Electrical imaging of hydraulic fracturing fluid using steel-cased wells and a deep-learning method

Abstract

Steel-cased wells used as long electrodes (LEs) in a surface-electric or electromagnetic survey can enhance anomalous signals from deep hydraulic fracturing zones filled by injected fluid. Although recent research has been published on the algorithms designed for the simulation of the effect of casings, feasibility studies on resolving the small-scale fracturing fluid flow from surface data are lacking. We have carried out a detectability and recoverability study for a top-casing electric source configuration. Applying a fast 3D DC modeling code using the concept of the equivalent resistor network, the detectability study shows that favorable conditions for detecting the fluid flow direction include multiple electrically coupled wells and different electric conductivities above and below the shale layer. The observed behavior is then modeled through a circuit analog. For the purpose of fracturing fluid imaging, the model for recovery is simplified as a distribution of full fluid saturation on a 2D fracture plane. A deep-learning (DL) framework is adopted to solve the imaging problem, which is difficult for conventional regularized inversions. Our DL implementation uses a supervised deep fully convolutional network to learn the relationship between data patterns on the surface and the model of fracturing fluid distribution, encoded in a large number of synthetic data-model pairs. Once trained, the neural network can make a real-time and pixel-wise prediction of the fluid distribution. The robustness of our DL approach is tested in the presence of ambient noise and inaccuracies in casing conductivity. A reasonably consistent prediction performance has been observed. Our numerical feasibility study results demonstrate that electric surveys using steel casings as LEs have great potentials in real-time monitoring of fracturing fluid flow.