Identification of Multi-class and Nonlinear Systems

An algorithm for the identification of multi-class systems which can be described by a class of models over different operating regions is presented. The algorithm involves partitioning the raw data set using discriminant functions followed by parameter estimation. An orthogonal least squares algorithm coupled with a backward elimination procedure are employed for the parameter estimation and data partitioning processes. Provided the data elements are linearly separable, the proposed algorithm will correctly partition the data into the respective classes and parameter estimation algorithms can then be applied to estimate the models associated with each different class. Simulation studies are included to illustrate the algorithm.