用新陈代谢极限学习机实现电池健康状态估算

Battery State⁃of⁃Health Estimation by Using Metabolic Extreme Learning Machine

针对因电池内部电化学反应的复杂性、算法泛化性差或可用已知数据量少导致的锂离子电池SOH估算精度下降的问题,提出使用极限学习机(ELM)构建强泛化性电池退化状态模型来描述不同电池的共性退化规律;引入新陈代谢机制来更新退化状态模型的输入数据进而实现对SOH的新陈代谢估算,在保证估算精度的同时降低对输入数据量的需求。利用两种不同材料电池在不同工况下所测数据对所提出方法进行验证,结果表明该方法能在仅用4个数据样本的情况下准确估算电池SOH,估算误差不超过2.18%。

In view of the low state of health(SOH)estimation accuracy of lithium⁃ion battery due to the complexity of internal electro⁃chemical reaction of battery,the poor generalization of the algorithm or the small amount of available data samples,a strongly generalized degradation state model of batteries is constructed based on the extreme learning machine(ELM)to describe the common degradation regularities of different batteries.More⁃over,the metabolic mechanism is introduced to update the input data of the degradation state model and then to real⁃ize the metabolic estimation of SOH,which reduces the demand for the amount of input data,while ensuring the ac⁃curacy of estimation.The method proposed is verified by using the measured data of two batteries with different mate⁃rials under different working conditions.The results show that the method proposed can accurately estimate the SOH of battery with only four data samples,and the estimation errors are no more than 2.18%.