Assisted history-matching is now widely used to constrain reservoir models. The objective of history match process is to improve the model in order to reproduce the production history while honoring the structural properties of the geological model. For that purpose, an objective function is defined in order to measure the mismatch between the simulation results and the production history. This objective function is usually minimized using a gradient-based optimization algorithm. However, history-matching of a large number of parameters in hydrocarbon reservoirs is a challenge because of several reasons: scarcity of available measurements relative to the number of unknowns, computational effort required for large reservoir and the need of insure that solutions are geologically realistic. All of these problems can be helped by using algorithms that rely on efficient and parsimonious descriptions (or parameterizations) of reservoir properties.
In this work, a history matching methodology is presented. First, a sensitivity study is performed for identifying the most relevant inversion parameters affecting the history-matching. The gradual deformation method is applied for the parameterization. Then, a new optimization technique, based on data partition for the gradient calculations, is studied for regional and well level history matching. The objective function is first split into local components, and the principal parameters are reduced for each component. In this context, we can propose perturbation designs with a smaller number of perturbations for the gradient computation. The proposed new technique is successfully applied on a real case in Libya in an integrated workflow, which makes history-matching with a large number of parameters tractable.