Discrete-Fracture-Network Generation From Microseismic Data by Use of Moment-Tensor- and Event-Location-Constrained Hough Transforms

Summary Reservoir simulation and prediction of production associated with hydraulic-fracturing require the input of the fracture geometry and the fracture properties such as the porosity and retained permeability. Various methods were suggested and applied for deriving discrete fracture networks (DFNs) from microseismic data as a framework for modeling reservoir performance. Although microseismic data are the best diagnostics for revealing the volume of rock fractured, its incompleteness in representing the deformation induced presents a challenge to calibrate and represent complex fracture networks created and connected during hydraulic-fracture stimulation. We present an automated method to generate DFN models constrained by the microseismic locations and fracture plane orientations derived from moment-tensor analysis. We use a Hough-transform technique to find significant planar features from combinations of the microseismic source locations. We have modified the technique with an equal-probability voting scheme to remove an inherent bias for horizontal planes. The voting mechanism is a general grid search in the space of fracture strike, dip, and location (φ,θ,r, respectively) with grid-cell sizes scaled by uncertainty estimates of φ,θ,r. We constrain fracture orientations with weighting on the basis of the moment-tensor orientations of neighboring events and their associated uncertainties. With two case studies, we demonstrate that our automated technique can reliably extract the complex fracture network on the basis of good matches with the event-cloud trends and the input moment-tensor orientations. We also tested the sensitivity of the technique to event-location uncertainty. With increasing location uncertainty, the details of the fracture network extracted are diminished with events grouping to larger-scale features, but the general shape and orientation of the fracture network obtained are insensitive to the location uncertainty.