Neural Networks Versus Classic Mathematical Models to Estimate Gasoline Blending Properties

Abstract

Nowadays, gasoline must comply with a number of quality standards, whose purpose is the pollution reduction due to the produces that result from gasoline burning. These quality standards have a connection with the gasoline properties. The determination of gasoline properties is being made according to the standards. The standard methods’ drawbacks are related to determination duration, the fact that some methods are destructive and the obtained results are useful as long as the components’ properties are constant. To offset these drawbacks, the possibility of gasoline properties estimation became attractive.
In this paper, a comparative study between two estimation methods for gasoline properties is presented. The first method is based on a mathematical model which has 8 equations, model that can be found in the literature. The second method uses artificial neural networks for estimation. To be able to compare these methods, 60 blendings based on 60 different blending recipes were prepared. These blendings’ components are: FCC gasoline, catalytic reforming (CR) gasoline, iC5 fraction and bioethanol. The obtained results from this comparative study can be used to determine which of these two methods offers the most precise estimation in a particular case.