Deep learning in geophysics
Kriging, first developed in the 1970s, has traditionally been the go-to method in geophysics and soil science for geospatial properties interpolation.
Kriging suffers from many practical issues due to its linear and deterministic nature and is particularly unsuitable for big data analysis. Our project aims to investigate the feasibility of using Deep Learning (DNN) for geostatistical estimation, as the method has proven extremely powerful in predicting non-linear relationship.
A DNN model was developed for the Dullingari Complex to derive correlations between a large volume of seismic attributes and recorded well logs and construct a 3D full-field petrophysical property block based on the interpolation model. A sequential stochastic simulation method was also developed to transform the deterministic result to probabilistic model suitable for decision making analysis.
Theme
Future energy and resources
Booth
FE03
School
Australian School of Petroleum
Exhibitors
Hoang Son Le
Alvin Chin