Departmental Research Colloquium: Deconvolution in well testing: From Least Squares to Bayesian Statistics
2 November 2016 15:00 in CM101
Deconvolution is a technique perhaps more usually applied to problems such as the de-noising of images, but is equally useful in the more esoteric problem of well test analysis in the world of petroleum engineering. Using only information on the pressure and the rate of production we can deconvolve a unique signature for the reservoir that gives insight into the geometry and geology of the system underground. Getting such rich information from a very simple data set is highly desirable, but reliable methods for deconvolution for this problem are not common.
In this talk, I will give a general introduction to the context in earth sciences, the mathematics underlying this problem (arising from the flow of fluids in porous media), and then focus on the statistical method to arrive at a solution. I will illustrate the complications that arise from working with progressively more complex and realistic data, and the adjustments required by the methodology. Ultimately, while we still arrive at a solution we find that (unsurprisingly to some) it might just have been better to be Bayesian from the start.
Contact email@example.com for more information