Multi-Method Statistical Analysis of Factors Influencing Predictive Maintenance of Electric Vehicle Fleets

Abstract

Accurate estimation of predictive maintenance is important for effective electric vehicle fleet management. Existing approaches often fail to account for the complex relationships between diverse influencing factors. In this study, we propose a multi-method statistical analysis framework that integrates Spearman correlation, Mutual Information, and ElasticNet regression to quantify these relationships. The findings show that charge cycles and load weight exhibit the strongest positive correlations with Remaining Useful Life (RUL), while route roughness and battery temperature demonstrate significant negative impacts. Additionally, the Mutual Information analysis identified battery temperature as having the strongest non-linear relationship with RUL, underscoring its unique predictive relevance. Interestingly, maintenance records were found to have small predictive value across all analytical methods. The ElasticNet regression also refined the analysis by identifying 11 critical predictive factors, successfully eliminating redundant variables, and showing how corresponding statistical methods can improve predictive maintenance. These results can help the fleet operators to prioritize monitoring efforts on the most impactful factors and develop more precise RUL prediction models.