Hybrid STO-FERNN technique for enhancing battery performance in electric vehicles

An accurate estimation of lithium-ion Battery State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL) is crucial for safety, reliability, and performance in an electric vehicle Battery Management System (BMS). Nevertheless, battery aging is extremely nonlinear and dependent on spatio-temporal interactions related to electrochemical, thermal, and operational factors, so accurate prediction is difficult. This paper presents a Hybrid Spatio-Temporal Optimization-based Fuzzy Ensemble Recurrent Neural Network (STO-FERNN) framework to improve the accuracy of battery state estimation with low computation cost. Subsequently, the Spatio-Temporal Optimization (STO) block executes constraint-based feature selection and hyperparameter optimization by considering operational bounds for voltage, temperature, and SOC, resulting in predictions that are physically compatible. The selected features are subsequently fed into Fuzzy Ensemble Recurrent Neural Network (FERNN), a model that combines fuzzy lagged logic to manage uncertainty, recurrent architecture to model temporal dependencies, and the ensemble approach that ameliorates robustness and generalization. The model is tested on two publicly available datasets of lithium-ion batteries, which include long-term degradation as well as real-time charging information. Experimental results indicate that the proposed model outperforms traditional machine learning and state-of-the-art deep learning models with RMSE as low as 0.92%, MAE of 0.71%, MAPE of 0.84%, and R² of 0.994. Ablation study and residual analysis demonstrate the competence of each module and the robustness of the model. Due to these advantages, the STO-FERNN architecture is a promising algorithm for online EV BHS monitoring and prognostics, which bridges prediction accuracy, robustness, interpretability, and computational feasibility with great potential in practical EV applications.