Well Location Optimization (WLO) plays a critical role in maximizing recovery from hydrocarbon reservoirs. Identification of well locations typically relies on expert subsurface knowledge and ad-hoc flow simulation scenarios. Such approaches, due to the complexity and uncertainty of subsurface descriptions, can easily miss highly profitable possibilities. This situation has motivated the development of model-based optimization workflows that can assist subsurface specialists in choosing improved well locations. In practice, model-based optimization workflows are implemented in a deterministic fashion and applied to low, mid and high simulation scenarios. The workflow proposed in this paper takes into account model uncertainty in the optimization outcomes.
Deterministic WLO entails optimizing the easting and northing coordinates of wells with respect to an objective function (e.g. net present value or ultimate recovery). The workflow applies to fixed well trajectories. For each new proposed well location, there is a mapping of the new well connections to the reservoir simulation grid. The approach has shown resilience in handling realistic scenarios including non-uniform areal gridding, pinch out features, local grid refinements or tartan gridding, and non-uniform layering dimensions.
Deterministic optimization techniques may anchor the results to particular portions of the uncertainty space, while robust optimization techniques propagate the effect of uncertain parameters on the optimization outcomes. The trade-off between the investigated span of uncertainty space and available computational capability needs to be made. Due to the continuous growth in computational capabilities, it is ever more attainable.
Robust optimization incorporates the effect of uncertain parameters in the optimization process. It is a natural extension of the history matching process in brownfields, but it is also applicable to green fields where dynamic data is yet to be received. For such cases, a robust optimization workflow will yield outcomes that take into account the uncertainty of parameters described by a prior distribution, e.g. only conditioned to static data.
The WLO workflow involves (i) executing deterministic WLO, (ii) examining the deterministic WLO results on a set of reservoir simulation models that capture the uncertainty space, and (iii) executing robust WLO. This workflow was tested with gradient-based optimization methods, derivative-free methods, and hybrid methods, the latter being the most optimal of the three.
The robust implementation of this workflow handles realistic reservoir simulation scenarios by exploiting coarse-grained parallelism and has been applied successfully to field cases with up to fifty wells. In this paper, WLO is showcased on one synthetic case and one field case to illustrate the capabilities of the workflow.