Abstract: To improve the fuel economy of hybrid vehicles, plug-in hybrid electric vehicles are chosen as research subject. The deep deterministic policy gradient (DDPG) algorithm is used to optimize the equivalence factor and battery state of charge (SOC) in equivalent fuel consumption minimization strategy (ECMS). By integrating deep learning's perception abilities with reinforcement learning's decision-making processes, the optimization control of energy management in the hybrid electric vehicle is accomplished. A complete vehicle simulation model is created in MATLAB/ Simulink for experimental purposes. The outcomes demonstrate that, under condition of meeting the vehicle's normal driving power demands, fuel consumption based on DDPG algorithm optimized equivalent fuel consumption minimization algorithm is significantly lower than that of double deep Q network (DDQN) and traditional ECMS, while maintaining battery SOC balance, ensuring multi-objective optimization.

Key words: plug-in hybrid electric vehicle; deterministic policy gradient algorithm; equivalent fuel consumption minimization strategy; equivalent factor; multi-objective balance

摘要： 为提高混动汽车的燃油经济性，以插电式混合动力汽车作为研究对象，采用深度确定 性策略梯度（DDPG）算法对等效燃油消耗最小策略（ECMS）的等效因子和电池荷电状态 （SOC）进行优化。将深度学习的感知能力与强化学习的决策能力相结合，解决了对混合动 力汽车的能量管理优化问题。在 MATLAB/Simulink 中搭建整车仿真模型进行试验，结果表明， 采用新欧洲驾驶循环特定工况，在满足车辆正常行驶动力需求下，基于 DDPG 算法优化的等 效油耗极小值算法燃油消耗明显低于双深度 Q 网络（DDQN）和传统的 ECMS，同时维持电 池 SOC 的平衡，保证了多目标平衡性。

关键词: 插电式混合动力汽车；确定性策略梯度算法；等效燃油消耗最小控制策略；等效因 子；多目标平衡