An Improved Salamander-Optimized Dual-Attention-Based Residual Bidirectional LSTM Deep Convolutional Maxout Network for EV Battery SOC and SOH Prediction

Abstract

Lithium-ion batteries (LIBs) are widely used in various sectors including electronic devices, electric vehicles (EVs) and energy storage systems due to their high energy density, low self-discharge rate and long cycle life. Despite widespread challenges, the inability of traditional estimation methods to accurately monitor the State of Charge (SOC) and State of Health (SOH) of batteries systems from the complex, non-linear degradation patterns that occur under various operating conditions. To resolve these problems, this study proposes a novel Dual-Attention-based Residual bidirectional long short-term memory Deep Convolutional Maxout Neural Network (DA-Res-Bi LSTM-MaxNet) for precise co-estimation of SOC and SOH in LIBs. Here, the Improved Salamander Optimization Algorithm (ISOA) is used for tuning the hyperparameters and weight parameters, ensuring maximum predictive accuracy. The ISOA is improved using the sine cosine algorithm (ISOA-SCA). The MATLAB/Simulink results show that the proposed values of Root mean square error (RMSE) of 0.131%, Mean absolute error (MAE) of 0.132%, Mean Squared Error (MSE) of 0.085% and Mean Absolute Percentage Error (MAPE) of 13.5%. As a result, these very low error measurements conclude that the proposed ISOA-SCA-optimized DA-ResBiLSTM-MaxNet architecture provides superior accuracy and robustness for simultaneous co-estimation of SOC and SOH under complex, nonlinear battery degradation conditions.