Nonlinear Identification for Control by Using HARMAX Models

The purpose of this article is to perform a comparison between two identification models: a nonlinear one referred to as HARMAX and a classical linear one, of Box-Jenkins (BJ) type. The first model modifies a classical ARMAX model by applying Hammerstein polynomials on input, output, and noise signals. Identification of models was performed by means of Multi-Step Least Squares Method, as described into the article. The simulations on a real-world plant, namely ASTANK2, a fluidic system with two inputs, two outputs and nonlinear static characteristic, have proven that the HARMAX model performs better than the BJ model and provides more accurate useful models to be integrated into closed loop configurations.