3D real-time imaging for electromagnetic fracturing monitoring based on deep learning

The electromagnetic method has a proven physical basis and advantages in subsurface fluid detection. The result of fracturing operation can be evaluated by monitoring the electromagnetic anomalies from low-resistivity fracturing fluid before and after the fracturing and inferring the range of fracturing fluid distribution. However, the traditional electromagnetic 3D inversion is time-consuming and cannot meet the requirement of real-time imaging during fracturing. In this paper, we use an improved supervised deep fully convolutional network (FCN) to learn the relationship between surface electromagnetic data patterns and the underground fracturing fluid distribution models. The relationship is encoded in many synthetic "data-model" pairs obtained through 3D forward modeling. By completing the forward modeling and neural network training on the computer cluster in advance, we successfully carried out a field experiment of 3D real-time imaging of fracturing fluid.