Frequency reconstruction oriented EMD-LSTM-AM based surface temperature prediction for lithium-ion battery

With the development of electric vehicles, safety concerns, especially thermal runaways, have garnered widespread attention. Accurate temperature prediction is essential to avoid the occurrence of thermal runaway. Herein, this paper proposes an innovative frequency reconstruction oriented EMD-LSTM-AM based surface temperature prediction method for lithium-ion batteries, which combines empirical mode decomposition (EMD), long short-term memory (LSTM), and attention mechanism (AM). First, this method employs EMD to extract the overall data trends from the raw data and restructures them into three different frequency components. Subsequently, the LSTM-AM network is utilized to learn long-term dependencies within the features and predict future temperatures. Finally, the accuracy and robustness of the proposed prediction method were validated under five distinct environmental temperatures and four complex driving conditions. The experiments indicate that, at 25 degrees C, the temperature prediction maximum absolute error achieved by the proposed network is 0.31 degrees C, exhibiting a 0.23 degrees C improvement compared to LSTM. Furthermore, under various operating conditions and temperatures, the proposed network demonstrates performance improvements across different metrics. These results suggest that the method effectively handles high-frequency non-stationary noise, mitigates prediction deviations caused by changes in battery physical characteristics, and provides a reliable reference for battery thermal management systems.