Redes neurais artificiais e redes complexas: aplicações em processos químicos

Abstract

In the first part of this work we apply an artificial neural network (ANN) trained with a back-propagation algorithm to predict the volatility of liquefied petroleum gases (LPG) produced from a fractionation tower of natural gas liquid (NGL). Our analysis indicate that the ANN scheme provides better results than a simulator developed based phenomenological which is currently being implemented in the plant under study. In the second part, our primary objective is to demonstrate that flowsheets of oil refineries can be intrinsically associated to complex network topologies, which are scale-free, display small-word effect and have hierarchical organization. The emergence of these properties artificial networks is explained as a consequence of the design principles used in the process’ design, which include heuristics rules and algorithmic techniques. We expect these results to be also valid for chemical plants of different types and capacities.