STEADY-STATE MODELING OF CHEMICAL PROCESS SYSTEMS USING GENETIC PROGRAMMING

Complex processes are often modelled using input-output data from expe rimental tests. Regression and neural network modelling techniques are commonly used for this purpose. Unfortunately, these methods provide minimal information about the model structure required to accurately r epresent process characteristics. In this contribution, we propose the use of Genetic Programming (GP) as a method for developing input-outp ut process models from experimental data. GP performs symbolic regress ion, determining both the structure and the complexity of the model du ring its evolution. This has the advantage that no a priori modelling assumptions have to be made. Moreover, the technique can discriminate between relevant and irrelevant process inputs, yielding parsimonious model structures that accurately represent process characteristics. Fo llowing a tutorial example, the usefulness of the technique is demonst rated by the development of steady-state models for two typical proces ses, a vacuum distillation column and a chemical reactor system. A sta tistical analysis procedure is used to aid in the assessment of GP alg orithm settings and to guide in the selection of the find model struct ure. (C) 1997 Elsevier Science Ltd.