基于并行深度强化学习的混合动力汽车能量管理策略优化

Energy Management Strategy of Hybrid Vehicle Based on Parallel Deep Reinforcement Learning

提出了能量管理策略优化方法,通过深度强化学习中的深度确定性策略梯度(deep deterministic policy gradient,DDPG)算法调整等效因子,以提高燃油利用率,达到SOC保持与油耗降低的目标。受到边缘计算架构启发,建立了基于并行的深度强化学习算法以加快学习速度。在FTP72工况的仿真结果表明:提出的算法使油耗相对基于PID控制器的传统A-ECMS算法降低了7.2%,而以边缘计算架构建立的并行深度强化学习算法使收敛速度提高了334%。

In this paper,the proposed energy management strategy optimization method is adopted with the Deep Deterministic Policy Gradient(DDPG)algorithm to adjust the equivalent factor,in order to improve fuel efficiency and achieve the objectives of SOC maintenance.In addition,inspired by the edge computing framework,a parallel deep reinforcement learning algorithm is established to accelerate the learning speed.The simulation results under FTP72 driving cycle show that the proposed algorithm can reduce fuel consumption by 7.2%compared with the traditional A-ECMS algorithm based on a PID controller,and the parallel deep enhancement learning algorithm based on edge computing framework increases the convergence speed by 334%.