The invention discloses a method for classifying power quality disturbance events. According to the method, a stochastic gradient descent algorithm based SVM (Support Vector Machine) classifier is applied to classification identification of the power quality disturbance events, so that the problem in classification of finite samples of big power data and nonlinear and high-dimensional modes can be effectively solved; linear SVMs are approximate to x<2> kernel SVMs, so that the classifier has high calculation efficiency of a linear kernel and high classification accuracy of a nonlinear kernel function; the classification can be performed only by utilizing symbols of data after projection, so that during data storage, only symbols of preprocessed data need to be stored, and the challenge to the increasing power grid data volume nowadays is greatly relieved; and a possibility is provided for accurately analyzing the big power data in real time.