Comparison of optimization algorithms for reservoir management with constraints—A case study

Abstract

Two important goals of reservoir management are responsible stewardship of the resource and finding a high net present value during the forecast period of reservoir assessment. The first step of reservoir management is characterization of the reservoir through history matching. Even with a history matched model, a robust production optimization algorithm is required to find an optimal value of net present value (NPV). Field-scale optimization problems consist of a highly complex reservoir model with many unknown control variables. Finding a high value for NPV in a reasonable time depends on the efficiency of the optimization algorithm. There are many optimization algorithms and in this paper three efficient algorithms for optimization are investigated. These are steepest ascent (SA), sequential quadratic programming (SQP) and the interior point (IP) methods. We found that although each iteration of the steepest ascent approach is relatively inexpensive, the method was unable to achieve a high value of the NPV because of the neglect of constraints in the gradient. In contrast, the sequential quadratic programming (SQP) often found a high NPV, but the computation cost of each iteration was substantial due to the need to identify active constraints.The interior point method is introduced for development planning in this paper. Although the final NPV achieved with the interior point method may not be as high as the NPV achieved with SQP, it is greater than the NPV obtained from the steepest ascent method with similar computation cost. Application of these methods to the Brugge field with 4560 total fluid rate constraints demonstrates the strengths and weaknesses of these algorithms. In all of these methods, gradient of controls has been computed using ensemble based method.