Connected Hybrid Electric Vehicles with Policy-Sharing: A Combination of Deep Reinforcement Learning and Federated Learning

Deep reinforcement learning (DRL) algorithms have shown impressive performance in managing single-hybrid electric vehicles (HEVs) during specific and predefined driving cycles. However, when it comes to real-world scenarios, groups of vehicles that must handle arbitrary driving patterns present a challenge for DRL methods, hindering their practical implementation. This study introduces a novel framework for achieving a generalized policy-sharing energy management strategy (EMS) across a group of connected HEVs. Using deep reinforcement learning (DRL), we first obtain the local optimal strategy and then incorporate policy sharing based on a federated learning (FL) algorithm. Our findings indicate that the generalized policy-sharing strategy can achieve a close total reward to the DRL strategy, which has access to driving cycle information in advance.