RLC: A Reinforcement Learning Based Charging Scheme for Battery Swap Stations

Abstract

Battery Swapping Station (BSS) is emerging as a promising solution to the prevalent issue of range anxiety among Electric Vehicle (EV) users. Typically, BSS replaces the drained battery of an incoming EV with a fully charged one. In this thesis, we propose a cutting-edge battery charging and swapping approach for BSS, termed Reinforcement Learning-based Charging (RLC). This innovative strategy enables the provision of partially charged batteries to EVs with lower energy requirements while simultaneously minimizing the overall energy expenditure of BSS. Technically, RLC employs an ensemble learning-based forecasting module to predict the electricity demand pertaining to EV battery swapping. Furthermore, it utilizes Deep Deterministic Policy Gradient (DDPG) to strategize the battery charging process within BSS. Specifically, the predicted electricity demand is fed into the DDPG agent, enabling it to adapt to the changing patterns of EV arrivals. Our experimental results indicate that RLC outperforms the baseline charging schemes in terms of overall electricity cost, average SOC discrepancy rate, and battery service rate. Our future work will focus on incorporating more real-life elements, such as dynamic electricity price and battery degradation, to further refine the proposed learning-based charging scheme.