双通道ELM在锂离子电池SOC估算的应用

A double channel ELM and its applications in SOC estimation of lithium-ion battery

为提高电动汽车的锂离子电池组荷电状态的预测精度,采用理论分析与实验相结合的方法,对传统极限学习机进行改进,在输入层与输出层间搭建直接通道,提高模型精度.针对系统噪声的时变性,应用自适应无迹卡尔曼滤波器估算电池SOC.研究结果表明:双通道ELM具有更强的泛化能力和极短的训练时间,AUKF对于锂离子电池组系统噪声的时变特性具有更强的适应能力,显著降低了SOC估算结果的平均误差和最大误差.

In order to improve state of charge prediction accuracy of lithium-ion battery pack, a method combined theoretical analysis and experimental research is adopted to reform the traditional extreme learning machine. Direct channels are built between the input layer and the output layer to improve battery model accuracy. For the system noise is time-varying, adaptive unscented Kalman filter is used for SOC prediction. The result shows that double channel ELM has strong generalization ability and extremely short training time, and AUKF is better adaptive for time-varying noise system. This improved method reduces the maximum error and mean error significantly.