【摘要】 锂离子电池健康状态(state of health, SOH)的准确评估是电池管理系统的核心功能，直接关系到电池的寿命及运行安全性。作为SOH估计的关键环节，健康因子(health factors, HFs)的提取与选择直接决定了退化表征的有效性，提出一种基于斑鱼狗优化算法(pied kingfisher optimizer, PKO)、极端梯度提升算法(extreme gradient boosting, XGBoost)及时间卷积网络(temporal convolutional network, TCN)的SOH估计方法。该方法从充电电压曲线和增量容量曲线中提取HFs，利用PKO优化XGBoost特征选择过程，筛选出更具代表性的HF组合；随后采用TCN对时间序列建模，捕捉长期依赖关系。最后，在两个公开数据集上验证了本方法的有效性，并与其他机器学习模型进行了比较。结果表明，所提出方法具有较高SOH预测精度，在所有数据集上的平均绝对百分比误差均小于1.30%，平均绝对误差均低于0.85%，且数据量减小时，仍具有优良的稳定性。更多还原

【Abstract】 Accurate estimation of lithium-ion battery state of health(SOH) is a core function of battery management systems(BMS), critically influencing battery lifespan and operational safety. As a pivotal component of SOH estimation, the extraction and selection of health factors(HFs) directly determine the effectiveness of degradation characterization. This paper proposed an integrated SOH estimation method based on the pied kingfisher optimizer(PKO), extreme gradient boosting(XGBoost), and temporal convolutional network(TCN). The proposed methodology extracted HFs from charging voltage curves and incremental capacity(IC) curves, employed PKO to optimize the XGBoost-based feature selection process for identifying more representative feature combinations,and subsequently utilized TCN to model temporal dependencies and captured long-term degradation patterns. Experimental validation on two public datasets demonstrates the method’s superiority through comparative analysis with other machine learning models. Results show that the proposed approach achieves high prediction accuracy, with mean absolute percentage error(MAPE) below1.30% and mean absolute error(MAE) under 0.85% across all datasets. Furthermore, it maintains excellent stability even under reduced data volume conditions.更多还原