基于深度强化学习的混合动力汽车能量管理策略

To resolve the problem of poor adaptability to varying driving cycles when energy management strategy for hybrid electric vehicles is running online, a design method of energy management strategy (EMS) with deep reinforcement learning ability is proposed. The presented method determines the optimal change rate of engine power based on the deep deterministic policy gradient algorithm and then establishes the power management strategy of the onboard energy system. The established control strategy includes a two-layer logical framework of offline interactive learning and online update learning. The control parameters are dynamically updated according to the vehicle operation characteristics to improve the vehicle energy-saving effect in online applications. To verify the proposed control strategy, the effectiveness of the algorithm is analyzed with the practical vehicle test data in Shenyang, and compared with the control effect of the particle swarm optimization algorithm. The results show that the proposed deep reinforcement learning EMS can achieve energy-saving effects better than particle swarm optimization-based strategy. Especially when the driving characteristics of vehicles change suddenly, deep reinforcement learning control strategy can achieve better adaptability.