基于动态规划与神经网络的增程式电动汽车能量管理策略研究  

Study on Energy Management Strategy for Extended-range Electric Vehicle Based on Dynamic Programming and Neural Network

作　　者：席利贺[1] 张欣[1] 吴建政[1] 宋雯[1] XI Li-he;ZHANG Xin;WU Jian-zheng;SONG Wen(Beijing Key Laboratory of Powertrain for New Energy Vehicle,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学新能源汽车动力总成技术北京市重点实验室,北京100044

出　　处：《公路交通科技》2018年第9期128-136,共9页Journal of Highway and Transportation Research and Development

基　　金：国家科技支撑计划项目(2015BAG05B00);高等学校博士学科点专项科研基金项目(20130009110029)

摘　　要：设计了一种具有实时控制能力的增程式电动汽车混合型能量管理策略。首先建立了面向能量管理策略优化的增程式电动汽车整车模型。根据能量管理策略特点,将优化目标设置为增程器系统燃油消耗及动力电池当前SOC值与目标值之间差值的总和。再采用动态规划算法求解增程式电动汽车在给定行驶工况下的能量管理优化问题,从而获得了增程器开启时刻与输出功率优化结果。但由于动态规划算法需要已知详细的工况信息,很难应用于实车实时控制,而且从动态规划优化结果中不易提取控制规则,因此利用BP神经网络算法对优化结果进行离线训练,建立了增程器输出功率与车辆行驶状态参数间的非线性映射关系,得到了具有实时控制能力的神经网络控制模型。在采用BP神经网络训练时,根据车辆各个状态参数在CAN总线中的传输精度,对神经网络输入层、输出层参数的精度进行了修正。仿真结果表明:神经网络模型能够获得类似动态规划的最优控制效果,能够控制动力电池SOC在目标值的3%误差带以内。采用NEDC工况对混合型能量管理策略进行了硬件在环仿真试验,试验结果表明:与实车采用的电能消耗-电能维持型控制策略相比,所提出的混合型能量管理策略使汽车的燃油经济性提高了9.5%。A blended EMS for EREV with real-time control is proposed. First,an EREV model for EMS optimization is developed. The extender fuel consumption and the difference sum between current SOC and target SOC of battery are set as the optimization object based on the characters of EMS. Second,the energy management optimization problem for a given driving cycle is solved by using Dynamic Programming（ DP）algorithm,and the optimized extender start time and output power are obtained. However,without the detailed driving cycle information, the DP algorithm could not apply for real-time control of EREV.Meanwhile,extracting the control rules from the dynamic programming optimization result is not easy,so BP neural network algorithm is applied to conduct the offline training of the optimization result,and the nonlinear relationships between extender output power and vehicle driving parameters are built,and the NN control model with real-time control ability is obtained. The parameters＇ accuracies in the input and output layers of the BP NN are adjusted according to the EREV condition parameters＇ accuracies transferred in CAN bus during the BP NN training. The simulation result shows that the NN control model could obtain the optimal control result similar to DP algorithm and control battery SOC varying along target curve within 3%error band. The hardware-in-loop test is carried out on the blended EMS adopting New European Driving Cycle（ NEDC）. The test result shows that the proposed blended EMS saved the fuel consumption by about9. 5% compared to that of Charging Deplete-Charging Sustain（ CD-CS） control strategy.

关 键 词：汽车工程 能量管理策略 动态规划 增程式电动汽车 神经网络 

分 类 号：U469.72[机械工程—车辆工程]