A Reservoir Fluid Identification Model Based On Boosting Tree Algorithm

Abstract

Summary The identification of reservoir fluids by well logs is essentially a highly nonlinear mapping function problem. Boosting Tree algorithm that combines multiple decision tree classifiers to complete classification tasks is one of the most popular methods for dealing with nonlinear classification problems. This study, for the first time, explores the application of Boosting Tree algorithm in building reservoir fluid identification model from well log data of Aryskum Oilfield of South Turgay Basin (the reservoir fluids are low resistivity oil and high resistivity water). For this purpose, a total of 1268 log data points derived from GR (natural gamma ray), SP (spontaneous potential), RT (true formation resistivity), DEN (density), CNL (compensated neutron log) and AC (acoustic interval travel time) are used as an input to the model to identify the reservoir fluids in three categories: oil, water, dry layer. Comparison of results given by Boosting Tree algorithm to those of Decision Tree and support vector machine (SVM) is also carried out. In the experiments, the model based on Boosting Tree algorithm achieves an accuracy of 88.5%, far higher than 61.8% given by both Decision Tree and SVM, showing that Boosting Tree algorithm is a powerful tool for reservoir fluid identification.