Estimation of Lithium Battery State of Health Using Hybrid Deep Learning with Multi-Step Feature Engineering and Optimization Algorithm Integration

Accurate State of Health (SOH) estimation is critical for the reliable and safe operation of lithium-ion batteries; this paper proposes an ORIME–Transformer–BILSTM model integrating multiple health factors and achieves high-precision SOH prediction. Traditional single-dimensional health factors (HFs) struggle to predict battery SOH accurately and stably. Therefore, this study employs Spearman and Kendall correlation coefficients to analyze multi-dimensional HFs and determine the key characteristics for quantifying SOH. The self-attention mechanism of the Transformer encoder extracts and fuses the key features of long-term sequences. A BILSTM network receives these input vectors, whose primary function is to uncover the temporal evolution of the SOH. Finally, the optimal random-weight-initialization meta-heuristic estimation (ORIME) algorithm adaptively adjusts the hyperparameters to optimize the model efficiently. Cycle data from four batteries (B5, B6, B7 and B18) provided by NASA are used for testing. The mean absolute error (MAE), mean absolute percentage error (MAPE) and root-mean-square error (RMSE) of the proposed method are 0.2634%, 0.4337% and 0.3106% Compared to recent state-of-the-art methods, this approach significantly reduces prediction errors by 33% to 67%, unequivocally confirming its superiority and robustness. This work provides a highly accurate and generalized solution for SOH estimation in real-world battery management systems.