Understanding the Facies Architecture of a Fluvial-Aeolian of Tensleep Formation Using a Machine Learning Approach

Many aeolian dune reservoirs are built from various dune types, and many may remain unrecognized in subsurface work. The challenge is to tackle the complex geological architecture of dune types within the Teapot Dome dataset caused by wind and water erosion. Machine Learning (ML) helps predict facies architecture away from boreholes using seismic attributes and facies logs. It provides a detailed understanding of the facies architecture analysis of the relationship between the fluvial–aeolian environment in Tensleep Formation based on seismic and well data. It allows operators to wisely assess their hydrocarbon reservoir, improve safety, and maximize oil and gas production investment. The data from the Teapot Dome field (Naval Petroleum Reserve No.3 - NPR-3) provides a good testing ground for Machine Learning, as it is easy to validate and prove its value. This study will show how the ML supervised learning method incorporating Neural Network Seismic Inversion (NNSI) can successfully create porosity log and facies volumes. Moreover, unsupervised learning using Multi-Resolution Graph-based clustering (MRGC) can be used to classify the facies logs. NNSI has 0.963 for the cross-correlation coefficients for all wells. The ML approach was used to help recognize the type of aeolian dune reservoirs in the subsurface and correlate the well log and facies volumes. In addition, ML allowed the distinct sequences and reconstruction of their depositional history in the Tensleep Formation. This study also refers briefly to other examples of fluvial-aeolian facies architecture worldwide. It successfully found the ancient model in an existing modern fluvial-aeolian environment, revealing hidden information about facies architecture based on the geometrical shape of geobodies in the oil-producing reservoir in the Tensleep Formation.

Discrete 3D fracture network extraction and characterization from 3D seismic data — A case study at Teapot Dome

Three-dimensional discrete fracture networks (DFNs) extracted from the seismic data of the Tensleep Formation at Teapot Dome successfully matched 1D fracture data from multiple boreholes within the area. The extraction process used four seismic attributes, i.e., variance, chaos, curvature, and spectral edge, and their multiple realizations to define seismic discontinuities that could potentially represent fractures within the Tensleep Formation. All of the potential fracture attributes were further enhanced using a fracture-tracking attribute for better extraction and analysis of seismic discontinuity surfaces and their network properties. A state-of-the-art discontinuity surface extraction and characterization workflow uniformly extracted and interactively characterized the seismic discontinuity surfaces and networks that correlate with borehole fracture data. Among the attributes, a fracture-tracking attribute cube created out of the high-resolution spectral-edge attribute provided the best match with the borehole fracture data from the Tensleep Formation. Therefore, the extracted discontinuity planes were classified as fractures and then characterized. The extracted fracture population also matched earlier published records of faults and fractures at Teapot Dome. Unlike the conventional method, which uses 1D borehole fracture data as primary input and 3D seismic data as a guide volume during DFN modeling, I used 3D seismic attributes as the primary data and the 1D borehole fracture data only for quality control. I also evaluated the power of converting seismic fracture attribute volumes into discrete surfaces and networks for effective correlation with 1D fracture logs from boreholes.