Abstract: Aiming at the problem of features overlapping and difficult recognition of complex power quality(PQ) disturbance, a PQ disturbance recognition method is proposed based on support vector machine(SVM) cascaded decision tree. Firstly, based on the initial characteristics of the disturbance extracted by the S-transform, 7 two-class SVMs are constructed for 7 single PQ disturbances such as voltage sag, harmonics, and voltage interruption. Then, in order to reduce the feature overlap of the complex disturbance, two-class statistical machine learning of the SVMs are used to reconstruct the features of disturbance. Finally, the constructed features are used to realize multi-class recognition of the complex PQ disturbance by the SVM cascade decision tree, combining with the two discrete feature processing of the CART decision tree. Simulation results show that the proposed method can effectively reduce the degree of feature overlap for composite PQ disturbances with strong noise immunity, and accurately identify 21 types of PQ disturbances including double and multiple complex disturbances.