区分的ＡＲＸモデルの同定におけるデータ分類の改良

Abstract

A piecewise ARX model is a typical model for identification of hybrid dynamical systems. This model consists of several ARX submodels which switch in accordance with the value of the regression vector. There have recently been reported many methods for piecewise ARX model identification based on data classification techniques. This approach first categorizes the observed data into several data sets and then estimates the parameters of the submodels and the switching hyperplanes based on the classification result. However, the result of the data classification procedure contains misclassified data in general, and they may have an adverse effect on the accuracy of the identified model. This paper presents two methods for refining the data classification by re-classifying the candidate for misclassified data based on piecewise linear separability or linear separability of the true data classification on the regression space, respectively. A numerical example also demonstrates the effectiveness of the present methods.