Seismic double-beam neural network approach to characterizing small-scale fractures in geothermal fields

Reliable seismic characterization of fracture zones is crucial but challenging in geothermal exploration and drilling. We image and characterize fractures with scales less than the dominant seismic wavelength using our novel machine learning (ML) method, called the seismic doublebeam neural network (DBNN). We shoot a source beam toward a fracture zone and obtain a receiver beam containing the multiply scattered waves arriving at the surface receivers from the fracture zone. The interference pattern created by these two beams is called a double-beam image. It carries information about the discrete fracture network pattern. We train a neural network to interpret the double-beam image as fractures. We verify our method using synthetic data and apply it to a 3D field seismic dataset acquired at the Soda Lake geothermal field to obtain discrete fracture images.