Special Report: Oil & Gas Computing
Not long ago, the playbook in unconventional operations called for drilling horizontal wells about anywhere in a “blanket” formation, so long as the wellbore stayed in zone to allow stimulation at regular intervals spaced along the lateral. The name of the game was breaking rock. While well productivity remains a function of creating fractures in low-permeability rock, oil and gas producers have come to appreciate the importance of how and where laterals are placed to ensure access to quality rock.
Advanced Attribute Analysis
Techniques that concentrate on a few key seismic attributes have proven highly effective in finding anomalies in subsurface datasets. But Tom Smith, chief executive officer of Geophysical Insights, says technological advances are making it possible to use seismic attributes in almost infinite combinations to delineate anomalies in unconventional plays.
There are hundreds of potential attributes of interest in seismic data, he notes. Reviewing them all to find the best attributes for analysis, and then using them to find sweet spots is a demanding task. “We set out to apply automated, unbiased analysis to this problem,” Smith says. “We developed Paradise™, an advanced geosciences
analytic software platform, to enable interpreters to apply these advanced pattern recognition methods to address this problem.”
Smith says Paradise provides workflows that guide geoscientists through the application of unsupervised neural networks (UNNs) and principal component analysis (PCA). “Paradise also takes full advantage of high-power, multicore processing using large-scale parallelism to accelerate the performance of these advanced techniques,” he says.
Unconventional reservoirs have introduced a new suite of rock mechanics properties, and Smith says the industry is still learning which ones provide the most valuable insights. UNNs have the advantage of running uninterrupted and unbiased by human assumptions.
“UNNs offer the advantage of operating on seismic data alone without the need for well logs. Where well logs are available, those data can be included in the UNN analysis, as can data from hydraulic fracturing,” Smith states. “The more data provided to the system, the more information we can discern from the results. We view this process as advantageous because we make no assumptions about linear or statistical combinations.”
Smith points out that even the most detailed well logs represent a tiny sampling of the subsurface. “It is not that obvious how to sort the properties,” he observes. “Running a supervised neural network is problematic in unconventional plays because the rock properties known at the borehole are an extremely limited sample set. Better tools are needed to lower exploration risk in unconventional plays. By applying both UNN and PCA on the seismic response, greater insights can be realized about the geology and sweet spots identified.”
UNNs look at the natural properties and find natural clusters that are not artificially biased in any way. “We are working in n-dimensional space, where n is the number of attributes,” Smith details. “Attributes can vary in data type and some parameters are predetermined.”
In unconventional formations, interpreters typically search for overpressured zones, sweet spots, AVO and fracture networks. They also look for anomalies and anything that is out of the ordinary. “Neural networks can scan large volumes to find areas of interest for further analysis,” Smith explains. “This capability enables interpreters to focus more effectively and efficiently.”
While the results of attribute analysis are presented in a 3-D cube, Smith says his team has built a 2-D color bar in the Paradise software to more effectively analyze and interact with the volume. The user selects a few neurons on the 2-D color bar, and the 3-D representation highlights only the regions in the volume that correspond to those neurons, enabling isolation of the classification results.
Geophysical Insights’ Paradise™ analytic software platform applies automated pattern recognition to analyze seismic attributes in almost infinite combinations to delineate anomalies in unconventional plays. The workflows use unsupervised neural networks and principal component analysis while taking advantage of high-power multicore processing using large-scale parallelism to accelerate performance. Shown here are attribute analysis results presented in a 3-D viewer with a 2-D color bar for interacting with the data volume.
Most Popular Papers
Systematic Workflow for Reservoir Characterization in Northwestern Colombia using Multi-attribute Classification