Tracking the state of stress evolution within geothermal reservoirs is important for optimizing energy production and characterizing seismic hazard. Borehole image analysis and hydraulic fracture tests can constrain principal stress directions and magnitudes at discrete locations, but cannot provide a full descrption of the evolving stress tensor throughout the reservoir. Interferometric synthetic aperture radar (InSAR) observations can reveal high-resolution, reservoir-scale, time-dependent surface displacements that can be used in inversion models to infer subsurface stress sources and changes in the stress tensor. However, such stress inversions are subject to various a priori assumptions, including source type, location, number, or magnitude. We develop a novel method to obtain a descrption of the evolving stress field that utilizes machine learning (ML) to quantify subsurface stress tensor changes from InSAR observations. We train a convolutional neural network (CNN) with data produced using analytical realizations of subsurface strain and surface displacement from sources representing volumetric and double couple deformation. From the strain values we calculate subsurface stress in the geothermal reservoir. We establish the range of earthquake parameters and fluid pressure sources that produce InSAR-measurable surface displacements and train the CNN to characterize the stress field from surface displacements. In parallel with our CNN solution to the relationship between ground deformation and interior stress, we develop an independent solution employing a variational procedure. These two problem solutions are used interactively to inform the development of each. We demonstrate the application of our ML-based solution for the Coso Geothermal field. The success of our ML-based approach suggests that with generalized training, our method could be employed broadly to geothermal fields to provide field operators with a near real-time model of stress field evolution.

Press the Back button in your browser, or search again.

Copyright 2021, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.