Deep learning approach for the prediction of Li-ion battery temperature under various discharge rates

Temperature control in lithium-ion batteries (LiBs) is critical for safety and performance issues. However, integrating temperature sensors into each cell can be challenging due to cost and space constraints. It is thus important to predict cell temperatures accurately through comprehensive battery analysis. This study aims to predict 18,650-type LiB cell temperatures in a 4S2P module under various discharge rates (4A, 8A, and 12A) using deep learning (DL) methods. Traditional models such as long short-term memory (LSTM) and gated recurrent unit (GRU) were tested alongside hybrid models like LSTM-convolutional neural network (CNN) and GRU-CNN. By analyzing temperature variations through time-series data, the study compares the predictive performance of different model architectures. The results show that all models achieved high accuracy, with R2 values up to 0.999 and mean absolute error (MAE) as low as 0.0151°C. Among them, hybrid models provided the most reliable predictions, with Nash-Sutcliffe efficiency (NSE) values reaching 0.8868 under an 8A discharge current. Results indicate that DL methods provide highly accurate temperature predictions, demonstrating promising potential for application in battery thermal management systems (BTMSs).