Statistics Seminars: Error Modelling for Flow in Porous Media
4 March 2008 14:15 in CM221
The task of 'history matching' oil reservoir models to observed data is complex and time consuming. It shares characteristics with inverse problems from many fields of science, and in particular often relies on many runs of computationally expensive finite difference fluid flow codes. The cpu demands mean that one often runs with many fewer simulations than one would like, and often at lower resolution.
There are a number of approaches that are used in history matching and uncertainty quantification. One approach consists of building emulators - computationally cheap approximations to the complex code - and using the emulators in place of the complex code.
The goal of this talk is to describe recent developments in solution error modelling. Solution error modelling aims to fit a statistical model to the difference between a highly resolved physically realistic model, which of necessity cannot be run often enough for use in history matching, and a simpler, reduced resolution or reduced physics model.
The talk will be illustrated with recent applications of the solution error model concept in the oil industry, in the Lorenz equations of atmospheric physics, and in gas dynamics.
Contact email@example.com for more information