Multiscale fracture prediction technique via deep learning, seismic gradient disorder, and aberrance: Applied to tight sandstone reservoirs in the Hutubi block, southern Junggar Basin

Abstract

Fracture prediction is an important and active area of research for oil and gas exploration in fractured unconventional reservoirs. Traditional seismic fracture prediction techniques come in one of two flavors, prestack anisotropy-based or poststack edge-enhancement attributes such as ant tracking and maximum likelihood. Inaccurate predictions may result from an apparent low signal-to-noise ratio in the prestack domain approaches or from cumulative effects that are misrepresented in poststack data; there also are shortcomings from using a single pre- or poststack domain. We propose a comprehensive multiscale prediction framework to delineate major faults (using deep learning), associated minor faults (using seismic gradient disorder), and fractures (using seismic aberrance). The principles of deep learning, seismic gradient disorder, and aberrance are introduced and their application effects are verified through the study of tight sandstone reservoir fractures in the Hutubi area, the southern margin of the Junggar Basin.