Integrating Differential Privacy and Reinforcement Learning for Efficient and Location Privacy-Preserving Electric Vehicle Charging/Discharging Scheduling

The surge in electric vehicle (EV) adoption under-scores the need for intelligent strategies to manage grid load while preserving user privacy. Addressing this dual challenge, this paper introduces a novel two-stage framework that seam-lessly integrates differential privacy and reinforcement learning to facilitate privacy-preserving EV charging and discharging scheduling. Our method navigates the intricacies of real-time, location-based EV charging station allocation, and scheduling in a dynamically evolving environment, while ensuring user location privacy. The first stage applies differential privacy to protect real-time location data during the charging station allocation process. The second stage employs reinforcement learning to formulate charging and discharging schedules that accommodate EV user needs and contribute to peak load shaving in the grid. Extensive experimentation validates the efficacy of our approach, which reduces total driving distance by approximately 80% compared to random allocation, ensures user 80% satisfaction ratio, with an average over 100% State of Charge (SoC) fulfillment, assists in grid load balancing by curtailing peak load by 15%, all while maintaining robust privacy protection with a location information leakage probability of less than 2%. This proposed framework provides a comprehensive, efficient solution to the challenges inherent in EV charging/discharging scheduling, paving the way for the next generation of smart grid management in intelligent transportation systems.