Asia Petroleum Geoscience Conference and Exhibition (APGCE)
Summary
Understanding the distribution of facies in a reservoir is important for reservoir characterization and can be used to aid in identifying the locations of hydrocarbons. However, due to the complexity of the earth’s subsurface, predicting the facies distribution information is often a very challenging task. Recently, the growth of computing power has enabled the application of Machine Learning models in service of studying many geoscience challenges, including the classification of facies using seismic data. Although, many previous studies have mainly relied on the use of multiple seismic attributes to build and train the model, in this work, we develop a Convolutional Neural Network (CNN) model using a U-Net architecture that can accurately map the facies distribution using only the original 3D Post-Stack seismic data. The approach was applied on a reservoir from a real oil field in the Malay Basin. We have improved the model using transfer learning that allows us to use learned parameters from a pre-trained model. The results obtained from the U-Net CNN model provide an accurate and realistic facies distribution compared with the true facies distributions. The model obtained 94% of accuracy for the training dataset and 89% of accuracy for the validation dataset.