Integrating FT-ICR MS and Machine Learning to Forecast Acid Content Across Boiling Cuts

Area of Science:

• Petroleum Chemistry
• Analytical Chemistry
• Computational Chemistry

Background:

• Total Acid Number (TAN) is a critical parameter for characterizing crude oil acidity and corrosivity.
• Traditional TAN determination methods, such as ASTM testing, can be time-consuming and require significant sample volumes.
• Understanding TAN distribution in True Boiling Point (TBP) fractions is essential for refining processes and asset integrity.

Purpose of the Study:

• To develop and validate a novel machine learning approach for predicting TAN in TBP distillation cuts.
• To leverage ultrahigh-resolution Fourier transform ion cyclotron mass spectrometry (FT-ICR MS) data for accurate TAN prediction.
• To establish a more efficient and less sample-intensive method for TAN analysis compared to conventional techniques.

Main Methods:

• Utilized advanced machine learning algorithms, specifically partial least-squares (PLS) regression and ordered predictor selection (OPS).
• Employed ultrahigh-resolution Fourier transform ion cyclotron mass spectrometry (FT-ICR MS) data for detailed chemical composition analysis.
• Analyzed 36 diverse crude oil samples to build and test predictive models.

Main Results:

• Achieved robust and highly accurate predictive models for TAN distribution in TBP cuts, validated by low RMSEC and high Rc.
• Identified nitrogen- and oxygen-containing compounds as key contributors to TAN variations using volcano plots.
• Demonstrated excellent agreement between predicted and actual TAN values, especially for high-TAN samples.

Conclusions:

• The developed machine learning approach accurately predicts TAN in crude oil fractions using FT-ICR MS data.
• This novel method offers a significant improvement over traditional ASTM testing, requiring smaller sample volumes.
• The approach provides a powerful tool for efficient molecular characterization and behavioral forecasting of complex petroleum mixtures.