A search algorithm for the identification of multiple inputs nonlinear systems using orthogonal least squares estimator

A search algorithm for the identification of multiple inputs nonlinear systems using the orthogonal least squares estimator is derived. Because of the high dimensionality of general nonlinear systems the forward regression algorithm is used to detect the plausible size of the final fitted model and a variation of the forward regression algorithm is proposed. Instead of choosing the best candidate term at each iteration, top few candidate terms which have the largest error reduction ratios are investigated at each iteration. A search algorithm coupled with the model predicted output is derived which will sort through all plausible candidate terms to produce an optimal solution for the problem. Simulated and experimental examples are included to demonstrate the effectiveness of the proposed algorithm.