Implementation of Adaptive Localization for Enhancing Ensemble-Based History Matching in Hydrocarbon Reservoir Management

Abstract

In reservoir management, the ensemble-based history matching is applied to quantify and update uncertainty in reservoir characterization with the main objective to support high quality decisions. However, the ensemble-based history matching could suffer from statistical problems that make the ensemble unable to quantify uncertainty statistically-correctly. Localization schemes can effectively solve the ensemble-based history matching problems. The aim of this work lies in the research about the practical pros and cons of adaptive over non-adaptive localization schemes, that has not been extensively tested in different settings for ensemble-based history matching of reservoir simulation models. This work proposes a general workflow to guide localization implementation and evaluation. The workflow is developed in five sections: i) the initial ensemble generation; ii) history matching with no localization; iii) history matching with non-adaptive localization (distance-based); iv) history matching with adaptive localization (correlation-based); and v) Comparative result analysis among history matching cases. The main conclusion from this work is that the history matching with the adaptive localization scheme overperformed the history matching with no localization and with the non-adaptive localization scheme. The history matching with adaptive localization demonstrated to estimate the reservoir dynamics and heterogeneities with more accuracy that can enhance decision quality in reservoir management. In addition, the implemented adaptive localization scheme supports standardization in the implementation of localization. However, the spatial regions for updating of the non-adaptive localization scheme are more intuitive than correlations of the adaptive localization scheme. Although correlations of the adaptive localization scheme embed reservoir dynamics, the understanding and interpretation of the correlations are not as straightforward as spatial regions are. Therefore, the adaptive localization scheme requires understanding of the statistical method and interpreting correlations to build history matching ownership. The study brought innovative guidelines to best practice for implementing non-adaptive and adaptive localization. The effectiveness of the workflows was tested, evaluated, and contributed to further developing and improving the history matching processes. The identified practical pros and cons will facilitate the adaptive localization implementation in Equinor.