Federated Learning for Geothermal Energy: A Decentralized Solution for Data Sharing and Sector Learning
conference paper
Geothermal energy faces challenges in design and operation due to subsurface uncertainties, high drilling costs, and complex maintenance requirements. Data and model sharing can enhance decision making processes in the geothermal sector by improving predictive maintenance, equipment and production monitoring during the operation to explore and design new geothermal installations by enabling advanced analytics and collaborative insights. However, concerns over data privacy and proprietary information hinder widespread adoption. This paper explores federated learning (FL) as a decentralized machine learning approach that facilitates secure collaboration across geothermal sites without sharing raw data. By aggregating model updates, FL improves anomaly detection, downtime reduction, and cost efficiency while preserving confidentiality. A case study demonstrates its effectiveness in predictive maintenance. The results showed that the FL models have a significantly higher accuracy and generalization compared to individual model of each stakeholder. The added value of FL models is more dominant for the parties with limited or small data. Findings highlight FL's potential to foster collaboration among operators, manufacturers, and researchers, enabling privacy-preserving, data-driven optimization in geothermal energy.
Topics
TNO Identifier
1015704
Publisher
SPE International
Source title
This paper was prepared for presentation at the SPE Europe Energy Conference and Exhibition held in Vienna, Austria, 10 - 12 June 2025
Pages
1-13
Files
To receive the publication files, please send an e-mail request to TNO Repository.