Closed Loop Reservoir Management Using the Ensemble Kalman Filter and Sequential Quadratic Programming

Abstract

Abstract As part of the SPE-ATW "Closed-loop reservoir management" held in Bruges, June 2008, a synthetic data set was made available to the participants. This paper describes our workflow and results for closed-loop reservoir management using this synthetic model. The closed-loop exercise is divided into two cycles, each consisting of data assimilation and production optimization. The data assimilation is done using the ensemble Kalman filter, and the optimization is done using sequential quadratic programming after a reformulation of the problem to reduce the number of variables. The methodology performs well on the case considered here, and high NPV values are obtained. In particular we have proposed a new strategy for production optimization that makes a large scale optimization problem feasible through a proper reduction of control variables based on analysis of the properties of the field and application of a state of the art optimization method. Introduction Closed-loop reservoir management is typically performed in two steps, an identification and updating step, and an optimization step where an updated reservoir model is used for long term production optimization. Over the last years there have been published a number of studies demonstrating the benefits of closed-loop reservoir management see e.g., Jansen et al. (2005); Nævdal et a. (2006); Sarma et al. (2006); Wang et al. (2007); Chen et al. (2008). In these studies, closed-loop reservoir management has been applied on small synthetic reservoir models, and different combination of approaches has been used for the updating and the production optimization step. As part of the SPE-ATW "Closed-loop reservoir management" arranged in Bruges, June 2008, the participants were invited to perform a closed-loop study on a large-scale model which was more realistic than the models in the previous studies. Another level of complexity included in this study was the fact that the "true" model was hidden to the participants and only known as TNO. The idea was to mimic the workflow performed applied while applying closed-loop reservoir management for a real reservoir. In this cases the reservoir is a synthetic model which consists of 450,000 grid cells. This true model was first used by TNO to generate ten years of initial production history for the field, and then also applied when running the production strategies provided by the participants for the next 2×10 years. (Closed-loop reservoir management is expected to be performed at a shorter time interval than ten years, but this constraint was included to make the workload reasonable for TNO).