Piecewise affine identification of MIMO processes

In this paper piecewise ARX (PWARX) model identification of a nonlinear MIMO process is discussed. PWARX models comprise several ARX models where each of them is valid over a polytope in the regressor space. The identification procedure simultaneously estimates both the polytopic regions and the ARX model coefficients in each region. Here we use the clustering-based identification procedure, that is designed for MISO processes, and proceed in a natural way to extend this approach to identification of nonlinear MIMO processes, A very important role in identification of process nonlinearities for each MISO process plays a suitable linear transformation in the regressor space. A new way for choosing that linear transformation is suggested, automatically from the identification data position in the regressor space. Using the proposed procedure, a PWARX MIMO model of a magnetic levitation laboratory setup is identified and validated