Neural network prediction of cetane numbers for isoparaffins and diesel fuel

A backpropagation neural network was used to correlate and predict the cetane numbers of isoparaffins and diesel fuels. For the isoparaffins, the correlation was made between the chemical structure of the branched paraffins and their cetane number. Thirteen branched paraffins were employed to train the network. The group additivity method was used to express the degree of branching of the isoparaffins. According to their positions in the molecule structure, three carbon groups (methylene [-CH2-] subdivided into methylene alpha, beta and gamma, or further, to a carbon that is not a methylene, methyne [ > CH-] and quaternary carbon [ > C<]) were regrouped into four categories, and included along with normal boiling points, as the network inputs. For the selected diesel fuels, different combinations of physical properties such as density, viscosity, aniline point, and distillation temperatures were tested as the neural network input. The best model was obtained using density, viscosity, aniline point and ASTM D86 measured distillation temperatures (IBP, 10%, 50%, 90% and FBP) as inputs. The trained network models, when applied to predict the cetane number of other isoparaffins and diesel fuel, were quite accurate.