Machine learning based real-time microseismic monitoring and stimulated fracture characterization at the Utah FORGE Geothermal site

In April 2022, a three-stage hydraulic stimulation was performed in a deep granite heat reservoir of low permeability at the Utah Frontier Observatory for Research in Geothermal Energy (FORGE). During the stimulation, around 1600 m³ pressurized fluids were injected into the target reservoir of ~2.4 km depth aiming at creating fracture networks and improving reservoir permeability for heat extraction. Microseismic monitoring is required to assess the stimulation efficiency and manage the induced earthquake risk during the stimulation. We perform near-real-time microseismic monitoring in a playback mode at the third stage of the stimulation where three deep monitoring boreholes equipped with three-component geophone chains were in operation. We apply machine learning (ML) techniques in combination with waveform back projection approaches to automate the microseismic event detection, increase microseismic event location accuracy, and promote the real-time capability of the monitoring workflow. Due to a lack of labeled datasets for model training or transfer learning, we devise a rescaling technique to tune the continuous microseismic recordings of high sampling rates that enables the application of existing ML models pre-trained on tectonic earthquakes. Our benchmark tests show that the proposed rescaling approach achieves high precision and accuracy in detecting microseismic events and picking their phase arrivals.

With the proposed workflow, we compiled a high-resolution microseismic catalog containing around 36, 000 microseismic events with magnitudes of –3.0 to 0.5. Detected events are relocated using a double-difference relocation method and waveform cross-correlation-based arrivaltime refinement. We cluster the detected microseismic events according to their spatial distributions and identify the dominant stimulated fracture planes with principle component analysis of the different event clusters. The spatial distribution of the detected events nicely depicts the stimulated fracture networks which can be used to design the trajectory of the future production well. We analyze the spatio-temporal evolution of the induced microseismic events during and after the stimulation to illuminate the rupturing mechanisms responsible for the induced fracture networks. Induced microseismic events are analyzed together with the injection data to quantify the induced earthquake hazard and the hydraulic stimulation efficiency. The proposed microseismic monitoring workflow and the corresponding analysis provide more insights into the fracturing dynamics and the potential induced earthquake hazard in the Utah FORGE geothermal site, and would benefit the operation of other enhanced geothermal systems.