Cetane number prediction from molecular structure using artificial neural networks

The production of next-generation biofuels is being explored through a variety of chemical and biological approaches, all aiming at lowering costs and increasing yields while producing viable alternatives to gasoline or diesel fuel. Chemical synthesis can lead to a huge variety of different fuels and the guidelines for which molecules would yield desirable properties as a fuel are largely based on intuition. One such property of interest is the cetane number, a measure of the ignition quality of diesel fuel. The present work improves on existing models and extends them to more oxygenates (primarily ethers) as an interim step in extending the model for further compounds of interest such as furans and tetrahydrofurans. Various members of these classes are being considered as fuels of interest by chemists and engineers in biofuel production. The present model uses artificial neural networks (ANN's) as a tool for quantitative structure property relationship (QSPR) analysis. Predicting the cetane number of a fuel is especially important because a large volume of pure sample (100mL or more) is typically required for lab testing, the production of which can be difficult and time-consuming at the lab scale. To this end, a predictive model will allow chemists to eliminate unlikely targets and focus their attention on promising candidates.