| Title | Forecasting Induced Seismicity Using Temporal Fusion Transformer: A Case Study in the Utah FORGE and Geysers Geothermal Fields |
|---|---|
| Authors | Nori NAKATA, Zhengfa BI |
| Year | 2025 |
| Conference | Stanford Geothermal Workshop |
| Keywords | induced seismicity, machine learning, forecasting |
| Abstract | Induced seismicity poses critical challenges for sustainable geothermal energy production due to seismic hazard as well as information for reservoir characterization. This study leverages the Temporal Fusion Transformer (TFT), an advanced deep learning model capable of capturing complex temporal patterns and uncertainties, to forecast induced seismicity rates in geothermal fields. We use the Utah FORGE and Geysers geothermal fields as examples. For the Geysers, we utilize a comprehensive dataset spanning 2002 to 2023, incorporating both historical seismicity rates and well injection rates as key inputs. The TFT model is trained to predict future seismicity rates with high precision by effectively learning the temporal dynamics and relationships between injection activities and seismic responses. Our results demonstrate that TFT outperforms traditional time series forecasting models, providing enhanced predictive accuracy and robust uncertainty estimates. This work advances the understanding of induced seismicity mechanisms and offers a novel tool for proactive seismic risk management in geothermal and other subsurface energy systems. |