Multiple Dispersion Spectra Inversion Using a Wide Neural Network

Summary Shear-wave velocity (Vs) is one of the geotechnical parameters that can reflect the elastic properties of the near-surface medium. The multichannel analysis of surface waves (MASW) method is commonly used to characterize the Vs structure in urban areas. The MASW method, however, has a weak ability to reflect the lateral changes of near-surface structures because of the assumption of a one-dimensional layered earth model. To improve the lateral resolution of inversion results without sacrificing the stability and efficiency of the MASW, we designed a wide neural network (WNN) to conduct dispersion spectrum inversion based on wave equation to replace the dispersion curve inversion in the MASW. Compared with the traditional deep-learning-based inversion method, our WNN adopted a new inversion mode taking multiple dispersion spectra as input and a two-dimensional Vs profile of the coverage area as output. Synthetic test results indicate the new WNN can achieve a high-precision Vs profile and provides a more accurate characterization of coverage areas of different dispersion spectra compared with the traditional method.