基于电池老化趋势重构与TCN-GRU-Attention网络的SOH估计  

SOH Estimation Based on Battery Aging Trend Reconstruction and TCN-GRU-Attention Network

作　　者：李士哲[1] 张天宇 谢家乐 LI Shizhe;ZHANG Tianyu;XIE Jiale(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《电力科学与工程》2025年第3期38-45,共8页Electric Power Science and Engineering

基　　金：国家自然科学基金资助项目(61973117);中央高校基本科研业务费专项基金资助项目(9160323007)。

摘　　要：针对噪声干扰导致锂电池老化过程中关键特征提取困难的问题,首先,在增量容量曲线中提取反应电池老化规律的峰值特征,捕捉电池性能随时间变化的关键信息;然后,通过改进的自适应噪声完备集合经验模态分解与小波阈值降噪对特征进行联合降噪,重构出更高精度的特征序列;最后,将该特征序列输入到时间卷积网络提取序列特征,并利用门控循环单元捕捉长时间依赖性,同时引入多头注意力机制进一步增强模型对关键特征的感知能力。实验结果表明,用该方法可有效提高锂电池健康状态估计的准确性,使均方根误差小于1.5%,平均绝对误差小于1%。To address the difficulty in extracting key features during the aging process of lithium batteries due to noise interference,the peak features that reflect the aging pattern of the battery are firstly extracted from the incremental capacity curve to capture the key information of battery performance changes over time.Then,the features are jointly denoised through an improved complete ensemble empirical mode decomposition with adaptive noise and wavelet threshold denoising to reconstruct a feature sequence with higher accuracy.Finally,the feature sequence is input into time convolutional network to extract sequence features,and gated recurrent unit is used to capture long-term dependencies.Meanwhile,a multi-head attention mechanism is introduced to further enhance the model’s perception ability of key features.Experimental results show that this method can effectively improve the accuracy of lithium battery state of health estimation,with a root mean square error less than 1.5%and an average absolute error less than 1%.

关 键 词：锂电池 电池健康状态 自适应噪声完备集合经验模态分解 小波阈值降噪 时间卷积网络 门控循环单元 多头注意力机制 

分 类 号：TM912[电气工程—电力电子与电力传动]