A new way of processing data from rock measurements could lead to a much more efficient oil extraction. During her PhD research, Sangeetika Ruchi developed a method to infer the most probable rock properties, based on only a few indirect measurements.
The best way to extract crude oil depends on many factors, in particular the rock properties of the oil reservoir. These properties are however notoriously difficult to assess. “The structure beneath the Earth’s surface is full of complexity and uncertainty”, says Ruchi.
Production wells provide measurements of rock properties, but only indirectly and only of a few locations. The amount of available measurements is much smaller than the amount of uncertain rock properties. Ruchi: “We cannot drill soil everywhere, so we need to make many assumptions regarding the structure of the surface and its petrophysical properties.”
Quite often this results in situations where the same measurements lead to different simulations of an oil reservoir – all with a different probability of occurring. These uncertainties hamper the development of optimal oil recovery strategies and, ultimately, lead to a far lower oil yield.
Accurate estimations based on limited knowledge
In order to make predictions of future oil extractions from a reservoir, the oil reservoir is simulated on a computer. “The challenge is to make accurate estimations of rock properties based on limited knowledge”, says Ruchi. “Moreover, limitations in computing power worsen the situation. For each and every configuration of an oil reservoir, you will need a computational simulation. These simulations are what we call ‘expensive’. They require a lot of computational resources and especially time.”
Scientists currently use two approaches to address the uncertainties of rock properties. The first approach is to simulate all plausible rock properties, even those that have a small probability of occurring. This approach is computationally very demanding. The second approach is to make strict assumptions about the rock properties – even if they might not hold in reality.
Ruchi now introduces a new approach to overcome these shortcomings. Her method is based on methods used in meteorology, called data assimilation. These methods combine measurements and computational weather models to give weather forecasts. When adapted, they are also suitable for making other types of predictions. The meteorological methods perform well, as long as the number of uncertain properties is small. However, when facing a large number of uncertain properties, like rock properties, the models struggle to give accurate predictions.
Fast without making assumptions
Ruchi refined the meteorological methods, to make them work for predicting oil reservoir properties. Her method requires only computational simulations of the most probable oil reservoir configurations, and provides accurate estimations of the rock properties.
Ruchi: “A major improvement is that it works relatively fast and makes no assumptions about uncertainties. These assumptions are a key ingredient of standard mathematical approaches, but don’t hold in practical scenarios. Being able to work without these assumptions, we can make far better estimations.”
The next step
The new method has proven to outperform standard mathematical approaches when tested on relatively simple subsurface structures. According to Ruchi, the next step would be to test its performance for more complex test cases. Also, the method can be further refined to reduce its computational expense.
Ruchi defended her thesis at the Utrecht University on 20 January 2020. Her research was supervised by prof. Jason Frank (Utrecht University) and dr. Svetlana Dubinkina of CWI's Scientific Computing group. During her PhD research, she visited the University of Nottingham as a guest researcher, as well as the University of Potsdam, that offered her a research fellowship for excellent PhD students. Ruchi’s project was part of the research programme Computational Sciences for Energy Research, organized by NWO and Shell. Her PhD thesis Parameter Estimation in Random Energy Systems using Data Assimilation is available to view and download at CWI’s online repository.
Ruchi now continues her career at the Dutch company ASML, the world’s largest supplier of photolithography systems for the semiconductor industry. At ASML she will focus on the development of accurate mathematical models for photolithography machines.