A Machine Learning-Based Data Augmentation Approach for Unconventional Reservoir Characterization Using Microseismic Data and EDFM

Abstract

Multi-stage hydraulic fracturing has recently gained strong interest in unconventional plays in the Middle East due to high natural gas production potential. However, prevalent characteristics of the area, including high-pressure / high-temperature (HPHT) conditions and presence of complex natural fracture networks, pose significant challenges to reservoir characterization. These challenges have motivated the development of an integrated workflow using microseismic data for the characterization of reservoir properties resulting from the interaction between natural and hydraulic fractures. This study proposes a reliable method for modeling hydraulic fractures from scarce microseismic data. Initially, a microseismic model—based on field records of microseismic data and natural fracture spatial characterization—was developed. Issues related to limited microseismic data availability were tackled through combination of a probabilistic algorithm, Gaussian Mixture Model, and a DFN model. Then, the resulting synthetic microseismic events enabled the generation of a hydraulic fracture model using the embedded discrete fracture model (EDFM) and an in-house microseismic spatial density algorithm that captured major hydraulic fracture growth tendencies. Next, the created hydraulic fracture geometries were validated against a physics-based hydraulic fracture propagation model. Lastly, a single-well sector model—based on a corner point grid that honored the original 3D discrete fracture network (DFN)—was history matched, confirming the successful application of the proposed methodology.