Refinement of data classification in the piecewise ARX model identification

Abstract

There have been reported some methods based on clustering techniques for the identification of a piecewise ARX model. In general, the result of clustering procedure contains misclassification near switching hyperplanes on the regression space. In order to obtain a good identification result, it is necessary to correct the misclassified data. This paper proposes a method for decreasing the misclassified data based on the linear separability of the obtained data on the regression space.