Aplicação de redes neurais no estudo do perfil do álcool etílico hidratado combustível comercializado em diferentes regiões no estado do Paraná

Abstract

Ethanol fuel must comply with quality control parameters such as density, alcohol content, pH and electrical conductivity. Its characteristics may vary according to its trade regions. Physical-chemical analysis data were collected, from 998 hydrous ethanol fuel samples traded in regions designated as “north”, “midwest" and “east” on the state of Paraná. The data fueled multilayer perceptron networks and selforganizing maps, both kinds of artificial neural networks, which classified the samples according to their trade regions. The perceptron networks learning rate was 0,10 and the samples were randomly divided, being 70% for training, 15% for testing and 15% for validation. One hundred networks were trained and the best performance was obtained by a network with six neurons in the hidden layer, which reached 85% of correction percentage for training, 82% for testing and 84% for validation. The selforganizing maps best configuration had a 45 x 45 topology and 5000 training epochs, with a final learning rate of 6.7x10-4, a final neighborhood relationship of 3x10-2and a mean quantization error of 2x10-2. This neural network gave origin to a topological map depicting three separated groups, each one corresponding to samples of a same region of trade. Four maps of weights, one for each parameter, were presented. Both kinds of neural networks made possible the separation of samples according do their region of trade and agreed the density was a relevant parameter for the classification.