Gappy POD algorithm for reconstructing the core temperature of energy-storage battery packs and its comparison with BP neural network prediction ability

The reliability and safety of energy-storage battery packs have always been an industry priority. Large energy-storage battery modules are characterized by high power, numerous built-in energy-storage batteries, complex structures, and a heightened risk of thermal runaway. Monitoring the core temperature in energy-storage battery packs in a noncontact, real-time manner is essential for preventing and controlling thermal runaway events. In response to the challenge of incomplete temperature data acquisition, especially in industrial settings where arranging multiple temperature measurement points inside the battery pack may not be feasible, this study introduced the Gappy POD reconstruction algorithm. Gappy POD is a data analysis method based on proper orthogonal decomposition (POD), which is commonly used in inverse heat transfer and fluid mechanics problems. This enables the prediction of the internal core temperature by monitoring the surface temperature of the battery pack. Considering the safety concerns of battery experiments, this study simulated battery temperature changes using a simplified simulated energy-storage battery experimental platform. The platform tests the stability and real-time reconstruction capabilities of the Gappy POD algorithm under stable and drastic changes in operating conditions. Although we did not introduce the equivalent circuit model in the experiment, this preliminary study verified the reconstruction ability of the algorithm under significant fluctuations in working conditions. Neural networks are renowned for their nonlinear solid prediction capabilities and have extensive applications in predicting the temperatures of energy-storage batteries. This study compares the reconstruction ability of the Gappy POD algorithm with the prediction capability of a back-propagation (BP) neural network. This study also explored the reconstruction and prediction capabilities of the Gappy POD and BP neural networks under varying database sizes for training. The research presented in this study indicates that the Gappy POD reconstruction algorithm exhibits high prediction accuracy, particularly under stable working conditions and with smaller training sample sizes. In these scenarios, it outperformed the BP neural network. Moreover, this algorithm demonstrates strong stability and low dependence on the volume of database data, providing a solid foundation for further applications in the thermal management of energy-storage batteries. It also presents a viable approach for noncontact monitoring of the core temperature of energy-storage battery packs. In conclusion, this study acknowledges areas for improvement in the current research and outlines prospects for future research.