← Back to Sessions

Developing a Machine Learning Model to Pick Phase Arrivals on Das Data at the Forge Site

Description: 

The UTAH FORGE site is a field laboratory for enhanced geothermal system with comprehensive geophysical monitoring network. The downhole DAS array recorded complete full wavefield of microseismicity occurred during the 2019 and 2022 stimulation, where clear P, S and S-P converted phases can be identified. The S-P converted phase occurred at the granite contract, where the seismic velocity significantly changes. We developed an interactive tool to manually pick and interpolate phase arrivals, and then analyze the frequency content and signal-to-noise ratio of different seismic phases. The phase picks with high signal-to-noise ratio are used to train a PhaseNet machine learning model using pytorch. Preliminary analysis of the predicted phase picks shows good agreement with manual picks, however, random noise was picked at channels further away from the microseismic events. Further improvement will be made to the training model. The delay time between S-P converted phase and P-phase will be used to better constrain event depth for microseismic events with nearly repeating waveforms (multiplets or repeating events) to identify spatial distributions of asperities. The current analysis is focused on the 2019 stimulation, and the machine-learning model will be applied to the 2022 stimulation to check the accuracy of predicted phase arrivals, and investigate the applicability of locating multiplets or repeating sequences during 2022 stimulation.

Session: De-risking Deep Geothermal Projects: Geophysical Monitoring and Forecast Modeling Advances [Poster]

Type: Poster

Date: 4/18/2023

Presentation Time: 08:00 AM (local time)

Presenting Author: Xiaowei Chen

Student Presenter: No

Invited Presentation: 

Authors