The fault diagnosis of steam turbine is of great significance to the operation of power plant. Models based on thermal parameters can be used to observe performance degradation of equipment and to predict its failure type. This paper uses the characteristic flow area method to establish the performance degradation model of the steam turbine system

simulates the system fault samples and test samples

and establishes the equipment fault sample library. Through the improved K-nearest neighbor(KNN) algorithm

based on the change law of thermal parameters of the steam and water system

the similarity of the current unit operating data samples to the equipment failure samples is calculated

and the probability of the current unit equipment failures is determined. The research shows that

the calculation error of the characteristic flow area under different operation conditions are all within 5%

which meets the calculation requirements. Compared with the conventional KNN algorithm

the improved KNN algorithm evaluates the weight of neighbors in the decision-making process through samples

and achieves a higher classification accuracy rate. The fault diagnosis results of the test sample show that

the improved KNN algorithm has a higher diagnostic accuracy rate than the conventional KNN

and the diagnostic accuracy of the test samples is 100%. The use of the improved KNN algorithm for steam turbine system fault diagnosis is feasible and consistent with the actual situation.