A Machine Learning-Based Multi-Scale Natural Fracture Characterization Approach Using Seismic Attributes in the Sichuan Basin

ABSTRACT: Multi-stage hydraulic fracturing is a crucial technology in the Sichuan Basin, China owing to its potential for natural gas production. However, reservoir characterization in a complex geological environment whose architecture is dominated by small- and large-scale fracture corridors poses significant challenges to an effective economic field development. Multistage fracturing uniformity indices rely on an optimal fluid distribution in every stage, which can be hampered by geological structures that serve as leakage sites in the reservoir. Recent field developments in the Sichuan Basin's shale reservoirs have motivated to characterize natural fracture properties and hydraulic fracture geometry based on geophysical data. This work employs an innovative method to extract three-dimensional (3D) natural fractures in fracture corridors from ant tracking and maximum curvature seismic features. The fracture corridors are directly included in reservoir simulation workflows. The results can be readily used for hydraulic fracturing completion optimization in unconventional reservoirs. 1. INTRODUCTION Gas production in Sichuan basin has made significant progresses in the last few years by adopting well-known technologies such as multistage fracturing. However, geological complexities still limit optimal deployment of stimulation treatments. During hydraulic fracturing propagation, numerous fracture corridors and pad-scale faults are encountered, which poses challenges to a uniform fluid distribution inside perforation clusters. Fluid leaking in such geological complexities reduce the efficiency of the stimulation treatment, leaving poorly stimulated stages. For those reasons, operators have put significant efforts in geological characterization with the hopes to efficiently model potential leakage sites and find strategies to avoid low performance. Literature shows work on automatic fault detection and fracture corridor characterization using seismic data. Fault detection has progressed in the last few years, resulting in accurate representations of such features (Hu et al., 2014; Fang et al., 2017). Fracture corridor characterization, however, is often simplified using faults to describe them (Astratti et al., 2010) or using upscaling methods in reservoir models (Ahmed et al., 2010; Zhang and Koutsabeloulis, 2010). The accurate modeling of both faults and fracture corridors are essential to optimize gas production in the Sichuan basin.