Multiobjective Stochastic Optimization of Dividing-wall Distillation Columns Using a Surrogate Model Based on Neural Networks

摘要

Surrogate models have been used for modelling and optimization of conventional chemical processes; among them, neural networks have a great potential to capture complex problems such as those found in chemical processes. However, the development of intensified processes has brought about important challenges in modelling and optimization, due to more complex interrelation between design variables. Among intensified processes, dividing-wall columns represent an interesting alternative for fluid mixtures separation, allowing savings in space requirements, energy and investments costs, in comparison with conventional sequences. In this work, we propose the optimization of dividing-wall columns, with a multiobjective genetic algorithm, through the use of neural networks as surrogate models. The contribution of this work is focused on the evaluation of both objectives and constraints functions with neural networks. The results show a significant reduction in computational time and the number of evaluations of objectives and constraints functions required to reaching the Pareto front.