Modeling the Solubility of Sulfur in Sour Gas Mixtures Using Improved Support Vector Machine Methods

Area of Science:

• Petroleum Engineering
• Computational Chemistry
• Machine Learning

Background:

• Sulfur solubility data is limited for sour gas reservoirs due to measurement challenges.
• Accurate sulfur solubility prediction is vital for the safe development of these reservoirs.

Purpose of the Study:

• To develop a machine learning model for accurate sulfur solubility prediction in sour gas.
• To address data scarcity and improve prediction accuracy across wide temperature and pressure ranges.

Main Methods:

• Utilized comprehensive gray relational analysis to identify key factors influencing sulfur solubility.
• Employed whale optimization algorithm (WOA) and gray wolf optimizer (GWO) to optimize support vector machine parameters.
• Developed WOA-WLSSVM and GWO-WLSSVM models using optimal kernel functions.

Main Results:

• The WOA-WLSSVM model demonstrated superior performance and reliability.
• Achieved an average absolute relative deviation of 3.45% and a determination coefficient (R²) of 0.9987.
• The model significantly outperformed other prediction methods.

Conclusions:

• The WOA-WLSSVM model offers a reliable solution for predicting sulfur solubility in sour gas.
• This approach enhances safety and efficiency in developing sulfur-containing gas reservoirs.
• Machine learning provides a powerful tool for overcoming experimental data limitations in geoscience.