Communications of the ACM
Stanford Scientists Train Computer Models to Accurately Simulate Nature's Variability
Stanford University scientists have devised a new method for generating "training images" that can be used to fine-tune models of uncertainty about underground physical processes and structures.
Credit: Shutterstock
Stanford University researchers have developed a technique for generating "training images," which can be used to refine models of uncertainty about subterranean physical processes and structures.
"We want to fundamentally understand uncertainty in naturally variable systems," says Stanford professor Jef Caers.
Multiple-point statistics (MPS) algorithms have become widely used because they can produce geologically realistic models of actual variability. The Stanford team has devised a way to select multiple training images, which can be ported into MPS algorithms to help model natural variability.
A miniature river basin environment was engineered to simulate millennia of river deposition in a few days, while a camera took pictures once a minute. The "demon algorithm," which calculates the strain acting on a moving image over time, was chosen to analyze a series of images and select those captured during periods when changes to the system were small and similar in nature. With this method, the Stanford team selected six images as training images for the MPS algorithm, and they discovered the variability generated by their geostatistical technique corresponded well with the "natural variability" represented by the snapshots of the basin.
"This research is a first step toward bridging the fundamental gap that exists between those fields of science focusing on the physical understanding of natural systems and the statistical representation of uncertainty when modeling complex systems," Caers says.
From Stanford Report
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found