Machine learning to optimise the petrophysical workflows in oil and gas exploration
Data architect - EPCC
Abstract: The oil and gas industry is awash with sub-surface data, which is used to characterize the rock and fluid properties of the sub-surface. This information in turn drives commercial decision making, exploration and exploitation planning. However, this wealth of data is poorly utilized and the full value seldom realized. The industry currently relies upon highly manual workflows, and-so a key question is how best to leverage data to complement the current activities of engineers in the search for hydrocarbons. In this talk we will be presenting work done in collaboration with Rock Solid Images, RSI, who sell reservoir characterisation services to oil and gas companies. Their software, RockAVO, supports the decision making of large companies about where to conduct oil and gas exploration. However, their software must be fed by well atlases, which require a complete petrophysical interpretation of each constituent well and this is a currently a manual, time consuming process.
Much of this petrophysical interpretation is based on the experience of specialists and there are some clear patterns, based on accepted geological knowledge, that one can see in the data. As such, a key question is whether making better use of the data, using modern analytics techniques, has the potential to dramatically reduce time to decision and the quality of the decision that are made. In this work we concentrate on supervised learning, where models are trained with labelled well log data and capture the underlying patterns driving mineral composition, the porosity of the rock and fluid saturations. In addition to talking about the methods and technologies we have used, we will also discuss the real world impact of machine learning in the oil and gas industry and explore some of the challenges we faced and solutions found.
Bio: Dr Nick Brown is a data architect at EPCC with research interests in parallel programming language design, compilers and runtimes. He has worked on a number of large scale parallel codes including developing MONC, an atmospheric model used by the UK climate & weather communities which involves novel in-situ data analytics. He is also interested in micro-core architectures developing ePython, a very small memory footprint Python interpreter with parallel extensions, for many core, low memory chips. Nick is a course organiser on EPCC's MSc in HPC course, as well as supervising MSc and PhD students.