Enhanced Geothermal Systems (EGS) is an efficient way to create geothermal reservoir in hot but insufficient natural permeable rock. In 2018, the U.S. Department of Energy developed a field laboratory in south Utah, Frontier Observatory for Research in Geothermal Energy (FORGE), to develop and test new technologies for characterizing and creating sustainable EGS. FORGE has generated huge amount of geologic, geophysical, petrophysical, and geochemical data to characterize geothermal resources. From FORGE database, we build a geothermal data matrix of attributes sampled at 249 locations. This study applies an unsupervised machine learning (ML) to explore and discover dominant attributes in the data matrix. Our ML approach, NMFk, couples non-negative matrix factorization (NMF) and customized k-means clustering. This technique explicitly discovers signals and dominant attributes from the complex geothermal data. The discovered geothermal features/signals are predominantly characterized by attributes related to borehole temperature at depth. This study found five dominant signals (A, B, C, D, and E), which are dominated by different attributes. Signal A is dominated by groundwater temperature and seismic velocity; Signal B is dominated by Na+, Cl-, and pH; Signal C is dominated by Mg2+ and InSAR slip rate; Signal D is dominated by HCO_3^- and SO_4^(2-); and Signal E is dominated by Ca2+ and K+¬. Also, discovered signals are representative of five spatial regions. These distinct signals allow us to make generalized prediction of attributes that characterize new geothermal resource. We will also provide examples of the proposed method to discover, quantify, and assess hidden geothermal resource in other Playfairway types. Examples include basins in Tularosa Basin, Tohatchi Hot springs, West Texas, southwest New Mexico, and Brady site in Nevada.
Machine Learning in Geothermal Development