When the electrical contact points of the main circuit of an electric vehicle have poor contact, it is extremely easy to generate arc faults, directly threatening the life safety of the occupants. This paper proposes an arc fault detection method based on Customized differential threshold filtering, segmented maximum standardization, and statistical numerical rules. Firstly, an electric vehicle arc fault experimental platform was built around the real electric vehicle Geely Emgrand EV450, to conduct arc fault experiments under different working modes. Then, taking the power supply terminal voltage as the object, the signal is subjected to wavelet decomposition. The low-frequency coefficients obtained from wavelet decomposition were subjected to Customized differential threshold filtering and segmented maximum value standardization. Finally, the number of identical values in the normalized data was counted, and series arc faults were detected using a threshold method. The paper conducted in-depth analysis of the model′s sample length, differential threshold ratio, number of segments in maximum normalization, and preprocessing method selection, optimizing the parameters to further improve the model performance. The results show that the accuracy of the constructed model for detecting electric vehicle arc faults is 98. 35% , with good real-time performance. Through generalization analysis, algorithm time complexity analysis, and comparative analysis with other arc fault detection models, it is proven that the proposed model has good applicability for arc fault detection in electric vehicles.