基于强化学习的P2结构混动车辆能量优化控制  

Reinforcement Learning-Based Energy Optimization Control for Hybrid Vehicles with P2 Structure

作　　者：胡作磊 童紫威 刘平 施伟[4] HU Zuolei;TONG Ziwei;LIU Ping;SHI Wei(College of Computer Science and Electronic Engineering,Hunan University,Changsha,Hunan 410082,China;Hunan Jinhe Landscape Construction Engineering Co.,Ltd.Limited,Changsha,Hunan 410148,China;China(Hunan)Pilot Free Trade Zone,Changsha Area,Changsha,Hunan 410137,China;School of Information Engineering,Changsha Medical University,Chang,Hunan 410219,China)

机构地区：[1]湖南大学信息科学与工程学院,湖南长沙410082 [2]湖南锦和园林建设工程有限公司,湖南长沙410148 [3]中国(湖南)自由贸易试验区长沙片区临空管理委员会,湖南长沙410137 [4]长沙医学院信息工程学院,湖南长沙410219

出　　处：《计算技术与自动化》2024年第3期1-7,30,共8页Computing Technology and Automation

基　　金：国家自然科学基金资助项目(201601420565)。

摘　　要：针对P2结构混动车辆,提出了一种基于强化学习的自适应等效燃油消耗最小策略(RL-ECMS),通过两个智能体实现等效因子的自适应更新和车辆扭矩的动态分配,以适应不断变化的驾驶需求。通过MATLAB/Simulink仿真平台对比了RL-ECMS与传统ECMS和基于规则的控制策略。结果表明,RL-ECMS在FTP75和FTP75-Highway两种典型驾驶工况下均能实现更低的燃油消耗,且不影响车辆性能。同时测试了未经训练的ECE典型工况,结果表明本文所提算法同样具有良好的泛化性与鲁棒性。A reinforcement learning-based adaptive equivalent fuel consumption minimization strategy(RL-ECMS)is proposed for P2-structured hybrid vehicles,which realizes adaptive updating of the equivalence factor and dynamic allocation of vehicle torque through two intelligences to adapt to the changing driving demands.The RL-ECMS is compared with the conventional ECMS and rule-based control strategies through the MATLAB/Simulink simulation platform.The results show that the RL-ECMS can achieve lower fuel consumption under both FTP75 and FTP75-Highway typical driving conditions without affecting vehicle performance.The untrained ECE typical conditions are also tested,and the result shows that the algorithm proposed in this paper also has good generalization and robustness.

关 键 词：混合动力车辆 能量优化 强化学习 自适应控制 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]