Doctor of Philosophy (PhD)
Craft and Hawkins Department of Petroleum Engineering
This study explores the application of data-driven modeling and prediction in reservoir characterization and simulation using seismic and petrophysical data analyses. Different aspects of the application of data-driven modeling methods are studied, which include rock facies classification, seismic attribute analyses, petrophysical properties prediction, seismic facies segmentation, and reservoir dimension reduction.
The application of using petrophysical well logs to predict rock facies is explored using different data analytics methods including decision tree, random forest, support vector machine and neural network. Different models are trained from a set of well logs and pre-interpreted rock facies data. Among the compared methods, the random forest method has the best performance in classifying rock facies in the dataset.
Seismic attribute values from a 3D seismic survey and petrophysical properties from well logs are collected to explore the relationships between seismic data and well logs. In this study, deep learning neural network models are created to establish the relationships. The results show that a deep learning neural network model with multi-hidden layers is capable to predict porosity values using extracted seismic attribute values. The utilization of a set of seismic attributes improves the model performance in predicting porosity values from seismic data.
This study also presents a novel deep learning approach to automatically identify salt bodies directly from seismic images. A wavelet convolutional neural network (Wavelet CNN) model, which combines wavelet transformation analyses with a traditional convolutional neural network (CNN), is developed and demonstrated to increase the accuracy in predicting salt boundaries from seismic images. The Wavelet CNN model outperforms the conventional image recognition techniques, providing higher accuracy, to identify salt bodies from seismic images.
Besides, this study evaluates the effect of singular value decomposition (SVD) in dimension reduction of permeability fields during reservoir modeling. Reservoir simulation results show that SVD is valid in the parameterization of the permeability field. The reconstructed permeability fields after SVD processing are good approximations of the original permeability values. This study also evaluates the application of SVD on upscaling for reservoir modeling. Different upscaling schemes are applied on the permeability field, and their performance are evaluated using reservoir simulation.
Zhou, Xu, "Data-Driven Modeling and Prediction for Reservoir Characterization and Simulation Using Seismic and Petrophysical Data Analyses" (2020). LSU Doctoral Dissertations. 5276.
Available for download on Friday, May 28, 2021