A method classifying power quality disturbances(PQD) based on modified S-transform and relevance vector
machine(RVM) is presented. The modified S-transform(MST) is achieved by adding three adjustable factors to the
Gaussian window function of the normal S-Transform. The adjustable factors change the velocity in which the width of
the window function varies with the frequency. The PQD sample eigenvectors can be extracted accurately by using the
modified S-transform with better time-frequency analysis performance than the S-Transform. Then the disturbance types
are identified through the multi-lay RVM pattern recognition classifier on hierarchical categorization and minimum output
coding. Numerical results show that the proposed MST-based RVM method can achieve higher classification accuracy
quickly, and requires substantially fewer relevance vectors and shorter test time than the SVM classifier.