SOH Prediction of High-Capacity Lithium-Ion Batteries under High-Rate Charging via a TCN-BiGRU Model with Multi-Head Attention

With the growing demand for charging efficiency of lithium-ion batteries, fast charging under high-rate conditions has emerged as a critical strategy. However, high-rate charging conditions may accelerate battery performance degradation, making it particularly important to monitor the state-of-health (SOH) of lithium batteries under such conditions. In this study, we propose a hybrid predictive model that integrates a Temporal Convolutional Network (TCN) and a Bidirectional Gated Recurrent Unit (BiGRU), further enhanced by a multi-head self-attention mechanism to better capture key temporal dependencies. Based on a dataset obtained from high-rate charging experiments on real vehicle power batteries in a laboratory environment, eight key features highly correlated with SOH were extracted and selected from multidimensional time-series curves, including voltage, current, and temperature, serving as model inputs to construct a TCN-BiGRU-Attention hybrid model for battery SOH prediction. Cross-validation results demonstrate that the proposed hybrid model achieves prediction errors within 0.6% across various evaluation metrics, significantly outperforming existing comparative models. This indicates that the model exhibits strong generalization capability and robustness in high-rate charging scenarios, providing a reliable solution for lithium-ion battery health management.