Prediction and Sensitivity Analysis of the Cetane Number of Different Biodiesel Fuels Using an Artificial Neural Network

The
cetane number (CN) of biodiesel fuels was predicted using an
artificial neural network (ANN). A data set with 156 measured biodiesel
CN data points was first collected from the literature. Then, two
input sets were introduced for training the ANN including the fatty
acid methyl ester (FAME) composition and the functional group of FAMEs.
In the composition-based method, the input set includes the mass fractions
of the 14 FAME components from C10:00 to C24:00. In the functional
group-based method, the input set contains three improved functional
group parameters of n(−CH2−)/n(C), n(CC)/n(C),
and the position index of CC. For the composition-based method
and the functional group-based method, the best mean absolute errors
are, respectively, 1.70 and 1.72, and the best mean relative errors
are, respectively, 3.13 and 3.24% for the test set. To deeply understand
the correlations between the CN and the composition and molecular
structure of the FAMEs in biodiesel, an analysis method for calculating
the single-factor and double-factor sensitivity coefficients between
the input set and the output set was first implemented for the fuel
property prediction study. It was found that C18:01, C18:02, and C18:00,
as well as n(−CH2−)/n(C) and n(CC)/n(C), provide the largest sensitivity coefficients.