| Abstract |
Discovery, exploration, and development of hidden geothermal resources have many risks and challenges because of the complex and uncertain subsurface conditions. To mitigate these risks, we have developed a tool called GeoThermalCloud, which utilizes unsupervised machine learning (ML) and physics-informed machine learning (PIML) methods to process the data and guide geothermal exploration and development efficiently. The unsupervised ML automates the data analyses and interpretations by extracting hidden signatures characterizing geothermal resources/exploration/development. It also enables practitioners to identify observations that are important to represent the discovered hidden signatures. PIML adds physical constraints such as mass balance, constitutive relationships, models, and data in the ML processes to characterize hidden geothermal resources better. GeoThermalCloud capabilities include (1) analyzing large field datasets, (2) assimilating model simulations (large inputs and outputs), (3) processing sparse datasets, (4) performing transfer learning (between sites with different exploratory levels), (5) extracting hidden geothermal signatures in the field and simulation data, (6) labeling geothermal resources and processes, (7) identifying high-value data acquisition targets, and (8) guiding geothermal exploration and production by selecting optimal exploration, production, and drilling strategies. The GeoThermalCloud is an open-source tool and can be found at https://github.com/SmartTensors/GeoThermalCloud.jl. We have used GeoThermalCloud on ten geothermal datasets, including a large and sparse dataset of the Great Basin, and all of them show promising results. Most of the data and analyses are available on GitHub as well. Obtained results can be reproduced and further expanded by adding additional data. Practitioners and researchers are welcome to utilize GeoThermalCloud to solve other geothermal problems. |