Improving predictions of shale wettability using advanced machine learning techniques and nature-inspired methods:...

Area of Science:

• Petroleum Engineering
• Geochemistry
• Machine Learning

Background:

• Carbon Capture, Utilization, and Storage (CCUS) is vital for mitigating climate change and enhancing hydrocarbon recovery.
• Shale wettability is a critical parameter influencing the success of CCUS operations in unconventional reservoirs.

Purpose of the Study:

• To evaluate shale wettability using machine learning techniques.
• To identify key factors affecting shale wettability in CCUS environments.

Main Methods:

• Utilized multilayer perceptron (MLP) and radial basis function neural networks (RBFNN) for wettability prediction.
• Trained MLP with five algorithms and optimized RBFNN with three algorithms.
• Analyzed 229 contact angle datasets from shale/oil/brine, shale/CO2/brine, and shale/CH4/brine systems.

Main Results:

• The RBFNN-MVO model demonstrated exceptional predictive accuracy (RMSE: 0.113, R²: 0.999993).
• Sensitivity analysis identified theta zero, TOC, pressure, temperature, and salinity as the most influential features.
• The study confirmed the efficacy of ML in assessing shale wettability.

Conclusions:

• The RBFNN-MVO model is highly effective for evaluating shale wettability in CCUS applications.
• Accurate wettability assessment supports enhanced hydrocarbon production and cleaner production strategies.
• This research contributes to the advancement of CCUS technology in shale formations.