Optimizing Well Placement with Quality Maps Derived From Multi-fidelity Meta-models (SPE 154416)

This paper presents a new methodology for locating infill wells so as to improve reservoir performance and value. The methodology centers on the determination of both qualitative and quantitative quality maps, quality being a measure of how good an area is expected to be for production. The determination of the best infill well locations is a highly nonlinear optimization problem. Solutions can be found using optimization algorithms. However, they usually correspond to local optima and require a few thousands of fluid flow simulations. This strongly penalizes the use of optimization algorithms for designing field production schemes. In this paper, we proposed a practical solution to handle infill well placement. First, various physical attributes are computed without any flow simulator to approximate the production capability of the reservoir. They are classified and used to delineate regions with poor or favorable potentials for well placement. Second, a few well locations are sampled on the basis of the defined regions and a flow simulation is performed for each of them to estimate how oil production evolves when an infill well is drilled at these locations. The resulting oil production responses are used to approximate oil production at unsampled locations. The specific feature of this method is to consider that grid blocks are characterized by their spatial coordinates plus the distance to the closest existing well. This third coordinate accounts for the wells already drilled and can be easily updated when a new one is implemented. This approach makes it possible to account for well interferences while calling for a reduced number of flow simulations. The proposed method is expedient. It does not yield the optimal locations of the wells to be added. Yet, it provides a useful first-pass set of well locations. An application case is presented to illustrate its potential.