Full Waveform Inversion Method for Roadways Based on Wave Velocity Structure Correction and Regularization Constraints

Concealed geological formations that induce disasters frequently result in incidents, such as water and gas outburst, during coal mine roadway excavation, thereby underscoring the critical necessity for precise advanced exploration of geological structures and their spatial positions to ensure safe excavation practices. Full waveform inversion (FWI) presents several advantages, including high-resolution imaging of complex geological structures and the effective inversion of essential physical parameters. However, the unique observational conditions, such as the linear arrangement of vertically excavated working faces in the roadway whole space and the implementation of single-offset detection, significantly exacerbate the challenge of non-uniqueness in the inversion process. To mitigate the problem of cycle skipping in the FWI of seismic advanced detection within roadways and to enhance inversion accuracy, a study was undertaken to investigate the impact characteristics of FWI under these unique observational conditions present. Initiating from the model structure and incorporating a time-domain multiscale inversion strategy, a novel approach was proposed for constructing a single-scale initial model based on wave velocity structure correction, thereby yielding a more accurate initial model for subsequent scales of inversion. Furthermore, a convolutional wavelet-independent inversion method, along with a total variation regularization approach, was introduced to effectively tackle the challenges of wavelet estimation in actual data inversion, as well as to more effectively mitigate the nonlinearity and ill-posedness inherent in the inversion process. Research findings indicate that, in comparison to conventional elastic wave FWI methods, the newly developed roadway FWI methodology can enhance inversion accuracy by approximately 20% in roadway scenarios, thereby providing high-precision imaging results.