Hybrid ESC-LSTM-BiGRU deep learning model for multi-state estimation of lithium-ion batteries

The state of charge (SOC) and state of energy (SOE) of lithium-ion batteries (LiBs) are critical parameters that influence both the safety and driving range of electric vehicles (EVs). However, SOC and SOE are highly sensitive to factors such as temperature variations, nonlinear behavior, and dynamic load conditions, making direct measurement impractical. To achieve accurate and stable estimations, this paper based on variations in temperature, driving conditions, noise interference, battery compositions, and hybrid conditions, proposes a hybrid deep learning model that integrates the Escape algorithm (ESC) with Long Short-Term Memory and bidirectional Gated Recurrent Unit (LSTM-BiGRU) for joint SOC and SOE estimation. In this framework, an anomaly detection model based on the quartile-Median Absolute Deviation (MAD) combination is established to identify and remove anomalies in the data, while the Savitzky-Golay (S-G) filtering method is used for smoothing. The LSTM network effectively addressing regression tasks in long-sequence data and overcoming the vanishing gradient problem. The BiGRU mitigates information decay during forward propagation unidirectional GRU, and its simplified structure with fewer parameters improves processing efficiency for sequential data and accelerates convergence. Additionally, the ESC is employed to optimize the four hyperparameters of the LSTM-BiGRU model, further improving prediction accuracy. The results show that the proposed model's prediction error metrics-including mean absolute error (MAE) and root mean square error (RMSE)-are limited to 1.08 % and 1.15 %, respectively. The model demonstrates high prediction accuracy and robust generalization capabilities, contributing to the safe and efficient operation of electric vehicles under wide temperature ranges and different working conditions.