储能科学与技术2025，Vol.14Issue(5)：2081-2097,17.DOI:10.19799/j.cnki.2095-4239.2025.0046
基于Soft-DTW算法与多源特征融合的实车动力电池SOH估算
SOH estimation of real-world power batteries based on Soft-DTW algorithm and multisource reature fusion
摘要
Abstract
The State of Health(SOH)of electric vehicle(EV)power batteries is a critical factor in ensuring efficient operation and extending service life.However,accurately assessing SOH is challenging owing to the complex and variable charging patterns in real-world EV usage,large sampling intervals in individual charge-discharge cycles,and missing feature data.To address these challenges,this study proposes a multisource feature fusion method for SOH estimation method using real-world vehicle operation data.The proposed method utilizes the soft dynamic time warping(Soft-DTW)algorithm to dynamically fuse parameters from weekly charging segment incremental capacity(IC)curves,generating an overall weekly IC fusion feature.By integrating these fused IC curve features with statistical features,a multisource feature set is constructed.Furthermore,a real-world SOH estimation model based on the Bidirectional Gated Recurrent Unit-eXtreme Gradient Boosting(BiGRU-XGBoost)is proposed.The model was tested using a real-world dataset comprising 20 EVs.K-fold cross-validation results demonstrates that the proposed SOH estimation method achieves a root mean square error(RMSE)within 1.21%and a mean absolute error(MAE)below 0.9%.Comparative experiments with GRU-XGBoost and long short-term memory(LSTM)models further validate the superiority of the BiGRU-XGBoost model,showing 36.1%and 47.6%reductions in RMSE.These findings highlight the enhanced robustness and generalization capabilities of the BiGRU-XGBoost model.关键词
电动汽车/健康状态估计/容量增量Key words
electric vehicle/state of health estimation/incremental capacity分类
信息技术与安全科学引用本文复制引用
丁萍,李涛涛,郑林锋,吴伟雄..基于Soft-DTW算法与多源特征融合的实车动力电池SOH估算[J].储能科学与技术,2025,14(5):2081-2097,17.基金项目
国家自然科学基金(52476200,52106244,52102424),广东省基础与应用基础研究基金(2024A1515030124),南方电网公司科技项目资助(GDKJXM20230246(030100KC23020017). （52476200,52106244,52102424）
