Presentation Information
[2A15]Development of an AI-Based Structural Identification Method for Petroleum Residual Fractions Using Extrapolative Component Prediction
○Yuya Murakami1, Kotaro Matsumoto2, Hiromasa Arai2 (1. Shizuoka University, 2. Japan Petroleum and Carbon Neutral Fuels Energy Center)
Keywords:
machine learning,property calculation,FT-ICR MS
Crude oil contains a highly diverse mixture of more than several hundred thousand chemical compounds, including ultra-high-boiling components with boiling points exceeding 1000 degC. Conventional analytical techniques are difficult to apply to such high-boiling fractions, and many aspects of their molecular structures remain unclear. In this study, we aimed to identify chemical species present in the ultra-high-boiling region by developing an algorithm that automatically generates candidate components and a model that predicts their probability of occurrence in crude oil. The results suggested the presence of polycyclic compounds that are difficult to identify using conventional methods.