Abstract:A synthetic fault diagnosis method based on improved rough set theory and probabilistic neural network for electric power transformer is proposed. The redundancy information is deleted by using decision table reduction technique of RS. The discernibility matrix is introduced into the reduction of decision table,which more quickly removes fault redundant attributes,and reduces complexity reduction process. The training stylebook of PNN is minimal decision table,thus the training speed and the accuracy of diagnosis are effectively improved. The diagnosis examples showthat the model can not only effectively diagnose incomplete information but also improve the diagnosis rate and correct rate.