In order to solve the problem that the power quality disturbance is diverse and the recognition rate is not high， the purpose of this paper is how to apply the multi-class classification problem to the support vector machine. Firstly， the disturbance signal eigenvalue is extracted to train model by S transform and FFT transform. Secondly， the generalized KKT decision condition is combined with the sample space distribution sequence to introduce interclass recognition degree. The hyperplane function with the highest interclass degree is used as the root node of the classifier to overcome the shortcomings of traditional Decision-oriented Non-cyclic Graph Support Vector Machine Classifier (DDAGSVM) randomization in the order of classification generation， and the improved DDAGSVM is applied to the classification of the energy disturbance signal. The experimental results show that the proposed algorithm has better effect and better robustness than the traditional DDAGSVM algorithm.