Study of correlations for physicochemical properties of Brazilian gasoline

Artificial neural networks (ANNs) were used to find correlations between the chemical composition of Brazilian gasoline and some its properties including specific gravity, distillation curve and Reid vapor pressure (RVP). The neural networks were trained with supervised learning by the use of a backpropagation algorithm. After preliminary studies and experimental planning, 35 samples of Brazilian gasoline from a universe of 1284 samples were chosen and submitted to standardized laboratory assays. The chemical composition of the samples was obtained by chromatographic analysis, which included software for detailed hydrocarbon analysis (DHA). These chemical compositions were the inputs and the standard laboratory assays were the outputs for the neural networks. The networks obtained are able to predict the gasoline properties within an average error of 1%.