| CPC G01V 1/306 (2013.01) [E21B 47/14 (2013.01); G01V 1/50 (2013.01); G06N 3/08 (2013.01); E21B 2200/22 (2020.05); G01V 2210/6226 (2013.01)] | 15 Claims |
|
1. A system for drilling a new well in a subterranean formation by estimating a rock property away from a drilled well, the system comprising:
one or more hardware processors configured to access acquired three-dimensional (3D) seismic data that includes seismic traces from a 3D seismic survey of an area of interest;
a multi-head Convolutional Neural Network (CNN) model including a plurality of kernels of various sizes for determining spatial and temporal relationships of the acquired 3D seismic data at different resolutions, the multi-head CNN model being trained to generate an estimated rock property value of a formation zone away from the drilled well and included in the area of interest; and
a rig for drilling the new well in the area of interest and according to a drilling program for a production system,
wherein the multi-head CNN model includes a plurality of heads and a plurality of layers,
wherein each head of the plurality of heads is an input channel that reads the acquired 3D seismic data at a different resolution per input channel,
wherein each head includes a kernel of a different size in a one-dimensional (1D) convolution layer of the plurality of layers,
wherein a first layer of the plurality of layers receives the seismic traces as input,
wherein a second layer of the plurality of layers is the 1D convolution layer which scans the seismic traces, the second layer comprising sixty-four neurons, a kernel of a particular size, a stride of one, and a nonlinear activation function,
wherein a third layer of the plurality of layers performs a dropout procedure on one or more nodes of the multi-head CNN model at a drop rate of 0.3,
wherein a fourth layer of the plurality of layers performs a batch normalization and a concatenation of the output of the plurality of heads,
wherein a fifth layer of the plurality of layers is a densely-connected layer with one neuron,
wherein a sixth layer of the plurality of layers outputs the estimated rock property value,
wherein the one or more hardware processors are further configured to update the drilling program for the production system based on the estimated rock property value, the drilling program being executed on a computing device of the production system.
|