Split and merge algorithm for identification of Piecewise Affine systems

This paper adresses the identification of a class of hybrid dynamical systems which can be represented by a Piecewise Affine Autoregressive Exogenous (PWARX) model. These systems are composed of an usually unknown number of ARX sub-models, each of which corresponds to a polyhedral region of the regression space. It is proposed a Split and Merge clustering algorithm, used under a clustering based identification framework, to estimate the correct number of sub-models. The main advantages of this clustering algorithm is that it requires no initialization and there is only one tuning parameter to be adjusted. The resulting identification procedure is applied in a practical example in the identification of a DC motor with dead zone and saturation.