Permeability in hydrothermal reservoirs is typically conceptualized as a branching network of variably discontinuous vertical and horizontal pathways. These pathways represent primary and secondary porosity and permeability associated with the stratigraphic section and fault system, respectively. Within this network, there are typically relatively few through-going flow pathways among a multitude of minor and/or dead-end pathways. The through-going pathways allow the vertical movement of fluids from depth to the near-surface, facilitating advective transport of heat. Geothermal production exploits one or more of these through-going flow paths. Still, other through-going flow paths (i.e., under-developed parts of the reservoir) might exist. Here, we present analysis aimed at identifying these under-developed parts of geothermal reservoirs as part of a DOEGTO-funded machine learning project lead by National Renewable Energy Laboratory (NREL). Focusing on the Brady geothermal system in western Nevada, we develop twelve 3D geologic proxies that may control the spatial distribution of through-going flow paths in the reservoir. By using principal component analysis (PCA), we compare the distribution of these proxies to the distribution of porosity and permeability values from an existing, robust, and history-matched reservoir model. PCA reveals which geologic proxies most closely correlate with the distribution of reservoir porosity and permeability values. At Brady, 3D fault density and 3D fault intersection/fault termination density most closely correlate with the permeability distribution in the Brady reservoir model. Spatially mapping the distribution of these geologic characteristics may help improve performance of the reservoir model. These geologic characteristics may also help identify under-developed volumes in this and other hydrothermal reservoirs.
Machine Learning in Geothermal Development