A new attribute reduction method of rough set was proposed to improve the efficiency of fault diagnosis. First, the attribute reduction problem was transformed into the set covering problem and the correlation coefficient matrix was built based on the correlation matrix. Then based on the selection principle of high quality feature set, the random mutation mechanism was introduced, and the disturbance searching algorithm was designed and the result of attribute reduction was obtained. At last, the UCI data set and the power system of one complex equipment were used to validate the algorithm. The simulation results indicated that the redundancy attribute was eliminated effectively and the fault diagnosis time was shortened on the basis of meeting the requirements of the classification accuracy.