基于改进ELM的锂离子电池RUL预测

RUL prediction of Li-ion battery based on improved ELM

针对传统的极限学习机(ELM)算法对锂离子电池剩余使用寿命(RUL)预测效果不佳,以及已有改进ELM算法中鲜有关注输入层与隐藏层的“局部”连接等问题,提出改进算法A,即把ELM输入层与隐藏层之间的全连接改为卷积、池化;由于改进算法A的预测结果存在不确定性,向改进算法B引入全局平均池化的思想,直接把ELM输入层与隐藏层之间的全连接改为池化。将马里兰大学高级生命周期工程研究中心(CALCE)和美国国家航空航天局(NASA)的两组数据用于仿真实验,发现两种改进算法的预测精度均比常见的几种改进ELM算法更好。以B7电池为例,当预测起始点T=100时,改进算法A和改进算法B的均方根误差分别可达到0.0232和0.0090。

In view of the poor effect of traditional extreme learning machine(ELM)algorithm in predicting the remaining useful life(RUL)of Li-ion battery and the existing improved ELM algorithms paid little attention to the“local”connection between the input layer and the hidden layer,the improved algorithm A was proposed.That was,the full connection between the input layer and the hidden layer of ELM was changed into convolution and pooling.Due to the uncertainty of the prediction results of the improved algorithm A,the idea of global average pooling was introduced in the improved algorithm B,which directly changed the full connection between the input layer and the hidden layer of ELM into pooling.Two sets of data from the Center for Advanced Life Cycle Engineering(CALCE)of the University of Maryland and National Aeronautics and Space Administration(NASA)were used in the simulation experiments.It was found that both of the two improved algorithms had better prediction accuracy than several common improved ELM algorithms.Taking B7 battery as an example,when the starting point of prediction T was 100,the root mean square error of the improved algorithm A and the improved algorithm B could reach to 0.0232 and 0.0090,respectively.