Identification of Dynamic Nonlinear Systems using Computational Intelligence Techniques

A novel approach based on computational intelligence techniques for the identification of nonlinear dynamic systems is presented in this paper. The technique encompasses both the properties of the Karhunen-Loeve transform in representing stochastic processes and the approximation capabilities of multi-layer neural networks. Experimental results on nonlinear systems governed by difference equations demonstrate the effectiveness of the proposed approach that is based on a real-time learning algorithm. Exhaustive experimentation on specific case studies was performed and some experimental results were compared with other existing techniques such as the Lee-Schetzen method, least mean square (LMS), recursive least square (RLS) and normalized least mean square (NLMS) algorithms. A better identification-accuracy was also achieved, and a reduction of some orders of magnitude in training-times compared with the well-known Lee-Schetzen method was obtained, thus making the proposed methodology one of the current best practices in this field