采用门控循环神经网络估计锂离子电池健康状态

锂离子电池健康状态(State of Health,SOH)描述了电池当前老化程度,对于提前对电池的故障及失控做出预警避免电池的不安全行为具有重要意义。其估计难点在于难以确定数量合适、相关性高的估计输入以及设计合适的估计算法。通过对现有电池老化数据集的研究发现,电池充电过程中电压曲线数据相对稳定,且随着电池的老化出现规律性变化。因此,文中直接采用充电过程中电压数据作为估计SOH的输入,并在数据驱动的框架下,提出了一种基于门控循环神经网络(Recurrent Neural Networks with Gated Recurrent Unit,GRU-RNN)的锂电池SOH估计方法。该方法能够挖掘出一维电压数据中的时序特征和SOH之间的映射规律。在两个公开的电池老化数据集上的实验结果表明,提出的方法达到了1.25%的均方绝对误差和低于5.62%的最大误差,在估计精度上达到现有技术发展水平。

State of health estimation for lithium-ion batteries using recurrent neural networks with gated recurrent unit

As one of the key state parameters of the battery, state of health (SOH) represents the degrees of battery degradation, which is very significant for predicting of battery failure and avoiding unsafe behavior of the battery. The difficulty is to determine the appropriate and high correlation input, and design an appropriate estimation algorithm. Through the study of existing battery aging datasets, it is found that the voltage data during charging is relatively stable, which are regular changes with the aging of lithium-ion batteries. Therefore, the voltage data in the charging process were used as the input for estimating SOH, and under the framework of data-driving, an SOH method based on Recurrent Neural Networks with Gated Recurrent Unit (GRU-RNN) was introduced, which could establish the mapping relations between the time series features of one-dimensional voltage data and SOH. The experimental results on two public battery aging datasets show that the proposed method achieves a mean absolute error of 1.25% and a maximum error of less than 5.62%, which is higher than the existing SOH estimation methods in estimation accuracy.