Low-Rank Tensors Applications for Dimensionality Reduction of Complex Hydrocarbon Reservoirs

Abstract

The significant non-uniformity of petrophysical properties and ultra-tight permeability distribution in unconventional reservoirs make them critically complex to investigate. Unconventional reservoirs are usually found in thin layers that expand over hundreds to thousands of miles spatially resulting in huge multidimensional data collection at well locations, which make them numerically challenging for the field development studies and operational analysis.To address such problems, the concept of low-rank tensor decomposition is introduced in this study, to be applied for unconventional reservoir modeling to target issues such as huge dataset management and missing data generation. Low-rank tensor decomposition is a powerful tool that can model a wide range of heterogeneous and multidimensional data. It works by extracting the most useful latent information out of several multidimensional data tensors and reconstruct the entire dataset in a compressed format using low-rank tensors. It is a novel idea to be applied in petroleum/ reservoir engineering to optimize reservoir models as well as field development strategies to significantly improve recovery efficiency and minimize uncertainties.