There are errors in the fault signal decomposition of three-phase AC power transmission lines which affect the accuracy of fault classification. In order to improve the precision of fault signal decomposition and classification accuracy an improved variational mode decomposition VMD - permutation entropy PE fault feature extraction classification method based on fault voltage signal was proposed. MATLAB / Simulink was used to build fault simulation lines and generate fault data sets. In order to obtain the ideal and most effective combination for decomposition whale optimization algorithm WOA was used to optimize the penalty parameters of fault voltage signal VMD and the number of decompositions to find the optimal solution combination increasing the decomposition accuracy of each component. The same variable method was used for comparative experimental analysis. The VMD and EMD were used to decompose the fault voltage to obtain the intrinsic mode function IMF and the permutation entropy PE was used to calculate each IMF to obtain the corresponding feature vector which was used as the basis for classification. Then the decision tree DT of Gaussian optimized support vector machine SVM was used for fault classification verification. The simulation results show that the improved VMD-PE has more separability on the fault voltage decomposition and the recognition rate is greatly improved compared with EMD-PE which avoids the occurrence of chaos to a large extent. The accuracy of fault recognition is up to 96. 7% which can be used as the basis for the decomposition and classification.