Unless the capacity is exceeded, we will not respond to your email. Attendance is free, just bring along your lunch.
RSVP NOW for the next Microseismic User Group (MUG) event.
If you have any questions, please contact:
Paige Mamer, Paige.Mamer@tgs.com,
Johnny Wentzel, Johnny.Wentzel@esgsolutions.com
Abstract
This talk goes over the use of unsupervised machine learning techniques (i.e. clustering) to try and group microseismic events into discrete fracture networks. An anonymized industry hydraulic fracturing program is used as a case study to compare several “traditional” techniques like partitional, hierarchical, and density-based clustering with a model-based technique. The talk will be approximately 30 to 40 minutes. It will briefly cover the theory for the clustering techniques and some exploratory data analysis to select attributes for clustering. It will then dive into a comparison of different techniques and show why some of the common validation metrics don’t work with hydraulic fracturing programs. The results of the gaussian mixture model will then be presented for the entire microseismic catalogue, along with some R and Python code snippets to get the audience started with these techniques using open-source software.
Biography
Scott is a PhD student in geomechanics at the University of Calgary Department of Civil Engineering. He isn’t a geophysicist but got excited about microseismic data when working with Dr. David Eaton’s group studying induced seismicity. This led him into the dark realm of machine learning from which he emerged with this talk. The rest of his PhD studies center around numerical modeling, subsurface characterization, and Bayesian statistics.