Applying machine learning to geothermal exploration
Sebnem Duzgun, professor and Fred Banfield Distinguished Endowed Chair of Mining Engineering at Colorado School of Mines, has been awarded funding from the U.S. Department of Energy to apply new machine learning techniques to geothermal exploration.
Specifically, Duzgun and her team plan to use machine learning techniques to analyze remote-sensing hyperspectral images, with the goal of developing a way to identify the presence of geothermal resources based on surface characteristics.
To do that, the researchers will develop a new methodology to automatically label data from hyperspectral images using existing geological, geophysical and drill hole data and use this data in developing convolutional neural networks (CNN) as deep learning models (DLM) that can predict the presence of geothermal resources based on surface characteristics with high accuracy.
“Currently, no model exists that has utilized dense subsurface data and data for surface manifestations as labels and hyperspectral data as features – this DLM will be the first of its kind in its ability to identify potential geothermal targets from hyperspectral data,” Duzgun said. “DLM necessitates a large number of labeled data. The proposed automatic data labeling model will have a transformational impact on geothermal exploration with reduced uncertainty. It is evident that the same approach can be applied to mineral exploration or other predictive modeling needs of the mining industry.”
Collaborating on the project are Ge Jin, assistant professor of geophysics; Sid Saleh, associate director of the Center for Entrepreneurship & Innovation; and Hilal Soydan, a postdoctoral fellow in the Mining Engineering Department.