Reservoir Properties Prediction from Well Logs Using Supervised Machine Learning Algorithms, Random Forest in Malaysia Fields

This study explores the feasibility of predicting subsurface rock properties, including T2 log distributions, using well log data such as neutron porosity (NPhi) and density (RhoB). These logs are commonly recorded in most wells, unlike T2 logs, which are more challenging to acquire. Developing a predictive tool for T2 logs from NPhi-RhoB data can assist engineers in making informed decisions about subsurface delineation, production, and storage injection targets.

The problem is inherently non-linear, with strong interactions between NPhi-RhoB and reservoir properties. To address this, a supervised machine learning workflow is developed using data processing, feature augmentation, and Random Forest model deployment. Random Forest, an ensemble method comprising multiple decision trees, is employed for regression tasks, calibrating predictions against core and NMR logs.

Using training data from four wells in the West Baram Delta, the model captures trends and provides physically consistent predictions across geophysical logs, including porosity, permeability, Vshale, and T2 logs. Validation in both the West Baram Delta and Malay Basin demonstrates the model’s ability to predict results for different lithologies, such as sand, silty-sand, and shale. This workflow shows promise for improving subsurface characterization and enhancing decision-making in reservoir management.