Machine Learning for High Energy Density Hydrocarbons for Sustainable Aviation Fuel

Abstract

Sustainable aviation fuels (SAFs) are crucial for addressing carbon emissions in the aviation industry. With a focus on SAFs, the research aims to establish a quantitative structure-property relationship for polycyclic hydrocarbons (PCHCs) and their net heat of combustion (NHOC) using the innovative approach of machine learning (ML). The model trained with support vector machine (SVM) algorithms in ML is selected as it demonstrates superior performance over other available algorithms with a high coefficient of determination (R2) and low mean absolute error (MAE) of 27.821 KJ/mol for 20% test data. Using the optimum SVM model, thirty-five potential PCHCs are identified as SAF candidates from C6 to C15 sourced from reputable scientific literature and databases. Furthermore, structural analysis revealed that high-performance PCHCs typically consist of saturated alkanes with multiple 3, 4, and 5-membered rings, suggesting that strained energy plays a role in their high energy density. The model obtained from ML can be employed to screen new hydrocarbons for their suitability as SAF candidates before costly experiments and ASTM evaluations.