TSINGHUA SCIENCE AND TECHNOLOGY

ISSN 1007-0214 12/20 pp1776−1784

DOI: 1 0 . 2 6 5 9 9 / T S T . 2 0 2 3 . 9 0 1 0 0 9 4
Volume 29, Number 6, December 2024

Artificial Intelligence Enabled Future Wireless Electric Vehicles with

Multi-Model Learning and Decision Making Models

Gajula Ramesh, Anil Kumar Budati*, Shayla Islam, Louai A. Maghrabi, and Abdullah Al-Atwai

Abstract: In the contemporary era, driverless vehicles are a reality due to the proliferation of distributed
technologies, sensing technologies, and Machine to Machine (M2M) communications. However, the
emergence of deep learning techniques provides more scope in controlling and making such vehicles energy
efficient. From existing methods, it is understood that there have been many approaches found to automate
safe driving in autonomous and electric vehicles and also their energy efficiency. However, the models focus on
different aspects separately. There is need for a comprehensive framework that exploits multiple deep learning
models in order to have better control using Artificial Intelligence (AI) on autonomous driving and energy
efficiency. Towards this end, we propose an AI-based framework for autonomous electric vehicles with multi-
model learning and decision making. It focuses on both safe driving in highway scenarios and energy
efficiency. The deep learning based framework is realized with many models used for localization, path
planning at high level, path planning at low level, reinforcement learning, transfer learning, power control, and
speed control. With reinforcement learning, state-action-feedback play important role in decision making. Our
simulation implementation reveals that the efficiency of the AI-based approach towards safe driving of
autonomous electric vehicle gives better performance than that of the normal electric vehicles.

Key words: wireless vehicles; deep learning; multi-model learning; reinforcement learning; Artificial Intelligence (AI)

1 Introduction are conducing to it. The technological innovations pave

for such vehicles to become reality rather than a dream.
Autonomous vehicles are the vehicles where self- Researchers contribute towards safe autonomous
driving is realized. It is essentially based on Artificial driving and also energy efficiency using deep learning
Intelligence (AI) approaches. This kind of approach in approaches.
vehicles is useful in certain roads and conditions that Li et al.[1] proposed a methodology using unscented
Gajula Ramesh is with Department of Computer Science and Engineering, Gokaraju Rangaraju Institute of Engineering & Technology,
Hyderabad 500090, India. E-mail: [email protected].
Anil Kumar Budati is with Institute of Computer Science and Digital Innovation (ICSDI), UCSI University, Kuala Lumpur 56000,
Malaysia, and also with Department of ECE, Koneru Lakshmaiah Education Foundation, Hyderabad 500090, India. E-mail:
[email protected].
Shayla Islam is with ICSDI, UCSI University, Kuala Lumpur 56000, Malaysia. E-mail: shayla@ucsiuniversity. edu.my.
Louai A. Maghrabi is with Department of Software Engineering, College of Engineering, University of Business and Technology,
Jeddah 21448, Kingdom of Saudi Arabia. E-mail: [email protected].
Abdullah Al-Atwai is with Department of Computer Science, Applied College, University of Tabuk, Tabuk 47512, Kingdom of Saudi
Arabia. E-mail: [email protected].
* To whom correspondence should be addressed.
Manuscript received: 2023-07-10; revised: 2023-08-03; accepted: 2023-08-28

© The author(s) 2024. The articles published in this open access journal are distributed under the terms of the
Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/).
 Gajula Ramesh et al.: Artificial Intelligence (AI) Enabled Future Wireless Electric Vehicles with Multi-Model... 1777

Kalman predictor and deep learning for autonomous parameters like energy, safe driving, AI, computations,
electric vehicles towards better motion control. Strugar awareness, and so on. Ahmed et al.[19] explored the
et al.[2] focused on micropayments in electric vehicles mass usage of electric vehicles and the importance of
that are autonomous in nature. They implemented a AI in their growth. Yu et al.[20] implemented a deep
system based on Machine to Machine (M2M) learning based framework for traffic safety solutions
distributed architecture. Jeong et al.[3] investigated on for autonomous electric vehicles as part of intelligent
self-diagnosis of autonomous vehicles based on deep transportation system. Qi et al.[21] proposed a method
learning and Internet of Things (IoTs). Min et al.[4] using deep reinforcement learning towards safe driving
modeled a system to characterize individual drivers and efficient self-learning. Sorlei et al.[22] investigated
with parameter-learning to control the vehicle with on fuel cell electric vehicles and energy efficiency.
braking using vehicle deceleration prediction model. From the literature, it is understood that there have
With respect to energy efficiency, Qi et al.[5] focused been many approaches or frameworks found to
on energy efficiency of autonomous electric vehicles automate safe driving in autonomous and electric
based on deep reinforcement learning. Miglani and vehicles and also their energy efficiency. However, the
Kumar[6] investigated on deep learning models that are models focus on different aspects separately. There is
suitable for traffic flow prediction automatically. Dixit need for a comprehensive framework that exploits
et al.[7] studied on various deep learning methods to multiple deep learning models in order to have better
identify anomalies in autonomous electric vehicles. Li control using AI on autonomous driving and energy
et al.[8] proposed methodology for hybrid battery efficiency. Our contributions in this paper are as
systems for electric autonomous vehicles to have follows.
energy efficiency. They used deep reinforcement • We propose an AI-based framework that has
learning approach towards achieving this. Jahangir multiple deep learning models to support decision
et al.[9] proposed a deep learning based approach with making in autonomous vehicles.
clustering to model behaviour of electric vehicle. • An algorithm named Multi-Model Learning and
Bhatti et al.[10] explored a digital twin technology for Decision Making (MML-DM) which exploits many
smart electric vehicles. Wan et al.[11] focused on deep deep learning models to arrive at right decision in
learning based approaches towards safe driving of autonomous driving.
autonomous vehicles. Super resolution Direction-of- • A simulation prototype has been made with an
Arrival (DOA) estimation approach is used towards it. open source dataset collected from Ref. [23].
Jinil and Reka[12] explored on power distribution and The dataset when used as a highway map, it helps in
power optimization in electric vehicles using deep simulation of navigation. The dataset contains data
learning models. pertaining to waypoints that reflect the trajectory of the
Wei et al.[13] focused on direct torque control in driving track in highway.
autonomous electric vehicles using deep reinforcement The remainder of the paper is structured as follows.
learning. Sanguesa et al.[14] investigated on different Section 2 presents the proposed framework and
aspects of electric vehicles. Their focus is on underlying methods to realize decision making in
challenges in usage of batteries, AI techniques for autonomous vehicles. Section 3 presents experimental
communication, power distribution, and eco-friendly results. Section 4 concludes the paper and gives
approaches in autonomous vehicles. Tien[15] focused on directions for future scope of the research.
smartness of autonomous vehicles. He investigated on
2 Proposed Methodology
how sensor technologies can leverage decision making
in such vehicles. Kim et al.[16] proposed an approach The proposed methodology is meant for building an
for fuel economy optimization in autonomous vehicles AI-based framework for Electric Autonomous Vehicle
using deep reinforcement learning. Jayasuria et al.[17] (EAV) for safe driving and energy efficiency. The
proposed methodology for object detection and framework is realized with multi-model learning based
localization for safe driving in autonomous vehicles. on deep learning approaches. As the vehicle is
Damaj et al.[18] studied the development of connected equipped with sensor devices, the data collected are
and autonomous electric vehicles considering different exploited by deep learning approaches to understand
 1778 Tsinghua Science and Technology, December 2024, 29(6): 1776−1784

runtime situations and make well informed decisions decisions with heuristic knowledge gained from time to
towards safe driving. In the process, the framework time. The deep learning models help in improving
exploits different computing resources, such as edge accuracy in making decisions. The data collected from
computing, fog computing, and cloud computing. As sensors help in data analytics and make well informed
the cloud computing does not provide good latency but decisions. Modular Recurrent Neural Network (RNN)
provides resources required, temporarily, the is used in order to facilitate distribution of power and
framework uses fog and edge computing resources for ensure optimization in power consumption. The vehicle
real-time decision making. The framework has is equipped with both safe driving and energy
different modules that are based on deep learning. The efficiency. Very important technique used in the
localization module uses deep learning to localize the proposed system is reinforcement learning. It has
vehicle, so that from that standpoint it can make correct mechanisms to know the state of the system and get
decisions. The high level path planning module gains reward/feedback. It incorporates both action-space and
knowledge on overall path to be followed. The low reward-space, and learns current developments
level path planning is meant for understanding subtler dynamically towards making most accurate decision.
details of the runtime traffic in order to make driving As presented in Fig. 2, the Long Short-Term
decisions. The reinforcement learning module is meant Memory (LSTM) is the underlying module used in
for revising knowledge from time to time, while the different learning models. As the LSTM has memory
transfer learning module is meant for reusing already cell, it works well with the time series data that are
acquired know-how efficiency. The speed control useful in autonomous driving. The LSTM model is
module is meant for controlling speed of the vehicle, used in the proposed system to learn temporal data and
while the power control module is meant for power make appropriate decisions. For instance, LSTM model
distribution and energy efficiency. can have memory cell in order to keep track of time-
As presented in Fig. 1, the vehicle takes knowledge linked data, so as to understand situations and temporal
from the framework and makes decisions on moving. dimensions.
There is continuous learning process in order to make The operations associated with front gate (ft) (as in

Edge cmputing Cloud computing

Deep learning Deep learning

Deep learning for path for low
for localization planning at level path
Power
highlevel planning
control

Runtime
scenario and
sensor data Electrical
Reinforcement Transfer
autonomous
learning learning
vehicle Modular
RNN

Speed control
Adaptive-Network-based Direct Torque
Fuzzy logic
Fuzzy Inference System (ANFIS) Control (DTC)

Fig. 1 Overall architecture of the proposed system.

 Gajula Ramesh et al.: Artificial Intelligence (AI) Enabled Future Wireless Electric Vehicles with Multi-Model... 1779

Ht Table 1 Notations and description.

Ct−1 X + Ct Notation Description
T Ct Current LSTM cell at given time step t
X
C̄ t Candidate layer at given time step t
σ σ σ X
T Ct−1 Previous LSTM cell’s memory
ft Forget gate at given time step t
+ + + + Ht Current block’s output
Ht−1 Output of previous block
Ht
0 1 2 3 It Input gate at given time step t
Ot Output gate at given time step t
Ht−1
W, U Weight vectors
X Input vector
Xt T Tanh neuran network
σ() Variance function
Fig. 2 Single LSTM cell used by different deep learning
modules, ⓪–③ are the speed level representations.
and there is co-energy strategy that extracts stage
Eq. (1)), candidate layer (C̄ t ) (as in Eq. (2)), input gate torque attributes. There is a phenomenon known as
(I t ) (as in Eq. (3)), output gate (Ot) (as in Eq. (4)), self-arranging learning calculation that is used to help
memory of current LSTM cell (Ct) (as in Eq. (5)), and RBFN-AFS to learn SRM faster including its
output of the current LSTM cell (Ht) (as in Eq. (6)) are electromagnetic attributes. The main architecture and
computed as follows. The LSTM model we used is the speed controlling modules are seamlessly
inspired by the one in Refs. [23, 24], integrated to have a combined system that works
ft = σ (Xt · U f + Ht−1 · W f ) (1) efficiently.
As presented in Fig. 4, the modular RNN is one of
C̄t = tanh (Xt · Uc + Ht−1 · Wc ) (2) the deep learning methods that is employed to obtain
power requirements of the vehicle. Here Gi (i = 1, 2, 3)
It = σ (Xt · Ui + Ht−1 · Wi ) (3) are the blocks representing recurrent layer, G(k) is the
input to the layer, Zi (k) (i = 1, 2, 3) is the output of the
Ot = σ (Xt · Uo + Ht−1 · Wo ) (4)
layer, and u−1 is the bias weight vector. Based on the
Ct = ft · Ct−1 + It · C̄t (5) information about power supply possibilities, it makes
important observations and coordinates to function
Ht = Ot · tanh (Ct ) (6) certain power-consuming modules. The aim of the
where the variables and function are explained in Table modular RNN is to consider electricity-driven context
1. All these equations are used as part of LSTM cell. of the EAV and ensure that power is managed
They reflect functionality of different gates associated efficiently. Here the usage of RNN is very important
with the LSTM cell. They take care of taking input, because it has memory structure that is reused, and thus
processing it, and then moving on to the next cell until time-series data are processed more efficiently when
the process is completed.
The Switched Reluctance Motor (SRM) of EAV SRM of EAV

needs to be monitored and controlled in terms of speed.

The module presented in Fig. 3 takes care of this.
There are different techniques used in order to SRM
Forecasting
characterize and control the speed of the vehicle in characterization
accordance with traffic conditions. Broadly, fuzzy
logic, DTC, and Radial Basis Function Network-based Fuzzy
RBFN-AFS DTC
Adaptive Fuzzy System (RBFN-AFS) work together in logic
order to achieve better control of the speed of the EAV.
There is a technique to quantify flux linkage of SRM Fig. 3 Speed controlling module.
 1780 Tsinghua Science and Technology, December 2024, 29(6): 1776−1784

u−1 technology-driven and real-time knowledge required.

u−1 Simulation of autonomous driving of electric vehicle is
made using highway driving environment. As far as
highway environment is concerned, the simulation
provides lanes on highway and ability of vehicle to
G1 G2 G3
change lanes safely based on surrounding traffic in
Z1(k) Z2(k) Z3(k)
order to minimize jerks and eliminate possible
collisions. Making changes in lane is one of the
essential things in the navigation of autonomous
G (k)
vehicle. Safe navigation on virtual highway is
Fig. 4 Modular RNN for predicting power requirements of indispensable for success of the project. The speed
the vehicle. limit set for vehicle is 40 Miles Per Hour (MPH).
Localization of car and analytics of its sensors’ data are
compared with other deep learning techniques. part of the system. It makes use of map of waypoints
As presented in Algorithm 1, it exploits many deep on the virtual highway. The car needs to use the 40
learning models to arrive at right decision in MPH speed when possible. And it drops speed based
autonomous driving. It has the mechanisms and on surrounding traffic. It has to consider passible lower
workflow to realize the proposed system shown in speed vehicles, manage change of lanes, and observe
Fig. 1. A simulation prototype has been made with an the lane changes of other vehicles as well. The car is
open source dataset collected from Refs. [23, 24]. The expected to follow these rules. (1) It should follow lane
dataset is used as a highway map, which helps in driving all the time. (2) It should never hit any other
simulation of navigation. The dataset contains data vehicles or objects. (3) It should safely change lanes
pertaining to waypoints that reflect the trajectory of the with indicators. (4) It should use the maximum speed
driving track in highway. The algorithm gains when there is a chance. The simulation work is carried
heuristics required to move the vehicle safely with real- out initially with an open source dataset collected from
time decision making. Section 4 presents results of Ref. [23]. The dataset is used as a highway map, which
driving with appropriate decisions. helps in simulation of navigation. As it is the
simulation study, highway map data which are pre-
3 Results and Discussion existed are used to train the models and accordingly,
they make decisions in vehicle movements.
This research is aimed at building an AI framework
The dataset contains data pertaining to waypoints
that has dual goal of “monitoring and regulating power
that reflect the trajectory of the driving track in
usage” and facilitating autonomous driving with
highway. Each instance in the dataset is a waypoint.
Algorithm 1 MML-DM Each waypoint has five different values. Part of the
Inputs: Trajectory of the driving track in highway data from highway map dataset are listed in Table 2. In
Output: Driving decisions Table 2, the first two data x and y reflect global map
1. Start position, the third is the distance (mile) travelled by the
2. For each instance Inst in trajectory Traj vehicle up to that waypoint, the fourth and fifth are x
3. loc←Localization (Inst); and y components respectively of a Frenet unit normal
4. p←HighLevelPathPlanning (Traj); vector. The first waypoint’s third value (distance) is
5. P←LowLevelPathPlanning (Inst, Traj); zero as it is the starting point. This data help electric
6. knowhow1←ReinforcementLearning (loc, p, P, Traj); vehicle to get simulated as per the waypoints given in
7. knowhow2←TransferLearning (knowledge history); the dataset, which is shown in Fig. 1. The results of
8. decisions←GetRealtimeDecisions (loc, p, P, knowhow1, self-driving are presented in the form of visual
knowhow2); representations of the simulations with cars on the
9. AEV follows decisions like moving to different lane, track. The data are related to highway map having pre-
straight, left, right, etc.; existing traffic scenario which is useful for simulation
10. End For study. That data are used to train models and help
11. End
driverless car in making self-driving decisions.
 Gajula Ramesh et al.: Artificial Intelligence (AI) Enabled Future Wireless Electric Vehicles with Multi-Model... 1781

Table 2 Excerpt from highway map dataset for path planning of autonomous vehicle.
x y Distance (mile) x compoment y compoment
815.2679 1134.9300 30.674 478 530 883 −0.010 994 79 −0.999 939 6
844.6398 1134.9110 60.046 371 459 961 −0.002 048 373 −0.999 997 9
875.0436 1134.8080 90.450 414 657 593 −0.001 847 863 −0.999 998 3
905.2830 1134.7990 120.689 735 412 598 0.004 131 14 −0.999 991 5
934.9677 1135.0550 150.375 551 223 755 0.059 043 82 −0.998 255 4
964.7734 1138.3180 180.359 313 964 844 0.167 776 10 −0.985 825 2
995.2703 1145.3180 211.649 354 934 692 0.307 788 80 −0.951 454 7
1025.0280 1157.8100 243.922 914 505 005 0.382 557 80 −0.923 931 7
1054.4980 1169.8420 275.754 606 246 948 0.381 560 30 −0.924 343 9
1079.2190 1180.1790 302.548 864 364 624 0.319 190 20 −0.947 690 7
1102.0470 1185.8570 326.072 883 605 957 0.183 314 70 −0.983 054 3
1127.1490 1189.1160 351.385 223 388 672 0.098 716 02 −0.995 115 7
1160.1700 1191.6230 384.501 911 163 330 0.064 293 68 −0.997 931 1
1182.1230 1192.6580 406.479 455 947 876 0.046 639 20 −0.998 911 9
1208.3600 1193.8730 432.744 598 388 672 0.039 875 33 −0.999 204 7
1235.6230 1194.7930 460.022 594 451 904 −0.003 609 78 −0.999 993 5
1271.5980 1193.6440 496.015 548 706 055 −0.038 438 56 −0.999 261 0

As presented in Fig. 5, the self-driving AEV follows

the first lane based on the prevailing traffic conditions.
As presented in Fig. 6, the self-driving AEV is on the
track and it is found to drive safely without any
untoward incidents. The car which shows dotted green
lines in front of it indicates the driverless vehicle being
tested. As of now, the system has provision for safe
driving. However, if any vehicle comes from behind
intentionally to hit the driverless vehicle, it has no
alerting system. The present work focuses on path
planning and changing lanes as needed based on the
traffic in front of driverless car.
As presented in Fig. 7, the self-driving AEV is
moving in a particular lane and it is traveling the Fig. 6 Self-driving car on the track.

Fig. 5 Self-driving car on track in the first lane. Fig. 7 Self-driving car in the centre lane.
 1782 Tsinghua Science and Technology, December 2024, 29(6): 1776−1784

distance without any incidents

As presented in Fig. 8, the self-driving AVE is
moving in a particular lane and it is traveling the
distance without any incidents. It also demonstrates
lane change as needed.
As presented in Fig. 9, the self-driving AVE is
moving in a particular lane and it is traveling the
distance without any incidents. It also demonstrates
going straight in the lane as it is having no obstacles.
As presented in Fig. 10, the self-driving AVE is
moving in a particular lane and it is traveling the
distance without any incidents. It also demonstrates
going straight in the lane with centre of the lane
movement.
Fig. 10 Self driving car showing centre lane movement.
4 Conclusion and Future Work
framework has provision for different modules for
In this paper, we propose and implement an AI-based learning and gaining knowledge. The modules include
framework based on deep learning comprising of localization, high level path planning, low level path
multiple models to arrive at safe driving decisions. The planning, reinforcement learning, transfer learning,
power control, and speed control. We propose an AI-
based framework that has multiple deep learning
models to support decision making in autonomous
vehicles. MML-DM algorithm leverages multiple deep
learning models to make accurate decisions in
autonomous driving scenarios. A simulation prototype
has been made with an open source dataset. The dataset
is used as a highway map, which helps in simulation of
navigation. The dataset contains data pertaining to
waypoints that reflect the trajectory of the driving track
in highway. The implementation results reveal that the
AI framework is working towards safe driving. In
future, we improve it further to have energy efficient
Fig. 8 Self-driving car involved in the process of left lane
models and optimize energy distribution.
change. Acknowledgment
The authors would like to express their gratitude to the
Ministry of Higher Education Malaysia for funding this
research project through Fundamental Research Grant
Scheme (FRGS) (No. FRGS/1/2022/TK02/UCSI/02/1)
and also to UCSI University, Malaysia.
References

[1] Y. Li, G. Yin, W. Zhuang, N. Zhang, J. Wang, and K.

Geng, Compensating delays and noises in motion control
of autonomous electric vehicles by using deep learning
and unscented Kalman predictor, IEEE Trans. Syst. Man
Cybern. Syst., vol. 50, no. 11, pp. 4326–4338, 2020.
Fig. 9 Self-driving car in the middle lane going straight in [2] D. Strugar, R. Hussain, M. Mazzara, V. Rivera, J. Y. Lee,
the lane. and R. Mustafin, On M2M micropayments: A case study
 Gajula Ramesh et al.: Artificial Intelligence (AI) Enabled Future Wireless Electric Vehicles with Multi-Model... 1783

of electric autonomous vehicles, in Proc. 2018 IEEE Int. and Y. Zhang, Deep reinforcement learning based direct
Conf. on Internet of Things (iThings) and IEEE Green torque control strategy for distributed drive electric
Computing and Communications (GreenCom) and IEEE vehicles considering active safety and energy saving
Cyber, Physical and Social Computing (CPSCom) and performance, Energy, vol. 238, p. 121725, 2022.
IEEE Smart Data (SmartData), Halifax, Canada, 2018, pp. [14] J. A. Sanguesa, V. Torres-Sanz, P. Garrido, F. J. Martinez,
1697–1700. and J. M. Marquez-Barja, A review on electric vehicles:
[3] Y. Jeong, S. Son, E. Jeong, and B. Lee, An integrated self- Technologies and challenges, Smart Cities, vol. 4, no. 1,
diagnosis system for an autonomous vehicle based on an pp. 372–404, 2021.
IoT gateway and deep learning, Appl. Sci., vol. 8, no. 7, [15] J. M. Tien, The Sputnik of servgoods: Autonomous
p. 1164, 2018. vehicles, J. Syst. Sci. Syst. Eng., vol. 26, no. 2, pp.
[4] K. Min, G. Sim, S. Ahn, M. Sunwoo, and K. Jo, Vehicle 133–162, 2017.
deceleration prediction model to reflect individual driver [16] H. Kim, H. Pyeon, J. S. Park, J. Y. Hwang, and S. Lim,
characteristics by online parameter learning for Autonomous vehicle fuel economy optimization with deep
autonomous regenerative braking of electric reinforcement learning, Electronics, vol. 9, no. 11, p.
vehicles, Sensors, vol. 19, no. 19, p. 4171, 2019. 1911, 2020.
[5] W. Qi, Y. Luo, G. Wu, K. Boriboonsomsin, and M. J. [17] M. Jayasuriya, G. Dissanayake, R. Ranasinghe, and N.
Barth, Deep reinforcement learning-based vehicle energy Gandhi, Leveraging deep learning based object detection
efficiency autonomous learning system, in Proc. 2017 for localising autonomous personal mobility devices in
IEEE Intelligent Vehicles Symp., Los Angeles, CA, USA, sparse maps, in Proc. 2019 IEEE Intelligent
2017, pp. 1228–1233. Transportation Systems Conf., Auckland, New Zealand,
[6] A. Miglani and N. Kumar, Deep learning models for
2019, pp. 4081–4086.
traffic flow prediction in autonomous vehicles: A review, [18] I. W. Damaj, D. K. Serhal, L. A. Hamandi, R. N. Zantout,
solutions, and challenges, Veh. Commun., vol. 20, p. and H. T. Mouftah, Connected and autonomous electric
100184, 2019. vehicles: Quality of experience survey and taxonomy, Veh.
[7] P. Dixit, P. Bhattacharya, S. Tanwar, and R. Gupta,
Commun., vol. 28, p. 100312, 2021.
Anomaly detection in autonomous electric vehicles using [19] M. Ahmed, Y. Zheng, A. Amine, H. Fathiannasab, and Z.
AI techniques: A comprehensive survey, Exp. Syst., vol. Chen, The role of artificial intelligence in the mass
39, no. 5, p. e12754, 2022.
adoption of electric vehicles, Joule, vol. 5, no. 9, pp.
[8] W. Li, H. Cui, T. Nemeth, J. Jansen, C. Ünlübayir, Z. Wei,
2296–2322, 2021.
L. Zhang, Z. Wang, J. Ruan, H. Dai, et al., Deep
[20] K. Yu, L. Lin, M. Alazab, L. Tan, and B. Gu, Deep
reinforcement learning-based energy management of
learning-based traffic safety solution for a mixture of
hybrid battery systems in electric vehicles, J. Energy
autonomous and manual vehicles in a 5G-enabled
Storage, vol. 36, p. 102355, 2021.
[9] H. Jahangir, S. S. Gougheri, B. Vatandoust, M. A. Golkar, intelligent transportation system, IEEE Trans. Intell.
A. Ahmadian, and A. Hajizadeh, Plug-in electric vehicle Transp. Syst., vol. 22, no. 7, pp. 4337–4347, 2021.
[21] X. Qi, Y. Luo, G. Wu, K. Boriboonsomsin, and M. Barth,
behavior modeling in energy market: A novel deep
Deep reinforcement learning enabled self-learning control
learning-based approach with clustering technique, IEEE
for energy efficient driving, Transp. Res. Part C, vol. 99,
Trans. Smart Grid, vol. 11, no. 6, pp. 4738–4748, 2020.
[10] G. Bhatti, H. Mohan, and R. R. Singh, Towards the future pp. 67–81, 2019.
of smart electric vehicles: Digital twin technology, Renew. [22] I. S. Sorlei, N. Bizon, P. Thounthong, M. Varlam, E.
Sustain. Energy Rev., vol. 141, p. 110801, 2021. Carcadea, M. Culcer, M. Iliescu, and M. Raceanu, Fuel
[11] L. Wan, Y. Sun, L. Sun, Z. Ning, and J. J. P. C. Rodrigues, cell electric vehicles—a brief review of current topologies
Deep learning based autonomous vehicle super resolution and energy management strategies, Energies, vol. 14,
DOA estimation for safety driving, IEEE Trans. Intell. no. 1, p. 252, 2021.
Transp. Syst., vol. 22, no. 7, pp. 4301–4315, 2021. [23] Cityscapes dataset, https://www.cityscapes-dataset.com/,
[12] N. Jinil and S. Reka, Deep learning method to predict 2023.
electric vehicle power requirements and optimizing power [24] T. Chen, R. Xu, Y. He, Y. Xia, and X. Wang, Learning
distribution, in Proc. 2019 5th Int. Conf. on Electrical user and product distributed representations using a
Energy Systems, Chennai, India, 2019, pp. 1–5. sequence model for sentiment analysis, IEEE Comput.
[13] H. Wei, N. Zhang, J. Liang, Q. Ai, W. Zhao, T. Huang, Intell. Mag., vol. 11, no. 3, pp. 34–44, 2016.
 1784 Tsinghua Science and Technology, December 2024, 29(6): 1776−1784

Gajula Ramesh received the BEng degree Shayla Islam received the BEng degree in
in information technology from RGMCET, computer science and engineering from
Kurnool, India, and the MEng and PhD International Islamic University
degrees from JNTUA College of Chittagong, Bangladesh, the MEng and
Engineering, Ananthapuramu, India. He is PhD degrees from International Islamic
currently working as an associate professor University Malaysia (IIUM), Malaysia in
at Department of Computer Science and 2012 and 2016, respectively. She is
Engineering, Gokaraju Rangaraju Institute currently an associate professor at UCSI
of Engineering and Technology, India. His main research University, Malaysia. In consequences, she was awarded the
interests include software engineering, machine learning, and Young Scientist Award for the contribution of research paper at
big data. He has been teaching since 2008. He has published the Second International Conference on Green Computing and
several papers in various international journals/conferences. He Engineering Technologies (ICGCET’16) in 2016, organized by
is life member in IAENG. Aalborg University, Denmark. She is a senior member of IEEE.

Anil Kumar Budati received the BEng Abdullah Al-Atwai received the BEng
degree from JNTU, India 2007, the MEng degree in computer engineering from
degree from JNTU, India in 2010, and the KFUPM University, Kingdom of Saudi
PhD degree in electronics and Arabia, the MEng degree in network
communication engineering from GITAM engineering from RMIT University,
Deemed to be University, India in 2019. Australia, and the PhD degree from Brunel
He is currently working as an associate University, UK in 2020. He is currently
professor at Department of ECE, Koeru chairing Department of Computer Science,
Lakshmaiah Education Foundation (deemed to be University), Applied College, University of Tabuk, Kingdom of Saudi
India, and a postdoctoral researcher at Institute of Computer Arabia. His research interests include Wireless Sensor Networks
Science and Digital Innovation (ICSDI) UCSI University, (WSN), IoTs, and Software Defined Networking (SDN).
Malaysia. He has more than 13 years of experience in teaching
and six years of experience in research. Besides, he has
published 50 research articles in highly reputed publisher
journals and conferences indexed in SCIE, SCOPUS, etc.

Louai A. Maghrabi received the BEng

degree in computer science from Lebanese
American University, Lebanon, the MEng
degree in information technology from
University of West of England, UK, and
the PhD degree in cybersecurity from
Kingston University, UK. He is currently
an assistant professor at Department of
Software Engineering, College of Engineering, University of
Business and Technology, Kingdom of Saudi Arabia. His
research interests include cybersecurity, risk assessment,
cryptography, artificial intelligence, machine learning, the IoT,
blockchain, drones, metaverse, quantum computing, and game
theory. He received the best research paper award in 2021.