Streamlined assessment and optimization of sustainable aviation fuels via few-shot learning and tailored genetic algorithm

• An innovative framework is developed for streamlined assessment and optimization of sustainable aviation fuels (SAFs). • A multi-scheme semi-supervised few-shot learning framework is proposed that reconstructs a limited jet fuel database with respect to the information on ASTM-specified jet fuel properties. • An accurate and reliable reciprocating mapping between jet fuel compositions and 14 ASTM-specified jet fuel properties is established. • A genetic algorithm-based optimization framework is developed to maximize the performance of SAFs. Sustainable aviation fuel (SAF) is considered the primary technological pathway to decarbonize the global civil aviation in the short to medium term. However, the scale-up of SAFs has been hampered by the high risk and high cost associated with assessing SAFs. Therefore, this study aims to develop an innovative framework to enable streamlined assessment and optimization of SAFs. A multi-scheme semi-supervised few-shot learning (MSSFSL) framework is first proposed and demonstrated to reconstruct a limited jet fuel database with respect to the information on ASTM-specified jet fuel properties. The database is further expanded via a novel synthetic data generation framework, where the database is enriched from only 48 jet fuel samples to over 2000 jet fuel samples. With the reconstructed and enriched database in conjunction with stacking ensemble learning, an accurate and reliable reciprocating mapping between jet fuel compositions and 14 ASTM-specified jet fuel properties (including density, derived cetane number, six distillation temperatures, flash point, freezing point, heat of combustion, surface tension, and viscosities) is established, based on which excellent prediction accuracy is achieved for practical petroleum-derived jet fuels, SAFs and their blends, with R 2 values above 0.8 achieved for all properties and relative prediction error reduced by approximately 2 orders of magnitude. Correlation analysis between jet fuel properties and compositions is subsequently conducted using a case study of ATJ, which further reveals the commendable contribution of iso-paraffins and di-cycloparaffins to the performance of SAFs (e.g., heat of combustion and freezing point). A genetic algorithm-based optimization framework is further developed and demonstrated through a case study where the 14 ASTM-specified jet fuel properties of three SAF candidates (i.e., one 100 % SAF and two 50 %/50 % SAF/Jet A blends) are optimized to fall within the ASTM tolerance and the respective property ranges of Jet A and Jet A-1. The proposed framework enables real-time, risk-lean and cost-effective SAF assessment and fine-tuning at lab-scale based on jet fuel standards and feedback from end-users before platform scale-up, which is expected to facilitate SAF development considerably. The developed model is released as a module of the SAFRA_Master_1.0 software package, hosted free of charge at the UHPC laboratory website ( https://uhpc-lab.org/downloads/ ).