Efficient automatic history matching by reducing the observed data

Automatic history matching assumes that by integrating dynamic data in reservoir description process, the description can be improved. However, the method can be made more efficient by only selecting production data which is more "informative" compared to other production data. A methodology has been developed in this work to identify the important production data and discard those 'less useful' data in the automatic history matching program coupled with the previously developed adjoint method. This enables us to save significant computational time without sacrificing much of the quality of the history matching. In this work, a study was conducted on the interrelationship between the sensitivity coefficients and the observed data. The optimal production data are selected based on the information provided by the sensitivity coefficients.; Two methods are developed in this work, the gradient voting method and the singular value decomposition method, to select the optimal observed data. To avoid computing the sensitivity coefficients for all the production data over a fine grid reservoir description, which is the most time consuming part in the history matching, the data selection procedures are combined with a scale-splitting algorithm. The history matching starts with a coarse grid model using reduced observed data, then the history matching program is repeated over a series of gradually downscaled finer grids. This procedure is repeated and the results from different grid scale levels are compared until no more improvement on the inverted result is found. This will improve the quality of the final results and save computational time.; The objective of this work is to find methods to save the computational time and make the automatic history matching practical. The comparisons of computational time, the quality of the inverted results and the quality of the future prediction were made in this work. The results indicate that this method is promising and practical for the automatic history matching.