The U.S. Department of Energy (DOE) announced today the selection of four projects to receive up to $3.5 million to apply machine learning techniques to geothermal exploration and production datasets. This work constitutes the Phase 2 portion of research and development (R&D) conducted under the DOE Geothermal Technologies Office’s (GTO) FY 2018 Machine Learning for Geothermal Energy funding opportunity.
These four projects, led by Los Alamos National Laboratory (LANL), Colorado School of Mines (CSM), University of Houston, and Penn State University, will build on previous work on machine learning algorithms and large geothermal datasets. Machine learning – the use of advanced algorithms to identify patterns in and make inferences from data – could assist in finding and developing new geothermal resources. If applied successfully, machine learning could lead to higher success rates in exploratory drilling, greater efficiency in plant operations, and ultimately lower costs for geothermal energy operators.
Vast potential exists for geothermal in the United States, but only 3.7 gigawatts electric (GWe) of energy capacity are currently installed. GTO’s 2019 GeoVision study concludes that with technology improvements, including the exploration techniques funded through this initiative, geothermal power generation could increase 26-fold from today, representing 60 GWe by 2050.
“Machine learning and artificial intelligence can provide impactful insights on large, complicated datasets, —like those used to analyze geothermal energy,” said Acting Assistant Secretary for Energy Efficiency and Renewable Energy Kelly Speakes-Backman. “As evidenced by the progress made by these four teams over the past eighteen months, geothermal developers will soon be the newest researchers able to use this cutting-edge technology in their chosen field.”
During Phase 1, LANL researchers developed GeoThermalCloud, a flexible, open-source, cloud-based machine learning framework designed to 1) discover new hidden geothermal signatures within existing large datasets, and 2) fuse big data and multi-physics models to develop data acquisition strategies. Phase 2 will expand GeoThermalCloud to accommodate a new range of datasets, including public, proprietary, satellite, airborne survey, and seismic data.
The CSM team will expand on its Phase 1 research by developing an explainable deep learning model (DLM) for the Coso Geothermal Area in California using surface and subsurface data sets. This will aid in detecting potential geothermal exploration sites from hyperspectral images to reduce the uncertainties associated with geothermal exploration, as well as provide a more in-depth understanding of the relation between surface and subsurface indicators of geothermal sources.
Houston’s Phase 2 research will focus on detecting and characterizing fracture zones, which is of great interest to operators of existing geothermal fields in western Nevada, with future application potential across various regions of geothermal resource activity. Models of faults and fractures captured and imaged in Phase 2 can support geothermal operators in their efforts to probe fluid flow pathways. This can help provide answers to key questions such as where to drill, permeability factors, and how much a resource can produce.
During Phase 1, Penn State researchers developed machine learning methods, using both field data and lab data, to locate and predict lab-scale seismicity, predict fluid injection characteristics, and model the evolution of reservoir permeability. Phase 2 work will expand on this by further exploring linkage between seismicity and permeability, improving machine learning models for enhanced geothermal systems (EGS) permeability and production, and building a flexible, baseline model for earthquake prediction and subsurface stress monitoring.
To stay up to date with GTO news and updates, sign up for the office’s monthly newsletter, The Drill Down.