基于变分模态分解和多头注意力的锂电池寿命预测

Prediction of the RUL for Lithium Batteries Based on Variational Mode Decomposition and Multi-Head Attention

锂离子电池被认为是迄今为止最高效的储能设备之一,以锂离子电池为核心的电动汽车凭借节能环保、经济实惠和安静舒适等特点受到了用户的喜爱,但随着电动汽车使用次数增加,因电池充放电所导致的车辆里程下降和电池使用安全问题越来越受到人们的关注。因此,本文提出了一种利用变分模态分解(VMD)和多头注意力机制(MHNN)的方法估算锂离子电池的健康状态(SOH)和剩余使用寿命(RUL)。该方法结合了VMD处理数据的优秀特性和MHNN提取不同变量之间交互作用的优势,解决了常规方法在遇到波动数据时预测不准确的问题,利用变模态分解(VMD)、网格搜索算法(GridSearch),对MHNN的参数进行优化。实验结果表明,本文所提出的VMD-MHNN方法在预测锂离子电池剩余使用寿命时均优于传统神经网络模型,具有较高的鲁棒性和更加稳定的预测性能。

Lithium-ion battery is considered as one of the most efficient energy storage devices so far. The electric vehicle with lithium-ion battery as its core has been very popular with users because of its energy conservation, environmental protection, economic benefits, quiet and comfortable charac-teristics. However, with the increase of the number of electric vehicles used, the reduction of vehicle mileage caused by battery charging and discharging and the safety problem of battery use have at-tracted more and more attention. Therefore, this paper proposes a method to estimate the state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries using variational mode de-composition (VMD) and multi-head attention mechanism (MHNN). This method combines the excel-lent characteristics of VMD in processing data and the advantages of MHNN in extracting the inter-action between different variables, and solves the problem of inaccurate prediction of conventional methods when encountering fluctuating data. The parameters of MHNN are optimized using varia-ble mode decomposition (VMD) and grid search algorithm (GridSearch). The experimental results show that the VMD-MHNN method proposed in this paper is superior to the traditional neural net-work model in predicting the remaining service life of lithium-ion batteries, and has higher robust-ness and more stable prediction performance.