Modeling Gilliland Correlation using Genetic Programming

Abstract

The distillation process is one of the most important processes in industry, especially petroleum refining. Designing a distillation column assesses numerous challenges to the engineer, being a complex process that is approached in various studies. An important component, directly affecting the efficient operation of the column, is the reflux ratio that is correlated with the number of the theoretical stages, a correlation developed and studied by Gililland. The correlation is used in the case of simplified control models of distillation columns and it is a graphical method. However, in many situations, there is the need for an analytical form that adequately approximates the experimental data. There are in the literature different analytical forms which are used taking into account the desired precision. The present article attempts to address this problem by using the technique of Genetic Programming, a branch of Evolutionary Algorithms that belongs to Artificial Intelligence, a recently developed technique that has recorded successful applications especially in process modeling. Using an evolutionary paradigm and by evolving a population of solutions or subprograms composed of carefully chosen functions and operators, the Genetic Programming technique is capable of finding the program or relation that fits best the available data.