Comparative study of parametric and structural methodologies in identification of an experimental nonlinear process

Abstract

Presents a comparative study of parametric and structural identification methodologies when applied to the identification of an experimental nonlinear process. Several approaches for parametric identification are presented, such as: (i) linear mathematical model obtained through recursive least-squares (RLS), (ii) linear model with estimation algorithm using multi-step-ahead, (iii) Hammerstein model, (iv) Volterra model and, (v) bilinear model. Two structural approaches for neural network configuration are used: (i) multilayer perceptron, and (vii) radial basis function. An experimental evaluation is performed on a fan-and-plate process which exhibits complex features. The main characteristics of each identification methodologies and experimental results are assessed and compared using performance indices and validation response curves.