Abstract

The identification of high-quality areas with relatively high porosity and high permeability zones in tight sandstone reservoirs has always been an enormous challenge beneath the background of the ultra-low petrophysical properties. Meanwhile, machine learning has made enormous breakthroughs in many fields as an important branch of artificial intelligence. Hence, the present study is focused on examining how machine learning technology can help improve the accuracy of reservoir quality assessment. To this end, a machine learning method known as the gradient boosting decision tree (GBDT) algorithm is used to design and perform reservoir quality evaluation. First, a typical data set for reservoir quality evaluation is constructed on the basis of the observed lithofacies and diagenetic facies characteristics, and is used to set six logging curves reflecting reservoir characteristics as the input variables and the corresponding reservoir quality categories of the samples as the output results. Then the intelligent GBDT algorithm is applied in order to train the constructed data set. Through confusion matrix analysis, the accuracy of the GBDT model is found to be about 89%. In addition, five commonly-used prediction methods (namely, factor analysis, Bayesian discriminant method, cluster analysis, random forest, and artificial neural network (ANN)) are executed for the lithofacies or diagenetic facies to confirm the reliability and application effectiveness of the GBDT model. Finally, the GBDT model is used to predict the reservoir quality in the study area, which is also well verified in terms of production data. The findings of this study demonstrate that, among the many available machine learning algorithms, the GBDT algorithm has the potential to become an essential tool for fast real-time reservoir quality evaluation.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call