History matching is a model validation procedure that consists in simulating the past performance of a reservoir and comparing the results with historical data. Differences are evaluated as an objective function. When differences are found, modifications are iteratively made to the input data for improving the match. This iterative process is often performed manually and can be very time consuming as well as frustrating. Moreover, an equally satisfactory history matching can be obtained from different reservoir descriptions. Historically, several methods of assisted history matching were studied. Gradient techniques were the first optimization tool to be looked in. Even though these widely used techniques are efficient in terms of convergence between the simulation and the production data, they can easily get trapped in local minima of the objective functions, resulting in a premature convergence of the search algorithm. Evolutionary algorithms were also considered very interesting because of their ability of avoiding local minima, but they considerably slow down the simulation convergence. They have been fully tested in many research fields as well as in reservoir engineering but only on analytic objective functions. In this study, a typical genetic algorithm and evolution strategies with different parameter combinations were employed in both single and multi-objective optimization techniques. The merging of the multi-objective optimization technique with evolutionary algorithms as presented and discussed in this paper represents an innovative methodology for performing assisted history matching. A synthetic reservoir model with production and injection data was generated to test the different approaches. Results were analyzed in terms of mean and minimum objective functions and distance from the true solution. They showed that a better performance of the evolutionary algorithms was obtained when the multi-objective optimization technique was adopted. Although the algorithms did not always approach the optimal solution due to the non-linearity of the problem, they proved to be very powerful in order to identify the most representative reservoir model based on a satisfactory history matching. Additionally, the application of a multi-objective optimization technique can help finding multiple optima and thus identifying multiple scenarios, which then become the input for a subsequent assessment process of the uncertainty associated to forecast output

This content is only available via PDF.
You can access this article if you purchase or spend a download.