Deep Learning Based 3D Velocity Field Nonlinear Multiple Regression

In this work we describe a Deep Neural Network workflow for 3D Velocity Field Nonlinear Multiple Regressive method, the algorithm has been designed to work even with a relatively small training set or large amount of features. By examining the time-velocity pair generated by supervised DNN, we arrive at that the workflow can well predict the velocity function all around the 3D seismic survey under the supervised seeds manual-picking time-velocity pairs. The use of this type of tool in the future will help geophysicist potentially reduce manual-picking control point density, reduce time spent on manual-picking and obtain the relatively accurate time-velocity pair through the Nonlinear Multiple Regressive prediction thus driving down the cost of processing a dataset. This method can also be used as an attribute interpolation tools such as seismic trace interpolation and recovery, well log interpolation and so on.