Supervisors: Rolf Johan Lorentzen, Olwijn Leeuwenburgh, and Jakub Wiktor Both
Short description of the project:
The Norwegian offshore oil and gas industry is required to reduce its emissions significantly over the next decade. Offshore oil and gas production is a highly energy-intensive process, particularly due to the energy demands of gas compression and water injection. Consequently, gas turbines, which are traditionally used to meet this high energy demand, contribute the most to emissions on the Norwegian Continental Shelf (NCS).
One possible approach for emission reduction is to utilize power from nearby offshore wind farms. The research conducted in this thesis focuses on developing methodologies for finding optimal operational strategies (often referred to as production optimization) for wind-powered reservoir management. In this context, the operational strategy of a reservoir field refers to the production and injection schedule of the operation. A given strategy affects both the profitability and emissions of the production, and studying the trade-off between the two is therefore possible.
The research conducted in this thesis is divided into two focus areas. The first area focuses on further developments of ensemble-based optimization (EnOpt), which is a popular method used for production optimization. EnOpt uses Gaussian sampling of production strategies to estimate a gradient direction, which is then used in optimization. This thesis presents GenOpt, a novel methodology that samples production strategies from a generalized non-Gaussian distribution, with a particular emphasis on distributions bounded on a domain. GenOpt is also extended to include ensemble-based Hessian approximations. This thesis also investigates the application of popular momentum methods from machine learning, combined with EnOpt, for production optimization.
The second area focuses on developing workflows for optimizing production strategies in reservoir management, where production is powered by wind energy combined with traditional gas turbines. The trade-off between long-term profits and emissions is investigated by performing the analysis over an ensemble of realistic time series of long-term wind power. Additionally, it is investigated how emissions can be reduced in the short term by utilizing a closed-loop workflow with a wind power forecast. The methodology developed in this thesis is demonstrated on benchmark reservoir test models.