Neural-Driven Multi-Sensor Data Fusion for Fracturing-Induced Seismicity Diagnostics: Electromagnetic, Distributed Fiber-Optic, and Tiltmeter

It is well-established scientifically that Hydraulic Fracturing (HF) can induce seismicity and should be monitored to reduce this seismicity hazard. In the past decades, numerous fracture diagnostic techniques have been developed to assess HF and failed to do so because of nonlinear fracture growths, massive scale of HF operations, and the complexity associated with the inversion of signals from each imaging technique. We introduce a deep-learning based sensor-fusion model in Phase I of this study to generate a proof of concept for joint estimation of far-field stress induced by fracturing, using multi-physics HF diagnostic data. This sensor-fusion model will be developed and validated using annotated datasets of ground-truth field experiments, which was collected by our proposal collaborator (Dr. Ahmadianĺs DOE Project DE- FE0031785) at the Devine Geophysical Test Site. Based on these data, a hydro-geomechanical model will be developed and calibrated, and a synthetic training dataset will be generated using this calibrated model and fed into the sensor-fusion model to assist its training. This feasibility study will evaluate a sensor fusion model developed based on advanced deep learning techniques predicting the fracture topology and far-field stress as a proxy for fracture-induced seismicity. In Phase II, the fusion method will undergo rigorous testing, and a hosted version of the data pipeline will be deployed on the web, enabling seamless data ingestion and fusion. This tool addresses a big gap in the market for a fast seismicity alert system to fracturing operators to plan in advance based on the available field data. This tool represents a significant leap forward in the field of induced seismicity monitoring using stimulation injection program in oil and gas fields or geothermal fields, providing scientists with a powerful and intuitive tool to unlock the full potential of unconventional hydrocarbon reservoirs by de-risking stimulation practices through a fast monitoring technique, ultimately advancing the national energy security. Additionally, in Phase I, as part of the PIER plan, MetaScape will hire a graduate student as a summer intern from UT-Austin, one of the largest Hispanic-serving institutions (HSIs) in the nation. The intern will be selected from minority groups and/or marginalized populations and will receive rigorous training on ML models for complex data modeling and interpersonal skills. UT-Austin will hire and mentor a postdoctoral fellow from underrepresented communities and run an outreach visit and presentation to educate at least 25-50 ethnically diversified students on topics about subsurface fracture monitoring and entrepreneurial skills.