Scatter Search Metaheuristic Applied to the History Matching Problem

Abstract

Abstract Reservoir simulation plays an important role in managerial decisions during the life of a reservoir. Thus, it is vital that the simulations present an accurate prospect of real reservoir performance. The main goal of the History Matching (HM) process is to improve the quality of numerical models by contrasting simulated with observed data. A typical HM process evaluates a chosen objective function (OF) comprised by non-linear functions on the uncertain reservoir properties that measure the current mismatch. Thus, the HM problem is a minimization problem where the decision variables are reservoir characterization properties and the goal is to find a set of values for the decision variables that minimize the OF. This paper focuses on the use of an optimization technique called Scatter Search (SS) to perform the HM. The main feature of SS is that it works on a set of solutions called the reference set. The goal is to improve the overall quality of the reference set after each new iteration. New solutions are generated by a non-convex combination of reference solutions. The proposed methodology was used in both automatic and assisted HM processes and the analyzed reservoir models presented increasing challenges to the algorithm. This paper describes how the HM problem was solved with the SS framework, the expected performance of the methodology in dealing with complex solution spaces and also points how it can be expanded to better scale to a growing number of optimization parameters. The application of the SS methodology to the HM problem is a recent trend. Unlike most metaheuristics, SS can be effective even when the simulation time of each instance of the problem is long. Most of the current HM methodologies do not perform well when the solution space is both large and complex.