An Efficient Procedure for Expensive Reservoir-Simulation Optimization Under Uncertainty

Abstract

Summary Many oilfield optimization problems concern expensive numerical-reservoir-simulation-based objective functions. These are computationally demanding and can require a significant amount of time to evaluate. With the addition of uncertainty in the underlying models, the computational cost is exacerbated and the collective cost necessary to realize the efficient frontier for the decision maker can become considerable. Although approximation methods are often used to alleviate the cost of optimization, methods to improve the efficiency of the entire process in the presence of uncertainty are less established. This paper presents a scheme that, unlike the conventional approach, ensures that a convex efficient frontier is generated while also reducing the number of simulation evaluations necessary. This is achieved by reuse of sampled data through a procedure referred to as "recasting." Here, existing sample data in the solution space are mapped to a particular objective space of choice dictated by the weights assigned to the mean and standard-deviation terms. Results from an analytical test case, a numerical reservoir-simulation model, and a coupled reservoir plus surface-facility-simulation model are presented. The results show that the proposed workflow can help reduce the high computational cost associated with expensive simulation based function optimization under uncertainty.