Toward investigating the application of reservoir opportunity index in facilitating well placement optimization under geological uncertainty

There have been several attempts to reduce the computational costs associated with well placement optimization under geological uncertainty by selecting a representative subset of realizations or by reducing the number of function evaluations to speed up the optimization. Although previous researchers have studied this issue, they have used static or dynamic properties of the reservoir to select a reduced subset of realizations or shrank the search space of the optimizer to reduce the computational burden of the task. Using static parameters is fast but lacks the accuracy of the dynamic approach. On the other hand, dynamic parameters require a computationally expensive full flow simulation. In this paper, a workflow based on the Reservoir Opportunity Index (ROI) maps was proposed to address the two issues mentioned above. The ROI maps were constructed using both the static and dynamic reservoir properties. The dynamic feature was obtained by the computationally efficient Fast Marching Method instead of full reservoir simulation. In this workflow, ROI density (DROI) maps were constructed by averaging the ROI values of adjacent cells to incorporate the effect of neighboring cells in finding potential regions to constrain the search space of the optimizer. In addition, these DROI maps were used to select a representative subset of realizations. Furthermore, the effect of re-selection of the representative realizations was investigated in this study. The proposed workflow was applied to optimize the location of production wells under geological uncertainty and was compared with the Full Case (using all realizations in optimization) in terms of three criteria, namely the expected value, the standard deviation, and the cumulative distribution function (CDF) of the objective function. Results of optimization showed that the constrained search with a fixed number of realizations and the statistical clustering technique outperformed other approaches with minimum differences of 1.5% and 8.9% from the objective function and its standard deviation of the values obtained from the Full Case, respectively. Also, it had 28.64% less difference between its CDF and the Full Case in comparison to the second-best technique. In addition, it was shown that the constrained search required fewer function evaluations to achieve similar results as the global search with approximately three times fewer function evaluations. The results of this paper can be used as a guidance for reservoir engineers to select a proper approach for optimization (constrained or unconstrained), clustering the geological realizations (statistical or K-means), and proper implementation of ROI maps for shrinking the search space of the optimizer and selecting a representative subset of realizations.