基于WOA-XGBoost的锂离子电池剩余使用寿命预测

Prediction of residual service life of lithium-ion battery using WOA-XGBoost

使用早期数据准确预测电池剩余使用寿命(RUL)可以加速电池的改进和优化。然而电池退化过程是非线性的,且在早期阶段容量衰减可忽略不计,使得RUL预测具有挑战性。为解决这一问题,本工作使用电池早期循环数据,并构建WOA算法和XGBoost算法的混合预测模型预测RUL。文章首先对电池实验数据进行预处理,观察放电电压-容量退化曲线和容量增量曲线的变化,选取与实际容量状态相关性较高的潜在特征,并将其时间序列数据作为XGBoost预测模型的输入,然后采用WOA算法对模型进行参数优化。最后使用由丰田研究所提供的84个在多步充电和恒流放电条件下的锂离子电池数据进行验证,结果表明所提出模型仅使用前100个周期循环数据即可对整个电池寿命预测,测试误差低于4%。

Using early data to accurately predict the remaining service life(RUL)of a battery can accelerate the improvement and optimization of the battery.However,the battery degradation process is nonlinear,and the capacity attenuation can be neglected in the early stage,which makes the RUL prediction challenging.To solve this problem,this paper uses the early cycle data of batteries,and constructs a hybrid prediction model of the WOA algorithm and the XGBoost algorithm to predict RUL.In this study,the experimental data of batteries are preprocessed,and the changes in discharge voltage-capacity degradation curve and capacity increment curve are observed.Then,the potential characteristics with high a correlation as well as actual capacity state are selected,and the time series data are used as the input of the XGBoost prediction model.Then,the parameters of the model are optimized by the WOA algorithm.Finally,84 battery data provided by Toyota Research Institute using multi-step charging and constant current discharging are used to verify the model.The results show that the proposed model can predict the whole battery life only using the data of the first 100 cycles,and the test error is 4%.