基于特征提取与NRBO-BiTCN-DTransformer融合模型的锂电池SOH估计

SOH Estimation of Lithium-Ion Batteries Based on Feature Extraction and NRBO-BiTCN-DTransformer Fusion Model

摘要: 针对现有锂离子电池健康状态(SOH)估计方法中特征提取维数多、神经网络学习时序数据长期依赖以及计算成本高的问题，本文提出一种融合特征提取与轻量化BiTCN-Transformer架构的锂离子电池SOH估计方法。首先，从电池充放电数据中提取并筛选12个静态与动态健康因子，采用堆叠自编码器进行降维融合，得到5个高相关性特征。其次，构建双向时序卷积网络BiTCN与Transformer混合模型，利用BiTCN捕捉局部波动特征，Transformer建模全局老化趋势。之后将动态通道剪枝(DCP)技术引入Transformer编码器的注意力模块，使用全连接层替代原Transformer的复杂解码器，用GELU激活函数替换RELU激活函数以提升性能。并结合NRBO智能优化算法进行超参数自动寻优，在保证估计精度的同时，显著提升模型的运行效率与跨电池泛化能力。实验结果表明，所提模型在独立的测试电池上SOH估计的精度以及计算效率均优于传统RNN、LSTM、Transformer等模型。

Abstract: To address the issues of high-dimensional feature extraction, long-term dependencies of neural networks learning on time series data, and high computational cost in existing lithium-ion battery state-of-health (SOH) estimation methods, this paper proposes a lithium-ion battery SOH estimation method combining feature extraction and a lightweight BiTCN-Transformer architecture. First, 12 static and dynamic health factors are extracted and selected from battery charge-discharge data and fused by dimensionality reduction via stacked autoencoder into 5 highly correlated features. Then, a hybrid model of bidirectional time series convolution network BiTCN-Transformer is built, where BiTCN captures local fluctuation features and Transformer models global degradation trends. Dynamic channel pruning (DCP) is introduced into the attention modules of the Transformer decoder, the original complex decoder of the Transformer is replaced by a fully connected layer, ReLU activation function is replaced by GELU activation function to improve performance, and NRBO intelligent optimization algorithm is used for automatic hyperparameter optimization to ensure the estimation accuracy while significantly improving the model’s running efficiency and cross-battery generalization ability. Experimental results show that the proposed model outperforms traditional RNN, LSTM, and Transformer models in both SOH estimation accuracy and computational efficiency on independent test batteries.