基于变分模态分解与Transformer-GRU的锂离子电池健康状态估计

Estimation of State of Health for Lithium-Ion Batteries Based on Variational Mode Decomposition and Transformer-GRU

摘要: 针对锂离子电池健康状态(SOH)估计中非线性时序特征建模难度大、长期依赖捕捉不充分等问题，本文提出一种融合变分模态分解(VMD)与粒子群优化(PSO)的Transformer-GRU联合模型。通过VMD对电池容量序列进行多尺度分解，筛选有效模态分量；采用PSO对Transformer-GRU超参数进行优化，利用Transformer的全局依赖捕捉能力与GRU的时序动态建模优势，分别对主趋势与高频子序列进行预测并融合。实验结果表明，该模型在NASA锂电池数据集上的平均绝对误差(MAE)和均方根误差(RMSE)分别低于0.62%和1.19%，决定系数(R²)达87.1，显著优于单一模型及传统联合模型，为高精度SOH估计提供了新思路。

Abstract: Aiming at the problems of difficult modeling of nonlinear time-series features and insufficient capture of long-term dependencies in the estimation of Lithium-Ion Battery State of Health (SOH), this paper proposes a Transformer-GRU joint model integrated with Variational Mode Decomposition (VMD) and Particle Swarm Optimization (PSO). First, VMD is applied to perform multi-scale decomposition on battery capacity sequences and screen effective modal components. Then, PSO is adopted to optimize the hyperparameters of the Transformer-GRU model; the model leverages the global dependency capture capability of Transformer and the advantages of GRU in time-series dynamic modeling to predict the main trend and high-frequency subsequences respectively, and then fuses the prediction results. Experimental results show that the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of this model on the NASA lithium-ion battery dataset are lower than 0.62% and 1.19% respectively, with the coefficient of determination (R²) reaching 87.1%. This performance is significantly superior to that of single models and traditional joint models, providing a new idea for high-precision SOH estimation.