A novel supervised multi-model modeling method based on k-means clustering

Abstract

A supervised multi-model modeling method is proposed for the nonlinear system in this paper. In the traditional k-means clustering method, the error of modeling multi-model is always ignored or even not considered in the clustering process. So, this unsupervised clustering method has large modeling error. In the new modeling method, the initial clusters are firstly obtained by the k-means clustering, then the data of clusters are reclassified considering the modeling errors of the multi-model, at last the new precise model parameters are obtained. The paper has given the analysis of the rationality of the method. In the end of the paper, the simulation results of the wastewater treatment process show that the supervised multi-model modeling method can improve the modeling precision and predictive performance.