5th December 2019
Applications of Supervised deep learning for Seismic interpretation and inversion
With Dr York Zheng at BP
The adoption of modern machine learning techniques by the oil and gas industry has been rapid, but obtaining robust results from their applications to real seismic data is still challenging. In this talk, two different case studies are presented to demonstrate supervised deep learning as an alternative to conventional techniques for seismic interpretation and inversion. The first example is a classification problem, in which a convolutional neural network (CNN) is trained to pick faults automatically in 3D seismic volumes. The second example is on elastic model building, re-casting pre-stack seismic inversion as a machine learning regression problem. In both cases, I show that CNN models can be trained on synthetic data and subsequently used to make efficient predictions on field data, whilst also highlighting some of the challenges of applying supervised learning techniques to seismic data.
Kindly sponsored by Pays International Ltd
York Zheng has been working as a geophysicist in BP’s Complex Imaging team since 2012. Prior to BP, York obtained his PhD in geophysics from the University of Cambridge, focusing on Full Waveform Inversion. He also worked in the Canadian oil sands upon completion of his BSc in Geological Engineering from the University of Waterloo in Ontario, Canada. At BP, York has worked on various projects including land and marine seismic processing, velocity model building, FWI and seismic reservoir characterization of fields in the North Sea, North Africa, Middle East and GoM. His recent work involves using machine learning techniques for geophysical problems and Bayesian optimization.
PESGB - Training Room