Research on the application of crosswell seismic full-waveform inversion method in high-steep structures

Abstract

Fault detection from 3D seismic data is commonly performed using scalar fault attributes, such as fault likelihood, coherence, and related geometric measures. These attributes can highlight discontinuities, but the resulting fault traces often suffer from fragmentation, unstable responses near intersections, and ambiguous lateral continuity on time slices. This study proposes a graph-clustering workflow for seismic fault enhancement. The method first applies multidirectional component analysis to decompose a fault-attribute slice into local directional components, and then builds a sparse weighted graph from high-response candidate points. Edge weights integrate attribute strength, spatial distance, axial-direction consistency, multidirectional component labels, and local linear-extension compatibility. Strong-edge connected components are interpreted as fault-trace instances. A proof-of-concept test on the open F3 dataset shows that the proposed workflow suppresses isolated responses, improves trace continuity, and preserves intersecting fault trends after component-wise clustering. Application to a Tarim Basin field dataset further indicates that the enhanced results provide clearer fault patterns on time and horizon slices than the original thinned fault likelihood. The workflow provides an interpretable intermediate step between local directional fault enhancement and 3D fault surface interpretation.