Oct. 10, 2024


Description

Geological parameterization entails the representation of a geomodel using a small set of latent variables and a mapping from these variables to grid-block properties such as porosity and permeability. Parameterization is useful for data assimilation, as it maintains geological realism while reducing the number of variables to be determined. Diffusion models are a recent class of generative deep-learning models that are trained to "denoise", which enables them to generate new geological realizations from a random noise input. Specifically, latent diffusion models combine the generative capabilities of diffusion with an autoencoder low-dimensional latent space. Diffusion-generated models can accurately reproduce complex 2D and 3D geological features, and correctly capture the variability in their distribution. Due to the smoothness of the parameterization, diffusion models can also be used for data assimilation in the latent space. Our current application involves conditional 2D three-facies (channel-levee-mud) systems. Significant uncertainty reduction, posterior P10-P90 forecasts that bracket observed data, and consistent posterior geomodels, are achieved. 


Featured Speakers

Speaker: Guido Di Federico
Speaker Guido Di Federico

Guido is a PhD Candidate in the Energy Science & Engineering Department at Stanford University, working in Prof. Durlofsky's Smart Fields Consortium Lab. He holds a BSc. in Mechanical Engineering from the University of Bologna (Italy) and a MSc. in Energy Engineering from Politecnico di Milano (Italy).
At Stanford, his research …

Guido is a PhD Candidate in the Energy Science & Engineering Department at Stanford University, working in Prof. Durlofsky's Smart Fields Consortium Lab. He holds a BSc. in Mechanical Engineering from the University of Bologna (Italy) and a MSc. in Energy Engineering from Politecnico di Milano (Italy).


At Stanford, his research focuses on the parameterization of geological models for subsurface flow simulation using generative machine learning. Such methods can speed up geostatistical simulation by quickly generating large numbers of conditional realizations of a geological concept or training image. Furthermore, they enable the representation of the full geomodel space in terms of a low-dimensional latent variable with known distribution, allowing for latent space-based history matching.

Full Description



Organizer

Prithvi Singh Chauhan


Date and Time

Thu, Oct. 10, 2024

6 p.m. - 7 p.m.
(GMT-0500) US/Central

View Our Refund and Cancellation Policy

Location

Virtual Format



Meeting link will be shared to registered participants.

Group(s): Data Analytics