基于EKF-SVM算法的动力电池SOC估计

State of Charge Estimation for Traction Battery Based on EKF-SVM Algorithm

针对单一SOC估计算法无法同时满足多项指标要求的问题,提出一种扩展卡尔曼滤波(EKF)与支持向量机(SVM)相结合的算法。通过动态跟踪模型参数和对开路电压(OCV)的实时估计,由EKF算法得到初步SOC估计值;通过对EKF算法滤波输出的DST工况数据进行训练,得到SVM模型,并利用其回归预测能力对初步估计值进行误差补偿,进一步地降低了估计误差。仿真结果表明,相比EKF算法和EKF-BP算法,所提EKF-SVM算法具有良好鲁棒性和泛化性,可实现电池SOC的准确估计,其最大绝对误差为1%左右。

In view of that the single SOC estimation algorithm cannot concurrently meet the requirements of multi-indicators,an algorithm combining the extended Kalman filtering(EKF)and support vector machine(SVM)is proposed.By dynamically tracking the model parameters and estimating the open-circuit voltage in real-time,the preliminary SOC estimation is obtained by using EKF algorithm.Furthermore,by training the DST condition data output from EKF algorithm,SVM model is obtained and its regression prediction ability is utilized to perform error compensation on preliminary estimation,further reducing the error of SOC estimation.The results of simulation show that compared with EKF and EKF-BP algorithms,the proposed EKF-SVM algorithm has better robustness and adaptability and can achieve accurate estimation of battery SOC,with the maximum absolute error of about 1%.