Machine and Deep Learning Applications in the Permian Basin скачать в хорошем качестве

Machine and Deep Learning Applications in the Permian Basin

Speaker: Robin Dommisse, Senior Geomodeling Advisor, Bureau of Economic Geology, University of Texas at Austin Accurate 3D reservoir characterization models of subsurface reservoirs rely on the maximum extraction of knowledge from the 3D seismic and well log data sets. In this presentation we discuss a new machine learning (ML) powered methodology for 3D lithofacies prediction of clinoform ooid grainstones in a Midland Basin carbonate CCUS reservoir. This new approach is faster than traditional seismic inversion techniques, leveraging the power of Self Organized Maps (SOM), an unsupervised form of machine learning. Using SOM, we create detailed 3D volumes of classified data down to a single seismic sample interval. These volumes are then combined with lithofacies from well logs, merging two independent data sources for more accurate predictions. In a second example we show how to quickly generate fault probability volumes for several 3D seismic surveys in the Permian Basin. The AI-based fault detection workflow discussed here uses a combination of supervised Deep Learning (DL) and unsupervised Machine Learning (ML) technologies to produce fault geobodies which are converted to fault planes for use in 3D sealed, structural geomodels.