A Bayesian-Optimized Deep Learning Framework for State of Charge and Remaining Useful Life Estimation of Lithium-Ion Batteries in Electric Vehicles

Abstract

Electric vehicles (EVs) significantly depend on lithium-ion batteries with monitoring of the State of Charge (SOC) and estimation of the battery life time being crucial to safety, enhancing operating life, and confidence by the user. Is the first of its kind to present a joint framework of SOC estimation and battery health and Remaining Useful Life (RUL) forecast. The method combines modern deep neural networks, namely Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Gated Recurrent Unit (GRU), and two-dimensional Convolutional Neural Networks (CNN2D), which were all optimized along the Bayesian code structure and do not require trial-and-error, but perform optimally. Based on these, classical machine learning baselines, Linear Regression, Support Vector Machine (SVM), XGBoost are employed in benchmarking. The models are then trained using rich time-series data including voltage, current, temperature and engineered features so that the models can capture the short-term dynamics and long-term degradation trends. In addition to the SOC estimation, the framework unites capacity fade modeling, cycle life and calendar aging studies, helping create precise RUL estimates. Through experimental analysis, Bayesian-optimized deep learning models are found to outperform conventional approaches consistently in terms of both SOC accuracy and prediction of the lifespan of EV batteries, providing a robust, data-driven addition to the battery management system of an EV. This dual functionality contributes not only to efficient energy consumption, but enables proactive maintenance methodology, so as to minimize down-time and the costs of operation.