Machine Learning Approach for Exploration: A Case Study in the Central Sumatra Basin, Indonesia

Geologically, the Central Sumatra Basin represents a series of present-day back-arc basins that are present along the central region of Sumatra island. At the same time, Central Sumatra Basin is the biggest oil producer basin in Indonesia with large subsurface data sets. These large data sets need to be re-evaluated to optimize the exploration strategy. These studies demonstrate the effectiveness of machine learning methodologies in constructing a detailed model of facies and hydrocarbon distribution. This has been made possible using many fascinating algorithms especially supervised learning, such as K-Neighbors Classifier, AdaBoost Classifier, Random Forest Classifier, and other algorithms. The modeling process uses static and dynamic petro-technical data, including petrophysical, geological, geophysical, production, and reservoir data sets, which the existing contractor already interprets. The algorithm was trained and tested to data sets and then analyzed which feature was suitable for each machine learning method. The facies and hydrocarbon distribution results can help understand and identify the regions that are more likely to have oil and gas-producing rocks. In addition, it shows the importance of good data cleansing and labeling to get a better machine learning model. These machine learning results can be used in exploration strategies and get a valuable model to decide which prospect to be drilled.