Performance Diagnosis of Controller Based on Eigenvector Subspace K-mean Clustering

摘要

To diagnose the specific reasons that lead to deterioration of controller performance and improve the diagnostic accuracy, a performance diagnosis method based on eigenvector subspace K-mean clustering is proposed. Firstly, the number of standard deterioration performance subspace with different degrees of deterioration information is increased on the basis of the eigenvector subspace distance diagnosis method, in which multiple distance values between eigenvector subspaces are obtained. Secondly, K-mean clustering is used to divide these distance values into two classes, and the clustering center with a maximum number of points is selected as the representative values. Finally, current performance degradation factor is attributed to the category that represented by the nearest standard performance subspace. In simulation, a double-tank model is established, and simulation result shows that average accuracy of controller performance diagnosis reaches 99.4% approximately.