Prediction of gasoline properties from composition data

• Autor(a) principal: Hugo Leonardo de Brito Buarque
• Data de Publicação: 2006
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFC
• Texto Completo: http://www.teses.ufc.br/tde_busca/arquivo.php?codArquivo=172
• topic: gasolinas; contribuiÃÃo de grupos; redes neurais artificiais; gasolines; group contribution; artificial neural networks; FISICA ATOMICA E MOLECULAR

Commercial gasolines are normally produced by blending hydrocarbon fractions obtained from the distillation of crude oil or from other petrochemical or refining processes, and carried through in order to comply with a variety of legal and ambient specifications at minimum cost. The quality for the use and commercialization of gasolines is evaluated through certain characteristics specified by governmental regulation. Such characteristics are usually determined by different methodologies and experimental techniques, since those depend on their constituents and their respective concentrations with a high complexity. Thus, blending of gasolines in petrochemical and refining industries is sometimes a very laborious procedure. The prediction of fuel properties from composition data is growing in importance in the last few years. Methods of group contribution have been used in the last decades to predict properties of pure organic compounds and some mixture parameters (e.g., UNIFAC). However, most of the recent studies use artificial neural networks as a technique for prediction for fuel properties using the composition of classes of constituents or key-compounds as input data. The main advantage of a neural network is its capacity to extract general and unknown information for certain series of data (training), supplying useful and fast models for prediction. However, the use of neural networks trained to predict properties of fuels produced from one given combination of petroleum fractions can not be suitable in the prediction of the characteristics of other gasolines produced from other origins due to the complexity and variability of gasoline composition. In this study, methods of multiple linear regression and artificial neural networks have been evaluated in the correlation and prediction of gasoline properties from information of composition obtained by gas chromatography, as well as a methodology for prediction of properties using a hybrid method composed of neural networks and group contribution. The developed model is evaluated and compared to other methods, revealing to be sufficiently promising for prediction of properties of pure components and complex mixtures.