A reliable model to predict mercury solubility in natural gas components: A robust machine learning framework and...

Area of Science:

• Environmental Chemistry
• Chemical Engineering
• Data Science

Background:

• Mercury contamination in natural gas presents significant risks including equipment corrosion, safety hazards, environmental pollution, and economic losses.
• Accurate prediction of mercury solubility in various gas mixtures is vital for effective risk management and regulatory adherence.

Purpose of the Study:

• To develop and validate advanced machine learning models for estimating mercury solubility in natural gas systems.
• To identify key factors influencing mercury solubility and assess the reliability of the developed models.

Main Methods:

• Utilized multilayer perceptron (MLP), generalized regression neural network (GRNN), and extra trees (ET) machine learning algorithms.
• Trained and validated models using a high-quality dataset under diverse pressure and temperature conditions.
• Employed the Leverage approach to confirm the reliability and trust region of the dataset.

Main Results:

• The MLP model achieved superior predictive performance with a determination coefficient of 0.9998 and a root mean square error of 1.7430 ppb.
• Temperature was identified as the most significant factor influencing mercury solubility.
• 96.5% of the data points fell within the trust region, confirming dataset reliability.

Conclusions:

• A pioneering ML-based framework was established for mercury solubility estimation in natural gas.
• The developed models offer significant industrial potential for real-time monitoring and risk reduction.
• This approach enhances safety, operational efficiency, and environmental sustainability in the natural gas industry.