The aggregated flexibility of electric vehicle (EV) charging presents significant potential for mitigating the adverse impacts that EV mass adoption imposes on the power grid. A three-stage framework, consisting of aggregation, bidding, and disaggregation, is proposed to enable smart charging for multiple EVs. To address modeling errors and uncertainties, a data-driven reinforcement learning (RL) algorithm is employed to optimize the bidding decisions for the aggregated EVs and to coordinate the bidding and disaggregation processes. The effectiveness of the proposed method is validated using both simulated and real-world datasets. With RL-based smart EV charging, the daily costs for each season are reduced, and the overload on distribution transformers is alleviated. Using simulated EV charging data for charging at home, the average daily cost is reduced by approximately 11%. When the proposed algorithm is applied to the real-life EV chargers in a commercial hub in England, UK (where EVs predominantly charge during the day), the average daily cost is reduced by approximately 3.83%, and transformer capacity violations are eliminated for all seasons except Autumn. The cost reduction with the commercial hub example is lower than that of the home charging case, because most cars park in the commercial hub for a short period of time during daytime and provide little flexibility for rescheduling.
Reinforcement Learning Based Smart Charging for Electric Vehicle Fleet
Communic.Comp.Inf.Science
International Symposium on Intelligent Technology for Future Transportation ; 2024 ; Helsinki, Finland October 20, 2024 - October 22, 2024
2025-03-12
9 pages
Article/Chapter (Book)
Electronic Resource
English