Abstract: In order to meet the demand for efficient and accurate detection of lithium ion battery thermal runaway, this study designed a lithium battery thermal runaway experimental platform. STM32F103ZET6 single chip microcomputer was used to connect four sensors such as carbon monoxide, carbon dioxide, hydrogen and NTC to collect characteristic parameters in real time. At the same time, PyroSim is used to simulate the experimental environment and generate high-quality simulation data to supplement the experimental data. Based on the pytorch platform, we designed a Transformer neural network that can output the normal, early warning and thermal runaway states of lithium batteries. By using experimental data and simulation data for training, we successfully achieved fusion detection of thermal runaway data of lithium batteries, and verified the effectiveness of the algorithm.

Key words: thermal runaway, characteristic parameter, PyroSim, PyTorch, Transformer, data fusion