Designing high-performance fuels through graph neural networks for predicting cetane number of multicomponent surrogate mixtures

Cetane number (CN) is an important fuel property in designing high-performance fuels in recently diversifying compression ignition engines. We introduce graph neural networks (GNNs) that predict CNs of multicomponent surrogate mixtures when only 2D structures and mole fractions of molecules are given. It considers the influences of mixing multiple components and their chemical structures on CN, reproducing the non-linear blending behavior observed for certain mixtures. We trained the GNNs using the CNs of 1,143 mixtures, and reliable accuracy was achieved with mean absolute errors of 3.4-3.8 from the cross-validation. Lastly, we analyzed the chemical structural effects on non-linear blending behavior.