A Data Intense Challenge: The Instrumented Oilfield of The Future

The overarching goal of this project is to develop an efficient framework that will accommodate diverse physical problems, associated numerical simulations, and data acquired through simulations or experimental measurements in order to enable more effective decision making and reliable parameter estimation on physical domains. This is a joint project between teams of researchers from 1) OSU Biomedical Informatics Department, 2) Center for Subsurface Modeling, Institute for Computational Engineering and Sciences, University of Texas at Austin, 3) Institute for Geophysics, University of Texas at Austin, and 4) Department of Electrical & Computer Engineering, Rutgers, The State University of New Jersey.

Optimization for decision making and uncertainty estimation are very challenging tasks in many engineering and scientific applications, such as biomedicine, structural mechanics, energy and environmental engineering. The overall goal is both to generate good estimates of optimal parameter values and to reliably predict end results. Finding a solution to the objective function requires a systematic search of the parameter space and evaluation of corresponding scenarios within the physical domain. An example application scenario is oil reservoir management. The number and locations of wells in an oil reservoir have significant impact on the productivity and environmental effects of the reservoir (i.e., optimum economic revenue, minimum bypassed hydrocarbon, minimal environmental hazards/impact). If the oil extraction wells are not placed carefully, large volumes of bypassed oil may remain in the field. The amount of water that needs to be injected (in order to drive the oil toward extraction wells) and disposed of also depends on the number and locations of injection wells as well as the extraction wells. If the objective of decision making is to maximize profit, the main decision parameters are the number of injection and extraction wells and their locations. These variables can be controlled by the reservoir management team. Variables such as the geological and fluid properties of the reservoir and the economic parameters (e.g., the cost of pumping water), on the other hand, are uncontrolled variables that need to be taken into account. Finding a solution for the objective function with controlled and uncontrolled variables requires the evaluation of a large number of possible configurations. For each placement of the wells, the reservoir model has to be evaluated for many time steps in order to calculate required parameter values (i.e., the effective volume of extracted oil and the net value). Moreover, the values of uncontrolled variables are not known precisely, introducing a high level of uncertainty into the problem.

In this and similar scenarios, an exhaustive search of the space is often unfeasible, since the space can consist of thousands to millions of data points and requires the evaluation of a large number of potential scenarios corresponding to these points. Our approach is to couple optimization algorithms with simulations and experimental measurements to enable a systematic evaluation of the scenarios. Such an approach is characterized by dynamic interactions between complex numerical models, optimization methods, and data. Major challenges to efficient application of this approach include the large computational times required by the complex simulations, the challenges of integrating dynamic information into the optimization process, and the requirements of managing and processing large volumes of dynamically updated datasets (obtained either from simulations or sensors). To address these challenges, we have been developing in this project

• algorithms to support parallel multi-block numerical models coupled with global stochastic optimization algorithms,
• execution engines that implement adaptive runtime management strategies to enable efficient execution of optimization and simulation processes in distributed and dynamic computational environments, and
• systems to handle very large and potentially dynamic multi-dimensional, scientific datasets on large scale storage systems.

Project Researchers

Tahsin Kurc, Ph.D. (Lead Developer)

Project Funding Participation

ITR/AP&M Data Intense Challenge: The Instrumented Oilfield of the Future

Project Publications