Machine learning fosters geophysical inversion for quantified geothermal parameters

Weiwu Ma, Jiamu Lu, Guiyu Zheng, Shams Forruque Ahmed, Gang Liu,
Machine learning fosters geophysical inversion for quantified geothermal parameters,
Geothermics,
Volume 133,
2025,
103477,
ISSN 0375-6505,
https://doi.org/10.1016/j.geothermics.2025.103477.
(https://www.sciencedirect.com/science/article/pii/S0375650525002287)
Abstract: The geophysical inverse method is crucial for the exploration of geothermal energy resources. However, accurately determining geothermal parameters from limited geophysical data is a challenge due to the complexity of geothermal existence conditions. We aim to propose a research framework for the inversion of quantified geothermal parameters by reviewing the recent advances in geophysical methods for geothermal resource inversion. It is found that inversions based on single geophysical data and the same geophysical properties can still lead to ambiguous interpretations. A common limitation of geophysical inversion methods is their inability to quantitatively describe geothermal resource information. Additionally, it is necessary to conduct additional testing to determine the suitability of applying machine learning to geothermal resource inversion, although machine learning can capture important features from complex geophysical data that cannot be accounted for by traditional methods. We employed an artificial neural network-based research framework for inverting quantitative geothermal parameters using thermal data based on the background of the increasing number of abandoned oil wells. Through an experimental case, the relative error of heat source temperature and fracture location of the model inverted by ANN is about 15% or less. The proposed framework will guide inversion of quantified geothermal parameters and serve as a complement to the joint interpretation of multi-geophysical inversions to obtain detailed and accurate geothermal reservoir characteristics.
Keywords: Geophysics inversion; Geothermal energy; Machine learning; Multiple parameters estimation; Experimental validation; Artificial neural network