Learning and Unlearning to Operate Profitable Secure Electric Vehicle Charging

• Published in: IEEE transactions on industrial informatics 2024-09, Vol.20 (9), p.11213-11223
• Main Authors: Lee, Sangyoon, Choi, Dae-Hyun
• Format: Article
• Language: English
• Subjects: Data models; Data privacy; Electric vehicle (EV); Electric vehicle charging; electric vehicle charging scheduling; Learning; machine unlearning (MuL); Perturbation; Perturbation methods; Privacy; privacy preservation; Q-learning; robust deep reinforcement learning; Robustness; Storage systems; Task scheduling; Vehicle-to-grid
• ISSN: 1551-3203; 1941-0050
• DOI: 10.1109/TII.2024.3396524

This study proposes a two-stage learning and unlearning framework that ensures profitable and privacy-preserving charging at electric vehicle charging stations (EVCSs) integrated with solar photovoltaic and energy storage systems (ESSs). In Stage 1, a robust dueling deep Q -network method combined with an optimization-based reward function is employed to perform the following two tasks: 1) an increase in the EVCS profit via the selection of charging poles for profitable charging scheduling of the reserved EVs based on ESS operation and 2) enhancement of the robustness to adversarial perturbations. In Stage 2, a computationally efficient machine unlearning method is adopted to protect the data privacy of the reserved EVs by completely erasing their traces of private data during unlearning. The simulation results demonstrate the advantages of the proposed framework in terms of profitable charging pole utilization, robustness against adversarial perturbations, accuracy of the unlearned EV charging model, and training time.