基于BP-EKF算法的锂电池SOC联合估计  

Co-estimation of Lithium Battery SOC Based on BP-EKF Algorithm

作　　者：孔祥创 赵万忠[1,2] 王春燕[1,2] 

机构地区：[1]南京航空航天大学能源与动力学院,南京210016 [2]重庆大学机械传动国家重点实验室,重庆400044

出　　处：《汽车工程》2017年第6期648-652,共5页Automotive Engineering

基　　金：国家自然科学基金(51375007);重庆大学机械传动国家重点实验室开放基金(SKLMT-KFKT-2014010和SKLMT-KFKT-201507);江苏省普通高校研究生科研创新计划项目(SJZZ15_0038)资助

摘　　要：电池荷电状态(SOC)的准确估计是电动车辆进行整车控制优化的先决条件,也是合理实施电池管理的依据。本文中在确定1阶RC等效电路模型的基础上,采用含有遗忘因子的递推最小二乘算法和BP-EKF算法对模型参数与SOC进行在线联合估计,提出一种BP神经网络和扩展卡尔曼滤波(EKF)相结合的锂离子动力电池SOC估计方法,使用相应的滤波输出参数离线训练BP神经网络,进而将训练成功的BP神经网络用于补偿EKF算法的估计误差。通过仿真和电池动态工况试验验证,结果表明,与EKF算法相比,所提出的SOC估计方法具有良好的抑制发散和鲁棒性能,能有效提高SOC的估计精度。Accurate estimation of battery SOC is not only a prerequisite for control and optimization of elec-tric vehicle, but also the basis for the reasonable implementation of battery management. In this paper, based on the first-order RC equivalent circuit model, a SOC estimation method for lithium power battery is proposed by combining BP neural network with extended Kalman filtering （ EKF）, in which recursive least-squares with forgetting factor and BP-EKF algorithm are adopted to conduct a co-estimation of SOC and other model parameters. Corresponding filter output parameters are used to train BP neural network off- line,and succed trained BP neural network are used to compensate the estimation error in EKF algorithm. The results of simulation and dynamic state test show that com-pared with EKF algorithm,the SOC estimation method proposed has good performances in divergence suppression and robustness,and can effectively enhance the estimation accuracy of battery SOC.

关 键 词：锂电池 荷电状态 扩展卡尔曼滤波 BP神经网络 联合估计 

分 类 号：U469.72[机械工程—车辆工程]