Real-Time Distributed Charging Station Recommendation for Electric Vehicles: A Federated Meta-RL Approach

Abstract

This paper presents a real-time distributed charging station recommendation framework for electric vehicles (EVs) that integrates Federated Learning (FL), Meta-Reinforcement Learning (Meta-RL), and edge intelligence to enable intelligent, privacy-preserving, and adaptive charging decisions in dynamic urban environments. Traditional EV charging recommendation systems often rely on centralized architectures that suffer from high latency, scalability limitations, and privacy concerns due to the continuous sharing of sensitive user mobility and charging behavior data. To address these challenges, the proposed approach introduces a Federated Meta-RL framework in which distributed charging stations collaboratively learn an optimal recommendation policy without exchanging raw user data. Each charging station functions as an autonomous edge agent that locally observes real-time contextual information, including station occupancy, queue length, charging cost, traffic conditions, grid load, and user preferences, to generate personalized charging recommendations.The proposed system employs Meta-Reinforcement Learning to learn a generalized recommendation policy that can rapidly adapt to changing demand patterns, unseen traffic conditions, and heterogeneous charging environments. Through Federated Learning, local model updates from distributed charging stations are periodically aggregated into a shared global meta-policy, enabling collaborative learning while preserving data privacy and reducing communication overhead. This distributed framework supports low-latency, scalable, and real-time charging recommendations by combining local station intelligence with global knowledge transfer. Experimental results demonstrate that the proposed approach significantly improves charging efficiency, reduces waiting time, balances charging station utilization, and enhances recommendation accuracy when compared with conventional centralized and standalone reinforcement learning methods. The proposed framework offers a scalable and privacy-aware solution for next-generation intelligent EV charging infrastructure in smart transportation ecosystems.