Solid State Battery SOC Prediction Using Deep Learning Technique

Abstract

The soul of electric transportation is the batterymanagement system (BMS) and the remaining charge in an electric vehicle (EV) battery is paramount for strengthening the safety, consistency and performance of battery equipped vehicles. Standard model-based methods required knowledge of electrochemical whereas data-driven techniques engage machine learning (ML) tools to obtain non-uniform and complex electrochemical characteristics. The implementation of deeplearning approaches and optimization techniques in predicting the SOC of EV batteries has led to notable improvements in both the accuracy and robustness of the models employed. The principal objective of this research proposes a hybrid framework for the SOC estimation for solid-state $\mathbf{L i F e P O}_{\mathbf{4}}$ energy storage systems integrating a Long short-term-memory network with Polar Fox optimization (PFO) and Hyperband-based hyperparameter tuning approach. The solid-state $\mathbf{L i F e P O}_{\mathbf{4}}$ battery dataset derived from the simulation by considering real world driving cycle in India. The hybrid LSTM performance evaluated by error based metrices (RMSE and MAE) together with the determination coefficient $\left(\mathbf{R}^{\mathbf{2}}\right)$. The outcome of this research concluded that LSTM with Polar Fox optimization and Hyperband based hyperparameter tuning approach provides better convergence, robust and efficient for the prediction of SOC for the solid-state $\mathbf{L i F e P O}_{\mathbf{4}}$ battery. And this research added value to the Solid-State battery BMS improvement and easy adoption of EV technology. The proposed framework facilitates SDG 7, SDG 9 and SDG 11 by enhancing the efficiency, reliability and sustainability of solid-state battery systems.