A TWO-STAGE ALGORITHM FOR IDENTIFICATION OF NONLINEAR DYNAMIC SYSTEMS

A two-stage algorithm is proposed for fast identification of optimal linear-in-the-parameters models for nonlinear dynamic systems. In the first stage, an initial model is selected from a significant number of candidates, using a stepwise forward procedure. The significance of each selected model term is reviewed iteratively at the second stage using a fast review procedure and insignificant terms are then replaced, resulting in a locally optimised compact model. The contribution is that both the forward and backward model selection is performed within a well-defined regression context, leading to significantly reduced computational complexity. The computational complexity analysis confirms the arithmetic efficiency and the simulation results demonstrate the effectiveness.