Research on Multi-Attributes Neural Network Inversion in Coalmines’ 3D Seismic Exploration

Abstract

The conventional inversion can only get P-impedance data, but neural network inversion can obtain various kinds of well-log attributes data by getting the nonlinear relationship through training the data of seismic attributes and well-log attributes, and then apply the nonlinear relationship to the whole seismic data volume. In this paper, we use multilayer feed forward neural network, probability neural network and radial basis function neural network to carry out the log-density inversion research, and obtain three pseudo density data volumes. We compare the effect of the inversion results and analysis the density distribution in spatial domain through section and slice data, at last we predict the stability of the coal seam’s floor in more reasonable way.