Attention-based deep learning framework for battery charge, health, and lifetime estimation in lithium-ion batteries for electric vehicles

The need to ensure Li-ion batteries are long-lasting, reliable, secure, and efficient is growing as the use of Electric Vehicles (EVs) rises. Despite advantages like higher energy density, faster charging, and environmental benefits such as reduced carbon emissions, lower air pollution, and less noise pollution, their major challenges, like performance degradation, safety concerns, and life span prediction, pose a need for a robust Battery Management System (BMS) together with methodologies for the prediction of their state parameters. A hybrid simulation and data-driven approach for the unified prediction of Li-ion batteries' State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL) is presented in this research. A Panasonic NCR18650BD Li-ion battery has been tested, using electrical and thermal elements to provide an extensive dataset of charge/discharge and degradation characteristics. After processing the simulated data in Python, an LSTM Attention-Augmented network approach has been proposed to capture the short- and long-term temporal dependencies of battery dynamics. For SOC, SOH, and RUL estimation, the superior estimation performance of the recommended model results in Root Mean Square Error (RMSE) values of 0.0091, 0.00127, and 0.00123, respectively, which correspond to coefficients of determination (R 2 ) of 0.9976, 0.9998, and 0.9989. Validation using the NASA battery aging dataset shows the applicability of the proposed framework to experimentally measured battery data and provides supporting evidence of its potential generalization capability under real-world operating scenarios. It can be observed that the proposed attention-based LSTM network efficiently integrates simulation and data-driven learning for reliable multi-state estimation, offering a scalable solution for advanced predictive BMS applications.