Reinforcement learning based adaptive power sharing of battery/supercapacitors hybrid storage in electric vehicles

The battery lifetime of Electric vehicles (EVs) is affected by hot temperatures and high charging and discharging effective battery current. Hybrid energy storage systems (HESS) coupling the best attributes of the battery with supercapacitors (SCs) help extend the battery lifetime and improve the EV storage performances. The key to a successful HESS at extending the battery lifetime is to adopt the appropriate Energy Management System (EMS) that ensures the best power sharing between battery and SCs. This paper proposes an innovative real-time optimization-based EMS with low computational costs and high adaptability to variable and commute driving profiles. The proposed EMS is organized in two levels. The lower level implements a rule-based frequency power sharing control. The upper level performs Reinforcement Learning (RL) optimizations to learn and adapt the best power sharing configuration considering real-time information and actual load conditions. An experimental test bench is developed and experimental measurements are conducted. The obtained results confirmed the effectiveness of the proposed EMS to provide the best trade-offs between simple implementation, computation time, solution optimality, real-time performance, and good adaption to variable driving conditions.