基于DDQN的燃料电池混动车辆能量管理策略研究

Research on Energy Management Strategy of Fuel Cell Hybrid Vehicle Based on DDQN

以燃料电池混合动力公交车为研究对象,建立相关模型针对燃料经济性为目标对该车的能量管理策略展开研究,提出一种基于专家经验引导优化方法的DDQN(Double Deep Q-Network)能量管理策略并与基于规则、动态规划能量管理策略进行仿真对比。分析相关数据后发现其等效氢耗量313.16g,动态规划与基于规则方法等效耗氢量分别为311.45g与330.37g,所提出的DDQN方法与动态规划方法仅相差0.55%;又在离线DDQN所训练的强化学习智能体基础上,针对测试工况(CHTC-B)继续对该智能体进行训练更新并在线应用,其中怠速功率以上到20kW高效区间占比从22.92%提升到32.85%,而40kW以上工作区间从5.65%下降到2.36%。修正更新后的DDQN策略燃料等效消耗为283.64g,相比离线DDQN提高了2.5%,研究证明了所提出的DDQN能量管理策略的有效性。

The fuel cell hybrid bus is used as the research object and a model is developed to study the energy management strategy of the bus with the objective of fuel economy.A DDQN(Double Deep Q-Network)energy management strategy based on an expert experience-guided optimisation method is proposed and compared with a rule-based,dynamic planning energy management strategy.Analysis of the relevant data shows that the equivalent hydrogen consumption is 313.16g,while the equivalent hydrogen consumption of the dynamic planning and rule-based methods are 311.45g and 330.37g respectively,with a difference of only 0.55%between the proposed DDQN method and the dynamic planning method.Based on the reinforcement learning intelligences trained by the offline DDQN,the intelligences continued to be trained and updated for the test operating conditions(CHTC-B)and applied online,where the efficient range from above idle power to 20kW improved from 22.92%to 32.85%,while the operating range above 40kW decreased from 5.65%to 2.36%.The updated DDQN strategy after correction for battery SOC has a fuel equivalent consumption of 283.64g,an improvement of 2.5%compared to the offline DDQN,and the study demonstrates the effectiveness of the proposed DDQN energy management strategy.