FMF

When?: February 21, 2017 4:00 PM
Where?: X5114.0001

What?

With the advent of increasingly large data sets, the growing amount of computational resources and recent algorithmic progress, data driven and machine learning techniques are perceived as potential game changers and are receiving unprecedented attention in many industries. Also the oil and gas industry is beginning to evaluate where data driven techniques can speed up, complement and in the long term potentially even replace existing workflows. In this talk, we first sketch the different data sources and data types that feed into the subsurface modelling process. The consistent and timely integration poses a major challenge since the data exists on various scales and is also subject to measurement noise. To make matters worse, large parts of the subsurface are hardly covered by the available data, so substantial uncertainties remain which need to be to be adequately captured in the modelling process. A common approach to deal with this problem is to work with a representative ensemble of models that tries to capture the known uncertainties. Managing and extracting relevant information from this ensemble is another non-trivial task, which we are nowadays trying to approach via machine learning techniques. Hence, in this presentation, we will cover some (basic) machine learning techniques and briefly mention their mathematical and algorithmic properties. We conclude with some high-level sketches of how those techniques can be used to accelerate the process of understanding and effectively handling subsurface uncertainties.

Bio: Jan joined the Quantitative Reservoir Management team of Shell in June 2013. In his current role he is working primarily on applications of statistical and machine learning techniques to complement more traditional reservoir engineering workflows, which are concerned with quantifying and adequately handling uncertainties in the subsurface. Before joining Shell, he studied at the University of Passau in Germany, where he acquired a Diploma in Computer Sciences and a PhD in Applied Mathematics.