Abstract: For the high-voltage circuit breakers(HVCBs) mechanical fault diagnosis process, the original feature dimension is too high, which leads to over-fitting. However, the traditional Relief-F algorithm is not specific to the diagnosis model in the process of screening features, and support vector machine (SVM) diagnosis algorithm is limited by parameter selection, resulting in poor diagnosis accuracy. This paper proposes an integrated SVM diagnosis algorithm with AM-ReliefF feature selection. The algorithm sorts and effectively filters the original feature space, generates an optimal feature subset of the adaptive model, and effectively integrates SVM as a base learner with the AdaBoost algorithm to improve the accuracy of mechanical fault diagnosis of HVCBs. Firstly, features were extracted from the closing vibration signals of LW30-252 SF₆ HVCB under six typical working conditions to form the original feature space, and then used the AM-ReliefF feature selection algorithm to generate the optimal feature subset matching the integrated SVM model, and finally used Integrated SVM model for diagnosis. The comparison with the diagnosis accuracy of the single SVM algorithm under the original features shows that the proposed method improves the fault diagnosis accuracy from 83.0% to 98.9%, which provides a new idea for the mechanical fault diagnosis of HVCBs.