Mode Partitioning Method Based on Eigenvector Analysis in Spectral Clustering

The multimode characteristics of the process data in actual production process will have a certain impact on the data modeling.Moreover,k-means,c-means and other clustering are several commonly used methods on mode analysis.However,these algorithms may not perform well in mode partitioning of the transition process.In this work,a general mode division method is proposed,in which the spectral clustering analysis of the similarity matrix is utilized.Moreover,by means of the relationship between the eigenvector of the similarity matrix and the involved classification information,a Gauss Manhattan distance is constructed for indicator variable such that the mode partitioning is achieved via the small window.Finally,the effectiveness of the proposed algorithm is verified by the experiment of multimode data with transition and nontransition process.