Assessment of the prediction capability of biodiesel cetane number by Artificial Neural Networks

Abstract

Artificial Neural Networks for the estimation of the cetane number of biodiesel from their fatty acid methyl ester composition were evaluated in this work. The study covered 64 neural networks varying the number of nodes in the hidden layer and using five topologies and six different algorithms for the second training. An experimental data from literature reports that covers 48 and 15 biodiesels in the modeling-training step and validation step respectively were taken. After the training step, the best two ANN concerning selected statistical parameters were used in a validation step, using an external data. A model to predict cetane number using an artificial neural network was obtained with better accuracy than 95%. The best neural network to predict the cetane number was a backpropagation network (11:4:1) using the Conjugate Gradient Descend algorithm for the second step of the networks training and showing 93.8% of correlation. The proposed network is useful for prediction of the cetane number of biodiesel in a wide range of FAME composition but
keeping the percent of total unsaturation lower than 80%.