LANL scientists enlist AI to find geothermal energy sources

Scientists at Los Alamos National Laboratory are using AI and machine learning to simplify the complex, labor-intensive search for "hot rocks" from which geothermal energy an be generated.

Energy companies drill deep underground in order to access reservoirs of water that has been superheated by underground magma. That superhot water and steam gush upward through another well into a generating station, where they drive turbines to make electricity. This form of geothermal energy produces no pollution, is renewable and sustainable (the water is recycled over and over), and compares favorably in cost to other forms of renewable energy.

There’s just one drawback – finding it.

Rather than rely on humans to ascertain the key subsurface characteristics that make for ideal geothermal prospects, scientists at Los Alamos National Laboratory aim to dramatically improve geothermal exploration through machine learning – computer programs that can process vast amounts of data, learn from it, and then automatically modify their algorithms to analyze it with increasing accuracy and efficiency.

With funding from the Department of Energy’s Geothermal Technologies Office to apply machine-learning techniques to geothermal exploration and production, the Los Alamos team has worked on determining which data are ideal for a computer to learn from, as well as developing the fundamental algorithms, or computer instructions, and statistical models that will serve as the computer’s brain.

Instead of having a team of scientists and engineers pore over huge stacks of images, maps and other data to hypothesize which sites are likely the best, these mountains of information are instead fed to a computer. New and powerful algorithms and statistical models – simplified and mathematically formalized ways to approximate reality – learn how to accurately and quickly identify new geothermal locations ideal for further exploration.

The computer’s learning never stops. As more information comes in, computer scientists feed it into the computer, which assimilates it and incorporates it into the existing data. As a result, the computer automatically improves its assessments based on new experiences, thus improving the odds when it comes to finding sources of hot dry rock that can produce sustainable geothermal energy for long periods of time.

The team applied these techniques to data collected in a study area located in southwestern New Mexico and found unique signatures – various characteristics of the geology that are critical for discovering geothermal resources. Moreover, the algorithms the team used identified an association between New Mexico’s geothermal data and different geothermal provinces, such as the Colorado Plateau and the Basin and Range. Establishing these associations enables artificial intelligence to discover unknown geothermal resources in New Mexico.

According to the Energy Information Administration, nine western states, including New Mexico, together have the geothermal resources to provide more than 20% of the country’s electricity needs. With machine learning applied to geothermal exploration as one way to unearth harder-to-discover resources, the Department of Energy anticipates a significant increase in production from geothermal reservoirs. Having a better idea of where to look is a great place to start.