A New Energy Vehicle Thermal Runaway Data Processing Model Based on Machine Learning Algorithm

Abstract

By monitoring the battery and vehicle state data of electric vehicles in different states, we established the thermal runaway data model by using the streaming data based on time series as the data source. Firstly, we preprocessed the outliers, null values and numerical conversion for voltage, temperature, and other data. Secondly, we also carried out the multi-dimensional expansion for indicators that can effectively reflect thermal runaway, such as voltage, temperature, SOC and entropy. Besides, we obtained an additive model based on the characteristics of the indicators. Through the real thermal runaway test on the data of 45 normal vehicles, we calculated the thermal runaway score in the whole life cycle, and evaluated the effect according to the established thermal runaway threshold. The results show that the model and algorithm described in our paper can effectively carry out thermal runaway warning and realize false positives within the range of very low probability, which has important theoretical significance and application value.