Predicting methane solubility in water and seawater by machine learning algorithms: Application to methane transport...

Area of Science:

• Environmental Science
• Geochemistry
• Computational Chemistry

Background:

• Fracking operations can release methane, potentially contaminating groundwater.
• Accurate methane solubility data is vital for subsurface transport simulations.
• Understanding methane migration is key to assessing environmental risks.

Purpose of the Study:

• To develop and validate machine learning models for predicting methane solubility.
• To cover a wide range of temperatures (273.15–518.3 K) and pressures (1–1570 bar).
• To assess model performance against established equations of state.

Main Methods:

• Employed four machine learning algorithms: regression tree (RT), boosted regression tree (BRT), least square support vector machine (LSSVM), and Gaussian process regression (GPR).
• Utilized experimental data from literature for model training and validation.
• Optimized model hyperparameters using Grid Search (GS), Random Search (RS), and Bayesian Optimization (BO).

Main Results:

• The Boosted Regression Tree with Bayesian Optimization (BRT-BO) model demonstrated the highest accuracy.
• Achieved a coefficient of determination (R²) of 0.99 and a mean squared error (MSE) of 1.19 × 10⁻⁷.
• The BRT-BO model showed excellent agreement with experimental data and outperformed existing equations of state.

Conclusions:

• Machine learning, particularly the BRT-BO model, provides a robust method for predicting methane solubility.
• The model's accuracy is suitable for simulating methane transport in diverse aquatic environments, including deep marine settings.
• This research contributes to better environmental risk assessment related to natural gas extraction.