Personalized Federated Management and Load Balancing for Multiple Charging Stations

The rapid growth of charging stations (CSs) creates grid challenges, especially with load fluctuations. Optimizing CS management and electricity trading for electricity retailers (ERs) is key for stability. Centralized methods risk privacy leakage, while distributed methods may overfit due to limited electric vehicle (EV) data. To address this, we propose a hierarchical personalized federated reinforcement learning (HPFRL) method for CS management. HPFRL enables CSs to optimize EV pricing and scheduling while balancing load for the ER. It trains local models on private data to ensure personalized decision-making, with a CS aggregator updating models using a similarity-based strategy, protecting privacy and reducing overfitting. The global ER influences CS decisions via real-time electricity pricing, stabilizing the system. We prove the existence of a Nash equilibrium for the ER-CS optimization problem. Experiments show HPFRL achieves high rewards, effectively performing peak shaving and valley filling, demonstrating its practicality.