End-to-end seismic inversion of geostatistically complex reservoir facies models with deep convolutional neural networks

WA Tech night

Title: End-to-end seismic inversion of geostatistically complex reservoir facies models with deep convolutional neural networks

Presenter: Anshuman Pradhan, Stanford University

Date & Time: 6th August 2020; 12 – 1PM AWST

Summary: We present a framework for performing end-to-end seismic inversion of reservoir facies models under complex geostatistical models of prior uncertainty. In our methodology, we directly learn the end-to-end inverse mapping between 3D seismic data and reservoir facies using deep 3D convolutional neural networks. Our training dataset is simulated from the forward generative model comprising of the geostatistical prior on facies and geophysical model relating seismic to facies through elastic properties. To ensure reliability during prediction with real data, a method for performing data-based falsification of prior uncertainty is presented. Using a real case study from an offshore deltaic reservoir, we demonstrate the efficacy of our approach by inverting a large-scale facies model from 3D post and partial stack seismic data.

Biography: Anshuman Pradhan is a PhD candidate in the department of Energy Resources Engineering at Stanford University. He is a research assistant associated with the Stanford Center for Earth Resources Forecasting, Stanford Rock Physics and Borehole Geophysics project and the Stanford Basin and Petroleum System Modeling consortia. Anshuman obtained his M.S. and B.S. degrees in Applied Geophysics from Indian Institute of Technology (Indian School of Mines), Dhanbad, India. Anshuman has several industry and academic internship experiences where he has worked on applications related to reservoir modeling, seismic inversion and machine learning.