A Bayesian approach to identification of hybrid systems

In this paper we present a novel procedure for the identification of hybrid systems in the piece-wise ARX form. The procedure consists of three steps: 1) parameter estimation, 2) classification of data points and 3) partition estimation. Our approach to parameter estimation is based on the gradual refinement of the a-priori information about the parameter values, using the Bayesian inference rule. Particle filters are used for a numerical implementation of the proposed parameter estimation procedure. Data points are subsequently classified to the mode which is most likely to have generated them. A modified version of the multi-category robust linear programming (MRLP) classification procedure, adjusted to use the information derived in the previous steps or the identification algorithm, is used for estimating the partition or the PWARX map. The proposed procedure is applied for the identification of the component placement process in pick-and-place machines.