A GENETIC PROGRAMMING APPROACH FOR CONSTRUCTION OF SURROGATE MODELS

摘要

Surrogate models, response surface models or meta-models are black-box models that describe a system with high accuracy. We present a methodology that combines iterative Design of experiments (DOE) with Genetic Programming (GP) in order to obtain surrogate models. GP is an evolutionary technique to create computer programs. In the context of surrogate modelling, the programs are possible functional forms of the model, that are used to fit experimental data. Therefore, unlike most approaches, non-1inear combinations of the basis functions are possible. The iterative DOE provides a methodology to choose data points to test current programs and build the next generation. Data is obtained from Aspen Plus based simulations and the process of data acquisition is automatized via Python. The methodology is applied to a RadFrac distillation column which is part of a corn to ethanol process and considers three input and three output variables. The results indicate that the proposed methodology is able to provide accurate surrogate models for the variables.