基于扩展卡尔曼神经网络算法估计电池SOC  

Estimation of battery SOC based on extended Kalman filter with neural network algorithms

作　　者：韩忠华[1,2] 刘珊珊[1,2] 石刚[2] 董挺[3] 

机构地区：[1]沈阳建筑大学信息与控制工程学院,辽宁沈阳110000 [2]中国科学院沈阳自动化研究所,辽宁沈阳110000 [3]中国电子技术标准化研究院,北京100007

出　　处：《电子技术应用》2016年第7期76-78,82,共4页Application of Electronic Technique

基　　金：国家重大科技专项项目(2011ZX02601-005);校涵育项目(XKHY2-61)

摘　　要：针对汽车锂电池的荷电状态(SOC)的问题,基于Thevenin电路为等效电路并且应用扩展卡尔曼算法(EKF)结合神经网络算法进行估计。在进行卡尔曼滤波算法估算过程中,需要用到实时的估算模型参数值(最新值),即在不同的SOC下模型的参数不同。传统做法是把SOC与各个参数的关系进行普通的拟合,这种方法在拟合过程中存在较大误差。为了解决这个问题,利用神经网络拟合各个电路模型参数与SOC关系曲线。试验结果表明,与单纯的扩展卡尔曼算法相比,该方法能够准确估计电池剩余电量,误差小于3%。An extended Kalman filter algorithm（EKF） with neural network is used to estimate the state of lithium battery（SOC）,which is based on Thevenin equivalent circuit.In the process of extended Kalman filter estimation,the real-time model parameters should be updated with the different SOC regard to the different SOC the different model parameters.The traditional approach which has a big error is that the fitting curve between SOC and the various separate parameters is common.To solve this problem neural net-work is applied to fit curve between the parameters of circuit model and the SOC separately.Finally,the results with the error less than 3 % show that compared with the pure extended Kalman algorithm,the method can realize the more accurate estimation of the remaining battery power.

关 键 词：锂离子电池SOC 扩展卡尔曼算法 神经网络 RC电路模型 

分 类 号：TM92[电气工程—电力电子与电力传动]