Prediction of temperature profile and extracted energy of lithium-ion batteries using machine learning approach

• PPCM-based-BTMS is proposed for cooling lithium-ion batteries (LIBs) • Machine learning approach is employed to predict the time-dependent profiles (TDPs) • Four ML algorithms are selected and evaluated using the dataset of experiments. • Results Showed that Random Forest Regressor is the best predictive model for TDPs. • XG-Boost exhibits mean absolute error of 0.9849 and a mean squared error of 1.8101. Efficient thermal management of lithium-ion batteries (LIBs) is essential for the safe and reliable operation of electric unmanned underwater vehicles (UUVs). In this study, we propose a pressurized phase change material (PPCM) integrated battery thermal management system (BTMS), which is well-suited for UUVs since they naturally operate under high-pressure conditions. Applying pressure reduces interfacial thermal resistance and enhances the cooling efficiency of PCMs. To model and predict the performance of this PPCM-based BTMS, we develop a two-loop machine learning (ML) framework. In the first loop, discharge time is predicted; in the second loop, the time-dependent profiles (temperature evolution, extractable energy, and capacity) are estimated. Four ML algorithms were evaluated, with the Random Forest Regressor providing the most accurate predictions of time-dependent profiles (R 2 ≈ 0.90). In addition, the XG-Boost algorithm was applied to predict final discharge-stage values, achieving superior accuracy (99 %) with a mean absolute error of 0.98 and mean squared error of 1.81. The results demonstrate that while Random Forest is preferable for profile prediction, XG-Boost is more reliable for final numerical outputs. The proposed framework offers a novel approach for integrating PPCM-based BTMS with ML prediction in UUVs, enabling both accurate thermal modeling and improved energy management.