Revolutionizing Electric Vehicle Charging Stations with Efficient Deep Q Networks Powered by Multimodal Bioinspired Analysis for Improved Performance

Abstract

The rapid growth of electric vehicle (EV) adoption presents significant challenges in planning efficient charging infrastructure, including suboptimal station placement, energy consumption, and rising infrastructural costs. The conventional methods, such as grey wolf optimization (GWO), fail to address real-time user demand and dynamic factors like fluctuating grid loads and environmental impact. These approaches rely on fixed models, often leading to inefficient energy use, higher operational costs, and increased traffic congestion. This paper proposes a novel framework that integrates deep Q networks (DQNs) for real-time charging optimization, coupled with multimodal bioinspired algorithms like ant lion optimization (ALO) and moth flame optimization (MFO). Unlike conventional geographic placement models that overlook evolving travel patterns, this system dynamically adapts to user behavior, optimizing both onboard and offboard charging systems. The DQN enables continuous learning from changing demand and grid conditions, while ALO and MFO identify optimal station locations, reducing energy consumption and emissions. The proposed framework incorporates dynamic pricing and demand response strategies. These adjustments help balance energy usage, reducing costs and preventing overloading of the grid during peak times, offering real-time adaptability, optimized station placement, and energy efficiency. To improve the performance of the system, the proposed framework ensures more sustainable, cost-effective EV infrastructural planning, minimized environmental impacts, and enhanced charging efficiency. From the results for the proposed system, we recorded various performance parameters such as the installation cost, which decreased to USD 1200 per unit, i.e., a 20% cost efficiency increase, optimal energy utilization increases to 85% and 92% during peak hours and off-peak hours respectively, a charging slot availability increase to 95%, a 30% carbon emission reduction, and 95% performance retention under the stress condition. Further, the power quality is improved by reducing the sag, swell, flicker, and notch by 2 V, 3 V, 0.05 V, and 0.03 V, respectively, with an increase in efficiency to 89.9%. This study addresses critical gaps in real-time flexibility, cost-effective station deployment, and grid resilience by offering a scalable and intelligent EV charging solution.

1. Introduction

As the global usage of electric vehicles (EVs) grows, the need for effective charging infrastructure has become a key concern. The proper planning and deployment of EV charging infrastructure are essential for the widespread adoption of EVs and the transition towards sustainable transportation. An effective EV charging infrastructure should be able to meet the increasing demand for charging while reducing energy consumption and environmental impacts. The optimization of EV charging infrastructure involves several factors, such as user demand, infrastructural cost, energy consumption, and environmental impact. The integration of both onboard and offboard charging optimization can help to address these factors and improve the overall efficiency of EV charging infrastructure and deployments. Furthermore, the deployment of EV charging infrastructure should consider various charging strategies, such as dynamic pricing, demand response, and battery storage, to optimize charging efficiency and reduce energy costs. The placement of EV charging stations is also critical for optimizing charging efficiency and reducing environmental impact. To reduce charging demand, traffic congestion, and emissions, optimal station placement should take into account user behavior, travel patterns, and geographical factors. Furthermore, the optimal placement of charging stations should be able to meet the growing demand for charging and reduce the waiting timestamp for charging battery packs. This paper proposes a deep Q network (DQN) using multimodal bioinspired analysis (ALO and MFO) for the optimal planning and deployment of EV charging infrastructure that integrates both onboard and offboard charging optimization, charging strategies, and optimal station placement.

The remainder of this paper is structured as follows. Section 2 presents an overview of related work in EV charging station planning techniques. Section 3 discusses the design of the proposed DQN with multimodal bioinspired analysis. Section 4 presents the results and discussions of both conventional and proposed frameworks. Finally, Section 5 presents the conclusion of the research work.

2. Review of Related Literature Work

Electric vehicles have gradually emerged as an important component of modern transportation and energy systems. Accelerating the development of EVs is posing enormous opportunities and challenges not only for infrastructural development but also in energy management and environmental sustainability. A higher demand for vehicles increases their charging infrastructure efficiency and sustainability while being compatible with power grids, renewable energy sources, and distribution networks. Optimizing the design, placement, and operation of electric vehicle charging stations seeks to make them efficient, reliable, and capable, yet with reduced environmental impacts in the process. The literature that has been reviewed cuts across a variety of approaches and techniques concerned with diverse problems in EV charging infrastructure. Among the methods developed, optimization algorithms are the most important and have been demonstrated to facilitate site selection, capacity planning, energy management, and load balancing. Many of the studies have stressed the need to take an integrated approach to cover technical aspects, like grid integration or power quality as well as economic and environmental factors. This holistic perspective is key to ensuring that EV charging stations are congruent with the broader carbon reduction agenda and wider sustainable energy systems goals.

To enhance the EV charging infrastructure and network performance, several methods have been explored. The improved gazelle optimization algorithm was employed to optimize solar and battery integration in EV station design, leading to improved efficiency but exhibiting limited scalability for larger grids [1]. A Hybrid RERNN-SCSO technique was used for power quality control in microgrids, which enhanced EV charging stability despite its complex implementation [2]. Incorporating an automatic variac transformer improved smart grid performance by using EVs as smart loads, enhancing voltage control, although it required a costly system setup [3]. Neural network optimization techniques optimized logistics for battery recycling, resulting in increased efficiency, yet faced limitations in generalizing to diverse recycling needs [4]. The Jaya grey wolf optimization method achieved a power quality improvement in EV charging stations by reducing harmonic distortion, despite a complex algorithm structure [5]. The implementation of the IoT-based solar EV charging system offered user-friendly integration and increased efficiency, although it incurred high initial costs [6]. Similarly, the integration of distributed energy resources and FACTS devices improved EV station performance in deregulated environments, but real-world testing was limited [7].

The use of vehicular edge computing for charging station navigation reduced decision-making latency, although it depended on a reliable vehicular network [8]. Online estimators optimized charging stop planning based on real-time data, enhancing the user experience, but the accuracy dropped when data quality was poor [9]. The GJO-APCNN technique effectively balanced loads in charging stations, though its algorithm was computationally intensive [10]. The combined algorithms were genetic algorithms and the fuzzy analytic hierarchy (FAH) process improved EV station location and capacity but was computationally demanding [11]. Many-objective evolutionary algorithms facilitated site selection for EV charging stations, increasing service coverage while facing scaling challenges in larger urban areas [12]. The grey wolf optimization (GWO) approach enabled solar and energy storage integration for EV charging, though its benefits were limited to sunny regions [13]. The hybrid techniques for renewable energy integration enhanced sustainability and reduced grid strain, requiring a robust renewable energy infrastructure [14]. A study on Uber and Lyft EV infrastructure deployment found it feasible for ride-sharing, though synchronizing charging times posed a challenge [15]. The hybrid genetic algorithm-simulated annealing method improved distribution network resilience via optimized EV station placement, despite its high computational demands [16].

The IoT and cloud computing integration enhances the real-time management of EV stations but raises data security concerns [17]. Techno-economic planning is used in sector-coupled energy systems with improved energy efficiency and cost savings for EVs, albeit with high initial investment requirements [18]. Optimal path planning reduces the travel time and energy consumption for EVs, though it has limited adaptability to real-time traffic changes [19]. One-way car-sharing facility planning offers improved service availability, despite difficulties in predicting user demand [20]. Multi-objective optimization enhanced network reliability and efficiency, yet its applicability was limited to urban areas [21]. Distributed generation planning improved reliability and reduced losses in distribution networks, but with integration challenges persisting [22]. Multi-period investment planning for electricity distribution reduced costs through demand forecasting, though future demand predictions remained uncertain [23]. Urban planning for environmental development facilitated sustainable EV infrastructural deployment, lacking adaptability in planning models [24]. Meta-heuristic route modeling optimized energy-efficient routing for diverse road surfaces, with limited urban applicability [25]. Machine learning for renewable energy planning improved prediction accuracy and system efficiency, albeit with high computational requirements [26]. Microgrid scheduling with battery swapping lowered energy costs and enhanced grid stability but incurred high maintenance costs [27]. GIS-based site selection for EV charging improved accuracy, though data quality was inconsistent across regions [28].

The dynamic road network model accurately forecasted charging loads, but its application was limited in small or rural areas [29]. The patent citation network analysis identified trends in charging technologies, focusing on patent-registered innovations [30]. Multi-agent Q-learning optimized EV infrastructure planning considering battery life and improved sustainability but faced coordination challenges [31]. An EV charging transaction analysis dataset provided insights into charging patterns, despite regional data bias [32]. The pseudo-inspired gravitational search algorithm enhanced distribution network resilience and reduced energy losses, though the computational intensity was high for large networks [33]. Stochastic EV loading and natural disruption management enhanced grid reliability under uncertain conditions, with resource variability posing a challenge [34]. Smart decision-hunting optimization improved the EV scheduling efficiency, though struggling to adapt to dynamic conditions [35]. Valet charging service optimization reduced user wait times and improved service efficiency, although setup costs were high [36]. The KOA-DRN approach for microgrid charging stations improved the power quality and system reliability, but integrating the approach into the existing infrastructure was complex [37]. The hybrid techniques for energy management in EV stations and distribution systems enhanced energy efficiency but were operationally complex [38]. The hybrid optimization for network reconfiguration improved network reliability and efficiency, with significant computational demands for large networks [39]. Lastly, a sustainability assessment for EV charging station locations aligned site selection with green energy goals, although its application was restricted to regions with abundant renewable energy sources [40].

The fast deployment of electric vehicles (EVs) needs the creation of effective and long-lasting charging infrastructures. For EV charging station optimization, the genetic algorithm (GA) has more limitations that reduce the efficacy in dynamic and real-time applications like slow convergence and early stalling in local optima, making it less effective at solving high-dimensional, complex issues. Additionally, its scalability is limited by the high computing cost resulting from recurrent crossover and mutation processes. The GA is not flexible enough to adjust to changes in EV demand, traffic patterns, and grid conditions in real-time, which makes it unsuitable for dynamic energy management [41,42,43]. Although particle swarm optimization (PSO) is effective in solving many optimization problems, there are a few limitations for energy management and EV charging station placement. Premature convergence is a significant problem, as particles lose their diversity and become stuck in local optima, making it impossible to discover the optimal global solution. The lack of real-time adaptation limits its efficiency in managing dynamic energy prices, traffic flow variances, and grid load fluctuations [44,45,46].

Ant lion optimization (ALO) prevents premature convergence, provides a strong balance between exploration and exploitation, and effectively handles geographical restrictions and power limitations. ALO is well-suited to large-scale EV charging station networks where appropriate placement is critical to ensure optimal coverage and load balancing [47,48,49]. Moth flame optimization (MFO) is beneficial in dynamic and real-time optimization scenarios, improves the exploration in the early phases of optimization, and reduces the chance of becoming trapped in local optima for EV charging [50].

The selection of ALO and MFO over other bioinspired algorithms, such as GA and PSO, is due to their numerous advantages. The proposed DQN integration with ALO and MFO successfully optimizes large-scale EV charging networks by managing complex and high-dimensional limitations, adjusting charging capabilities and station placements dynamically in response to demand, preventing premature convergence, and allowing for quick real-time optimization for station placement.

3. Design and Implementation of Proposed Deep Q Network (DQN)

The design of an efficient DQN using a multimodal bioinspired algorithm (ALO and MFO) is proposed for improved power quality and performance in EV charging stations. This proposed system overcomes the issues of low deployment efficiency and high complexity of analysis, which are present in existing EV station planning methods. This proposed system is interconnected with ALO, MFO, utility grid, EVs, and charging stations. ALO receives the EV demand pattern from the charging stations and recommends the charging station location to the DQN. MFO receives the adjusting power balance threshold from the utility grid and recommends the optimized fine values to the DQN to maintain the power management/balance, to meet the requirement. The DQN plays a vital role in managing the charging slots and improving energy utilization. The proposed DQN receives all the parameters from the grid, MFO, charging stations, ALO, and EVs. The DQN optimizes all required parameters and controls the system, as observed in Figure 1.

3.1. Implementation of ALO Algorithm with DQN

The design of the proposed DQN for EV station planning incorporates the ALO algorithm to enhance charging efficiency and minimize energy costs, as represented in Figure 2. The ALO algorithm, inspired by the predatory behavior of ant lions in trapping their prey, is adept at solving optimization problems in complex and dynamically changing environments like EV charging station placement and utilization. This suitability arises from ALO’s ability to explore and exploit the search space effectively, which is critical in addressing the multifaceted challenges associated with EV infrastructural deployment. The fundamental operation of the ALO algorithm involves simulating the hunting mechanism of ant lions, which create pits in sandy environments to trap ants. In the context of EV station planning, the ‘ants’ represent potential solutions for station placement and management, while the ‘ant lion’ represents the optimal solution, seeking to ‘capture’ the most effective configuration. This bioinspired approach allows ALO to adapt to various parameters including user demand, cost implications, environmental considerations, and urban traffic patterns, thereby optimizing the deployment and operation of charging stations.

The mathematical model of the ALO algorithm involves several key operations that define its search and optimization capabilities. The stochastic walk of ants is represented via Equation (1).

$$X\left(\left(\left(t+1\right)\right)\right)=X\left(t\right)+{\displaystyle \frac{R\left(t\right)}{\alpha \left(t\right)}}$$

(1)

where (t) is the position of the ant at timestamp t, α(t) is a control parameter that increases over temporal instance sets to reduce the step size, and R(t) is a stochastic function that dictates the scope and magnitude of the step. The algorithm alters the ant’s position according to the relative position and fitness of the ant lion, as shown in Equation (2).

$$Y\left(\left(\left(t+1\right)\right)\right)=Y\left(t\right)+\beta \left(t\right).\left(X\left(t\right)-Y\left(t\right)\right)$$

(2)

where Y(t) is the position of the ant lion at timestamp t, and β(t) is a parameter that increases over temporal instance sets to enhance the trapping capability levels. The ant’s position is optimized through the interaction between the ant and the ant lion, which is controlled by Equation (3).

$$X_{new}\left(t\right)={\displaystyle \frac{X\left(t\right)+Y\left(t\right)}{2}}$$

(3)

A more optimal location of EV stations is reflected in this procedure, which guarantees that the ant progresses towards better solutions over iterations. The fitness evaluation is a key component for determining the effectiveness of each solution by using an integral form of the cost function, which is represented via Equation (4).

$$F(X)={\displaystyle \int \left(C_d(X)+C_e(X)+C_t(X)\right)}dx$$

(4)

where $C_d\left(X\right)$, $C_e\left(X\right),$ and $C_t\left(X\right)$ represent the cost due to demand, energy consumption, and traffic congestion, respectively. To ensure that the solution complies with environmental and regulatory constraints, the optimization process is further refined using a derivative-based modification and is estimated via Equation (5).

$${\displaystyle \frac{dF}{dX}}=0$$

(5)

This condition finds optimal solutions by identifying points where the cost function remains constant or decreases. A differential operation is used to control the system’s temporal evolution in dynamic circumstances when environmental and demand elements vary dramatically, which is represented via Equation (6). The final stages of optimization involve a convergence criterion via Equation (7).

$${\displaystyle \frac{dX}{dt}}=\gamma \left(X,t\right)$$

(6)

$$\underset{t\to \infty }{\lim } \sigma \left(X(t)\right)=0$$

(7)

This final stage of the proposed system determines the best locations for EV charging stations. This improves coverage and accessibility for users. As a result, the charging network performs and operates more efficiently.

3.2. Implementation of MFO Algorithm with DQN

The integration of ALO with the DQN and other bioinspired algorithms like MFO forms a comprehensive framework for the strategic planning and operation of EV charging stations, ensuring sustainability and efficiency in the growing domain of electric vehicles for real-time scenarios. The MFO algorithm is intricately designed to identify optimal locations for new EV charging stations by leveraging patterns in user behavior, travel dynamics, and geographical characteristics. MFO inspired by the navigation method known as transverse orientation used by moths in nature employs a population-based approach where moths (potential solutions) fly through the problem space by maintaining a fixed angle relative to the moon (optimal solution) to minimize their paths to light sources (optimal locations). This methodology is particularly effective for the spatial distribution challenges inherent in EV station planning due to its ability to explore large search spaces and converge towards global optima, which represents the most feasible station locations considering user demand, access routes, and geographic constraints. The first key operation in the MFO algorithm denotes the position update of the moths, given by Equation (8).

$$X\left(i+1\right)=X_i+r\cdot \left(Y_i-X_i\right)$$

(8)

where X_i is the current position of moth i, Y_i is the position of the flame (optimal solution), and r is a stochastic number in the range [−1, 1] that dictates the variability of the moth’s flight path, ensuring exploration of the search space. Adjusting the moth’s trajectory towards the optimal solution and flame’s updating mechanism is controlled effectively through Equation (9).

$$Y_{new}={\displaystyle \frac{Y_{old}+Best\left(Y_{history}\right)}{2}}$$

(9)

where Y_old is the current position of the flame, and Best (Y_history) represents the best position encountered by any moth up to that iteration, facilitating a collective memory feature. The moth’s positions are further refined through an adaptive function based on their proximity to the flame, which is represented by Equation (10).

$$r\left(adaptive\right)=r_0\cdot e^{-\lambda t}$$

(10)

where r₀ is the initial stochastic factor, λ is the decay constant, and t is the current iteration number, reducing the step size as the algorithm converges to ensure precise exploitation near the optimal levels. The spatial fitness of each of the moths, representing its suitability for a station location, is evaluated using a multi-objective function via Equation (11).

$$F\left(X\right)={\displaystyle {\int }_A^B\left[fu\left(X\left(u\right)\right)+ft\left(X\left(t\right)\right)+fg\left(X\left(g\right)\right)\right]}du$$

(11)

where fu, ft, and fg assess the user demand, traffic accessibility, and geographical suitability of location X, respectively. A derivative-based approach is employed to find the optimal gradient for station setup, given by Equation (12).

$${\displaystyle \frac{\mathit{\partial }F}{\mathit{\partial }X}}=0$$

(12)

This ensures that the selected points represent stationary points in the fitness landscape, potentially indicating optimal locations. The geographic constraints are modeled as barriers or boundaries, and a constraint function (X) is included, necessitating the condition represented by Equation (13).

$$C\left(X\right)\le 0$$

(13)

where X represents limitations such as no-build zones or restricted areas. The standard deviation of the population’s fitness is used to track the variability of solutions as the optimization process moves forward to achieve the condition shown in Equation (14).

$$\sigma \left(F\right)\to minimum$$

(14)

This indicates convergence to the most suitable station locations. Finally, the environmental impact is evaluated by integrating an environmental cost function over the proposed station areas, which is represented by Equation (15).

$$E\left(X\right)={\displaystyle \int \left[ec\left(X\left(c\right)\right)+ee\left(X\left(e\right)\right)\right]dc}$$

(15)

where ec and ee represent the construction and operational environmental costs associated with each location. The selection of the MFO method is justified by its proficiency in handling complex, nonlinear optimization tasks that are typical in geographical and infrastructural planning. Its capability to mimic the moth’s natural navigation method allows for an effective balance between exploration and exploitation, making it highly suitable for the dynamic and multi-faceted problem of EV charging station placement. By integrating this method with other optimization frameworks like the DQN and ALO, a comprehensive, robust solution for infrastructural development for sustainable transportation is achieved, thereby fulfilling the increasing demand for efficient and strategically located EV charging facilities.

The integration of a DQN into the ALO and MFO methods aims to harness deep learning capabilities for enhancing decision-making processes in the optimization of EV charging station placements, as shown in Figure 3. The DQN method is particularly well-suited for this task due to its ability to learn optimal policies directly from high-dimensional sensory inputs and execute long-term strategic planning. This deep reinforcement learning approach is essential for dynamically adapting to changing transportation and user behavior patterns, making it invaluable for complex spatial–temporal optimization tasks. This system maximizes the cumulative reward, which is related to the cost-efficiency, accessibility, and sustainability of charging station placements.

The primary component of the DQN is the Q-function, which estimates the value of taking a particular action in a given state. The Q-function is represented by a neural network, which approximates the action value function with high efficiency. The first crucial operation for the DQN is the update rule for the Q-function, defined by the Bellman process via Equation (16).

$$Q_{new}\left(st,at\right)=Q\left(st,at\right)+\alpha \left[r\left(\left(t+1\right)\right)+\gamma ma{x}^{a}Q\left(s\left(\left(t+1\right)\right),a\right)-Q\left(st,at\right)\right]$$

(16)

where st and at represent the current state and action, ((t + 1)) is the reward received after taking action at, α is the learning rate, and γ is the discount factor that moderates the importance of future rewards. The optimization of the Q-function requires the calculation of the loss during training, which is typically achieved through the mean squared error (MSE) between the predicted Q values and the target Q values, which is represented via Equation (17), where yi is the target value for the Q-function, which is represented via Equation (18).

$$L={\displaystyle \frac{1}{N}}{{\displaystyle \sum \left(y_i-Q\left(s_i,a_i;\theta \right)\right)}}^2$$

(17)

$$y_i=r_i+\gamma \stackrel{a\prime }{\max }Q\left(s\left(i+1\right),a\prime ;{\theta }^{-}\right)$$

(18)

This is the target value for the Q-function at the next state (i + 1). Here, θ and θ⁻ represent the parameters of the current and target neural networks, respectively. The DQN incorporates experience replay to improve learning stability and efficiency. This process involves storing transitions (st, t, rt, ((t + 1))) in a replay buffer and stochastically sampling mini-batches from this buffer to update the network, as given via Equation (19).

$$D\leftarrow D\cup \left(st,at,rt,s\left(\left(t+1\right)\right)\right)$$

(19)

This stochastic sampling reduces correlation in the observation sequence and smooths over changes in the data distribution. To efficiently explore the action space and prevent the method from becoming stuck in local optima, a ϵ-greedy policy is utilized for this process. The DQN framework aims to maximize the action value function (s, a; θ) concerning actions a, balancing exploration and exploitation. This is achieved by selecting the action with the highest Q value with probability 1 − ϵ, while with probability ϵ, a stochastic action is chosen for this process. The value of ϵ decreases over temporal instance sets to gradually shift the focus from exploration to exploitation. The iterative training of the DQN involves adjusting the neural network weights θ to minimize the loss (θ) using gradient descent via Equation (20).

$$\theta \left(t+1\right)=\theta t-\eta \nabla \theta L\left(\theta \right)$$

(20)

where η represents the learning rate, and ∇(θ) represents the gradient of the loss function concerning the network parameters for this process. To ensure stable convergence during training, a target network is utilized, which is a replica of the DQN but updated less frequently via Equation (21).

$${\theta }^{-}\leftarrow \tau \theta +\left(1-\tau \right){\theta }^{-}$$

(21)

where τ controls the rate of update, promoting convergence stability. During deployment, the trained DQN method is employed to make decisions by selecting the action with the maximum Q value via Equation (22).

$$a*=argma{x}^{a}Q\left(s,a;\theta \right)$$

(22)

This deployment phase is crucial for identifying optimal locations for new EV charging stations and integrating user demand and geographic constraints to ensure resource allocation efficiency. The selection of a DQN method complements ALO and MFO techniques by providing a robust framework to handle dynamic optimization problems, adapting to temporal variations in user behavior and traffic patterns. The integration of the DQN with these bioinspired methods signifies a progressive approach towards sustainable transportation solutions, optimized through advanced computational methodologies. The performance parameters of the proposed system are compared with existing methods in terms of different metrics.

3.3. Method Configuration and Resource Requirement

The implemented pseudocode algorithm for the proposed model is as follows (Algorithm 1):

Algorithm 1. Deep Q-Network Implementation Process

Input: ▪ Traffic data (T). ▪ Geographical data (G). ▪ User demand (D).

Output: ▪ Optimal charging station locations (X, Y).

Process: Initialize DQN with ALO and MFO parameters. Load traffic data, geographical data, and user demand. Execute ALO to identify high-demand zones. Feed ALO output to the DQN for further optimization. Execute MFO to optimize energy utilization. Update Q values based on rewards from energy efficiency and user satisfaction. Repeat steps 3–6 until convergence is achieved. Output optimal charging station locations.

The proposed method evaluated the corresponding computational complexity of DQN-integrated ALO and MFO in terms of processing time and hardware consumption during its training phase. The DQN model took around 10 h to train on a computer system with an Intel i7 processor and 16 GB of RAM, processing close to 100,000 data samples. In contrast, the time of training for the conventional GWO method was 14 h, thus showing a 28% saving on computational time via the proposed method. The real-time testing indicated an average decision time of 0.8 s compared to conventional approaches that took approximately 1.4 s, demonstrating its excellent computational effectiveness.

The method configuration adjusts DQN, ALO, and MFO algorithm settings for the best performance by striking a balance between exploration, convergence, and efficiency to increase the accuracy of the system. All vital hyperparameters of the deep Q-network (DQN) were configured to optimize and ensure that the model needs were performance and more rapid convergence. The learning rate was set up to start with 0.01 but reduced by 0.005 during every 50 succeeding episodes to avert overshooting and keep the training steady. The discount factor (γ) was set at 0.95 to encourage a long-term accumulation of rewards during the decision-making process. For learning stability improvements, the size of the replay buffer was kept at 10,000 transitions along with the mini-batch size, which was 32 for purposes of balancing computational efficiency and accuracy in learning. To this end, the epsilon was set at 1.0 at the start of 100 training episodes of decay: the linear decay leading to 0.01 incidentally gave an asymmetric balance in exploration and exploitation, which improved the accuracy of optimal charging station placement.

Table 1. Method configuration of various optimizers.

Deep Q-Network (DQN)	Ant Lion Optimizer (ALO)	Moth Flame Optimizer (MFO)
Learning Rate (α): 0.01	Population Size: 50 ant lions	Population Size: 50 moths
Discount Factor (γ): 0.95	Maximum Iterations: 100	Maximum Generations: 100
Replay Buffer Size: 10,000 transitions	Exploration Factor: Starts at 0.5 and decays by 5% per iteration	Flame Number Reduction Factor: Linear reduction from 50 to 1 over generations
Mini-batch Size: 32	-	-
Target Network Update Frequency: Every 5 episodes	-	-
Exploration Strategy: ϵ-greedy, starting from 1.0 and decaying to 0.01 over 100 episodes	-	-

shows the parameters considered for the optimization process.

3.4. Description of Dataset Used for Training and Evaluation

The data for the training and evaluation of this proposed system were gathered from sources of real-time traffic data, geographical information, and demand profiles from users among the urban electric vehicle service providers and traffic management systems. For training and evaluation of the proposed system, the key assumptions, preprocessing steps, and data sources are discussed to improve the reproducibility as follows:

3.4.1. Key Assumptions

• Data Depiction: Collect accurate data for EV charging patterns.
• Similarity of Station: Assume that charging stations work under comparable conditions unless mentioned otherwise.
• Time-based Uniformity: Make the assumption that trends seen during the time of data collection will remain constant throughout time.
• User Performance: Assume that user charging patterns do not change much due to other influences that are not taken into consideration in the data.

3.4.2. Preprocessing Steps

• Cleaning of Data: Fix discrepancies to guarantee the quality of the data.
• Aggregation of Temporal: Adjust data to regular intervals of time.
• Data Processing: Determines the energy consumption pattern, peak usage period, and charging average duration time.
• Standardization: Improves the performance of machine learning algorithm.
• Resemblance: Select the model.

3.4.3. Dataset Samples

The dataset samples are captured from various fields, as shown in Figure 4, to direct the algorithms towards efficient solutions. The dataset includes traffic patterns from urban areas with varied data on a high vehicle density in peak hours from traffic management systems. The geographic data highlight elevation and urban planning restrictions. The user demand was derived from electric vehicle registration databases/EV service providers across several cities. The dataset samples were constructed as follows.

The traffic data comprised nearly 120,000 movements of vehicles across important locations on a given day, where 40% consisted of peak-hour demands. The geographic data were obtained based on the grid network over a 10 km² area, with granularity down to 50 m accuracy. The data preparation activities are normalization, imputation of missing values, and denoising, ensuring that inputs to the DQN model are clean and standardized. The EV charging demand data are subdivided into three levels concerning demand, low (50 EVs/day), medium (250 EVs/day), and high (500 EVs/day), so the model generalizes over such different demand scenarios. The off-peak hours are assumed to maintain a uniform distribution of demand, as well as a 25% surge during peak hours, simulating a realistic demand scenario for urban EV charging stations [51,52,53,54].

4. Results and Discussion

To validate the proposed DQN framework, the proposed system is integrated with ALO and MFO algorithms. The simulated environment replicating urban and rural scenarios for EV charging station deployment was developed. This environment incorporates diverse variables, such as traffic density, geographic features, and user demand profiles, to closely mimic real-world conditions. Each method performance was tested under a variety of charging demand scenarios from the low density of EVs, i.e., 50 EVs per day, to a high density of EVs, i.e., 500 EVs per day, with incremental adjustments to test scalability and responsiveness. The simulations were run on a synthetic grid representing a mixed urban–rural landscape, approximately 10 km² in size, to validate the effectiveness of the system tested under various performance metrics.

4.1. Performance Characteristics

The proposed system can ensure that the EV charging infrastructure [55,56,57] improves several performance metrics, including the installation cost efficiency, peak and off-peak energy utilization, average wait time, availability of charging slots, user satisfaction, environmental impact, carbon emission reduction, scalability, and operational robustness [58,59,60,61,62,63]. The comparisons are made with respect to various performance parameters, as discussed in the following subsections.

4.1.1. Validation and Testing of Installation Cost Efficiency

We examined the total cost of installation per charging unit versus average usage intensity. The proposed DQN-integrated ALO and MFO method predicts historical available data and is benchmarked against conventional placement strategies for optimal planning of EV charging infrastructure in diverse environments. An iterative feedback loop allowed for continuous method refinement based on simulated performance outcomes. The results were cross-validated using K-fold validation to ensure the robustness and generalizability of the findings.

Table 2. Installation costs for different EV scenarios.

Method	Average Cost per Unit (USD)	Usage Intensity (Users/Day)
Proposed Method	1200	450
Refs. [58,59]	1500	400
Refs. [60,61]	1600	350
Refs. [62,63]	1700	300

gives a comparison of the installation cost efficiency of the proposed method with other conventional methods. The proposed method frees up daily resources through real-time demand analysis to place its stations wisely, thereby lowering the average cost per unit. Conventional methods, on the other hand, generally produce over-provision in low-demand markets or under-provision high-demand zones, thus raising per-unit costs. The DQN integration with ALO and MFO, in contrast, clusters high-demand regions for optimum resource allocation, which reduces per-unit installation costs by 20%. Furthermore, lesser infrastructure and energy losses also serve to minimize costs, improving deployment efficiency in general. The proposed method produces a significant reduction in installation costs per unit charge and high usage intensity, i.e., USD 1200 and 450 users per day, respectively.

4.1.2. Validation and Testing of Energy Utilization

The energy utilization considered the percentage of energy drawn from sources and peak demand-shaving capabilities. The proposed method outperforms other methods by leveraging optimized placement strategies, which reduce redundant expenditures and focus investments on high-demand areas.

Table 3. Energy utilization.

Method	Energy Utilization (%)
	Peak Hours	Off-Peak Hours
Proposed Method	85	92
Refs. [58,59]	75	85
Refs. [60,61]	70	80
Refs. [62,63]	65	75

compares the percentages of energy drawn from sources across different methods during peak and off-peak hours. Figure 5 represents a clear understanding of energy utilization for EV charging, especially during peak and off-peak hours for the conventional and proposed systems. The conventional systems’ energy utilizations are 65%, 70%, and 75% during peak hours. Similarly, 75%, 80%, and 85% are at off-peak hours. The proposed system enhances energy utilization effectively for EV charging, i.e., 85% of energy utilized during peak hours and 92% of energy utilized during off-peak hours. From the quantitative analysis of the results, we surmised that the proposed method more effectively utilizes energy than the other methods, which is vital for sustainable operation.

4.1.3. Validation and Testing of User Satisfaction and Environmental Impact

Table 4. User satisfaction.

Method	Average Wait Time (Min)	Charging Slot Availability (%)
Proposed Method	5	95
Refs. [58,59]	10	90
Refs. [60,61]	15	85
Refs. [62,63]	20	80

assesses user satisfaction based on wait times and the availability of charging slots during peak hours. The proposed method scores higher in user satisfaction by predicting peak demand and adjusting supply dynamically. From the observations, the proposed system has less of a waiting time to recharge EV batteries, and the availability of charging slots for the next EV is very accurate, i.e., 5 min and 95%, respectively. So, this approach is user-centric in station planning and management.

EV adoption and resulting carbon emission reduction is the best solution for sustainable transportation and to support an eco-friendly environment. The proposed method effectively minimizes environmental impact through strategic station placement by reducing 30% of carbon emissions, as observed in

Table 5. Environmental impact.

Method	Carbon Emissions Reduction (%)
Proposed Method	30
Refs. [58,59]	22
Refs. [60,61]	18
Refs. [62,63]	15

. The claimed 30% reduction in carbon emissions resulted from appropriate charging station placement, minimizing unnecessary distances for travel and optimizing energy consumption. The baseline was compared using historical data obtained from charging station placements, averaging about a 4.5 km distance traveled by each user. The suggested approach decreased each user’s average distance traveled to 2.8 km, resulting in savings of roughly 1.7 kg of CO₂ per charging cycle. The emission factor is calculated from the power grid’s energy mix, assuming 0.5 kg CO₂/kWh. The total carbon emissions were lowered from 15 tons per day to 10.5 tons per day by improving energy consumption and reducing trip distances, representing a 30% reduction. The method has a superior capacity to reduce carbon emissions, which reflects its environmental benefit.

4.1.4. Scalability with EV Deployment and Operational Robustness

Table 6. Scalability.

Method	Scalability Score (1–100)	Supported Increase in EV Adoption (%)
Proposed Method	90	50
Refs. [58,59]	70	35
Refs. [60,61]	60	30
Refs. [62,63]	50	25

evaluates the scalability of each method by simulating increased EV adoption rates and the corresponding impact on EV infrastructure and deployments. The proposed system has the highest scalability, i.e., 90, and EV adoption increases to 50%. Figure 6 presents statistical charts for the proposed and other existing methods. The proposed method exhibits robust scalability, which indicates its ability to accommodate future increased EV adoption without significant performance degradation.

Table 7. Operational robustness.

Method	Robustness Rating (1–10)	Performance Under Stress (%)
Proposed Method	9	95
Refs. [58,59]	7	85
Refs. [60,61]	6	80
Refs. [62,63]	5	75

analyzes the operational robustness and focuses on the method’s ability to maintain performance under varying conditions such as unexpected spikes in demand or temporary disruptions. The proposed method scores high in operational robustness, i.e., 9, showcasing its capacity to handle stress and maintain high levels of performance, i.e., 95%, crucial for the reliability of EV charging infrastructure and deployments. These evaluations collectively demonstrate the superior performance of the proposed method across multiple dimensions critical to the success of EV charging station deployment.

4.2. Power Quality Characteristics

The proposed system confirms that the EV battery current remains in a predetermined range, leading to a more consistent and stable charging operation. This further leads to improved power quality, by reducing voltage sag/swell, flickers, and notches, and maintaining an appropriate frequency and improved efficiency [64,65,66]. These power quality characteristics and their comparisons are discussed in the following subsections.

4.2.1. Voltage Characteristics

Based on these optimizations, an EV charging infrastructure was designed, and it was implemented with MATLAB Simulation R2021a. The results were obtained from various operating characteristics like voltage characteristics, frequency characteristics, and power characteristics of the EV charging process. Figure 7 depicts the voltage characteristics of the charging circuit under sag and swell conditions across a period of 10 s. The voltage profiles of an increasing pattern range from about 10 V to 70 V. The sag produced in the voltage waveform is 2 V in 2 s and the swell produced in the voltage waveform is 3 V in 1 s. The voltage profiles for several EV charging system optimization techniques are contrasted in this picture. Compared to un-optimized (purple and magenta) and GWO-based (red and blue) scenarios, the suggested approach (green and black with markers) maintains a higher voltage level, indicating better regulation. Slight variations are highlighted in the zoomed-in region (2.38–2.44 s), which demonstrates that the suggested approach has a sharper voltage rise, suggesting a better dynamic response. The fact that unoptimized cases continue to be at the lowest levels indicates laxer control. Overall, this figure confirms that the suggested strategy outperforms traditional techniques in terms of efficiency and voltage stability.

4.2.2. Voltage Characteristics—Overvoltage and Undervoltage Scenarios

Figure 8 presents the voltage characteristics of an EV charging circuit under sag, swell, overvoltage, and undervoltage for 10 s. At starting, all voltages will remain close to 12 V. At about the 2 s mark, it suddenly surges to around 13.08 V, thus depicting a swell at around 1.08 V, before springing back down to 12 V at about the 5 s mark to show an overvoltage situation. The un-optimized method still holds its trend and turns out to be responsive to voltage disturbances, with a bit more variability. The proposed method’s results remain very close to the GWO with minimal deviations from them and thus depict proper voltage control.

4.2.3. Voltage Characteristics—Flickers

Figure 9 shows the voltage characteristics of an EV charging circuit under flicker conditions for 10 s. There are three different levels on which the voltage profiles are shown. The proposed methods, however, always manage to hold the voltage slightly higher at about 12.05 V, meaning that flicker is effectively minimized. The un-optimized methods manifest a slight variability of about 11.95 V, thereby showing less efficient control of the flicker. The GWO methods indicate only moderate fluctuations of about 12 V. This view, therefore, allows for a side-by-side comparison, clearly depicting the superior performance of the proposed optimization techniques in terms of keeping the voltage stable under conditions of flicker.

4.2.4. Voltage Characteristics—Notches

Figure 10 illustrates the voltage characteristics of an EV charging circuit under notch conditions for 10 s. The voltages of the proposed methods remain a little higher at approximately 12.03 V, showing a superior performance in handling notches. The un-optimized methods give more variation at around 11.97 V, hence showing less effective notch mitigation. The un-optimized methods show more variation around the average value of 11.97 V. The GWO methods have fluctuations that are at a moderate level of about 12 V. This comparative presentation of the results shows that against notches, proposed optimization techniques can ensure better voltage stabilization compared with the un-optimized and original methods.

The proposed method is seen to have similar patterns of fluctuation during sag, swell, and notch conditions due to IEEE voltage regulation on all the tested methods. Nonetheless, the proposed method has a faster stabilization level, thereby reducing sag by 2 V and swell by 3 V, leading to improved power quality.

4.2.5. Frequency and Power Characteristics

The frequency and power characteristics can be observed in Figure 11 as follows. This shows a detailed analysis of the EV charging circuit for a time duration of 10 s. Figure 11a shows the frequency characteristics, which are constant during the entire simulation time and are held around a set point so no large frequency variations are present. To understand the power demand during charging, the simulation time is considered from 0 s to 10 s, as shown in Figure 11b. From the results, a stable growth of power can be observed as the power demand increases throughout charging. Figure 11c depicts that small voltage disturbances occur at different time durations. Based on these characteristics, it can be noted that the proposed method has better efficiency and can be deployed for real-time scenarios.

4.2.6. Efficiency and Output Voltage

Efficiency is the key parameter in power quality indices to indicate the performance of the system, as shown in Figure 12a. From the simulation, the proposed and existing system efficiencies were captured as 97.5% and 90%, respectively. From the observation of the output voltage waveform, the existing system output voltage was captured as 55 V, which deviated from the set value, and the proposed system output voltage was captured as 53 V, which is close to the set value depicted in Figure 12b.

4.3. Overall Performance of the Proposed System

Table 8. Ant lion optimizer (ALO) results.

Iteration	Best Cost (USD)	Location X (km)	Location Y (km)	Coverage Area (km²)
1	1200	5.2	7.3	2.5
10	1150	5.4	7.4	2.7
20	1120	5.5	7.5	3.0
50	1080	5.7	7.8	3.5
100	1050	5.8	8.0	4.0

delves into the quantitative data of the use case approach using data samples, reflecting the traffic patterns, user demand, and geographic characteristics for optimizing EV charging station placement. This bioinspired method explores potential sites for EV charging stations, focusing on minimizing infrastructural costs while maximizing geographical coverage.

ALO focuses on cost reduction and increased coverage area over iterations, indicating effective optimization of the initial conditions represented in Figure 13. Iteration 1 covers a distance of around 2.5 km², the optimal cost function is USD 1200, and the northward X and eastward Y distances are 5.2 km and 7.3 km, respectively. The number of iterations increases to 10, 20, 50, and 100 to obtain the optimal solution. At the 100th iteration, the proposed system obtained the best results. That is, the optimal coverage area increases to 4 km², and the optimized cost and locations are USD 1050, X 5.8 km, and Y 8 km, respectively.

Table 9. Moth flame optimizer (MFO) results.

Generation	Flame Fitness	Location X (km)	Location Y (km)	User Proximity Index
1	0.95	5.8	8.0	0.90
10	0.85	5.9	8.1	0.92
20	0.80	6.0	8.2	0.94
50	0.78	6.1	8.3	0.97
100	0.75	6.2	8.4	0.99

demonstrates that the quantitative analysis of the moth flame optimizer refines the selection of traffic data, flame fitness, and user behavior patterns. This stage is critical for aligning the EV charging stations with actual user needs and accessibility requirements. The results from MFO highlight an improvement in the user proximity index from 0.90 to 0.99 as the generations progress, suggesting that the moths (solutions) are converging towards the optimal flame (best solution). At iteration 1, the user proximity index is 0.90, the X and Y distances are 5.8 km and 8 km, and the flame fitness is 0.95. The number of iterations increases to 10, 20, 50, and 100 for the best solutions. At iteration 100, the user proximity index is 0.99, the locations of X and Y are 6.2 km and 8.4 km, and the flame fitness is 0.75, as represented in Figure 14.

The DQN optimizes both the cost and user satisfaction score by learning from each simulated deployment scenario. The quantitative analysis results are presented in

Table 10. Deep Q-network (DQN) results.

Episode	Reward	Total Cost (USD)	Location X (km)	Location Y (km)	Satisfaction Score
1	0.75	1050	6.2	8.4	0.95
50	0.85	1020	6.3	8.5	0.97
100	0.90	1000	6.4	8.6	0.99
200	0.95	980	6.5	8.7	1.00
300	1.00	950	6.5	8.8	1.00

. To obtain the best solution, this method conducts 300 episodes for which it records the total cost, rewards, locations, and user satisfaction. Figure 15 indicates that the method has achieved an optimal balance between cost reduction and user satisfaction, ensuring that the strategic placement of EV charging stations is both economically viable and highly efficient. In episode 1, the results indicate 0.75 rewards, USD 1050 total cost, X 6.2 km, Y 8.4 km, and user satisfaction 0.95. As the number of episodes increased to 300, rewards improved to 1, the total cost was reduced to USD 950, the X and Y distances increased to 6.5 km and 8.8 km, and the user satisfaction score was 1.

The final quantitative output is given in

Table 11. Overall results.

Method	Location X (km)	Location Y (km)	Final Cost (USD)	Coverage Area (km²)	User Satisfaction Score
Proposed method	6.5	8.8	950	4.0	1.00
Refs. [58,59]	7.0	9.0	1200	3.5	0.90
Refs. [60,61]	7.5	9.5	1300	3.0	0.85
Refs. [62,63]	8.0	10.0	1400	2.5	0.80

to consolidate the findings from the results. The superiority of the proposed method is in achieving a significant reduction in costs while maximizing user satisfaction and coverage area. When observing the statistics, it can be seen that the overall performance of the proposed system is superior to those of the existing methods.

Figure 16 presents a stochastic chart making comparisons of the proposed and existing systems with quantitative analysis. With the proposed system, the total cost is reduced from USD 1400 to USD 950, the covering area is increased from 2.5 km² to 4 km², and the percentage of user satisfaction is improved from 0.80 to 1. From the observations, it can be surmised that the strategic placement of EV charging stations as determined by the proposed method guarantees optimal infrastructural deployment, thereby enhancing the feasibility and sustainability of EV ecosystems. Thus, the strategic placement of EV charging stations as determined by the proposed method guarantees optimal infrastructural deployment, enhancing the feasibility and sustainability of EV ecosystems.

4.4. Summary and Discussion of Results

The suggested DQN integration with ALO and MFO systems performs better than traditional EV charging techniques in terms of sustainability, affordability, efficiency, and power quality. It achieves a steady frequency and 97.5% efficiency while reducing flicker (0.05 V), notches (0.03 V), swell (3 V), and voltage sag (2 V). We found 20% lower installation costs (USD 1200/unit) with 92% off-peak and 85% peak energy use. Moreover, 95% of charging slots were available, user happiness was at 1.0, and carbon emissions were 30% lower. Its 95% robustness and high scalability (90) make it the most adaptable, cost-effective, and efficient EV charging solution. The quantifiable results listed in

Table 12. Performance comparison between proposed and conventional systems.

Category	Parameter	Proposed Method	Conventional Method	Recommended Value
Power Quality	Voltage Sag Reduction	2 V (Superior)	1 V	Typically, voltage deviation should not be more than ±10% of nominal voltage
	Voltage Swell Reduction	3 V (Superior)	1.5 V
	Flicker Reduction	0.05 V (Superior)	0.08 V	Pst ≤ 1.0
	Notch Reduction	0.03 V (Superior)	0.06 V	Not explicitly defined by IEEE
	Frequency Stability	Constant (Superior)	Minor Variations	±0.2 Hz
	System Efficiency	97.5% (Superior)	92%	100%
Performance	Installation Cost	USD 1200/unit (Superior)	USD 1500–1700	Varies by region and technology
	Energy Utilization (Peak)	85% (Superior)	75%	Industry best practice ≥ 80%
	Energy Utilization (Off-Peak)	92% (Superior)	85%	Industry best practice ≥ 85%
	Charging Slot Availability	95% (Superior)	90%	Industry best practice ≥ 90%
	User Satisfaction Score	1.00 (Superior)	0.90	No std. value; target ≥ 0.95
	Carbon Emission Reduction	30% (Superior)	22%	-
	Scalability Score	90 (Superior)	70	-
	Operational Robustness	95% Under Stress (Superior)	85%	-

clearly demonstrate the efficacy of the proposed system. The proposed system performs better than the traditional system according to the quantitative analysis.

5. Conclusions

The placement of EV charging infrastructure/stations is essential for charging EV batteries to develop pollution-free transportation, thereby supporting an eco-friendly environment. This article proposes the concept of a deep Q network using multimodal bioinspired analyses like ALO and MFO for optimal planning and deployment of EV charging infrastructure to enhance performance and power quality. Several performance metrics are measured and examined for the proposed scheme, including installation cost efficiency, peak and off-peak energy utilization, average wait time, availability of charging slots, user satisfaction, environmental impact, carbon emission reduction, scalability, and operational robustness.

Further, various power quality parameters such as voltage characteristics (under sag, swell, overvoltage, undervoltage, flickers, and notches), power characteristics, frequency characteristics, and efficiency are measured and examined. To provide the optimal station location and effective energy use in real-time applications, the DQN combined with the ALO and MFO frameworks is used, which reacts dynamically to changes in EV demand, traffic congestion, and energy grid conditions. The following gives a summary and the major benefits of the proposed system based on the overall cumulative findings.

▪

The proposed DQN, ALO, and MFO integration improves availability and demand responsiveness by dynamically adjusting charging station distribution based on past and present traffic patterns using real-time learning.

▪

To ensure effective and responsive placement, the proposed DQN, ALO, and MFO integration adaptively fine-tunes charging station sites to correspond with new road networks, urban expansion, and changing grid circumstances.

▪

The proposed method has an average installation cost of USD 1200 per unit, which is significantly lower than competing alternatives (USD 1500, 1600, and 1700). This leads to a 20% cost efficiency increase over the least efficient method.

▪

The optimal energy utilization is 85% during peak hours, i.e., 10% higher than the next best method, and 92% during off-peak hours, i.e., 7% higher than the next best method.

▪

The proposed method increases customer satisfaction by reducing the wait time to 5 min and ensuring 95% charging slot availability, which is much higher than other methods.

▪

The proposed method reduces carbon emissions by 30%, demonstrating its significant environmental benefits when compared to competing methods that lower emissions by just 22%, 18%, and 15%, respectively.

▪

The proposed method is exceptionally robust, retaining 95% performance under stress, ensuring resilience and reliability in dynamic environments.

▪

The proposed system provides voltage waveforms that show stable performance, with a 2 V sag lasting 2 s and a 3 V swell lasting 1 s, exhibiting effective voltage control and operational stability.

▪

The proposed system stabilizes the voltage at 12.05 V, reducing flicker to only 0.05 V and improving overall system reliability and performance. The proposed method shows a better performance in managing voltage notches, keeping voltages at 12.03 V with only 0.03 V notches, indicating improved stabilization.

▪

The voltages of the proposed methods remain a little higher at approximately 12.03 V, which means the 0.03 V notches found show a superior performance in handling notches. This can ensure better voltage stabilization compared with the un-optimized and original methods.

▪

The proposed system ensures the frequency characteristics remain consistent during the simulation period and are maintained at a set point, preventing significant frequency fluctuations.

▪

The proposed method greatly increases system efficiency from 90% to 97.5%, hence improving overall system performance and energy consumption.

▪

The proposed approach supports a 50% increase in EV adoption and has an exceptional scalability score of 90.

Thus, based on the results for the proposed system, it is concluded that the proposed deep Q network with multimodal bioinspired analysis, such as ALO and MFO, outperforms existing methods. Therefore, it is recommended for optimal planning and deployment of EV charging infrastructural applications.

Limitations and Future Scope

▪

This work focused on simulation-based validation, and future work will explore city-wide deployment models with extensive EV fleets and distributed charging hubs to further assess scalability.

▪

The present study leverages the DQN integration with the ALO and MFO framework to optimize EV charging stations in real-time. The usefulness of the proposed approach has been established via actual performance comparisons, including enhanced energy utilization, charging slot availability, and cost efficiency. However, statistical significance testing may be used in subsequent research to provide additional validation in extensive real-world implementations.

▪

The limitations in grid infrastructure are among the possible hardware restrictions.

▪

To guarantee successful market integration, regulatory obstacles including dynamic pricing rules and cybersecurity threats must be taken into account.

▪

To enable large-scale deployment, legislators, grid operators, and EV manufacturers must work together across several stakeholders due to practical feasibility issues like cybersecurity risks, user adoption obstacles, and economic sustainability.

▪

The proposed scheme, despite scoring all the goals concerning setting up EV charging infrastructure, was not without limitations. The first and foremost condition essential for the model performance is proper input in the form of real-time traffic and demand patterns. In this regard, access to very high computational power to train DQNs also poses another challenge for small-scale deployment. Therefore, future work can focus on integrating carbon emissions by using renewable energy sources, whereas lightweight deep-learning-models will enable real-time deployment in resource-constrained environments. Furthermore, the extension of the model to a multi-agent system would allow the collaborative optimization of several charging stations for better overall grid stability.