Reservoir facies classification based on random forest and geostatistics methods in an offshore oilfield

Abstract

Machine learning methods are increasingly employed in various seismic and petrophysical methods for parameter estimation, interpretation, prediction, and classification. Reservoir facies classification assists the interpretation of seismic data as an important step in petroleum exploration and production monitoring. In this study, we estimate a reservoir facies model by integrating random forest (RF) algorithms and geostatistics modeling. The Surmeh Formation with the Jurassic age is known as one of the most important hydrocarbon reservoirs in the Middle East. The upper part of the Surmeh Formation is equivalent to the Arab Formation, which includes sequences of evaporitic carbonate facies in the study area. Well log data including DT, GR, RHOB, and PHI are used in the RF method for reservoir facies classification. Cross-validation verifies the high accuracy of our classification, with an average accuracy of 95%. The predicted reservoir facies consistently describe the carbonate and evaporitic facies with the geological information of this formation. The decision tree diagrams of the RF algorithm give valuable information on decision limitations and how to select features for efficient computation. We use the classification results for facies modeling. The comparison between facies models and drilling core data shows that the APE value of the sequential indicator simulation model is less than that of the indicator kriging model.