic
flow
flow prediction.
prediction.AndAnd then, based
then, on the
based traffic
on the flowflow
traffic prediction results,
prediction the improved
results, queu-
the improved
ing model is used to predict the charging load of FCS. The charging load prediction
queuing model is used to predict the charging load of FCS. The charging load prediction results
are shown
results in Figure
are shown 8. To perform
in Figure further
8. To perform quantitative
further analysis,
quantitative the MAE
analysis, and and
the MAE RMSE are
RMSE
selected as metrics
are selected to compare
as metrics the the
to compare prediction accuracy
prediction of the
accuracy twotwo
of the models.
models.

Figure 8. The prediction results of different

different models.

From Figure
From Figure 8,
8, it
it can
can be
be intuitively
intuitively observed that the
observed that the results
results of
of the
the WSTGCN-based
WSTGCN-based
charging load prediction model are closer to the actual values than
charging load prediction model are closer to the actual values than those of those of the
the ASTGCN-
ASTGCN-
based model. Moreover, from Table 2, it can be found that the WSTGCN-based
based model. Moreover, from Table 2, it can be found that the WSTGCN-based traffic flow traffic flow
prediction model has better predictive performance compared to the benchmark
prediction model has better predictive performance compared to the benchmark ASTGCN ASTGCN
model. These
model. These results
results indicate
indicate that
that the
the charging
charging load
load prediction
prediction accuracy
accuracy isis influenced
influenced by by
the traffic flow prediction results significantly. More accurate traffic flow prediction
the traffic flow prediction results significantly. More accurate traffic flow prediction re- results
contribute
sults to more
contribute to accurate chargingcharging
more accurate load prediction results. The
load prediction improved
results. The WSTGCN
improved
WSTGCN provides more accurate traffic flow prediction results, thus greatlythe
provides more accurate traffic flow prediction results, thus greatly improving accuracy
improving
of charging load prediction.
the accuracy of charging load prediction.
Table 2.
Table Performance comparison
2. Performance comparison of
of different
different traffic
traffic flow
flow prediction
prediction models.
models.
TrafficFlow
Traffic Flow Model
Model ASTGCN
ASTGCN WSTGCN
WSTGCN
Charging model
Charging model Improvedqueuing
Improved queuing model
model
MAE
MAE 15.32
15.32 9.01
9.01
RMSE
RMSE 17.76
17.76 10.47
10.47

4.2.3. Impact of Improved Queuing Model

To validate the impact of the queuing model
model on charging
charging load
load predictions,
predictions, the
the im-
im-
queuing model
proved queuing model and the traditional queuing model are both used to predict the
Energies 2024, 17, x FOR PEER REVIEW 14 of 17
charging load
load of
ofFCS
FCSbased
basedononthe same
the traffic
same trafficflow prediction
flow results
prediction fromfrom
results WSTGCN.
WSTGCN. The
results are shown in Figure 9.
The results are shown in Figure 9.

Figure 9.
Figure The charging
9. The charging load
load prediction
prediction results.

From Figure
From Figure9,9,ititcan
canbebeintuitively
intuitively seen
seen that
that thethe charging
charging loadload predicted
predicted by im-
by the the
improved queuing theory model is lower than that of the traditional queuing
proved queuing theory model is lower than that of the traditional queuing theory model. theory model.
This may
This may be
be because
because the
the subjective
subjective behaviors
behaviors of of EV
EV owners
owners areare fully
fully considered
considered inin the
the
improved queuing theory model, affecting the prediction results significantly.
improved queuing theory model, affecting the prediction results significantly. Due to the Due to the
limited capacity
limited capacity of
of the
the FCSs,
FCSs, some
some EVsEVs leave
leave the
the FCS
FCS without
without receiving
receiving charging
charging service,
service,
resulting in a loss of charging load.
resulting in a loss of charging load.

4.2.4. Sensitivity Analysis

To explore the impact of key parameters on prediction results, sensitivity analyses
are conducted for the number of servers C, the maximum capacity of the FCS N, the refusal
 From Figure 9, it can be intuitively seen that the charging load predicted by the im-
proved queuing theory model is lower than that of the traditional queuing theory model.
This may be because the subjective behaviors of EV owners are fully considered in the
improved queuing theory model, affecting the prediction results significantly. Due to the
limited capacity of the FCSs, some EVs leave the FCS without receiving charging service,
Energies 2024, 17, 4798 14 of 17
resulting in a loss of charging load.

4.2.4. Sensitivity Analysis

4.2.4. Sensitivity Analysis
To explore the impact of key parameters on prediction results, sensitivity analyses
To explore the impact of key parameters on prediction results, sensitivity analyses are
are conducted for the number of servers C, the maximum capacity of the FCS N, the refusal
conducted for the number of servers C, the maximum capacity of the FCS N, the refusal to
to join intensity parameter σ, and the impatience to leave intensity parameter δ.
join intensity parameter σ, and the impatience to leave intensity parameter δ.
As shown in Figure 10a, when the number of servers C or the maximum capacity N
As shown in Figure 10a, when the number of servers C or the maximum capacity
increases, the predicted charging load correspondingly increases. This is because the lim-
N increases, the predicted charging load correspondingly increases. This is because the
ited capacity of the FCS significantly influences the EV arrival rate. If the service capacity
limited capacity of the FCS significantly influences the EV arrival rate. If the service
of the FCS is small, charging servers will easily become fully occupied; thus, some EVs are
capacity of the FCS is small, charging servers will easily become fully occupied; thus,
forcedEVs
some to leave, resulting
are forced in aresulting
to leave, loss of charging
in a loss load. The results
of charging load. indicate thatindicate
The results the service
that
capacity
the of capacity
service the FCS deeply influences
of the FCS deeplythe chargingthe
influences behaviors
chargingofbehaviors
the EVs. The insufficient
of the EVs. The
service capacity
insufficient will
service limit the
capacity widespread
will use of EVs to
limit the widespread usesome extent.
of EVs Therefore,
to some extent. construct-
Therefore,
constructing adequate charging facilities can greatly facilitate EV charging andEV
ing adequate charging facilities can greatly facilitate EV charging and promote devel-
promote
opment.
EV development.

Figure
Figure 10. (a) The
10. (a) The impact
impact of parameters C
of parameters and N;
C and N; (b)
(b) the
the impact
impact of
of parameters σ and
parameters σ δ.
and δ.

Figure 10b shows the impact of the refusal parameter σ and the impatience parameter
Figure 10b shows the impact of the refusal parameter σ and the impatience parameter
δ. The results show that if drivers are more sensitive to waiting times, reflected by a larger
δ. The results show that if drivers are more sensitive to waiting times, reflected by a larger
refusal parameter σ and impatience parameter δ, the charging load decreases accordingly.
refusal parameter σ and impatience parameter δ, the charging load decreases accordingly.
Additionally, it can be observed that the reduction in charging loads is particularly notice-
Additionally, it can be observed that the reduction in charging loads is particularly no-
able during peak periods. This is because, during the peak time, the number of queuing
ticeable
EVs duringresulting
increases, peak periods. Thisleaving
in a larger is because, during
rate and lowerthe peakrate
arrival time,
thanthe number
other of
periods,
thus leading to a larger loss in charging load.

4.2.5. Impact of FCS Scale on Service Performance

Through the analysis of the queuing model, the service performance of the FCS can
be further assessed. This service performance is directly linked to user experience at the
FCS, which is crucial for enhancing customer satisfaction and promoting the adoption of
electric vehicles. We choose two metrics to quantify the service performance of different
scale FCSs, i.e., average waiting time and average queue length of EVs. The results are
shown in Figure 11.
From Figure 11, it can be found that the service performance of the FCS is directly
related to its scale. As the number of charging servers C and the capacity N decrease, the
average queue length and average waiting time for EVs increase. Additionally, it can be
seen that when the scale of FCS exceeds 15 servers and can accommodate up to 20 EVs, the
improvements in service performance become marginal. Therefore, it can be concluded that
the installation of FCS should be planned for an optimal scale. If there are too few servers,
service quality suffers, leading to excessively long wait times; conversely, if there are too
many, the improvement in service quality is negligible while investment costs rise. The
 periods, thus leading to a larger loss in charging load.

4.2.5. Impact of FCS Scale on Service Performance

Through the analysis of the queuing model, the service performance of the FCS can
Energies 2024, 17, 4798 be further assessed. This service performance is directly linked to user experience 15atofthe
17
FCS, which is crucial for enhancing customer satisfaction and promoting the adoption of
electric vehicles. We choose two metrics to quantify the service performance of different
scale FCSs,
proposed i.e., average
model waitingmodel
can effectively time the
andcharging
average process
queue length of EVs.strategic
and provide The results are
insights
shown in Figure 11.
for the planning of charging stations.

Figure 11.
Figure 11. (a)
(a) Average waiting time;
Average waiting (b) average
time; (b) average queue
queue length.
length.

5. Conclusions
From Figure 11, it can be found that the service performance of the FCS is directly
related to paper
This its scale. As the anumber
proposes of charging servers
deep learning-based model C for
andpredicting
the capacity theNcharging
decrease,loadthe
average
of FCSs toqueue length
quantify and
the average
impact waiting
of EV time on
charging forthe
EVspower
increase.
grid.Additionally,
The main work it cancan
be
seen
be that whenasthe
concluded scale of(i)FCS
follows: An exceeds
improved 15 servers
WSTGCN andmodel
can accommodate
is proposed up to 20 EVs,
to model the
the traffic
EV improvements in servicethe
flow, considering performance become marginal.
inherent spatial–temporal Therefore, itofcan
characteristics be flow.
traffic con-
cluded thatto
Compared thetemporal
installation of FCS
models should
(HA, be planned
ARIMA, LSTM,for andanGRU),
optimal thescale. If therenot
WSTGCN areonly
too
considers
few servers, theservice
temporal feature
quality inherent
suffers, in the
leading traffic data,long
to excessively but wait
also times;
captures the spatial
conversely, if
correlation
there are too of many,
traffic the
dataimprovement
at different nodes in the
in service trafficisnetwork.
quality negligibleBywhile
fullyinvestment
exploiting
the
costsdynamic
rise. Thespatial–temporal
proposed modelcorrelations,
can effectively themodel
WSTGCN model performs
the charging process andmuch more
provide
accurately than temporal
strategic insights models. (ii)
for the planning By modifying
of charging the residual modules and integrating
stations.
the weight fusion mechanism into the WSTGCN model, the prediction performance has
been significantly enhanced. The experimental results show that the proposed WSTGCN
5. Conclusions
model exhibits
This papersuperior
proposesaccuracy in predicting traffic
a deep learning-based modelflow compared the
for predicting to other benchmark
charging load of
models. Compared to the second-best ASTGCN model, the
FCSs to quantify the impact of EV charging on the power grid. The main work can beWSTGCN model further
reduces
concluded theasMAE and (i)
follows: RMSE by 7.5% and
An improved 8.6%. (iii)
WSTGCN An improved
model is proposed queuing
to modeltheory
the EV model
traf-
is used to predict the EV charging load of FCSs with special consideration
fic flow, considering the inherent spatial–temporal characteristics of traffic flow. Com- of subjective
EV user behaviors. By considering the behaviors of EV users refusing to join and leaving
pared to temporal models (HA, ARIMA, LSTM, and GRU), the WSTGCN not only con-
impatiently, the predicted charging load of the FCS is lower than the traditional queuing
siders the temporal feature inherent in the traffic data, but also captures the spatial corre-
theory model, especially during peak times. Additionally, it should be noted that the
lation of traffic data at different nodes in the traffic network. By fully exploiting the dy-
proposed charging load prediction method is based on traffic flow prediction, which
namic spatial–temporal correlations, the WSTGCN model performs much more accu-
can successfully address the challenge of lacking historical data, thus holding important
rately than temporal models. (ii) By modifying the residual modules and integrating the
practical value in charging load prediction for newly established FCSs.
weight fusion mechanism into the WSTGCN model, the prediction performance has been
significantly
Author enhanced.Conceptualization,
Contributions: The experimentalJ.Z.results showsoftware,
and H.C.; that the J.Z.;
proposed WSTGCN
methodology, model
H.Z., Z.W.
exhibits superior
(Zhiqiang Wang) and accuracy in predicting
Z.W. (Ziyi traffic flow
Wen); visualization, Z.W.compared
(Ziyi Wen); towriting—original
other benchmark models.
draft, Z.W.
(Ziyi Wen); formal analysis, X.Z.; supervision, X.Z.; writing—review and editing, X.Z. All authors
have read and agreed to the published version of the manuscript.
Funding: This research was funded by the National Natural Science Foundation of China, grant
number 72001058; Guangdong Basic and Applied Basic Research Foundation, grant numbers
2023A1515010724 and 2022A1515240051; General Program of Foundation of Shenzhen Science and
Technology Committee, grant number GXWD20231130154831002; Major Science and Technology Spe-
cial Projects in Xinjiang Autonomous Region, grant number 2022A01007; and Xinjiang Autonomous
Region Key Research and Development Task Special Project, grant number 2022B01016.
Energies 2024, 17, 4798 16 of 17

Data Availability Statement: The contributions presented in the study are included in the article.
Conflicts of Interest: Author Jun Zhang, Huiluan Cong, Hui Zhou, Zhiqiang Wang were employed
by the company State Grid Shandong Electric Vehicle Service Company Ltd. The remaining authors
declare that the research was conducted in the absence of any commercial or financial relationships
that could be construed as a potential conflict of interest.

References
1. Amini, M.; Kargarian, A.; Karabasoglu, O. ARIMA-based decoupled time series forecasting of electric vehicle charging demand
for stochastic power system operation. Electr. Power Syst. Res. 2016, 140, 378–390. [CrossRef]
2. Louie, H.M. Time-series modeling of aggregated electric vehicle charging station load. Electr. Power Compon. Syst. 2017, 45,
1498–1511. [CrossRef]
3. Akshay, K.C.; Grace, G.H.; Gunasekaran, K.; Samikannu, R. Power consumption prediction for electric vehicle charging stations
and forecasting income. Sci. Rep. 2024, 14, 6497. [CrossRef] [PubMed]
4. Wang, S.; Zhuge, C.; Shao, C.; Wang, P.; Yang, X.; Wang, S. Short-term electric vehicle charging demand prediction: A deep
learning approach. Appl. Energy 2023, 340, 121032. [CrossRef]
5. Yi, Z.; Liu, X.; Wei, R.; Chen, X.; Dai, J. Electric vehicle charging demand forecasting using deep learning model. J. Intell. Transp.
Syst. 2022, 26, 690–703. [CrossRef]
6. Jahangir, H.; Gougheri, S.; Vatandoust, B.; Golkar, M.A.; Ahmadian, A.; Hajizadeh, A. Plug-in electric vehicle behavior modeling
in energy market: A novel deep learning-based approach with clustering technique. IEEE Trans. Smart Grid 2020, 11, 4738–4748.
[CrossRef]
7. Sasidharan, M.; Kinattingal, S.; Simon, S. Comparative analysis of deep learning models for electric vehicle charging load
forecasting. J. Inst. Eng. Ser. B 2023, 104, 105–113. [CrossRef]
8. Dabbaghjamanesh, M.; Moeini, A.; Kavousi-Fard, A. Reinforcement learning-based load forecasting of electric vehicle charging
station using Q-learning technique. IEEE Trans. Ind. Inf. 2021, 17, 4229–4237. [CrossRef]
9. Mekkaoui, D.; Midoun, M.; Shen, Y. LA-RCNN: Luong attention-recurrent-convolutional neural network for EV charging load
prediction. Appl. Intell. 2024, 54, 4352–4369. [CrossRef]
10. Li, Y.; Huang, Y.; Zhang, M. Short-term load forecasting for electric vehicle charging station based on niche immunity lion
algorithm and convolutional neural network. Energies 2018, 11, 1253. [CrossRef]
11. Wang, S.; Chen, A.; Wang, P.; Zhuge, C. Predicting electric vehicle charging demand using a heterogeneous spatio-temporal
graph convolutional network. Transp. Res. Part C Emerg. Technol. 2023, 153, 104205. [CrossRef]
12. Su, S.; Li, Y.; Chen, Q.; Xia, M.; Yamashita, K.; Jurasz, J. Operating status prediction model at EV charging stations with fusing
spatiotemporal graph convolutional network. IEEE Trans. Transp. Electrif. 2023, 9, 114–129. [CrossRef]
13. Chen, Y.; Wang, M.; Wei, Y.; Huang, X.; Gao, S. Multi-encoder spatio-temporal feature fusion network for electric vehicle charging
load prediction. J. Intell. Robot. Syst. 2024, 110, 94. [CrossRef]
14. Li, M.; Zhu, Z. Spatial-Temporal Fusion Graph Neural Networks for Traffic Flow Forecasting. Proc. AAAI Conf. Artif. Intell. 2021,
35, 4189–4196. [CrossRef]
15. Li, F.; Feng, J.; Yan, H.; Jin, J.; Yang, F.; Sun, F.; Jin, D.; Li, Y. Dynamic graph convolutional recurrent network for traffic prediction:
Benchmark and solution. ACM Trans. Knowl. Discov. Data 2023, 17, 1–21. [CrossRef]
16. Zhang, Y.; Liu, C.; Rao, X.; Zhang, X.B.; Zhou, Y. Spatial-temporal load forecasting of electric vehicle charging stations based on
graph neural network. J. Intell. Fuzzy Syst. 2024, 46, 821–836. [CrossRef]
17. Shi, J.; Zhang, W.; Bao, Y.; Gao, D.W.; Wang, Z.H. Load forecasting of electric vehicle charging stations: Attention-based
spatiotemporal multi-graph convolutional networks. IEEE Trans. Smart Grid 2023, 15, 3016–3027. [CrossRef]
18. Zhang, X.; Chan, K.W.; Li, H.; Wang, H.; Qiu, J.; Wang, G. Deep-learning-based probabilistic forecasting of electric vehicle
charging load with a novel queuing model. IEEE Trans. Cybern. 2020, 51, 3157–3170. [CrossRef]
19. Wu, T.; Ji, X.; Wang, G.; Liu, Y.; Yang, Q.; Bao, Z.; Peng, J. Hydrogen energy storage system for demand forecast error mitigation
and voltage stabilization in a fast-charging station. IEEE Trans. Ind. Appl. 2021, 58, 2718–2727. [CrossRef]
20. Arias, M.B.; Kim, M.; Bae, S. Prediction of electric vehicle charging-power demand in realistic urban traffic networks. Appl. Energy
2017, 195, 738–753. [CrossRef]
21. Feng, J.; Chang, X.; Fan, Y.; Luo, W. Electric vehicle charging load prediction model considering traffic conditions and temperature.
Processes 2023, 11, 2256. [CrossRef]
22. Tang, S.; Mu, Y.; Zhou, Y.; Dong, X.; Jia, H. A spatial-temporal electric vehicle charging load forecasting method considering the
coordination among the multiple charging behaviors. In Proceedings of the 2021 Power System and Green Energy Conference
(PSGEC), Shanghai, China, 20–22 August 2021; pp. 629–634.
23. Guo, S.; Lin, Y.; Feng, N.; Song, C.; Wan, H. Attention-based spatial-temporal graph convolutional networks for traffic flow
forecasting. Proc. AAAI Conf. Artif. Intell. 2019, 33, 922–929. [CrossRef]
24. Williams, B.M.; Hoel, L.A. Modeling and forecasting vehicular traffic flow as a seasonal ARIMA process: Theoretical basis and
empirical results. J. Transp. Eng. 2003, 129, 664–672. [CrossRef]
25. Sutskever, I.; Vinyals, O.; Le, Q.V. Sequence to sequence learning with neural networks. arXiv 2014, arXiv:1409.3215.
Energies 2024, 17, 4798 17 of 17

26. Chung, J.; Gulcehre, C.; Cho, K.H.; Bengio, Y. Empirical evaluation of gated recurrent neural networks on sequence modeling.
arXiv 2014, arXiv:1412.3555.
27. Yu, B.; Yin, H.; Zhu, Z. Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting. arXiv
2017, arXiv:1709.04875.
28. Our World in Data. Number of New Cars Sold, by Type, World. Available online: https://ourworldindata.org/energy (accessed
on 18 September 2024).
29. IEA. Global Electric Car Stock, 2013–2023. Available online: https://www.iea.org/data-and-statistics/charts/global-electric-car-
stock-2013-2023 (accessed on 18 September 2024).

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.