Estimating the existence probability of cavities using integrated geophysics and a neural network approach

The existence probability (Ep) of lime silicate cavities, which affect ground stability, was analyzed using borehole, electrical resistivity, and gravity data. A neural network training function was used to estimate the contribution coefficients of physical properties that play a part in determining the Ep. For this process, the resistivity and density values of areas for which borehole data exist were used as training input vectors, and a target probability vector was constructed using the borehole data. The pseudo-Ep was then calculated using the estimated contribution coefficients. The range of pseudo-Ep was adjusted for conversion to the Ep of cavities, and this new approach was verified by comparison with previous researches and commensurate with previous results. The new probability approach is useful not only for detecting cavities, but also for imaging underground structures.