We present a data-driven dispersion curve inversion network (DispINet) that directly maps the multimode dispersion curves to the S-wave velocity model. DispINet is trained with synthetic samples that consist of multimode dispersion curve data and the corresponding S-wave velocity models. The S-wave velocity models are generated according to prior information in a specific study area. Multimode dispersion curves are calculated by the generalized reflection-transmission coefficient method for each model. The well-trained DispINet could be used to predict the S-wave velocity model for other dispersion curve data never seen by DispINet. Testing data and Qademah field data results show that DispINet has the ability to retrieve a reasonable underground S-wave velocity structure in real time.
ASJC Scopus subject areas