A failure risk assessment method for lithium-ion batteries based on big data of after-sales vehicles

Abstract

Accurate prediction of battery failure has always been a challenge in the field of electric vehicles. In existing prediction methods, battery current, voltage and temperature of a single vehicle sample are often analyzed as the main characteristic parameters, however, the independent analysis of these parameters can hardly achieve the early warning of battery failure. This paper proposes a failure risk assessment method based on big data analysis, which can evaluate the failure risk level of the battery pack in advance. Specifically, the characteristic parameters strongly related to battery failure are extracted through the correlation analysis of after-sales data. Then, the synthetic minority oversampling technique is employed to expand the number of failure samples for model training. Finally, the vehicle risk coefficient prediction model is established by the random forest algorithm. According to the prediction results of field data of extensive vehicles, the model can successfully pick out the high-risk vehicles before battery failure, occupying roughly 7% of all vehicles, which is of great value for engineering application. In practical application, these vehicles with high-risk coefficient can be paid more attention by companies, so as to further ensure the driving safety of vehicle users.