AI-Driven Smart Charging System Optimizes EV Experience in Coupled Power-Transport Networks
As electric vehicle (EV) adoption accelerates across China, urban infrastructure faces mounting pressure from rising charging demand. Traffic congestion at charging stations, prolonged waiting times, and grid instability during peak charging hours have become persistent challenges. In response, researchers from State Grid Fujian Electric Power Co. and Fuzhou University have developed a groundbreaking solution that leverages artificial intelligence to transform how EVs are guided to charging stations. The new strategy, rooted in a multi-agent graph reinforcement learning (MAGRL) framework, promises to enhance user experience, balance charging station loads, and support grid stability—all in real time.
The research, led by Liang Huabin from State Grid Fujian Electric Power Co., Lin Jian from the same organization, and Yuan Yujuan from Fuzhou University’s School of Electrical Engineering and Automation, introduces a novel approach to EV charging guidance within complex power-transportation coupling systems. Published in Hubei Electric Power, the study presents a dual-timescale framework that dynamically adapts to fluctuating traffic and power grid conditions, offering a scalable and privacy-conscious solution for urban EV ecosystems.
The core innovation lies in shifting the decision-making paradigm from vehicle-centric to road-centric intelligence. Traditional methods often treat each EV as an isolated agent, optimizing routes based on static criteria such as distance or pre-set pricing. However, this approach fails to capture the dynamic interplay between traffic flow, charging station utilization, and electricity pricing. The MAGRL model addresses this by treating roads as intelligent agents, enabling a more holistic and coordinated response to real-time conditions across the entire network.
At the heart of the system is a multi-objective optimization model that simultaneously considers four critical performance indicators: user time cost, economic cost, charging station load balance, and grid voltage stability. These metrics reflect the interests of all stakeholders—drivers seeking convenience and affordability, charging station operators aiming for efficient resource use, and utilities striving to maintain grid reliability. By balancing these competing objectives, the model ensures that no single aspect is optimized at the expense of others.
The implementation of this strategy unfolds through a two-layered temporal architecture. On the slow timescale—updated every 10 minutes—the system computes node marginal prices (LMPs) using second-order cone optimization of the distribution network’s optimal power flow. This process integrates real-time data on power generation, load demand, and renewable energy output to determine the cost of electricity at each node in the grid. The resulting LMPs serve as dynamic price signals that guide EVs toward stations where charging is not only cheaper but also less likely to stress the local grid.
On the fast timescale—operating at one-minute intervals—the MAGRL algorithm takes over, making real-time routing decisions for EVs navigating the transportation network. Unlike conventional shortest-path algorithms that prioritize distance alone, this AI-driven method evaluates a broader set of factors, including current road speeds, queue lengths at charging stations, expected waiting times, and real-time electricity prices. By processing this multidimensional data, the system can recommend routes that minimize total cost—both in terms of time and money—while avoiding congestion and supporting grid health.
A key technical advancement is the use of graph convolutional networks (GCNs) to model the complex topology of urban transportation and power networks. Roads are represented as nodes in a graph, with connections reflecting intersections and possible route transitions. This structure allows the system to efficiently propagate information across the network, enabling intelligent agents on one road segment to anticipate conditions on downstream segments. The integration of GCNs with deep Q-networks (DQNs) enables the system to learn optimal policies through continuous interaction with the environment, refining its decision-making over time without requiring explicit programming for every possible scenario.
One of the most significant contributions of this work is its emphasis on user privacy. As AI systems increasingly rely on personal data, concerns about data security and misuse have grown. The research team addressed this by designing an independent information module that processes sensitive EV data—such as state of charge (SOC) and location—locally before transmitting only anonymized features to the central system. This design ensures that raw user data never leaves the vehicle or local device, significantly reducing the risk of privacy breaches while still allowing the AI to make informed routing decisions.
The system’s ability to protect user data without sacrificing performance marks a critical step forward in the ethical deployment of AI in public infrastructure. It demonstrates that intelligent systems can be both effective and respectful of individual rights, setting a precedent for future smart city applications.
To validate the effectiveness of their approach, the researchers conducted simulations on a realistic testbed combining a 25-node urban traffic network with the IEEE 33-node power distribution system. Four fast-charging stations were integrated into the network, each equipped with varying numbers of charging points to reflect real-world diversity in station capacity. The simulation scenarios varied the proportion of EVs requesting charging—from 12% to 28% of the total fleet—allowing the team to assess the system’s scalability and robustness under different demand levels.
The results were compelling. Compared to the traditional shortest-path method (Dijkstra’s algorithm), the MAGRL-based system consistently outperformed across all key metrics. Users experienced reduced waiting times, with average queue durations decreasing significantly, especially during peak demand periods. Economic savings were also notable, as the system successfully steered EVs toward stations with lower real-time electricity prices, reducing overall charging costs.
More importantly, the system demonstrated a remarkable ability to balance the load across charging stations. In the shortest-path scenario, EVs naturally converged on centrally located stations, leading to severe congestion and underutilization of peripheral facilities. In contrast, the MAGRL system distributed demand more evenly, ensuring that all stations operated closer to their optimal capacity. This balanced utilization not only improved service quality but also extended the lifespan of charging infrastructure by preventing overuse of specific locations.
Grid stability was another area of significant improvement. Large-scale EV charging can cause voltage fluctuations, particularly in distribution networks with limited reactive power support. By strategically directing EVs to stations connected to stronger grid nodes and avoiding simultaneous charging surges, the MAGRL system reduced average voltage deviation by up to 30% compared to the baseline method. This outcome is crucial for utilities, as it reduces the need for costly grid reinforcements and enhances the reliability of power delivery.
The adaptability of the system was further confirmed through its consistent performance across different demand scenarios. As the number of charging requests increased, the advantages of the MAGRL approach became even more pronounced. While the shortest-path method struggled with congestion and inefficiency under high load, the AI-driven system maintained its ability to optimize routing decisions, demonstrating strong scalability and resilience.
Beyond technical performance, the study highlights the importance of interdisciplinary collaboration in solving modern energy challenges. The integration of transportation engineering, power systems analysis, and machine learning expertise was essential to developing a solution that works across domains. The success of this project underscores the need for holistic thinking in urban planning, where siloed approaches can no longer address the interconnected nature of mobility and energy systems.
From a policy perspective, the findings suggest that governments and utilities should invest in intelligent charging infrastructure that goes beyond basic connectivity. Real-time pricing signals, dynamic routing guidance, and AI-powered coordination can play a vital role in managing the transition to electric mobility. Moreover, the privacy-preserving design of the system offers a blueprint for how public services can leverage data without compromising individual rights—a critical consideration in the era of digital governance.
For EV owners, the implications are clear: smarter charging guidance means less time spent waiting, lower bills, and greater confidence in the reliability of the charging network. For city planners, it offers a tool to manage urban mobility more efficiently, reducing traffic congestion and environmental impact. For grid operators, it provides a mechanism to integrate growing EV loads without compromising system stability.
Looking ahead, the research team plans to expand the system to include bidirectional vehicle-to-grid (V2G) capabilities, allowing EVs to not only draw power but also supply it back to the grid during peak demand. This evolution could turn millions of vehicles into distributed energy resources, further enhancing grid flexibility and renewable energy integration.
In conclusion, the work by Liang Huabin, Lin Jian, and Yuan Yujuan represents a major leap forward in the intelligent management of electric mobility. By combining advanced AI techniques with a deep understanding of power and transportation systems, they have created a solution that is not only technically sound but also socially responsible and economically viable. As cities around the world grapple with the challenges of decarbonization and digital transformation, this research offers a compelling vision of what is possible when innovation is guided by real-world needs and ethical principles.
The study was published in Hubei Electric Power, a leading journal in power system research, and has been recognized for its methodological rigor and practical relevance. Its findings are expected to influence the design of next-generation charging networks, paving the way for a more sustainable and user-friendly electric transportation future.
Liang Huabin, Lin Jian, Yuan Yujuan, Hubei Electric Power, DOI: 10.19908/j.hep.2024.03.002