Clustering-based identification of a piecewise affine electronic throttle model

Abstract

Piece wise affine (PWA) model comprises several affine dynamics defined over polyhedral regions in the regressor (state+input) space. Identification of a PWA model is very often a starting point for the controller synthesis of hybrid systems. In this paper we extend the clustering-based procedure for identification of a piece wise autoregressive exogenous (PWARX) model proposed in [Ferrari-Trecate et al., 2003]. By exploiting a priori process knowledge we choose an appropriate linear transformation of the regression vector for a better and more efficient identification of the process nonlinearities. We significantly reduce the computational complexity of the classification algorithm for finding the complete polyhedral partition of the model domain. This modified clustering-based procedure is used to identify a PWARX model of the electronic throttle-a highly nonlinear component that regulates air inflow to the engine of a car.