Uncertainty is an intrinsic property of the oil & gas exploration & production industry. A major goal in any interpretation project is to reduce uncertainty to the minimum, and one way to do that is to work in an integrated fashion, using as much data as possible.
With the huge amounts of subsurface data available to interpreters (well logs, core samples, prestack and poststack seismic data, multiple attributes, etc.), it is virtually impossible for the human mind to integrate and extract all the available information in a timely manner.
This is where machine learning comes in. Machine learning technology boosts interpreters’ confidence by extracting unprecedented amounts of information from vast and heterogeneous data sources for use in geological classifications, rock property predictions, and anomaly detections.
The newest version of SeisEarth (Paradigm 18) offers several state-of-the-art machine learning-based technologies in the form of automated and easy-to-use guided workflows embedded in the integrated interpretation platform. These include:
Waveform Classification Workflow
Waveform classification and amplitude blending in SeisEarth 3D Canvas
Waveform classification and amplitude blending in SeisEarth 3D Canvas
Allows interpreters to easily perform seismic trace-based classification to create facies maps while interpreting data and performing reservoir characterization and modeling.
This workflow uses the neuronal (SOM1D) method, an advanced artificial intelligence process that excels at pattern recognition. Interpreters can perform unsupervised classification, resulting in a set of facies determined purely by the neural network algorithm, or a semi-supervised classification using logs, specific seismic traces or traces from a wedge model.
Rock Type Classification Workflow
Geobodies detected in rock type classification
Geobodies detected in rock type classification
A supervised classification algorithm to predict facies distribution and probability of occurrence away from well control.
The workflow uses the Democratic Neural Network Association (DNNA) method to generate a probabilistic lithologic model calibrated to wells.
It incorporates lithofacies information from wells and any poststack or prestack seismic attributes.
Attribute Clustering Workflow
Classification of structural attributes to highlight faults as clusters with opacity rendering
Classification of structural attributes to highlight faults as clusters with opacity rendering
A new unsupervised classification method used to create classification volumes from poststack and/or prestack seismic data.
The algorithm uses a Self-Growing Neural Network that extends the use of classification techniques beyond seismic facies analysis.
The method can also be used to isolate data outliers for AVO analysis and structural delineation.
Lorena Guerra is a Paradigm E&P software product manager for Interpretation. She is an expert in 3D interpretation and visualization, having worked for Marathon Oil and Landmark Graphics before coming to Emerson’s Exploration & Production Software business. Lorena has a B.Sc. degree in Geophysical Engineering from Simon Bolivar University, Venezuela.
From Jim: You can connect and interact with other oil & gas E&P experts in the Oil & Gas group in Emerson Exchange 365 and/or at the October 1-5 Emerson Exchange conference in San Antonio, Texas.
