Enhancing operational decision-making in hydrocarbon exploration drilling using machine learning for gas data...

Area of Science:

• Petroleum Geoscience
• Machine Learning Applications
• Geochemical Analysis

Background:

• Mud-gas interpretation in hydrocarbon exploration is complex due to factors like mud properties and drilling parameters.
• Drill Bit Metamorphism (DBM) can alter gas signatures, complicating the identification of reservoir fluids.

Purpose of the Study:

• To develop a machine-learning-assisted workflow for assessing reservoir-fluid signals in Advanced Gas (AG) measurements.
• To improve the interpretation of gas signatures acquired during drilling operations.

Main Methods:

• Integrated a multi-domain dataset from 104 Brazilian exploration wells with PVT fluid compositions and geological interpretation.
• Developed predictive models using Kernel Ridge Regression, XGBoost, and LightGBM.
• Created Reservoir Affinity Curves and a DBM Severity Curve for fluid characterization and thermal alteration assessment.

Main Results:

• The workflow successfully distinguished between representative formation-fluid signatures and zones affected by DBM or operational artifacts.
• Reservoir Affinity Curves estimate the similarity between AG signatures and reference PVT fluids.
• The DBM Severity Curve quantifies drilling-induced thermal alteration.

Conclusions:

• The developed workflow enhances early fluid characterization in hydrocarbon exploration.
• It improves fluid-sampling decisions and reduces interpretation uncertainty before laboratory analysis.
• This AI-assisted approach offers a more reliable method for interpreting drilling-acquired gas data.