基于数据统计特性的GS-SVM电池峰值功率预测模型

Peak power prediction model for batteries based on data statistical characteristic and GS-SVM

以锰酸锂动力电池为研究对象,对电池处于不同温度和荷电状态下的情况进行10 s峰值功率测试,同时测量电池内阻。对实验测试得到的温度、荷电状态、内阻及峰值功率数据进行统计分析,包括测试变量间的相关程度评估和共线性检测,挖掘电池外特性参数与峰值功率数据间的统计关系。在此基础上,提出采用基于网格搜索的支持向量机(GS-SVM)建立电池的峰值功率预测模型。验证结果表明所提模型预测精度高,平均误差仅为3.65%;该模型训练时间短、响应速度快、操作性强,可以实现对动力电池峰值功率的快速估计,为电动汽车安全可靠运行提供有力保障。

Tests for 10 s peak power and internal resistance are carried out in different temperatures and SOC（State Of Charge） values for lithium manganese batteries. In order to obtain the relationship between external characteristic parameters and peak power data,the temperature,SOC,internal resistance and peak power data obtained from the test are statistically analyzed,including the correlation evaluation and collinear detection among test variables. Based on the analytical results,it is proposed to establish peak power predict model based on GS-SVM（Grid Search-Support Vector Machine）. The verification shows that the established model has high prediction accuracy and its mean error is only 3.65 %. With short training time,fast calculation speed and strong feasibility,the established model can rapidly estimate the peak power of power battery and ensure the reliable operation of electric vehicles.