Decision Tree Model based Fault Diagnosis Method for New Energy Vehicles

Abstract

The fault diagnosis method for new energy vehicles faces the dual problems of unstable accuracy and slow response speed, which restricts the practicality of its online intelligent diagnosis. This article applies a decision tree model and optimizes it based on a strategy that combines feature importance sorting with cost complexity pruning (CCP). This study combines the information gain rate and the Gini coefficient to comprehensively evaluate the multidimensional features and screen out the key features with the most discriminative power for fault diagnosis. On this basis, the minimum leaf node sample number limit and CCP strategy are adopted to simplify the decision tree structure. Finally, multiple sub-models with limited depth are constructed based on the simplified single decision tree, and the diagnosis results of each sub-model are fused to improve robustness and response speed. Experiments show that this method achieves more than 93.6% accuracy on the test set, and the model complexity is significantly reduced. The model maintains an average accuracy of 93.5% on different vehicle types, and the accuracy ranges from 92.52% to 94.72% under different working conditions, verifying its high precision, low latency, and generalization ability.