This is a presentation on the Probabilistic Estimation of Seismic Response Using Physics-Informed Recurrent Neural Networks by GTC Analytics, presented by Jesse Williams. This video slide presentation discusses the development of machine learning-based predictive tools to estimate the magnitude-frequency response of stimulation-induced seismicity. This presentation was featured in the Utah FORGE R&D Annual Workshop on August 15, 2024.

This is a presentation on the Probabilistic Estimation of Seismic Response Using Physics-Informed Recurrent Neural Networks by GTC Analytics, presented by Jesse Williams. This video slide presentation discusses the development of machine learning-based predictive tools to estimate the magnitude-frequency response of stimulation-induced seismicity. This presentation was featured in the Utah FORGE R&D Annual Workshop on August 15, 2024.

This is a technical report for the Probabilistic Estimation of Seismic Response Using Physics Informed Recurrent Neural Networks project. The report describes the process of designing a recurrent neural network (RNN) to predict induced seismicity. Background material is included to inform non-subject matter experts about the types of architectures available. The exact architectures (layers) of three models are discussed, which are being used to predict induced seismicity.

This is a presentation on the Cutting Edge Application of Machine Learning, Geomechanics, and Seismology for Real-Time Decision Making Tools During Stimulation by the University of Utah, presented by No'am Zach Dvory. This video slide presentation, by the University of Utah, discussed the technical objectives of developing a real-time decision-making platform to enhance seismic monitoring and risk management during stimulation activities. This presentation was featured in the Utah FORGE R&D Annual Workshop on August 15, 2024.

This submission is a technical report for the Probabilistic Estimation of Seismic Response Using Physics Informed Recurrent Neural Networks project. The report describes the process of extracting events from the borehole seismic sensors. To be effective once deployed, the process must be done in real-time. A summary of the methodology is as follows: bandpass filter, shift (via cross-correlation) and stack signals, envelope function, peak detection, transfer function from amplitude to magnitude, creation of magnitude-frequency distribution, and finally, extract MFD "a" and "b" parameters. The datasets used in this work are linked below and include the raw waveform data and the seismic event catalog used for magnitude calibration, also hosted on the GDR.

This report describes the development of a preliminary 3D seismic velocity model at the Utah FORGE site and first results from estimating seismic resolution in the generated fracture volume during Stage 3 of the April 2022 stimulation. A preliminary 3D velocity model for the larger FORGE area was developed using RMS velocities of the seismic reflection survey and seismic velocity logs from borehole measurements as an input model. To improve the accuracy of the model in the shallow subsurface, travel times phase arrivals of the direct propagating P-waves were determined from the seismic reflection data, using PhaseNet, a deep-neural-network-based seismic arrival time picking method. The travel times were subsequently inverted using the input velocity model. The results showed that the input velocity model needs improvement as the resulting model appears too fast in the easter region of the FORGE area. During the next phase of this work, we will update the input velocity model and generate P-wave arrival times for additional seismic source locations, to improve the horizontal resolution in the sedimentary layer and to obtain a model that better matches the sedimentary layer and the travel time observations.

This catalog contains microseismic event locations recorded during the April 2024 stimulation at the Utah FORGE site. Events were detected and located using a DAS-specific source imaging workflow that avoids conventional phase picking. Instead, STA/LTA-transformed DAS and downhole geophone traces were back-propagated through a calibrated 1D velocity model to generate a coherency field across a 3D subsurface grid. Events were identified at locations where P- and S-wave back-projections constructively aligned, and associated uncertainties were estimated from the spatial coherency distribution.

This report describes the development of a preliminary 3D seismic velocity model at the Utah FORGE site and first results from estimating seismic resolution in the generated fracture volume during Stage 3 of the April 2022 stimulation. A preliminary 3D velocity model for the larger FORGE area was developed using RMS velocities of the seismic reflection survey and seismic velocity logs from borehole measurements as an input model. To improve the accuracy of the model in the shallow subsurface, travel times phase arrivals of the direct propagating P-waves were determined from the seismic reflection data, using PhaseNet, a deep-neural-network-based seismic arrival time picking method. The travel times were subsequently inverted using the input velocity model. The results showed that the input velocity model needs improvement as the resulting model appears too fast in the easter region of the FORGE area. During the next phase of this work, we will update the input velocity model and generate P-wave arrival times for additional seismic source locations, to improve the horizontal resolution in the sedimentary layer and to obtain a model that better matches the sedimentary layer and the travel time observations.

This is a presentation on the Cutting Edge Application of Machine Learning, Geomechanics, and Seismology for Real-Time Decision Making Tools During Stimulation by the University of Utah, presented by Dr. No'am Zach Dvory. This video slide presentation, by the University of Utah, discussed the technical objectives of developing a real-time decision-making platform to enhance seismic monitoring and risk management during stimulation activities. This presentation was featured at the Utah FORGE R&D Annual Workshop on September 9, 2025. The workshop offered a valuable opportunity to review the progress of Research and Development projects funded under Solicitation 2022-2, which aim to improve our understanding of the key factors influencing Enhanced Geothermal System (EGS) reservoir and resource development.

This report reviews the training of machine learning algorithms to laboratory triaxial ultrasonic velocity data for Utah FORGE Well 16A(78)-32. Three machine learning (ML) predictive models were developed for the prediction of vertical and two orthogonally oriented horizontal stresses in the well. The ML models were trained using laboratory-based triaxial ultrasonic wave velocity (labTUV) data wherein wave velocities were measured with various combinations of true triaxial applied stress. The ultrasonic velocities data include compressional, fast shear, and slow shear velocities in each of three directions for a total of nine velocities for each stress combination. However, because the ultimate goal is to deploy the trained model for interpretation of field sonic log data where only the vertically propagating waves are measured, the work here focuses on just the wave velocities with vertical (z-direction) propagation. Also, because vertical (overburden) is often well constrained, one approach explored here is to take the vertical stress also as known and train the model to predict the two horizontal stresses. This work was done as part of Utah FORGE project 2439: A Multi-Component Approach to Characterizing In-Situ Stress at the U.S. DOE FORGE EGS Site: Laboratory, Modeling and Field Measurement.