Control-Switching Strategies for Reservoir Water-Flooding Management
MetadataVis full innførsel
Waterflooding is a commonly used operation in the oil industry. It is used to in-crease oil recovery from the reservoir, and is considered as a secondary recoverytechnique. In short, the reservoir is flooded with water to increase pressure, and todisplace oil from pore spaces. This thesis has explored three general methods foroptimizing this operation, with special attention paid to treatment of constraints.As means of background theory, a brief introduction to petroleum engineering isprovided, the concept of dynamic optimization is stated, and reservoir simulationis presented. The three methods of interest are gradient-based control optimization, reactivecontrol, and techniques based on simulator-embedded constraints. The popularityof gradient-based control optimization has grown during the last decade, and therehas been conducted a vast amount of research on the topic during this period. Suchoptimization is usually referred to as adjont-based optimization, as efficient com-putation of gradients serves as a prerequisite for the method to remain tractable forreservoir optimization. Next, reactive control is widely used in the industry, and itspopularity is among others due to simplicity, robustness, and model-independence.The last method considered is based on simulator-embedded constraints, which oftoday is a relatively unexplored area. A new heuristic has however been developedduring this work, which combines output unconstrained control optimization andsimulator-embedded constraints. The three general approaches are implemented using the Matlab Reservoir Simu-lation Toolbox. A series of four case studies are employed to assess the merits ofthe methods. In short, if geological uncertainty is put aside, gradient-based op-timization appeared as the preferred method. Furthermore, the popularity of thereactive control approach can be understood, as the method performed adequately- keeping its simplicity in mind. The heuristic that was developed showed somepotential, but there are three issues at the current stage that must be addressedand resolved.