Genetic Programming based Identification of an Industrial Process

Abstract

In the field of industrial automation, it is essential to develop and improve mathematical methods that assist in obtaining more accurate models of real-world systems. In the following paper, a machine learning tool is applied to the problem of identifying a model of an industrial process. Symbolic regression and genetic programming are a successful combination of methods using which one can identify a nonlinear model in analytical form based on data collected from a process during routine operation. In this paper, a detailed description of the method implementation as well as necessary data preprocessing steps are presented. Then, the resulting models are validated on an industrial data set and compared on the basis of performance metrics with more classical methods and previous results achieved by the authors. Finally, the encountered problems in the realization of the methods are reflected upon.