A Mini Review of the Impacts of Machine Learning on Mobility Electrifications

Abstract

Electromobility contributes to decreasing environmental pollution and fossil fuel dependence, as well as increasing the integration of renewable energy resources. The increasing interest in using electric vehicles (EVs), enhanced by machine learning (ML) algorithms for intelligent automation, has reduced the reliance on. This shift has created an interdependence between power, automatically, and transportation networks, adding complexity to their management and scheduling. Moreover, due to complex charging infrastructures, such as variations in power supply, efficiency, driver behaviors, charging demand, and electricity price, advanced techniques should be applied to predict a wide range of variables in EV performance. As the adoption of EVs continues to accelerate, the integration of ML and especially deep learning (DL) algorithms will play a pivotal role in shaping the future of sustainable transportation. This paper provides a mini review of the ML impacts on mobility electrification. The applications of ML are evaluated in various aspects of e-mobility, including battery management, range prediction, charging infrastructure optimization, autonomous driving, energy management, predictive maintenance, traffic management, vehicle-to-grid (V2G), and fleet management. The main advantages and challenges of models in the years 2013–2024 have been represented for all mentioned applications. Also, all new trends for future work and the strengths and weaknesses of ML models in various aspects of mobility transportation are covered. By discussing and reviewing research papers in this field, it is revealed that leveraging ML models can accelerate the transition to electric mobility, leading to cleaner, safer, and more sustainable transportation systems. This paper states that the dependence on big data for training, the high uncertainty of parameters affecting the performance of electric vehicles, and cybersecurity are the main challenges of ML in the e-mobility sector.

1. Introduction

Transportation infrastructure is undergoing a revolution thanks to electric mobility, which also addresses the need for sustainable solutions in the face of urbanization and climate change. By 2030, the size associated with the global electric mobility market is projected to reach USD 1229.47 billion [1]. In 2023, about 14 million new electric vehicles were registered worldwide, increasing the total number of electric vehicles on the road to 40 million [1,2]. Figure 1 shows the growth rate of electric cars in the world, which has accelerated in recent years [2].

This transition plays a vital role in global efforts to combat climate change by significantly reducing the carbon footprint associated with transportation [3]. Research studies have showcased the potential for substantial reductions in air pollutants and greenhouse gas emissions through the widespread adoption of electric vehicles.

Beyond its environmental impact, mobility electrification catalyzes technological innovation, particularly in crucial areas such as battery technologies, power electronics, and smart grid integration. This transformative impact extends to the automotive industry, fostering advancements beyond a mere change in power source. The integration of EVs into smart grids further enhances their significance by enabling bidirectional energy flow, contributing to grid stability and resilience.

In essence, mobility electrification emerges as a cornerstone in the broader transition toward a sustainable, technologically advanced, and economically vibrant future, with its effects rippling across environmental, technological, and economic domains [4,5,6,7,8]. Environmentally, electrified mobility promises to significantly reduce greenhouse gas emissions, mitigate air pollution, and curb dependence on fossil fuels, thereby fostering cleaner and healthier communities. Technologically, the integration of EVs and associated infrastructure drives innovation in energy storage, charging solutions, and smart transportation systems, paving the way for more efficient, interconnected, and autonomous mobility networks. Economically, the electrification of mobility opens new avenues for innovation, investment, and economic development, positioning nations and industries at the forefront of the burgeoning electric vehicle market. Consequently, these advancements underscore the pivotal role of mobility electrification in shaping a more sustainable and resilient for generations to come.

1.1. Why Machine Learning in Transportation Mobility

In the context of mobility electrification, several uncertain parameters may impact the planning, implementation, and operation of EV infrastructure and systems, including charging demand, energy supply, battery technology, and consumer behaviors, which should be addressed by advanced techniques. ML and DL are subsets of artificial intelligence that use neural networks with multiple layers to learn representations of data automatically. In electric mobility, ML/DL refers to the application of advanced neural network techniques to analyze and optimize various aspects of EVs and charging infrastructure. ML/DL plays a vital role in advancing electrical mobility by improving battery management, enabling autonomous driving (AD), optimizing energy efficiency, enhancing charging infrastructure, and facilitating vehicle design and performance optimization. For example, by predicting battery degradation, managing charging cycles, and optimizing energy usage, ML/DL helps enhance the efficiency and longevity of EV batteries.

Many researchers have studied the applications of ML/DL in e-mobility. The authors in [9] emphasize the significance of integrating ML/DL with the Internet of Things (IoT), encompassing various aspects of smart cities such as urban modeling, infrastructure, transportation, governance, sustainability, and more. Technology selection, administrative attitudes, real-time data processing, implementing lightweight ML techniques, and dealing with limited and heterogeneous datasets are known as challenges in these applications.

The study in [10] comprehensively analyzes the advantages and challenges related to smart cities, with a specific focus on the AI-centric smart mobility paradigm. This study concisely defines smart mobility, elucidates research methodologies and transition management, and offers invaluable insights for academics, policymakers, and urban scientists. Through strategic planning and fostering trust, the focus is placed on promoting sustainable urban growth. The research in [11] delves into Korea’s smart mobility strategy, with a focus on leveraging automated vehicle technologies and communication to enhance throughput, reduce traffic, and improve safety. The proposed strategy is divided into eight modules that cover various topics, including vehicle driving control technology (VDCT), traffic flow analysis, on-demand reserved lane system, congestion-free management, traffic signal management, ready-to-go pavement, and passage reservation.

The research in [12] explores methods of vehicle research, including open-innovation approaches, lead users, crowdsourcing, innovation with communities, and traditional market research methods, including focus groups, panels, and discrete innovation approaches (big data analyses). The authors investigate the impact of electrification and digitalization on traditional manufacturers as they confront challenges arising from evolving consumer expectations and innovations from companies like Apple and Tesla. The study assesses strategic choices while considering factors such as profitability and climate change, providing a comprehensive perspective on disruptive trends in the future of the automotive industry.

The research in [13] focuses on EVs as distributed energy resources in transactive energy systems, highlighting their potential. It delves into the challenges and benefits of vehicle-to-home (V2H) and vehicle-to-grid (V2G) technologies, emphasizing the necessity for innovation in communication, software, and hardware to ensure successful integration. The study explores the current state of the EV market, the impact of EV charging on grid load profiles, and the feasibility of EVs providing grid services.

The authors in [14] review optimizing energy consumption using IoT networks with a focus on the use of a hybrid metaheuristic strategy that combines the whale optimization algorithm (WOA) with simulated annealing (SA). The approach is designed to identify the best cluster head (CH) for IoT networks by taking into account several performance criteria such as the temperature, live node count, load, energy consumption, and cost function. The simulation results demonstrate the hybrid solution’s effectiveness in maximizing IoT network energy usage.

The work in [15] explores the critical roles of the IoT and ML in the development of data-centric smart cities. The study addresses applications and challenges, highlighting the critical importance of addressing security, privacy, and ethical concerns. In addition, it emphasizes the revolutionary influence of the IoT and ML in various urban domains, including mobility, environment, citizen welfare, and governance. The study highlights the challenges in adopting 5G-based features and emphasizes the need for balanced governance. It also underscores creative alternatives such as localization, talent development, finance models, and inclusive smart city design.

A research study on smart cities [16] investigates how artificial intelligence and ML can be used to manage data-intensive applications through the IoT. Deep reinforcement learning (DRL) techniques are being employed across various domains, such as smart grids, 5G/B5G communication, cybersecurity, smart transportation, and healthcare. The study emphasizes the many applications of supervised, reinforcement, and unsupervised learning in smart city contexts, with a focus on cybersecurity, big data standards in smart grids, enhanced intelligent transportation system efficiency, and consistent data gathering.

1.2. Main Contributions of the Work

Considering the importance of artificial intelligence and its ever-increasing developments, this paper provides a mini review of the applications and effects of ML/DL in mobility electrification. Electrification of transportation requires interdependency between the power grid and transportation. This includes various aspects such as battery management, energy management, efficient charging infrastructure, preventive maintenance, electricity markets, cybersecurity, and V2G, which require collecting, processing, and analyzing a large amount of information. This article reviews recent research on applying ML/DL models in the mentioned aspects. The characteristics and prerequisites of each of the studies in the field of e-mobility are examined in detail. In addition, the weaknesses and obstacles of each work are determined in line with new trends for the development of future works. This paper also briefly introduces training models in the field of transportation electrification.

This mini review serves as a valuable resource for engineers and researchers across various academia and industries, particularly in transportation, energy management, and smart grid technology. By systematically analyzing the applications of ML/DL in mobility electrification, this paper not only highlights current advancements but also identifies gaps and challenges in the existing research. Ultimately, the findings can assist researchers in developing more efficient, resilient, and integrated solutions for the electrification of transportation, paving the way for sustainable mobility advancements.

1.3. Paper Organization

The rest of this paper is organized as follows: Section 2 provides ML/DL models and all required steps for prediction based on the most-used models in e-mobility. All applications and impacts of ML/DL in mobility electrification, including energy management, traffic management, V2G, battery management, charging demand, range prediction, charging infrastructure optimization, AD, predictive maintenance, and fleet management, are evaluated comprehensively in Section 3. Section 4 discusses the advantages, disadvantages, and limitations of ML/DL in e-mobility. In Section 5, new trends and future works in the field of ML/DL impacts on mobility electrification are represented. Finally, Section 5 concludes the paper.

2. ML/DL Models

Since the introduction of the initial ML/DL models, there has been a significant progression in both intelligence and system modeling. ML methods, specifically DL techniques, quickly became popular and had more uses. The most commonly employed DL techniques encompass Convolutional Neural Networks (CNNs) [17], Recurrent Neural Networks (RNNs), Denoising Autoencoders (DAEs) [18], Deep Belief Networks (DBNs), and long short-term memory (LSTM) networks [19].

Table 1. The architecture of the most-used DL models.

Technique	Architecture
CNN
RNN
DAE
DBN
LSTM

shows the architecture of the most-used DL models, which can be applied to various aspects of e-mobility and transportation electrification.

Generally, DL models contain seven steps as follows:

• Data collection: gather data from various resources.
• Data preprocessing: Clean the collected data and preprocess them to train the DL model. This may involve removing outliers, normalizing the data, and splitting them into training, validation, and test sets.
• Model architecture: Design a DL model architecture suitable for predicting based on the collected data. This could be an RNN, CNN, DAE, DBN, or LSTM, depending on the nature of the data and the desired prediction accuracy (refer to Table 1).
• Training: Train the model using the preprocessed data. During training, the model learns to map the input data. This involves adjusting the parameters of the model (e.g., weights and biases) using optimization algorithms.
• Evaluation: evaluate the trained model on the validation set to assess its performance and fine-tune hyperparameters if necessary.
• Testing: Finally, the trained model will be tested on the unseen test data to evaluate its generalization performance. This step ensures that the model can accurately predict.
• Deployment: Once the model has been trained and tested successfully, deploy it for real-time prediction. The model continuously analyzes data and provides accurate predictions.

Figure 2 shows the overview of the DL model in the field of battery management, traffic management, and charging stations.

3. Mobility Electrification with ML/DL

ML/DL algorithms play a crucial role in optimizing various aspects of electromobility, from battery management to AD. By analyzing vast amounts of data generated by EVs, ML models can predict battery health, optimize charging strategies, and improve energy efficiency. These algorithms enable electric vehicle manufacturers to develop smarter and more reliable battery management systems, ultimately enhancing the performance and longevity of EV batteries.

Moreover, ML/DL techniques are being applied to optimize energy consumption, enhance V2G integration, and enable predictive maintenance in EVs. In this section, a comprehensive review of mobility and transportation electrification with ML/DL is represented. This survey includes various aspects of battery management, range prediction, charging infrastructure optimization, AD, energy management, predictive maintenance, traffic management, V2G, and fleet management and discusses the most-used ML/DL techniques in these aspects.

3.1. Battery Management

A battery management system (BMS) is crucial to guarantee the efficient, reliable, and safe performance of EVs. In other words, battery management is the main component in EVs that decides the optimal operation. Several review studies in the area of battery management of EVs use ML/DL methods. For instance, the review in [20] explores techniques for accurately predicting the capacity of lithium-ion batteries in electric and hybrid vehicles, emphasizing the importance of reliable battery management. It investigates various approaches, including EMF-based correlations, deep visual analytics (DVA)/incremental capacity analysis (ICA) methodologies, adaptive joint estimation techniques, electrochemical model approaches, Kalman filters, and ML methodologies such as relevance vector machine (RVM) and particle filter (PF). The research highlights challenges such as computational effort and precise voltage sample collection, underscoring the critical need for accurate capacity estimation in calculating range, energy storage, and monitoring battery health. Another review [21] centers on the evolution of a BMS into multifunctional integrated systems, highlighting the importance of accurate battery modeling for efficient control. The review explores critical technologies, including fault diagnostics, energy equalization, thermal control, and condition assessment. It investigates the role of a BMS in optimizing driving and discusses the advantages and disadvantages of energy storage.

The paper examines multiple modeling techniques and state estimation methodologies, emphasizing their significance for the economy and safety of EV batteries. A systematic review of battery modeling and state estimation methodologies for an advanced BMS is presented in [22]. The review encompasses various techniques, including measurement and analysis, empirical fitting, Bayesian methods, ML techniques, and more. The work in [23] conducts a comprehensive investigation into ML approaches for a BMS, focusing on techniques for remaining useful life (RUL) estimation and fault identification. The methods are categorized into supervised, unsupervised, and RL approaches. Various techniques are discussed, including classic neural network (NN), Modern NN, support vector machines (SVMa), RVM, Gaussian process regression (GPR), and model-based methods. Specific examples of ML applications in BMSs, such as the use of SVMs for battery health diagnosis and prognostics, as well as ML/DL algorithms for accurate multi-forward-step voltage prediction in battery systems, were investigated. Figure 3 shows a general schematic of ML application for the BMS area.

A particular focus on the SOC and state of health (SOH) of batteries in EVs was addressed in some research [24]. For instance, an alternative survey in [25] extensively explores ML algorithms for predicting the SOC and SOH of batteries in EVs. The methodologies discussed include feedforward neural networks (FNNs), RNNs, radial basis function (RBF) neural networks, extreme learning machines (ELMs), SVMs, and Bayesian networks (BNs). The accuracy of these methods is presented, with RBFs exhibiting the highest error rate at 2.2%. However, FNNs and RNNs show promising results, with average errors of 0.7% and 0.5%, respectively. Later on, in [26], an examination is conducted on the application of ML/DL algorithms for estimating the SOC, SOH, and RUL of lithium-ion batteries in EVs. The review encompasses a range of ML/DL algorithms, including DNNs, DBNs, RNNs, LSTMs, gated recurrent unit (GRU), autoencoder, and CNNs. To construct an SOC estimator for lithium-ion batteries and real-time battery management applications in EVs, [27] employ a two-step ML approach. In the first stage, a genetic-fuzzy clustering technique is utilized to determine the model topology and antecedent parameters. In the second stage, the recursive least-squares algorithm is employed to identify subsequent parameters. The backpropagation learning algorithm optimizes both antecedent and consequent parameters simultaneously during the second phase. The model demonstrates high accuracy, with a root mean squared error (RMSE) of 1.68% and a probability of 98.12%. Recently, the work described in [28] presents a comprehensive framework for EV battery management, with a specific focus on SOH and SOC prediction. The framework integrates a data interface, cloud-based virtual models, and an on-vehicle BMS, utilizing a digital twin on Microsoft Azure, continuous learning for SOH, and a Kalman filter for SOC estimation. The implementation achieves an MSE of 0.022 for SOH, and realistic simulations that combine data-driven, equivalent circuit, and electrochemical models contribute to accurate forecasts.

In conclusion, effective battery management is crucial for enhancing EV performance and safety, with ML and DL techniques playing a key role in capacity estimation and state assessment. Recent reviews highlight the evolution of BMSs toward multifunctional frameworks, focusing on integrating advanced modeling approaches for accurate SOC and SOH predictions. The ongoing development of sophisticated algorithms shows significant potential in optimizing battery performance, ultimately contributing to the broader adoption and efficiency of EV technologies.

3.2. Range Prediction

Electric vehicles face a significant challenge in terms of their driving range, which is a major obstacle to their widespread adoption. To alleviate drivers’ range anxiety, it is essential to predict the remaining distance that an EV can cover accurately. There is considerable interest in using ML/DL algorithms to predict the range of electric mobility, as this addresses a critical challenge in EV adoption. By analyzing large datasets containing real-time vehicle performance metrics, environmental variables, and driving patterns, ML/DL models offer superior predictive capabilities than traditional methods. The advancement of sensor technology and data analytics further enhances the accuracy and adaptability of these models, making them capable of learning from diverse driving scenarios. The use of ML/DL to predict EV range not only alleviates user concerns but also can potentially optimize vehicle efficiency and performance, paving the way for sustainable transportation solutions.

For example, an ML-based model for estimating an EV’s remaining range is presented in [29], which makes use of the combination of XGBoost and LightGBM algorithms. The aim is to obtain the input features, such as cumulative output energy of the motor and the battery, different driving patterns, and temperature of the battery, and to predict the driving distance. The results showed a small prediction range within [−0.8, 0.8]. A review in [30] describes several prediction techniques, including Control-Based Driving Style, Discretization Scheme, Data-Driven algorithm for Large-Scale Road Networks, and AI-based solutions with related accuracy metrics after looking at variables like the BMS and driver behavior to provide accurate EV range forecasts. With a focus on Kernel Adaptive Filtering (KAF) methods, the paper [31] investigates real-time ML approaches for power and mission energy prediction in EVs. In particular, it contrasts the NN and basic linear filter performance with that of the fixed-budget quantized kernel least mean squares (QKLMS-FB) and the kernel recursive least-squares tracker (KRLS-T). According to the published accuracy measures, the KRLS-T has a normalized root mean squared error (NRMSE) of approximately 90% after 8000 training steps.

A variety of approaches, including regression, physics-based models, and ML algorithms such as an RBF-NN, CNN, LSTM, and DNN, are explored in [32] for predicting the range of EVs. Notably, a two-stage LSTM-DNN mixture model that updates battery energy usage based on traffic circumstances and integrates long-term dynamics is developed and offers considerable gains in the accuracy of EV range prediction and valuable application in vehicle control systems. The model outperforms regression-based techniques with a remarkable 2–3 km range prediction accuracy, having been trained using 160,000 km of driving data.

In order to predict the demand for EV charging, the study in [33] examines a number of ML techniques, including federated learning, ensemble learning, temporal encoder–decoder–LSTM, hybrid deterministic-stochastic methodology, AI Model Predictive Control (MPC), RNNs, deep inference framework, and backpropagation model. With an accuracy of 97.14%, the recommended empirical mode decomposition (EMD)–arithmetic optimization algorithm (AOA)–distributed long short-term memory (DLSTM) predictor model is highly successful. This model outperforms in accuracy, error rate, and computing efficiency by combining the arithmetic optimization algorithm (AOA), LSTM recurrent model, and empirical mode decomposition (EMD).

Another work in [34] uses regression models, NNs (such as Bidirectional LSTM (Bi-LSTM), LSTM, and GRU), RBF-NNs, and feature scaling techniques (Min–Max and Z-score). Optimization methods like gradient descent and the Adam optimizer are investigated for sequential models like dense layers and simple RNN. With an impressive R-squared score of 0.99998 and a low MSE of 0.029 km, the Bi-LSTM model proves to be very accurate.

In a recent review [35], four main obstacles to predicting the remaining driving range (RDR) have been outlined. These challenges include battery state estimation, driving behavior classification and recognition, driving condition prediction, speed prediction, and RDR calculation methods. A proposed driving range prediction method is introduced to address these challenges, leveraging vehicle–cloud collaboration. This approach integrates the strengths of cloud computing and ML, offering potential avenues for future research in this field.

In summary, precise driving range predictions are vital for reducing range anxiety and encouraging wider adoption of EVs. The use of ML and DL algorithms has demonstrated great potential in improving range prediction accuracy by utilizing real-time data from multiple sources, such as vehicle performance and environmental factors. With ongoing progress in sensor technology and data analytics, these predictive models will be key in optimizing EV efficiency and facilitating the shift toward more sustainable transportation solutions.

3.3. Charging Infrastructure Optimization

The widespread adoption of EVs faces challenges stemming from an inadequate charging infrastructure, optimal positioning of charging stations, and efficient charge scheduling. Even aspects like vehicle price and driving range can be addressed to some extent with a robust charging network. In recent years, researchers have focused on optimizing the placement and sizing of EV charging stations, employing various approaches, objective functions, and optimization algorithms to tackle this issue.

For example, to identify optimal locations for slow-charging EV infrastructure in central Ohio, the study in [36] introduces a simulation-based optimization approach. The research incorporates a linear model to evaluate the probability of EV adoption by integrating EV flow analysis, a charging simulation model, and linear integer programming.

In the review in [37], the optimal locations for EV charging stations are thoroughly examined using modeling techniques such as node-based, tour-based, and path-based approaches. The review assesses a range of optimization algorithms, including Firefly, Ant colony, Particle Swarm, genetic, and Differential Evolution. Additionally, the study concludes with a detailed analysis of the trends in charging infrastructure development, highlighting different problem formulations and optimization techniques. The study described in [38] explores the optimal siting of EV charging stations in urban settings, utilizing a multi-objective model. This model concurrently addresses two optimization criteria: maximizing the number of accessible households and minimizing the overall energy costs associated with e-transportation.

Another study in [39] focuses on optimizing the EV charging infrastructure in Manhattan through the utilization of integer programming and a genetic algorithm. The analysis takes into consideration the M/M/sj/N queuing model, accounting for constraints on charging service capacity and finite queue length.

In the last few years, using ML methods has become increasingly interesting in many areas related to EVs. For example, the paper in [40] introduces a data-driven smart charging strategy for fleets of EVs, utilizing ML methods such as XGBoost, linear regression, and NNs. The study focuses on maximizing energy usage and presents a comprehensive approach for training regression models, data preparation, and integration into a smart charging algorithm. XGBoost exhibits superior accuracy compared to NNs and linear regression, achieving a mean absolute error of 126 W. Another study in [41] explores the application of RL techniques, specifically ANNs, double deep q-networks (DQNs), and dueling network topologies, for optimizing EV fleets in smart electricity markets. The study demonstrates that RL agents enable accurate forecasting, profitable bidding strategies, improved smart grid profitability, and effective virtual power plant portfolio management. Recently, the paper in [42] explores the integration of EV charging into smart grids using ML techniques, as shown in Figure 4. It emphasizes the significance of cutting-edge technologies and sustainable energy practices in addressing the challenges associated with the increasing popularity of EVs. The study highlights the importance of effective information management for energy sustainability and governance in smart cities. The study suggests utilizing the LSTM model for optimal EV fleet reconfiguration and charging, as it achieves the highest accuracy in classifying charging stations and vehicle speeds. The results indicate that the RF model achieves 94% accuracy in charging station classification, while the LSTM model achieves the same accuracy. For charging speed classification, the LSTM model achieves 93% accuracy. The LSTM model also achieves 94% accuracy in classifying the usage of a 10% global weighted network for EV stations and 93% accuracy in classifying pricing structures.

To sum up, optimizing charging infrastructure is crucial for the widespread adoption of EVs, as it addresses key challenges like station placement, sizing, and efficient charge scheduling. Recent progress in optimization algorithms and ML techniques has greatly improved the effectiveness of charging station strategies, allowing for better integration with urban areas and smart grids. By utilizing these innovative approaches, stakeholders can enhance accessibility and efficiency, ultimately supporting the growth of a more sustainable and resilient EV ecosystem.

3.4. Autonomous Driving (AD)

AD or self-driving is responsible for always managing and controlling the vehicle. Automatic control, sensor, and telematics technologies are the key to AD, which requires real-time prediction. A system known as Local Multiple CNN-SVM, LM-CNN-SVM, proposed in [43], presents a creative combination of multiple CNNs for local feature extraction along with an SVM for accurate classification in AD applications. This technique achieves remarkable accuracy on the Caltech-101 dataset, scoring 89.80% and 92.80% for item identification with 15 and 30 photos per class, respectively. The LM-CNN-SVM system surpasses existing SVM-based and ML/DL methods in addressing challenges such as background clutter, partial occlusion, and variations in illumination.

The importance of CNNs in the field of image recognition for AD, over more conventional methods since 2010, has been extensively studied in [44]. This research provides a comprehensive analysis of various techniques, including bag-of-features, scale-invariant feature transform (SIFT), histogram of oriented gradients (HOGs), and SVM, that enhance the effectiveness of CNNs in achieving high accuracy for image classification. CNNs outperform the HOG by 8%, achieving an impressive 3% miss rate in human identification. However, this research could further be enhanced by using the You Only Look Once networks for objective detections in images like those shown in Figure 5.

The survey [46] investigates the robustness and ethical implications of AD, while also listing possible advantages such as reducing pollution and preventing accidents. The research delves deeper into the use of sensors, system architecture, and several methods, including light detection and ranging (LiDAR)-based 3D object detection, simultaneous localization and mapping (SLAM), and GPS–inertial measurement unit (GPS-IMU) fusion. The effectiveness of perception strategies, including semantic segmentation, ML for 3D object detection, and image-based identification, is examined. Comprehensive research on AD is also conducted in another survey [47], which encompasses motion control, behavior arbitration, path planning, and scene perception. Notably, it highlights the remarkable success of the End-to-End CNN architecture, achieving an impressive 90.4% accuracy in real-time Tesla Autopilot assessments. The study investigates CNNs, RNNs, LSTMs, and DRLs. Furthermore, the paper explores the ongoing debate surrounding cameras and LiDAR sensors, with a specific focus on the role of ML/DL in positioning, semantic segmentation, and object detection.

ML/DL techniques for AD tasks, with a focus on robustness, online learning, fairness, and transparency, are considered in the survey [48]. The review achieves impressive area under the curve (AUC) values of 93.04, 88.64, and 79.27 in mainstream vehicle detection for simple, moderate, and hard examples. Furthermore, traffic sign detection demonstrates recall and precision scores ranging from 87.2% to 99.18%, while drowsiness detection achieves an accuracy of 75.57%. The review also explores prospective solutions for safety and computational complexity, including energy-efficient CNNs, sequence learning, generative adversarial networks, RL, and edge computing.

In another comprehensive review [49], thorough research is conducted on vehicle trajectory prediction for AD utilizing RNN and CNN models. Input representations include track histories, bird’s eye views, and raw sensor data. Output types encompass maneuver intention, unimodal/multimodal trajectories, and occupancy maps.

To enhance safety by ensuring accurate recognition of vehicles and pedestrians, the study [50] focuses on developing a high-fidelity virtual LiDAR sensor for AD simulators. The proposed model achieves this by adjusting the intensity based on empirical readings and simulated weather conditions, effectively reducing disparities between virtual LiDAR data and real-world LiDAR data on the computer-assisted real-time layout (CARLA) platform.

Recently, a study in [51] explores an integrated approach to motion planning and decision making in AD using a double-layer RL architecture. The proposed system combines MPC for motion planning with DRL for decision making, aiming to address the interdependency between these modules. Additionally, a comprehensive architecture for AD in the vehicle virtual reality metaverse was introduced, utilizing generative AI to generate realistic traffic data [52]. The proposed mechanism was aimed at enabling cooperative decision making among autonomous vehicles, remote sensing units (RSUs), and simulators. The system employs performance metrics such as the R2 score for trajectory predictions and generative score (ranging from 0 to 1) for AI models, emphasizing the importance of high accuracy in scenario modeling. Furthermore, research in [53] delves into the intricate connections between drivers’ perceptions of comfort and safety in conditional AD by studying their driving behaviors. The study systematically compares automated driving behavior (ADB), desired driving behavior (DDB), and personal driving behavior (PDB) by employing techniques such as data preprocessing, Likert scale grading, and verbal feedback.

To conclude, AD depends on advanced sensor technologies and ML/DL techniques for real-time vehicle control and management. Recent developments, including the LM-CNN-SVM for local feature extraction and various CNN architectures, have significantly improved object recognition and classification accuracy, which are critical for the safety and efficiency of AD systems. Additionally, ongoing research into ethical considerations, system robustness, and integrated decision-making frameworks underscores AD’s potential to enhance road safety and minimize environmental impact, paving the way for a more sustainable future in transportation.

A notable case study is Waymo, a leader in autonomous driving technology, which utilizes DL algorithms to process data from sensors and cameras. Their system has been trained on millions of miles of driving data, allowing it to navigate complex urban environments effectively. By employing RL techniques, Waymo’s vehicles can adapt to various driving conditions, improving safety and efficiency. This real-world application exemplifies how ML/DL can enhance autonomous driving capabilities, ultimately contributing to the overall efficiency of EV operations [54].

3.5. Energy Management

The energy management system of EVs is considered the most crucial part of EV scheduling, and it can influence both traffic and electrical grids. Advanced energy management can optimize the use of energy during charging, the cost of electricity purchased, and battery health; different energy management strategies could be applied to EVs, which are usually associated with high-level uncertainty.

The work in [55] explores multi-agent microgrid energy management by ML/DL with a CNN-GRU architecture to make precise day-ahead forecasts of market pricing, wind, solar, and load demand. To enhance convergence and achieve the optimal operating point of the microgrid in a distributed manner, an alternating direction method of multipliers (ADMM) is employed. Another work [56] presents an approach for the efficient management of renewable microgrid energy and highly accurate predictions of hybrid electric vehicle (HEV) charging. This is achieved through the combination of support vector regression (SVR) for predicting the amount of charging demand and modified dragonfly algorithm (MDA) for optimization. The SVR-MDA prediction results yield an MAPE of 0.978.

The work in [57] presents an energy management strategy in HEVs utilizing a DRL attached to a transfer learning technique. The approach focuses on applying the deep deterministic policy gradient (DDPG) algorithm to transfer knowledge between different types of HEVs, utilizing a network-based TL technique. This technique improves convergence efficiency by an average of 70% by initializing NNs in additional HEVs with front layers from a pre-trained DRL-based EMS for a Prius brand of EVs.

The authors in [58] introduce an EMS for plug-in HEVs using an enhanced MPC and ML/DL. Among the six prediction models considered, including exponential function, Markov chain, K-nearest neighbor (KNN), RF, DNN, and LSTM, the LSTM-based Improved Model Predictive Control (LSTM-IMPC) algorithm demonstrates the highest accuracy. Three EMSs, namely charge-depleting charge-sustaining (CD-CS), dynamic programming, and LSTM-IMPC, are evaluated under various driving cycles, with LSTM-IMPC showing superior fuel-saving rates for Worldwide light-duty test cycle (WLTC), new European driving cycle (NEDC), and real driving cycle (RDC): 3.81%, 5.6%, and 18.71%, respectively.

The review in [59] examines a DL-based energy management strategy for EVs in smart cities. The strategy utilizes RNNs such as LSTM and GRU. By leveraging real traffic data, RNNs accurately forecast EV trajectory and delay, making them effective for long prediction intervals. In another review [60], ML techniques for energy management in HEVs are examined and categorized into three groups: supervised, RL as semi-supervised learning, and unsupervised learning. Supervised approaches, including classification and regression algorithms such as NNs, SVMs, and RFs, and unsupervised approaches, including clustering and dimensionality reduction such as k-means, principal component analysis (PCA) [61], RL, utilizing DQN and DDPG, increase rewards based on actions [62].

The strategy introduced in [63] integrates DRL and computer vision into the EMS to enhance the fuel efficiency of HEVs. The DDPG-AN algorithm, which incorporates You Only Look Once (YOLO) for real-time visual input, generates optimal control policies for torque and speed based on traffic signals and nearby cars. This strategy achieves a fuel economy of 96.5%, surpassing conventional techniques such as DP and MPC.

A CNN is employed in [64] to predict pollutant emissions in HEVs and achieve high accuracy for both instantaneous and cumulative values. Additionally, the study introduces double deep-Q learning (DDQL) to optimize EMS using real-world driving data and DP. The results obtained from the co-simulation framework demonstrate that the final error in predicted CO₂ remains below 2.5%, while the final cumulative error for pollutants CO and HC emissions remains below 8.5%.

In conclusion, EMSs are pivotal for optimizing the performance of EVs, influencing both energy consumption and grid interactions. Recent advancements in ML and DL techniques, such as CNN-GRU architectures and LSTM-based predictive controls, demonstrate significant improvements in forecasting energy demands and optimizing charging strategies. By integrating these advanced methodologies, researchers are enhancing the efficiency and sustainability of EV energy management, ultimately contributing to the broader goals of smart grid integration and reduced environmental impact.

In the realm of battery management, Tesla’s advanced battery management system serves as an exemplary model. By leveraging real-time data analytics and ML, Tesla can predict battery health and optimize charging cycles based on driving patterns. This capability not only extends battery life but also enhances range prediction accuracy, allowing users to plan their journeys more effectively. The use of data from actual driving conditions illustrates the practical implications of theoretical concepts discussed in this review [65].

3.6. Predictive Maintenance

With its ability to provide preemptive intervention to minimize failures and provide insights into the health of vehicle components, predictive maintenance has become a vital part of EV management. The research in [66] examines the role of ML in predictive maintenance for automotive systems. This method employs data and analytics to forecast equipment failures, facilitating timely maintenance, minimizing downtime, and reducing costs. Based on the results, the ANN methods are suitable for complex automotive data and ensure precise and dependable maintenance predictions.

To remain profitable, the automotive industry depends on optimizing production time and delivering higher-quality products in less time and at a lower cost. However, maintenance time poses a significant challenge as it is one of the primary causes of efficiency loss. Hence, the research in [67] offers a comprehensive overview of predictive maintenance techniques in the automotive industry, to improve equipment reliability and reduce downtime. Advanced ML/DL techniques like LSTM and CNN are employed to achieve higher accuracy in predictive maintenance.

The research in [68] conducts a comprehensive review of predictive maintenance strategies that utilize ML techniques. The study identifies specific ML methods employed in various predictive maintenance contexts. For handling large datasets and preventing overfitting, RFs are chosen. ANNs excel in tasks based on historical data and demonstrate robustness. SVMs are valued for their high precision in data classification, and K-means clustering is also discussed.

As the world moves toward the electrification of vehicles, monitoring the health of lithium-ion batteries used in them becomes essential. Therefore, in the study conducted in [69], ML emerges as a powerful tool for addressing the challenges related to the prognostics and health management of batteries in real-world EV applications. The proposed models achieve impressive results, with an MAPE of 0.28% and a Root Mean Squared Percentage Error (RMSPE) of 0.55% when predicting the battery’s lifespan. The rapid increase in EVs presents a challenge for battery health prediction. In [70], an advanced predictive maintenance approach is proposed to achieve cost savings and increased profitability. This comprehensive method integrates maintenance data, geographic information system (GIS) information, the Cox Proportional Hazard Model (Cox PHM), and a Multiplicative LSTM (M-LSTM) network to enhance the accuracy of RUL prediction.

The research in [71] highlights that real-world scenarios involve dynamic distributions, which can cause model performance issues. To address this challenge, Continual Learning (CL) methods are introduced as a solution to adapt models over time and prevent knowledge loss. The study aims to seamlessly combine CL with PdM in order to mitigate the challenges posed by non-stationary environments efficiently.

Due to cost constraints, it is not feasible to continuously monitor all equipment, making it crucial to select an optimal inspection schedule. To this end, the methodology proposed in [72] dynamically determines the time interval between measurements (TIBeM) based on the criticality and reliability of each machine. This approach ensures accuracy by adapting the measurement intervals. The method introduces an objective tool for calculating the baseline TIBeM (referred to as TIBeMB) and allows for adjustments to specific intervals based on the machine’s status. The study in [73] delves into the role of ML in DTs for PdM, examining the existing literature and presenting practical applications. DT technology plays a crucial role in this research, enabling precise equipment status recognition and proactive fault prediction. The study addresses challenges and opportunities in integrating ML for fault diagnosis, RUL prediction, and health indicators.

To sum up, PdM has become a vital approach for improving the reliability and efficiency of EVs by using ML techniques to foresee equipment failures and optimize maintenance schedules. Recent research emphasizes the success of advanced ML models, including LSTM and CNN, in accurately predicting the health and lifespan of key components like lithium-ion batteries. By incorporating these predictive methods, the automotive industry can greatly minimize downtime and maintenance expenses, ultimately enhancing vehicle performance and promoting sustainability.

Furthermore, companies like Uber are utilizing predictive maintenance strategies to enhance fleet management. By analyzing historical data and employing machine learning models, Uber can predict vehicle maintenance needs before issues arise, thereby reducing downtime and operational costs. This application showcases how predictive analytics can lead to more efficient fleet operations, aligning with the broader goals of energy management and sustainability [74].

3.7. Traffic Management

Integrating intelligence transportation systems (ITSs) into roadway construction can effectively address transportation emissions and alleviate traffic congestion by optimizing existing infrastructure. The research in [75] emphasizes the enduring and significant role of ITS in transportation electrification, traffic control, safety, environmental sustainability, and real-time traffic monitoring.

Another work in [76] introduces the bundled causality engine (BCE) approach, which aims to characterize the causal relationships between electricity consumption, transportation networks, and weather data in urban areas. The study’s findings demonstrate that the integration of traffic data with historical electricity data significantly enhances the accuracy of 1-day-ahead load forecasting. The average error rate of the model can be reduced to 8% using the SVM method, while the DNN method achieves an error rate of approximately 18%.

The study in [77] proposes the utilization of XGBoost and Shapley Additive exPlanations (SHAP) algorithms to harness highway digitalization to achieve transportation infrastructure that is more easily accessible, reasonably priced, secure, and sustainable. The authors achieved a 99% accuracy rate and a 79% detection rate in real-time accident detection.

Accurate traffic flow forecasting is crucial for efficient smart city traffic management. Therefore, the research in [78] investigates various ML algorithms, including LSTM, BiLSTM, Prophet, and Transformer models, for short-term traffic flow prediction. Among the evaluated models, the Transformer model achieves the lowest MAE (5.613), while the LSTM and BiLSTM approaches have error rates of 5.722 and 5.638, respectively. In other research, Stacked AutoEncoder (SAE) [79] and LSTM [80] were used to learn generic traffic flow features and short-term traffic flow prediction.

The authors in [81] investigate the relationship between traffic conditions and energy/fuel consumption, specifically when transitioning from Internal Combustion Engine Vehicles (ICEVs) to EVs. The Virginia Tech Comprehensive Power-based EV energy consumption model (VT-CPEM) forecasts energy usage, exhibiting an average error of approximately 6% in comparison to the empirical data. To simulate fuel consumption in ICEVs, the CO₂MPAS model was employed. Remarkably, the model achieves an unbiased error of 4% in estimating fuel usage in 75% of scenarios.

The review in [82] explores ML/DL algorithms, including CNNs, which are adept at capturing spatial traffic dependencies, and RNNs, specifically LSTM. The review also discusses the benefits of Restricted Boltzmann Machines (RBMs) and SAE. Additionally, this study emphasizes the limitations of traditional models in capturing complex spatial–temporal connections and investigates the advantages and disadvantages of these ML/DL models in diverse spatial–temporal contexts.

The research in [83] introduces a DL-based car-following model called Memory, Attentiveness, and Prediction (MAP). This model incorporates various NN architectures to emulate human driving capabilities, including memory, attention selection, and prediction. ML techniques, such as Graph Neural Networks (GNNs), have proven to be valuable for traffic management. These techniques have been applied in Intelligent Modelling in Network Management and Orchestration [84] as well as Spatial-Temporal Cellular Traffic Prediction for 5G and beyond [85]. Building upon these advancements, a study [86] advocates for the integration of a Multi-Arm Bandit (MAB) algorithm and Software-Defined Networking (SDN) into a novel framework for traffic regulation in software-defined IOT networks. Through SDN orchestration, GNN, MAB, and an SDN controller, the architecture dynamically refines traffic management policies in response to real-time traffic patterns. The GNN model consistently achieves an accuracy level of approximately 97% across three datasets, demonstrating its remarkable ability to discern intricate traffic patterns.

To conclude, integrating ITS with advanced ML algorithms is crucial for optimizing traffic management and promoting the sustainability of transportation networks. Recent research showcases the effectiveness of ML techniques, including XGBoost, LSTM, and GNN, in improving traffic flow predictions and real-time accident detection, leading to reduced congestion and emissions. By adopting these innovative methods, cities can develop more efficient, responsive traffic management systems, ultimately fostering safer and greener urban environments.

3.8. V2G

With the use of V2G technology, energy from an EV’s battery may be returned to the power grid [87]. An EV battery may be depleted using V2G technology in response to various signals. This technology involves drawing surplus power from EVs into the grid. The work in [88], as one of the lead research projects, presents a detailed analysis of the global impact of V2G systems on the energy market by using a model called energy research and investment strategies) ERIS (. It investigates the implications on submarkets, alternate vehicle adoption, and renewable energy support while optimizing energy system costs across sectors. The study promotes an examination of the linkages between the energy system, V2G presence, and changes in global climate policy by presenting four possibilities: BaseV (with V2G system technologies and no-climate-policy), ClimateV (with V2G system technologies and climate policy scenario), BaseN (without V2G system technologies and climate policy scenario), and ClimateN (with climate policy scenario and without V2G system technologies).

A proposed model in [89] combines a microgrid optimization scheduling strategy based on deep q-learning with a mathematical representation of V2G technology, enabling the capture of user charging behavior and EV mobility patterns, which can effectively manage uncertainties and randomness. The research in [90] explores the advantages and challenges of integrating V2G technology into smart city systems. The study advocates for a decentralized information management approach, utilizing Map Reduce on the adoop architecture. Furthermore, this investigation presents a decision support system model that optimizes charging profiles and pricing, taking into account factors such as energy demand, grid stability, and environmental sustainability.

The review study in [91] aims to mitigate carbon emissions in passenger transportation by studying the integration of V2G in Europe. Through qualitative methodologies and literature analysis, the study uncovers various business models, highlighting the importance of examining innovation “activity systems”. The identification of 5 business model clusters and 12 stakeholder groups exemplifies the complexity of the V2G ecosystem. Additionally, the review explores policy implications and recommends incentives and user participation in mobility services. It also delves into prospective V2G services and technological challenges related to renewable energy storage.

The research explored in [92] examines V2G technology and specifically explores EV charging management using a DQN-based RL technique. The study surpasses existing approaches in tackling unpredictable travel patterns and fluctuating electricity bills, resulting in cost reductions of over 98%. This method offers a comprehensive and customizable V2G-oriented EV charging management solution, demonstrating its cost-effectiveness in real-world scenarios.

An optimal V2G control strategy employing a DRL-based DDPG algorithm to provide supplemental frequency regulation for EVs was presented in [93]. This strategy reduces frequency deviation and enhances the area control error (ACE) in a two-area power system by dynamically adjusting V2G power scheduling. The study provides valuable insights for grid optimization in response to the increasing adoption of EVs and the integration of renewable energy sources.

The analysis of V2G technology, with a specific focus on optimization techniques, objective functions, and constraints for economic and grid support services, was comprehensively studied in [94]. The review explores techniques, among others, such as active power assistance, peak shaving, valley filling, spinning reserve, and congestion management. Additionally, the study takes into account constraints such as battery capacity, charging station current limits, charge/discharge rates, and more.

In conclusion, V2G technology presents a transformative opportunity for enhancing energy management by allowing EVs to contribute surplus battery power back to the grid. Recent research highlights the potential of V2G systems to optimize energy costs, support renewable energy integration, and improve grid stability through advanced algorithms such as deep q-learning and DDPG. By addressing the complexities of V2G implementation and promoting innovative business models, this technology can significantly mitigate carbon emissions and facilitate the transition to sustainable energy systems.

Another compelling example is the pilot program conducted by Nissan and several utility companies in the UK, which explored V2G technology. In this initiative, Nissan Leaf owners were able to sell surplus energy back to the grid during peak demand periods. The implementation of ML algorithms allowed for a better prediction of energy supply and demand, optimizing the timing of energy discharge from EVs. This case not only highlights the practical application of V2G technology but also demonstrates how ML can facilitate the integration of renewable energy sources into the grid [95].

3.9. Fleet Management

There are multiple factors to effectively manage the EV fleet, including energy management, robust charging infrastructure, planning routes, and type of vehicle. The study in [96] formulated an autonomous mobility-on-demand (AmoD) model, employing DRL for collaborative decision making and examining static policies for baseline comparisons. ML/DL techniques were considered as critic-only techniques like SARSA and Q-learning, actor-only techniques such as the Williams REINFORCE algorithm, and actor–critic techniques, along with Trust Region Policy Optimization (TRPO) to investigate electric AMoD systems. ANN was employed to compare dynamic policies with static ones while considering operating costs. Later on, another study [97] introduces an integrated framework that combines DRL with a rule-based methodology to enhance battery swapping station (BSS) and AMoD fleet operations. The framework utilizes an MILP model for BSS management and a network flow model for modeling AMoD behavior. The MILP model is specifically designed to optimize swapping schedules and vehicle balancing, ultimately maximizing the profit of the AMoD operator.

The investigation in [98] explores fleet rebalancing in expanding shared e-mobility networks using Multi-Agent RL (MARL) and a novel policy optimization approach with Action Cascading (ac-PPO). The findings demonstrate that ac-PPO outperforms baselines such as Demand Gap Greedy (DMD), Revenue Greedy (REV), Random Rebalancing (RND), and No Rebalancing (NR), resulting in a 14% increase in demand satisfaction and a 12% improvement in net revenue. The study compares several algorithms while all have an approach, including Policy Gradient (PG), DQN, and Advantage Actor Critic (A2C). It acknowledges the challenge of rebalancing in ever-expanding EV sharing networks and presents an incentive-based problem within the context of MARL.

The study in [99] investigates shared e-scooter fleet usage prediction using open-source big data and ML approaches, including linear regression, SVR, gradient boosting DTs, and LSTM. The study highlights the significance of open data in promoting innovation in service operations. It emphasizes the impact of time-series variables on performance prediction, accounting for 67.0% of node breaks in trees. The final contribution is an effective framework combining feature engineering with public data to achieve optimal micro-mobility service prediction, which can be applied in urban transportation. The research [100] introduces an approach for mobility-aware charging scheduling in shared on-demand EV fleets, utilizing Binary Linear Programming (BLP) and DRLs such as DQN, DDPG, and Soft Actor–Critic (SAC). Real data from Haikou City are employed in the study, which formulates the problem as a Partially Observable Markov Decision Process (POMDP) and addresses the challenges of joint charging scheduling, order dispatching, and vehicle rebalancing in large-scale operations.

The predictive fleet management for on-demand mobility using Munich’s taxi dataset is addressed in [101]. It employs NNs, CNN, time-varying Poisson models, and LSTM networks for passenger demand prediction. Two strategies, Predictive Maximum Bipartite Matching (PR-BPM) and Predictive Nearest-Taxi/Nearest-Request (PR-NTNR), are implemented through an MILP solver. Additionally, Open Street Map (OSM) data and Dijkstra’s algorithm are utilized for traffic modules. The findings indicate a 30% reduction in fleet size, resulting in weekly savings of EUR 120,670 and a decreased distance of 19,199 km using PR-BPM.

In another study [102], an RL-based Markov Decision Process (MDP)-based charging coordination system for EV fleets is introduced. The system aims to construct a flexible RL-based framework, leading to significant improvements in intelligent EV fleet charging strategies. Compared to other solutions, this model outperforms by eliminating situations where the power limit is exceeded and reducing load variance by 65%.

To conclude, effective fleet management for EVs relies on the integration of advanced algorithms and data-driven strategies to optimize key operational aspects such as energy management, route planning, and vehicle rebalancing. Recent studies highlight the effectiveness of DRL and MARL in enhancing decision-making processes within AmoD systems and shared e-mobility networks. By utilizing innovative frameworks and predictive models, these approaches improve operational efficiency while also contributing to cost savings and enhanced service reliability in urban transportation systems.

3.10. Relationship Between the Various Mentioned Aspects

In the fast-changing realm of EVs, the interrelationship of various components, including battery management, range prediction, charging infrastructure optimization, AD, energy management, predictive maintenance, traffic management, V2G technology, and fleet management, is crucial for improving overall performance and sustainability. Each of these factors plays an important role in ensuring the efficient functioning of EVs and the larger transportation ecosystem.

Battery management is vital for the performance and lifespan of EV batteries, directly impacting range prediction. Reliable range prediction depends on effective battery management systems that track the health and state of charge of the battery, taking into account aspects such as driving conditions and energy consumption patterns. These data are essential for optimizing the charging infrastructure, as they guide the placement and operation of charging stations to effectively serve EV users. By combining real-time information from battery management and range predictions, charging infrastructure can be strategically developed to minimize wait times and enhance convenience for drivers.

Additionally, the incorporation of AD technology significantly boosts the efficiency of EV operations. Autonomous vehicles leverage advanced algorithms to navigate and make immediate decisions based on traffic conditions, which can be managed by traffic management systems. These systems can enhance vehicle flow, alleviate congestion, and improve energy efficiency, thereby aligning with energy management goals. Through predictive maintenance, fleet operators can anticipate maintenance requirements and reduce downtime, ensuring that vehicles remain in peak condition.

The concept of V2G technology introduces another layer of complexity and potential within this ecosystem. V2G enables EVs to not only consume energy from the grid but also return excess energy, which helps stabilize the grid and facilitates the integration of renewable energy. This two-way energy exchange is especially beneficial in fleet management, where operators can optimize energy consumption and costs while also participating in demand response programs. Therefore, the collaboration among these components—battery management, range prediction, charging infrastructure optimization, AD, energy management, predictive maintenance, traffic management, V2G technology, and fleet management—forms a comprehensive framework that improves the sustainability and efficiency of electric mobility.

Table 2. Comparison of AI/ML methods and performance metrics in e-mobility.

Reference	AI/ML Method	Application Area	Performance Metrics	Advantages	Weaknesses
[20]	RVM, PF, SVM	Battery management	RMSE: 1.68%, accuracy: 98%	High accuracy for SOC/SOH estimation	Requires extensive preprocessing
[25]	FNN, RNN, RBF	Battery management	Error: RBF 2.2%; FNN: 0.7%; RNN: 0.5%	Effective across diverse conditions	RBF models show higher error rates
[28]	LSTM, DBN, CNN	Battery management	MSE: 0.022 for SOH	Combines data-driven and physical modeling	Complex deployment in real systems
[29]	XGBoost, LightGBM	Range prediction	Prediction error: [−0.8, 0.8] km	High precision with minimal error range	Requires extensive driving data
[33]	Federated learning, DLSTM	Charging demand prediction	Accuracy: 97.14%	Combines multiple models for robust prediction	High computational resource needs
[34]	Bi-LSTM, GRU	Real-time range prediction	R2: 0.99998, MSE: 0.029 km	Very high accuracy for short-term predictions	Complex feature scaling techniques
[36]	Linear integer programming	Optimal charging station placement	Reduced energy costs in urban areas	Optimizes EV adoption in target regions	Limited scalability for large areas
[40]	XGBoost, neural networks	Smart fleet charging	Mean absolute error: 126 W	Effective for dynamic charging strategies	Requires real-time data integration
[41]	Reinforcement learning	EV fleet optimization	Improved bidding strategy accuracy	Enhances profitability in smart energy grids	Complexity in multi-agent systems
[43]	LM-CNN-SVM	Object recognition and classification	Accuracy: 92.8% for classification	Handles background clutter and occlusions	Limited scalability for larger datasets
[46]	End-to-End CNN	Real-time AD system evaluation	Accuracy: 90.4% on Tesla Autopilot	Robust decision making in real-world tests	High training data requirements
[51]	DRL + MPC	Motion planning and decision making	Improved fuel efficiency metrics	Integrates predictive control for AD systems	Complexity in system integration
[55]	CNN-GRU	Day-ahead forecasting	MAPE: <1%	Highly precise predictions for hybrid systems	High computational demands
[57]	DRL with transfer learning	EV energy optimization	Convergence: improved by 70%	Efficient transfer of pre-trained models	Limited adaptability to new systems
[59]	RNN (LSTM, GRU)	Trajectory and delay forecasting	Effective for long prediction intervals	High effectiveness in smart city integration	Dependence on real traffic data
[66]	ANN	Automotive predictive maintenance	Precise component health forecasting	Reduces downtime and maintenance costs	Data-heavy approach
[69]	Cox PHM + M-LSTM	Battery health prediction	MAPE: 0.28%, RMSPE: 0.55%	Highly accurate RUL estimations	Complexity in multi-parameter optimization
[72]	Dynamic inspection scheduling	Time interval between measurements	Improves inspection efficiency	Reduces unnecessary checks	Limited to pre-defined parameters
[77]	XGBoost, SHAP	Real-time accident detection	Accuracy: 99%, detection rate: 79%	Highly accurate accident detection	Requires highway digitalization
[78]	Transformer, LSTM, Bi-LSTM	Short-term traffic flow prediction	MAE: Transformer (5.613)	Effective for capturing spatial–temporal data	High computational requirements
[83]	GNN, Multi-Arm Bandit (MAB)	Intelligent traffic management	Accuracy: ~97% across datasets	Dynamic refinement of traffic policies	Limited real-world testing
[88]	Deep Q-Learning	V2G energy optimization	Cost reduction: over 98%	Highly effective for fluctuating energy costs	Complexity in multi-scenario modeling
[94]	Decentralized map reduce	Charging profiles and pricing	Optimized grid stability and demand	Promotes energy sustainability	Requires advanced infrastructure
[93]	DDPG	Frequency regulation	Reduces frequency deviation, ACE improved	Robust optimization for grid performance	High algorithmic complexity
[96]	DRL (AmoD System)	Fleet rebalancing	Demand satisfaction: +14%, revenue: +12%	Effective for shared mobility optimization	Limited scalability for large fleets
[99]	Linear regression, LSTM	E-scooter usage prediction	Effective time-series prediction	Promotes innovation with open data	Requires extensive data preprocessing
[102]	MDP, RL	Fleet charging coordination	Load variance reduced: 65%	Eliminates power limit exceedances	Challenging for large-scale systems

comprehensively compares AI/ML methods and their applications in e-mobility, highlighting performance metrics, advantages, and weaknesses. In battery management, techniques like SVM, RNN, and LSTM show high accuracy and adaptability but often require complex preprocessing or deployment. Range prediction and charging demand forecasting benefit from models such as XGBoost, Bi-LSTM, and federated learning, achieving minimal errors and robust predictions, though they demand extensive data and computational resources. AI applications for optimizing electric vehicles, managing fleets, and predicting traffic flow utilize reinforcement learning, deep learning, and hybrid methods. These approaches enhance efficiency and accuracy while tackling system complexity and scalability challenges. Emerging methods like GNN and decentralized systems excel in traffic and grid stability but face real-world testing and infrastructure readiness challenges. These methods significantly enhance e-mobility but often require trade-offs between performance, data requirements, and scalability.

4. Discussion

The role of ML/DL models in various fields of mobility electrification is strongly highlighted. According to the research reviewed in the previous section, the application of ML/DL can be summarized in Figure 6, from the perspective of different aspects such as battery management, range prediction, charging infrastructure optimization, AD, energy management, predictive maintenance, traffic management, V2G, and fleet management.

Despite the many advantages and strengths of ML/DL in the electrification of transportation, there are still limitations in this area, such as high-level uncertainty, data dependency, computation time, big data, cybersecurity, accessibility, diversity of EV technologies, need for complex optimization methods, and interdependency between electrical and transportation grids.

4.1. Challenges

4.1.1. General Challenges

In the case of battery management, gathering reliable data for training ML/DL models in the context of air temperature, electrical parameters, and mobility can be difficult due to factors such as restricted availability of practical driving data, fluctuations in driving conditions, and the need for diverse datasets that capture different driving scenarios and vehicle behaviors. Considering the involvement of cultural discussions and social welfare, high accuracy of learning models will be very important because the EV owner will decide to charge, use, or resell electricity by relying on accurate prediction and battery management. Therefore, it is necessary to consider models with very high accuracy.

In the case of preventive maintenance, due to the involvement of multiple parameters, environmental, and road conditions, ML/DL models strongly require sensors with high accuracy and speed. This dependence can reduce the attractiveness of ML/DL models due to the high price of the sensors.

Due to the role of communication infrastructures in storing data related to transportation and mobility, the importance of cybersecurity in ML/DL models will be very highlighted. ML/DL models in e-mobility may be targeted at various phases of their lifespan. During the testing process, attackers can exploit vulnerabilities using several types of attacks, including modeling extraction attacks, modeling inversion assaults, and adversarial crimes. In all applications, especially energy management, AD, and traffic management, the information received for training must be protected with a high level of security. This challenge makes the owners of EVs or fleets look more cautiously at ML/DL models. In summary, the ML/DL models must be robust against assaults. While aggressive assaults may not immediately threaten EVs, they can target EV infrastructure facilities such as charging stations or V2G communication networks. Ensuring the security and resilience of this equipment against hostile assaults is crucial to the ongoing viability of EV services.

The high volume of computation and optimization can lead to high energy loss, which evokes the need for a green data center. Training ML/DL models for EV-related tasks, such as battery management, energy efficiency, or self-driving vehicles, frequently necessitates large computational resources. This might be an issue for smaller EV producers or research institutions with limited access to high-performance computing resources.

One of the issues of ML/DL’s impact on mobility electrification is related to overdesigning. In the case of charging forecasting, battery management optimization, and managing EV dynamics, overfitting of models can reduce the effectiveness of ML/DL in specific applications. Realizing that models are applied efficiently throughout many driving scenarios and vehicle types is crucial for practical applications.

ML/DL models should be as understandable and interpretable as possible to facilitate their use for various applications. Such ML/DL models used in mobility electrification, like battery depletion forecasting and energy charging demand optimization, are critical for guaranteeing security and safety. Players including manufacturers, regulators, and customers seek information on how these types of vehicles make decisions and the rationale behind them.

The health and validity of the information as well as the bias in the training data can lead to unrealistic, optimistic, and unfair ML/DL models that strongly affect the safety of drivers. This importance doubles when it comes to making informed decisions about road and traffic conditions and the efficient operation of EVs.

Considering that, currently, ML/DL models are used locally and limitedly, the lack of global models at the macro level can increase the risk of using these models in mobility electrification. In other words, there is a need to generalize ML/DL models in a wide range of driving situations, different geographic and weather conditions, and different types of vehicles. Therefore, the models should be acceptable to the extent of taking into account different conditions, and they should have accurate and momentary decision making.

4.1.2. Detailed Challenges

In this section, more detailed challenges regarding some of the areas considered in this paper are discussed.

Battery Management: BMSs face several challenges that significantly impact their effectiveness in real-world applications. One of the primary challenges is the data quality and availability; inaccurate or incomplete data can lead to erroneous estimations of the SOC and SOH, which can compromise the safety and reliability of EVs. For instance, if the SOC is underestimated, drivers may experience unexpected battery depletion, leading to range anxiety. To address this issue, implementing advanced data validation techniques and employing machine learning algorithms that can learn from historical data patterns can enhance prediction accuracy. Additionally, the cybersecurity risks associated with BMSs must be taken seriously, as vulnerabilities can expose vehicles to unauthorized access and manipulation. To mitigate these risks, manufacturers should adopt robust cybersecurity protocols, including encryption and regular software updates, ensuring that BMSs are resilient against potential threats. By proactively addressing these challenges, the reliability and safety of battery management systems can be significantly improved, thereby fostering greater consumer confidence in electric mobility [103].

Range Prediction: In the context of range prediction for EVs, several significant challenges can adversely affect real-world applications. One prominent issue is data dependency; the accuracy of range predictions heavily relies on high-quality data from various sources, including driving patterns, environmental conditions, and battery performance metrics. Inconsistent or incomplete data can lead to inaccurate estimations, resulting in driver frustration and potential range anxiety. To mitigate this challenge, implementing a robust data collection framework that utilizes real-time data analytics and machine learning algorithms can enhance the reliability of predictions. Furthermore, the challenge of cybersecurity is critical, as the integration of connected vehicle technologies increases vulnerability to data breaches and manipulation. A successful cyber-attack could compromise the integrity of range predictions, leading to unsafe driving conditions. To address this risk, manufacturers should adopt stringent cybersecurity measures, such as end-to-end encryption and regular system updates, alongside developing a comprehensive incident response plan. By proactively addressing these challenges, the effectiveness of range prediction systems can be significantly enhanced, ultimately improving user confidence and promoting the adoption of electric vehicles [104].

Charging Infrastructure Optimization: The optimization of charging infrastructure for EVs encounters significant challenges that can impede its effectiveness in real-world scenarios. One major challenge is data dependency; accurate optimization relies on comprehensive data regarding vehicle usage patterns, charging behavior, and grid capacity. Inadequate or inconsistent data can lead to poor decision making regarding the placement and capacity of charging stations, resulting in insufficient access for EV users and increased range anxiety. To address this, it is crucial to implement robust data collection strategies that utilize real-time analytics from various sources, such as traffic patterns and user behavior, to inform infrastructure planning. Additionally, cybersecurity threats pose a critical risk as charging stations become increasingly connected to smart grids and cloud-based systems. Vulnerabilities in these systems can lead to unauthorized access, data breaches, and potential safety hazards for users. To mitigate these risks, stakeholders must prioritize the implementation of stringent cybersecurity measures, including encryption, secure authentication protocols, and regular security assessments. By proactively tackling these challenges through enhanced data strategies and fortified security protocols, the effectiveness and reliability of charging infrastructure can be significantly improved, thereby facilitating the widespread adoption of electric vehicles [105].

Autonomous driving (AD): The implementation of AD technology is fraught with significant challenges that can hinder its effectiveness in real-world scenarios. One prominent issue is data dependency; autonomous vehicles depend on continuous streams of data from various sensors and cameras to navigate and make decisions. Inconsistent or erroneous data can lead to critical failures in perception and decision making, potentially resulting in accidents. To mitigate this, it is essential to develop advanced sensor fusion techniques that combine data from multiple sources, ensuring a more accurate and reliable understanding of the vehicle’s environment. Additionally, the challenge of cybersecurity is paramount, as the interconnected nature of AD systems exposes them to potential hacking and malicious attacks. Such vulnerabilities could not only compromise vehicle safety but also jeopardize user privacy. To address these concerns, manufacturers must prioritize robust cybersecurity measures, including regular software updates, intrusion detection systems, and secure communication protocols. Furthermore, collaboration among industry stakeholders to establish standardized cybersecurity frameworks can enhance the overall safety of autonomous driving systems. By proactively tackling these challenges, the deployment of autonomous vehicles can be made safer and more efficient, ultimately fostering greater public acceptance and trust in this transformative technology [106].

Energy Management: EMSs face several critical challenges that can significantly affect their performance and effectiveness in real-world applications. A primary concern is data dependency; effective energy management relies on accurate and timely data regarding energy consumption patterns, generation sources, and grid conditions. Inaccurate or incomplete data can lead to suboptimal decision making, resulting in inefficient energy use and increased operational costs. To address this issue, organizations should invest in advanced data analytics and machine learning algorithms that can process large datasets to identify trends and optimize energy usage proactively. Additionally, cybersecurity remains a pressing challenge, as EMSs are increasingly interconnected with smart grids and IoT devices, making them vulnerable to cyber-attacks. A successful breach could disrupt energy distribution and compromise sensitive consumer data. To mitigate these risks, it is crucial to implement robust cybersecurity measures, including encryption, multi-factor authentication, and regular vulnerability assessments. Furthermore, establishing industry-wide standards for cybersecurity practices can help create a more secure energy management landscape. By addressing these challenges through enhanced data strategies and fortified security measures, energy management systems can become more reliable and efficient, ultimately supporting the transition to sustainable energy solutions [107].

Predictive Maintenance: PdM systems encounter several significant challenges that can impact their effectiveness in real-world applications. One of the primary issues is data dependency; the accuracy of predictive models relies heavily on the quality and quantity of historical and real-time data collected from machinery and equipment. Insufficient or poor-quality data can lead to inaccurate predictions, resulting in unexpected equipment failures and increased downtime. To address this challenge, organizations should focus on implementing comprehensive data collection strategies that utilize IoT sensors and advanced analytics to ensure high-quality, real-time data are available for analysis. Additionally, cybersecurity poses a critical risk, as PdM systems are often connected to corporate networks and cloud platforms, making them vulnerable to cyber threats. A successful cyber-attack could compromise sensitive operational data and disrupt maintenance schedules. To mitigate these risks, it is essential to adopt robust cybersecurity protocols, including regular software updates, network segmentation, and employee training on security best practices. Moreover, establishing a culture of cybersecurity awareness within the organization can further enhance protection against potential threats. By proactively addressing these challenges through improved data management and strengthened cybersecurity measures, predictive maintenance can be optimized, leading to increased operational efficiency and reduced maintenance costs [108].

Traffic Management: Traffic management systems face several critical challenges that can significantly affect their effectiveness in real-world applications. A key issue is data dependency; these systems rely on accurate, real-time data from various sources, including sensors, cameras, and GPS devices, to monitor and manage traffic flow. Inaccurate or delayed data can lead to poor decision making, resulting in increased congestion, longer travel times, and higher emissions. To address this, cities should invest in advanced data analytics and machine learning algorithms that can process and analyze data from multiple sources, enabling more accurate predictions of traffic patterns and more effective responses to congestion. Additionally, cybersecurity is a pressing concern, as traffic management systems are increasingly connected to the Internet and other networks, making them vulnerable to cyber-attacks that could disrupt traffic flow and compromise public safety. To mitigate these risks, it is essential to implement robust cybersecurity measures, including encryption, secure communication protocols, and regular security audits. Furthermore, fostering collaboration between government agencies, technology providers, and cybersecurity experts can help establish best practices and standards for securing traffic management systems. By proactively addressing these challenges through improved data strategies and enhanced security measures, traffic management systems can become more efficient and resilient, ultimately leading to safer and more sustainable urban mobility [109].

V2G: The implementation of V2G technology presents several significant challenges that can hinder its effectiveness in real-world applications. One major issue is data dependency; V2G systems require accurate and timely data regarding vehicle battery status, energy demand, and grid conditions to optimize energy exchange between EVs and the grid. Inaccurate or incomplete data can lead to inefficient energy management, resulting in suboptimal grid performance and reduced economic benefits for EV owners. To address this challenge, it is crucial to develop robust data management frameworks that utilize advanced analytics and real-time monitoring to ensure high-quality data are available for decision making. Additionally, cybersecurity poses a critical risk, as V2G systems involve the integration of numerous connected devices, making them vulnerable to cyber-attacks that could compromise grid stability and user safety. To mitigate these risks, stakeholders should implement comprehensive cybersecurity measures, including secure communication protocols, regular software updates, and intrusion detection systems. Furthermore, fostering collaboration among automakers, energy providers, and cybersecurity experts can help establish best practices and standards for securing V2G systems. By proactively addressing these challenges through improved data strategies and enhanced security protocols, V2G technology can become a reliable and efficient solution for integrating renewable energy sources and supporting grid resilience [110].

Fleet Management: Fleet management systems encounter several significant challenges that can impact their efficiency and effectiveness in real-world applications. A primary concern is data dependency; these systems rely on accurate and timely data from various sources, such as GPS tracking, vehicle diagnostics, and driver behavior analytics, to optimize operations and reduce costs. Inaccurate or incomplete data can lead to poor decision making, resulting in increased fuel consumption, unnecessary maintenance, and inefficient routing. To address this challenge, fleet operators should invest in advanced data integration platforms that aggregate and analyze data from multiple sources, enabling real-time insights and more informed decision making. Additionally, cybersecurity poses a critical threat, as fleet management systems are often interconnected with various devices and networks, making them vulnerable to cyber-attacks that could compromise sensitive operational data and disrupt fleet operations. To mitigate these risks, it is essential to implement robust cybersecurity measures, including encryption, regular system updates, and employee training on security protocols. Furthermore, establishing industry standards for cybersecurity practices can enhance the overall resilience of fleet management systems. By proactively addressing these challenges through improved data management and fortified security measures, fleet management can achieve greater operational efficiency, reduce costs, and enhance overall safety [111].

4.2. Future Trends

The role of AI and ML/DL models in the future of the electric transportation industry is undeniable. Figure 7 illustrates the trajectory of publications related to “electric mobility” and “ML/DL” as extracted from the Scopus database. Notably, the integration of ML/DL techniques in electric mobility research has seen a remarkable surge since 2019, with the number of documents surpassing 3000 by 2023. This exponential growth is vividly depicted by the trendline, underscoring the rapid advancement in this domain. The year 2012 was a pivotal moment, marking an “AI resonance” period [112]. This period was characterized by the emergence of the big data paradigm and the promising prospects of ML/DL in image processing [112,113]. Notably, the application of ML/DL in e-mobility subsequently gained momentum. This can be attributed to the increasing significance of electrifying transportation systems as a means to mitigate air pollution stemming from transportation emissions [114]. Given the upward trajectory observed in integrating ML/DL within mobility electrification, it is reasonable to anticipate further innovation and advancement in the coming years. As technology continues to evolve and researchers delve deeper into the possibilities offered by ML/DL algorithms, we can expect novel applications and refined methodologies to emerge, driving progress in electric mobility. The synergy between ML/DL and mobility electrification holds immense promise for addressing pressing environmental concerns and shaping the future of transportation.

Considering the growing trend of using EVs worldwide, there is a need to predict models for charging demand, traffic control, energy management, advanced techniques for interdependency of electricity and transportation networks, management of renewable resources, and sustainability [115]. This requires aggregation, processing, and management of big data under comprehensive algorithms. The main leaders of the electric car industry, such as Tesla, are looking for the range of their batteries by using ML/DL models to increase productivity and satisfy customers. With these descriptions, there is a bright future for ML/DL applications in the electrification of transportation. Among the trends that can be introduced in future works, the following are mentioned.

• V2G as a solution for renewable energy integration: Focusing on V2G technology as the most sustainable solution for integrating high-capacity EVs and renewable resources until 2030 is the main trend in the field of e-mobility. In other words, the integration of the high capacity of clean energy to pursue carbon-free policies by 2030 requires expensive network equipment such as high-voltage transmission lines and high-voltage substations, which certainly cannot be developed by the target year. Therefore, V2G technology can help reduce the stress on distribution networks while integrating wind and solar resources. This trend is due to the increasing need for weather forecasting, modeling the uncertainty of highly fluctuating wind and solar resources, allocation and sizing of charging stations, and bi/directional power exchange in real time, increasing the need for ML/DL models.

To enhance the effectiveness of V2G technology as a solution for renewable energy integration, it is crucial to focus on improving generalization in machine learning models, particularly when data are limited. This can be achieved by employing advanced techniques such as transfer learning and data augmentation to create more robust predictive models. Additionally, developing smarter and more adaptive charging infrastructure strategies can optimize energy exchange between electric vehicles and the grid, ensuring the efficient utilization of renewable energy sources while maintaining grid stability.

• Marine electrification: Although the concept of electric mobility is mostly associated with electric personal cars, zero-carbon policies have also gradually included the electrification of maritime transport and light and semi-heavy trucks. Cruise ships, Ro-Ro, and electric ferries need energy management in line with grid conditions when anchoring in ports due to constant load consumption and day-ahead planning. ML/DL models can help the integration of the marine electrical and power grid by predicting the load, weather conditions, and ship load management. On the other hand, the prediction of nodal voltage and participating in voltage response programs that the power grid suffers from in the presence of electrical ships can be controlled by prediction and preventive management by ML/DL models.

To improve marine electrification, addressing the challenge of improving generalization in machine learning models with limited data is crucial. This can be achieved by employing techniques such as transfer learning and data augmentation, which enable models to leverage existing data more effectively. Additionally, developing smarter and more adaptive charging infrastructure strategies, such as dynamic load management and real-time optimization based on vessel schedules and energy demand, can significantly improve the efficiency and reliability of charging operations, ultimately supporting the transition to electrified marine transportation.

• Battery swapping station: Reducing the charging time and traveling long distances is another important trend in the field of transportation electrification, which seems to be covered under the shadow of AL models in the future. Battery swapping stations that have recently been used as pilots have provided promising results in solving this problem, reducing the charging time and inconvenient stops for EVs [116]. Under these conditions, the driver can replace the battery with a full one in a short time (less than 5 min). This method requires the definition of the ID for each vehicle and each battery. On the other hand, since the replacement time is a matter of taste, it can destroy the life and health of the battery. The infrastructure of automatic battery replacement and physical/cybersecurity are among the problems of this challenge, which ML/DL models can propose a solution for with temporal and spatial prediction and integration with GAS-based models. On the other hand, due to the traffic network being affected by battery swapping stations, there is a need to collect and process a large amount of information. Therefore, robust training is needed to increase the efficiency of battery swapping stations so that the driver can decide the most optimal route, battery, and time of charging based on it.

To enhance the effectiveness of battery swapping stations, it is crucial to focus on improving generalization in ML models, especially when dealing with limited data. This can be achieved by employing techniques such as transfer learning and data augmentation to ensure that models can accurately predict battery performance and demand patterns. Additionally, developing smarter, more adaptive charging infrastructure strategies will allow for real-time adjustments based on usage trends and grid conditions, optimizing energy distribution and minimizing downtime for electric vehicles. These advancements can significantly improve the operational efficiency and user experience of battery swapping stations.

• Wireless Charging: ML/DL has transformative potential in advancing wireless charging and smart road systems for electric vehicles, offering multifaceted benefits across various domains. Beyond optimizing infrastructure placement and dynamically managing charging parameters, ML/DL algorithms can revolutionize wireless charging systems by refining energy transfer efficiency, adapting to diverse vehicle types and battery technologies, and mitigating charging infrastructure costs through predictive maintenance and intelligent energy routing. Moreover, ML/DL can enhance user experiences by adjusting charging schedules to individual preferences, integrating seamlessly with smart home and grid systems for holistic energy management, and facilitating convenient payment and authentication processes. Additionally, the application of ML/DL in wireless charging can extend to innovative areas such as dynamic charging lanes on highways, where vehicles can charge while in motion, and wireless charging pads embedded in parking lots, offering autonomous charging services.

To develop the effectiveness of wireless charging systems, it is crucial to focus on improving generalization in machine learning models, particularly when working with limited data. Developing adaptive charging infrastructure strategies that can dynamically respond to real-time usage patterns and energy demands will further optimize charging efficiency and user experience. By leveraging advanced analytics and AI-driven insights, we can create a more responsive and intelligent wireless charging ecosystem that meets the evolving needs of electric vehicle users.

• Autonomous Electric Vehicles: The field of autonomous electric vehicles (AEVs) is experiencing rapid advancement, driven by breakthroughs in artificial intelligence, sensor technology, and electric propulsion systems. As ML/DL algorithms continue to evolve, they enable autonomous vehicles to perceive and interpret their surroundings with unprecedented accuracy and efficiency, paving the way for safer and more reliable AD experiences. Moreover, the electrification of vehicle fleets is gaining momentum, spurred by the growing awareness of environmental sustainability and the declining costs of battery technology. This convergence of AD and electric propulsion is reshaping the future of transportation with a shift toward shared mobility services, such as autonomous robo-taxis and on-demand electric shuttles. Looking ahead, the future trends in autonomous EVs are likely to focus on enhancing vehicle autonomy levels, expanding deployment in urban environments, integrating with smart city infrastructure, and advancing toward fully autonomous, driverless transportation networks. As these technologies mature, autonomous EVs have the potential to revolutionize mobility, offering safer, greener, and more accessible transportation solutions for people and goods alike.

To augment the performance of AEVs, it is crucial to focus on improving the generalization of machine learning models, especially when working with limited datasets. This can be achieved through techniques such as transfer learning and data augmentation, which help models adapt to diverse driving conditions and environments. Additionally, developing smarter and more adaptive charging infrastructure strategies is essential; implementing dynamic charging solutions that respond to real-time demand and vehicle availability can optimize energy use and reduce operational costs, ultimately supporting the widespread adoption of AEVs.

• Cybersecurity: As the technologies of EVs and ML/DL models continue to evolve and improve, the risks of cyber-attacks and information security breaches pose increasingly significant threats. Consequently, there is a growing imperative to develop more robust and reliable ML/DL models to safeguard these systems. This trend toward enhancing the resilience and security of ML/DL models is poised to become a permanent fixture in the landscape of electric mobility and AI integration. By prioritizing the development of sophisticated security measures and incorporating robustness into ML/DL algorithms, stakeholders can fortify EVs and associated technologies against potential cyber threats, ensuring the continued safety and integrity of future mobility systems.

In general, the role of ML/DL in interconnected power and transportation networks has been highlighted with the emergence of new players such as electric cars, electric trucks, and marine electrification, along with the integration of renewable resources more and more. Reinforcement learning can help with the main challenge of these resources, i.e., charging/discharging infrastructure and increasing productivity. On the other hand, by developing ML/DL models based on technical, environmental, and social aspects, it is possible to develop and analyze the concept of sustainability in interconnected electricity and transportation networks.

To enhance cybersecurity in the context of ML applications, particularly when dealing with limited data, it is crucial to focus on improving generalization techniques. This can be achieved by employing advanced methods such as data augmentation and transfer learning, which allow models to learn more robust features from minimal datasets. Additionally, developing smarter, more adaptive charging infrastructure strategies can help mitigate vulnerabilities by dynamically adjusting to real-time threats and optimizing resource allocation, thereby ensuring a more secure and efficient operational environment.

• Application of New Framework for Input Features: While input features for ML and DL methods in electromobility can include a set of scalar features [117], using time-series input is also crucial, particularly for deep learning methods. Therefore, identifying the most appropriate input features before applying DL methods, such as CNNs [118], is essential. This approach will enable the model to uncover the most relevant features effectively.
• Battery Management: In the context of battery management, future research should focus on developing targeted strategies to enhance the performance and adaptability of battery systems. One critical area for improvement is the generalization of ML models, particularly when operating with limited datasets. Researchers could explore techniques such as transfer learning or data augmentation to enable these models to make more accurate predictions in diverse operating conditions [119]. Additionally, there is a pressing need for the development of smarter, more adaptive charging infrastructure strategies. This could involve the integration of real-time data analytics and IoT technologies to optimize charging cycles based on user behavior and grid conditions, thereby improving overall energy efficiency and extending battery life. By addressing these specific areas, future research can significantly contribute to the advancement of battery management systems.
• Range Prediction: In light of the evolving landscape of range prediction, future research should focus on several key areas to enhance the effectiveness and applicability of machine learning models. Firstly, improving generalization in ML models with limited data is crucial. This can be achieved through techniques such as data augmentation, transfer learning, and the development of more robust model architectures that can leverage small datasets effectively. Additionally, there is a pressing need to innovate smarter and more adaptive charging infrastructure strategies. Research could explore the integration of real-time data analytics and predictive modeling to optimize charging station placements and improve energy distribution. By focusing on these targeted recommendations, researchers can contribute significantly to the advancement of range prediction methodologies and their practical applications.
• Charging Infrastructure Optimization: Concerning the evolving landscape of EV adoption, it is imperative to focus on specific strategies for optimizing charging infrastructure. One critical area for future research is enhancing the generalization capabilities of ML models, particularly in scenarios where data are limited. By developing robust algorithms that can effectively learn from sparse datasets, we can improve the accuracy of demand forecasting and user behavior prediction, thereby optimizing charging station placement and capacity. Additionally, there is a significant opportunity to innovate smarter, more adaptive charging infrastructure strategies. This could involve integrating real-time data analytics and IoT technologies to dynamically adjust charging rates based on grid demand and user preferences, ultimately leading to a more efficient and user-friendly charging experience. Targeted research in these areas will not only advance the field but also facilitate the seamless integration of EVs into our existing transportation networks.
• Energy Management: Regarding energy management, it is essential to refine our approach to future research trends by emphasizing specific, actionable recommendations. One promising direction involves enhancing the generalization capabilities of machine learning models, particularly in scenarios with limited data. This could be achieved through techniques such as transfer learning or data augmentation, which allow models to learn more effectively from smaller datasets [120]. Additionally, developing smarter and more adaptive charging infrastructure strategies can significantly improve energy efficiency and user experience. This may include the integration of real-time data analytics to optimize charging schedules based on demand forecasts and user behavior patterns. By focusing on these targeted areas, future research can contribute to more robust and innovative solutions within the energy management sector.
• Predictive Maintenance: In the context of predictive maintenance, future research should focus on enhancing the generalization capabilities of machine learning models, particularly in scenarios where data are limited. This can be achieved by exploring advanced techniques such as transfer learning and few-shot learning, which enable models to leverage knowledge from related tasks or datasets, thereby improving their predictive accuracy in low-data environments [119]. Additionally, the development of smarter and more adaptive charging infrastructure strategies is crucial. Future studies could investigate the integration of real-time data analytics and IoT technologies to create dynamic charging solutions that adjust to the operational needs of various assets. By implementing these targeted approaches, researchers can significantly contribute to the efficiency and reliability of predictive maintenance systems.
• Traffic Management: In light of the evolving challenges in traffic management, future research should focus on developing more targeted strategies to enhance the effectiveness of ML models, particularly in scenarios with limited data availability. One promising avenue is the exploration of transfer learning techniques that allow models to generalize better across different traffic conditions and environments, thereby improving their predictive capabilities. Additionally, there is a pressing need to innovate smarter and more adaptive charging infrastructures for EVs. This includes integrating real-time data analytics to optimize charging station locations and availability based on traffic patterns and demand forecasts. By addressing these specific areas, researchers can significantly contribute to the advancement of traffic management systems, ultimately leading to more efficient and sustainable urban mobility solutions.
• Fleet Management: Future research on fleet management should focus on enhancing the generalization capabilities of ML models, particularly in scenarios where data availability is limited. One promising avenue is the development of transfer learning techniques that leverage pre-trained models on similar datasets, thereby improving the robustness and accuracy of predictions in diverse operational environments. Additionally, research should explore the design of smarter and more adaptive charging infrastructure strategies, which can optimize energy distribution and minimize downtime for electric fleets. This could involve the integration of real-time data analytics to anticipate charging needs based on fleet utilization patterns, as well as the implementation of dynamic pricing models that encourage off-peak charging. By addressing these specific areas, researchers can contribute significantly to the advancement of fleet management systems that are both efficient and sustainable.

5. Conclusions

Transportation electrification is one of the main zero-carbon policies worldwide that is being seriously pursued to mitigate total CO₂ pollution. However, due to the involvement of various social, environmental, economic, global warming, management, forecasting, and control infrastructures, it should be developed according to e-mobility. ML/DL, relying on advanced information processing and training, can influence the future of mobility transportation. This paper presented a mini review on the effects and applications of ML/DL in mobility electrification in various sectors, including energy management, battery management, preventive maintenance, traffic management, vehicle-to-grid, electricity market, autonomous driving, and fleet management. The most widely used ML/DL models in these areas were introduced, and it was stated how each aspect can be managed and controlled by ML/DL models. Due to the dependence on big data and the need for sensors to monitor weather, road, and driving conditions, among other things, appropriate solutions were introduced to increase productivity. The most recent research concluded that integrating mobility electrification with renewable resources, battery swapping stations, vehicle-to-grid technology, marine electrification, wireless charging, autonomous electric vehicles, and cybersecurity are the main ML/DL trends that require more accurate and higher-speed models. This article introduced uncertainty management, cybersecurity, and long computation time for large volumes of data as the main limitations of ML/DL models, which require the development of more robust and reliable models.