Markov链与Q-Learning算法的超轻度混动汽车模型预测控制  

Model predictive control of super-mild hybrid electric vehicle based on Markov chain and Q-Learning

作　　者：尹燕莉[1,2] 马永娟 周亚伟 王瑞鑫 詹森 马什鹏 黄学江 张鑫新 YIN Yanli;MA Yongjuan;ZHOU Yawei;WANG Ruixin;ZHAN Sen;MA Shenpeng;HUANG Xuejiang;ZHANG Xinxin(School of Mechatronics&Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Baotou Bei-Ben Heavy Vehicle Co.Ltd,Baotou 014000,China)

机构地区：[1]重庆交通大学机电与车辆工程学院,重庆400074 [2]包头北奔重型汽车有限公司,包头014000

出　　处：《汽车安全与节能学报》2021年第4期557-569,共13页Journal of Automotive Safety and Energy

基　　金：重庆市教委科学技术研究项目(KJQN201800718);重庆市技术创新与应用发展(重点项目)(cstc2020jscx-dxwtBX0025)。

摘　　要：为了同时兼顾能量管理策略的全局最优性与运算实时性,本文提出了基于Markov链与Q-Learning算法的超轻度混合动力汽车模型预测控制能量管理策略。采用多步Markov模型预测加速度变化过程,计算得出混合动力汽车未来需求功率;以等效燃油消耗最小与动力电池荷电状态(SOC)局部平衡为目标函数,建立能量管理策略优化模型;采用Q-Learning算法对预测时域内的优化问题进行求解,得到最优转矩分配序列。基于MATLAB/Simulink平台,对于ECE_EUDC+UDDS循环工况进行仿真分析。结果表明:采用Q-Learning求解的控制策略比基于动态规划(DP)求解的控制策略,在保证燃油经济性基本保持一致的前提下,仿真时间缩短了4 s,明显地提高了运行效率,实时性更好。A model predictive control energy management strategy for super-light hybrid electric vehicles(HEV)was proposed to take into account the global optimality of the energy management strategy and the real-time operation at the same time based on Markov chain and Q-Learning algorithm.The multi-step Markov model was used to predict the acceleration change process to calculate the future required power of HEV.An energy management strategy optimization model was established by taking the minimum equivalent fuel consumption and the local balance of the state of charge(SOC)of power battery as the objective function.The Q-learning algorithm was used to solve the optimization problem in the prediction time domain to obtain the optimal torque distribution sequence.The simulation analysis was carried out under the ECE_EUDC+UDDS cycle conditions on the base of MATLAB/Simulink platform.The results show that the control strategy solved by the Q-Learning solution reduces the simulation time by 4 s under the same fuel economy condition,comparing with the control strategy based on the dynamic programming(DP)solution.The proposed control strategy can significantly improve the operating efficiency and has better real-time performance.

关 键 词：超轻度混合动力汽车 模型预测控制 Markov链(Markov chain) Q-Learning算法 多步Markov模型 能量管理 

分 类 号：U469.72[机械工程—车辆工程]