(418f) Optimization of Well Placement and Geometry for Tight Natural Gas Production

The optimal placement of wells in tight gas reservoirs is a challenging task because these reservoirs are usually quite complex and reservoir characteristics are often not well understood due to scarce data. Yet optimal well placement has a significant effect on expected recovery, because drainage areas around wells in tight reservoirs are small compared to conventional gas fields. The well placement problem can be cast as a numerical optimization problem. The problem is challenging for a number of reasons: The number of possible scenarios is large, particularly when horizontal well geometries must be optimized; objective function evaluations are computationally expensive because they rely on running a reservoir simulator to estimate future production and recovery; and uncertainty in both reservoir properties and economics must be addressed. Well placement is a fundamental problem and many different approaches have appeared in literature towards its solution, usually for the conventional reservoirs. These approaches focus on two main issues, namely uncertainty and the large number of alternatives. In this work, we propose a methodology for finding optimal well types and its locations for a particular tight gas reservoir based on numerical optimization that relies on the combination of two interrelated methods: Design and Analysis of Computer Experiments (DACE) and Efficient Global Optimization (EGO). Cases like placement of deviated or horizontal wells are efficiently handled using this approach. The optimization is performed first for the starting point of the development of a field as well as at later points, as more wells are added. All available information (such as seismic data, past production data) is assumed to have been translated to reservoir properties that are used in a (rigorous or short-cut) reservoir simulator to provide estimates of future production given well location, geometry, and stimulation. The optimization algorithm proceeds by (a) running the reservoir simulator at a few points to perform corresponding evaluations of the objective function, (b) using the (computationally inexpensive) DACE approximation for evaluation of the objective function at several intermediate points to identify well coordinates and geometries with the maximum expected improvement in the objective, (c) running the simulator for well coordinates chosen in part b, and (d) repeating the process from step b, until convergence. Uncertainty is naturally incorporated in the preceding search as an integral ingredient of the DACE modeling approach.

To illustrate the proposed method, we show computer simulations based on simulated data for a tight gas field. The work presented is done in collaboration with Texas A&M University, Lawrence Berkeley Lab, and Anadarko under funding from RPSEA.