The "over fitting" and "false and leakage alarm" problems easily occur during the application of conventional moving window principal component analysis(PCA) in dealing with slow deterioration of engineering parameters. To solve this problem

this paper proposes an improved moving window kernel principal component analysis(KPCA) algorithm based on data block

according to the sliding window KPCA algorithm. This improved algorithm carries out data standardization by Matlab and applies the iterative method to eliminate the abnormal data of the samples

then it uses the data block as the unit to update the model during the detection. Through regulating the size of the data block

the fault diagnosis will be more rapid and accurate. Moreover

this improved algorithm and the conventional KPCA method were applied to analyze the test data for wind turbine unit

the result verifies the above improved algorithm can solve the problem of slow deterioration of parameters in a better way.