Operational Optimization of Oil Reservoirs in the Danish North Sea

Abstract

Long-term oil production optimization for oil reservoirs management is the main topic in this Ph.D. thesis. In a globalized world with a continued high demand for oil and gas, unpredictable oil prices and stringent environmental regulations, the industry is seeking improved oil recovery methods that reduce financial risks and ensure a minimal environmental footprint. By intelligent utilization of known production technologies and already existing infrastructure, production optimization methods have the potential to significantly improve the oil recovery factor for both new and mature oil and gas reservoirs, and to ensure a more sustainable production. The overall purpose of the work presented in this thesis is to bring the technology for oil production optimization forward to a level ready for industrial uptake. We focus on computing optimal open-loop control strategies for reservoir management by applying adjoint gradient-based methods for constrained optimization. The goal is to develop a framework that enables application of optimal control to operational optimization of industrial-scale reservoir models. In order to achieve this goal, we combine novel as well as existing methods and strategies for oil production optimization together with well-established proven simulation software and state-of-the-art gradient-based software for constrained optimization. This thesis has three main contributions: 1. Hierarchical multigrid optimization method. To enable application of production optimization procedures on high-resolution reservoir models, we introduce a hierarchical multigrid optimization method. The method utilizes a semi-automated model reduction procedure to generate a hierarchy of increasingly coarse models, based on the high-resolution model. The optimization procedure is initialized at the lowest (coarsest) level. Optimal solutions computed at lower levels are propagated through the model hierarchy, thereby utilizing the information to obtain faster convergence. 2. Handling of nonlinear output constraints. Upper bounds on total water and gas production rates due to limited topside capacity are examples of nonlinear output constraints naturally arising in production optimization. However, these types of constraints are difficult to handle in large-scale optimization problems. A formal treatment of nonlinear output constraints requires an additional adjoint simulation for each constraint, e.g. an upper bound on the gas production rate throughout the reservoir lifetime requires one additional adjoint simulation for each controlled time interval. Hence, a formal treatment of nonlinear output constraints in production optimization on real-life reservoir models will render the problem computationally infeasible. We propose to handle nonlinear output constraints by a soft constraint penalty method, where the constraint gradients are approximated using a simplifying fluid flow assumption. This method reduces the gradient computations to one additional adjoint simulation for each constraint type. Consequently, this method enables incorporation of nonlinear constraints in the optimization procedure. 3. Software for oil production optimization of industrial-scale reservoir models. We have developed a software optimization tool (RESOPT) that enables application of production optimization methodologies on industrial-scale reservoir models. The software tool merges the simulation power and robustness of well-established reservoir simulators with adjoint gradient capabilities and state-of-the-art gradient-based software for constrained optimization. By this software integration, we enable routine use of existing complex reservoir models in ensemblebased optimization strategies, for combined profit maximization and risk minimization. Consequently, the software constitutes a powerful tool to support and guide decision-making in a real-life reservoir management process. This thesis consists of a summary report and a collection of seven research papers written during the Ph.D. project period from December 2015 to May 2019. Five research papers are published and two paper manuscripts are in submission.