基于SAM-BiGRU网络的锂电池RUL预测

RUL prediction of lithium battery based on SAM-BiGRU network

针对现有锂电池剩余使用寿命(RUL)预测方法精度低等问题,提出一种基于自注意力机制(SAM)的双向门控循环单元(BiGRU)网络模型。将锂电池的容量数据作为该模型输入序列,通过自注意力机制捕捉到锂电池容量历史信息中的关键时间点,并为其分配权重,利用BiGRU模型学习其容量退化趋势,据此实现剩余寿命预测。所提方法应用于CALCE锂电池数据集的CS2系列35、36、37号锂电池,实验结果表明所提方法35、36号锂电池上的预测误差均在1.5%以内,37号锂电池预测误差为2.22%。

Aiming at the problems of low accuracy of existing lithium battery remaining useful life prediction methods,a bidirectional gated recurrent unit(BiGRU)network based on self-attention mechanism was proposed.The model taked the capacity data of lithium batteries as the input sequence.It could capture the key time points in the historical information of the lithium battery capacity and assign weights to them through the self-attention mechanism.Afterwards,the BiGRU model was used to learn its capacity degradation trend,based on which the remaining life prediction was achieved.The proposed method was applied to the lithium batteries No.35,No.36 and No.37 of CS2series of the CALCE lithium battery data set.The experimental results show that the prediction errors of the proposed method on the lithium batteries No.35 and No.36 are all within 1.5%,and the prediction error of the lithium battery No.37 is 2.22%.