Collaborative Design Framework for Geophysical Direct Inversion and Its Application in Radial Velocity Inversion(Radial Velocity Direct Inversion Training Dataset)

Description

This dataset contains 287,817 forward-modelled paired samples for radial shear-wave velocity inversion from dipole acoustic dispersion responses. Each MATLAB MAT file contains two paired arrays: param2 (N x 10), representing 10-layer radial shear-wave velocity model parameters in m/s, and velocity2 (N x 256), representing the corresponding dipole acoustic dispersion responses. Rows in the two arrays are paired samples. The dataset supports the experiments reported in the manuscript, “Collaborative Design Framework for Geophysical Direct Inversion and Its Application in Radial Velocity Inversion.” Large source pools are split into multiple files while preserving the original variable names and sample order. Matching variables can be concatenated along the first dimension to reconstruct each complete pool.

Steps to reproduce

The dataset was generated through numerical forward modelling rather than field acquisition. A radially layered formation model surrounding a borehole was parameterized as a 10-layer shear-wave velocity profile over the radial interval from 0.1 to 1.1 m. For each model realization, the shear-wave velocity parameters were sampled within predefined physical bounds and screened according to radial recovery trends, interlayer relationships, and other physical-feasibility constraints. The dipole flexural-wave dispersion equation for a radially layered borehole formation was then solved numerically to calculate the corresponding dispersion response. Each paired sample contains a 10-element radial shear-wave velocity profile (param2) and its associated 256-point dipole acoustic dispersion response (velocity2). The final dataset contains 287,817 forward-modelled model-response pairs. The workflow can be reproduced by constructing physically feasible 10-layer radial velocity models, solving the dipole flexural-wave dispersion equation for each model, and storing the resulting velocity parameters and dispersion responses as paired rows. The accompanying code repository provides the data-construction, training, and evaluation workflows used in this study.

Institutions

• China University of Petroleum, Beijing

  Beijing, Beijing

Categories

Licence