Geothermal has potential to be a major renewable energy source in Hawaii, with hot spot volcanism as the primary heat source. Yet the Hawaiian Islands remain largely underexplored from a geothermal perspective. The nearly complete Play Fairway Analysis (PFA) has provided a major step forward; however, this project did not attempt to identify robust relations between different attributes in the PFA data. To discover dominant geothermal attributes and signatures, an unsupervised machine learning (ML) method is used to analyze geothermal PFA data of the Big Island, Hawaii. The dataset includes 12 geothermal attributes at 5,00 locations. The applied unsupervised ML is based on non-negative matrix factorization with customized k-means clustering, called NMFk. This technique accelerates discovery of hidden signals from the complex PFA data, which is difficult with traditional data analytics tools. Through this ML-enhanced PFA study, we have identified signatures and dominant attributes in data to define the type of geothermal system (e.g., low/moderate/high temperature resource). These dominant attributes reveal characteristics of the geothermal system while hidden signatures allow us to intelligently guide data acquisition at new locations. The discovered geothermal features/signals are primarily characterized by hydraulic conductivity, fault, Cl cation. Through ML, we will discover overall dominant attributes in the Hawaii PFA, which was not performed previously. This combination of dominant signals and attributes can be applied to identify favorable data sources to explore new hidden geothermal resources.
Machine Learning in Geothermal Development