Database-driven and property-constrained inference of molecular composition of petroleum fractions from routine experimental data

Abstract

Molecular composition characterization of petroleum fractions underpins molecular-level modeling and simulation of refining processes, yet for heavy distillate fractions it typically requires specialized instrumentation and methodologies, limiting its accessibility under routine industrial analytical conditions. This study proposes a rapid approach for determining the molecular composition of heavy petroleum fractions based on routine and cost-effective analytical techniques commonly available in industrial laboratories, such as gas chromatography (GC) and elemental analysis. A multidimensional experimental database comprising 68 heavy gas oils was established, integrating molecular composition data from high-resolution mass spectrometry (HRMS) with bulk properties and multiple GC analyses. Preliminary molecular compositions are estimated by matching GC profiles of target samples with the database using non-negative least squares fitting, and are subsequently optimized and quantitatively refined through marginal distribution adjustment under property constraints. The final inferred molecular compositions show good agreement with HRMS results and accurately reproduce experimental properties. The proposed molecular composition inference framework bridges industrially accessible analytical techniques and research-grade molecular insights, enabling routine acquisition of molecular-level compositional data in industrial environments, with broader applicability anticipated upon further expansion of the dataset.