基于注意力机制的改进LSTM锂电池健康状态估计方法

Method of Improved LSTM Lithium Battery Health Estimation Based on Attention Mechanism

将车载式安全工器具柜应用于配网抢修车辆中,其供电电池性能直接影响了工器具柜的稳定运行,因此需要精确预估供能电池的健康状态(state of health, SOH).本文提出一种基于海鸥算法(SOA)和注意力机制优化改进长短时记忆神经网络(long short-term memory neural network, LSTM)的锂电池健康状态估计方法.首先提取出4种与电池老化特性强相关的健康因子,将其历史运行数据输入到结合SOA-LSTM思想设计的SOH估计方法;其次利用Attention机制对输入变量进行权重分配,以强调关键特征在SOH预测中的作用;最后利用已公开的电池充放电曲线数据集对所提算法进行测试验证,并与其他算法进行对比.结果表明,本文方法可实现高精度SOH预测,均方根误差为0.011,模型拟合度达到98%以上.

The vehicle-mounted safety tool cabinet is applied to the emergency repair vehicle in distribution network,and its performance of power supply battery directly affects the stable operation of the tool cabinet,so it is necessary to accurately estimate the state of health(SOH)of energy supply batteries.In this paper,a method of lithium battery health estimation based on seagull algorithm(SOA)and attention mechanism optimization and improvement of long short-term memory neural network(LSTM)is proposed.Firstly,four health factors strongly related to battery aging characteristics are extracted,and their historical operating data are input into the SOH estimation method designed by combining SOA-LSTM ideas.Secondly,the Attention mechanism is adopted to assign the weights to the input variables in order to emphasize the role of key features in SOH prediction.Finally,the published battery charge-discharge curve dataset is utilized to test and verify the proposed algorithm.Compared with other algorithms,the proposed method can achieve highprecision SOH prediction,the root mean square error is 0.011,and the model fit degree reaches more than 98%.