Identification and Inverse Modelling of Industrial Systems Using Regional Models

Abstract

With the advancement of technology, the speed of industrial processes has also shown great growth, accompanied by the need to obtain models and control them in a faster and more interactive way, however, the speed and ability to obtain data have also shown great advances, allowing the use of techniques capable of modeling processes reliably and quickly using the System Identification process. Generating a model from the input and output data of the systems, the System Identification has been the subject of many studies, with several techniques being proposed capable of generating reliable models in a short period of time. One of these techniques, presented in this article, is that of Regional Models and Robust Regional Models, which use Clustering techniques such as Self Organizing Map (SOM) and K-means capable of dividing the systems data space into more similar regions in order to produce more reliable models of the same using supervised neural networks too, the robust model also performs the treatment of Outliers in the data using the M-Estimation technique. The techniques presented will be tested in nonlinear industrial systems and evaluated based on their Normalized Mean Square Error (NMSE) and the autocorrelation function evaluated from the residues.