基于数据预处理和集成机器学习的锂离子电池剩余使用寿命预测

Prediction of remaining useful life of lithium-ion batteries based on data preprocessing and ensemble machine learning

针对锂离子电池容量退化存在局部再生现象,导致单一模型预测不准确问题,提出了一种基于数据预处理和集成机器学习的锂离子电池剩余使用寿命(RUL)预测方法。首先,利用自适应噪声完全集成经验模态分解(CEEMDAN)算法将锂离子电池容量退化序列分解成波动数据和主趋势两个部分。然后,使用时域卷积网络(TCN)和多头注意力机制(MHA)的组合模型预测波动数据部分。对于主趋势部分,选择差分自回归移动平均模型(ARIMA)进行预测。最后,集成各预测结果得到锂离子电池剩余使用寿命的预测结果。以NASA公开的数据集进行验证,实验结果表明:四组电池的均方根误差不超过1.85%,平均绝对误差在1.25%以内。证明了所提出的多模型融合方法具有良好的预测性能和鲁棒性。

In response to the phenomenon of local rejuvenation in the capacity degradation of lithium-ion batteries,which leads to inaccuracies in predictions by a single model,a lithium battery remaining useful life(RUL)prediction method based on data preprocessing and ensemble machine learning is proposed.Firstly,the adaptive noise-aided complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)algorithm is employed to decompose the lithium-ion battery capacity degradation sequence into fluctuation data and main trends.Then,a combined model utilizing temporal convolutional network(TCN)and multi-head attention(MHA)mechanisms is employed to predict the fluctuation data segment.For the main trend segment,the autoregressive integrated moving average(ARIMA)model is chosen for prediction.Finally,the predicted results from each model are integrated to obtain the prediction of the remaining life of the lithium-ion battery.Validation is carried out using the publicly available NASA dataset,and experimental results demonstrate that the root mean square error for the four battery groups is within 1.85%,and the mean absolute error is within 1.25%.This confirms the efficacy and robustness of the proposed multi-model fusion method in achieving accurate predictions.