An improved parameter identification approach for piecewise affine model

This study proposes a high-efficient approach to identify the piecewise affine (PWA) model. The proposed approach constitutes two major steps, initial estimation and refinement process. In the initial estimation, Hough Transform (HT) is adopted to generate a group of submodel candidates; then a variable-threshold technique is applied to pick up the real submodel. In the refinement process, not only the distance constraint between data points and submodel's hyperplanes but also the clustering constraint between data points in regression regions are considered. An efficient algorithm is presented to alternately refine the submodel's parameters and the subregression sets. In the case study, the proposed approach is used to identify the fault model of the track circuit in high-speed railway. Analysis shows that the proposed approach has linear time complexity and exhibits superior data availability in small-sample case.