Lithium Battery SOH Estimation Using the LPE_BiGRU_EBAM Model Based on the WMA

This study proposes an LPE_BiGRU_EBAM model based on the whale migration optimization algorithm. By integrating learnable positional encoding to rapidly capture temporal embedding information, BiGRU effectively captures attention dependencies. The enhanced bottleneck attention mechanism highlights key features while suppressing noise. Global optimization via the whale migration algorithm further improves stability and generalization capabilities. Experimental results demonstrate that the model achieves SOH estimation under single-start low-temperature (B0047, B0048), single-start high-temperature (B0031, B0032), and room-temperature multi-start (B0005, B0006, 2C, 3C) conditions. At these points, the model’s mean MAE falls below approximately 0.0049, with R 2 values predominantly exceeding 0.98. The model also maintains excellent performance for state of health estimation across varying charge rates, with MAE below 0.0046 and R 2 values exceeding 0.96. Overall, the model demonstrates outstanding accuracy and stability in long-term, short-term, extreme temperature, and cross-charge-rate predictions, validating its exceptional generalization capability and practicality. This provides reliable technical support for lithium-ion battery health management.