Uncertainty quantification with application of distance metric in naturally fractured reservoirs modelling

Characterization and modelling of naturally fractured reservoirs (NFR) is usually complicated with very high degree of heterogeneity and uncertainty related to fractures. A commonly used framework for uncertainty estimation such as Monte-Carlo modelling is straightforward but in case of NFR is highly time-consuming as it requires generation of a large number of realizations and their flow simulation. We propose a more efficient method in terms of time cost, for uncertainty estimation in NFR flow performance. The idea of the method is to select a subset of reservoir models reflecting the same uncertainty range in flow response as the full set. The large set of NFR models is generated capturing the variability of fractures parameters. We calculate Euclidean distance between flow responses obtained from results of fast but not accurate flow simulations and apply multidimensional scaling to map realizations into some space representing spatially their uncertainty. Grouping similar realizations in clusters we find those realizations which are located in their centers and hence the most different. Once the most diverse realizations are obtained, an accurate flow simulation is run and uncertainty is quantified using only selected small subset of realizations. We demonstrate the workflow on the synthetic but realistic example.