Abstract: Aiming at the problem that the classification accuracy of different types of power quality disturbance signals is not high, nine different models of power quality disturbance signals are built by MATLAB/simulink to make simulation analysis, and a kind of optimization method that an improved gravitational search algorithm（IGSA）optimize the penalty factor and kernel function parameters of support vector machine（SVM）is proposed.By optimizing the penalty factor and kernel function parameters of SVM,the IGSA-SVM classifier is constructed, and then the extracted feature vectors are normalized and input into the constructed IGSA-SVM classifier for training and classification. The simulation results show that the classification accuracy of IGSA-SVM classifier is better than that of SVM and GSA-SVM. It can realize the fast and accurate classification of 9 different power quality disturbance signals, which is helpful to solve practical engineering problems.