Machine Learning and Data Science applications in geothermal exploration: examples from the Upper Rhine Graben

  • Datum:

    23/01/2025

  • Referent:

    Thilo Wrona & Johannes Wiest (Deutsche ErdWärme)
     

  • Zeit:

    16:00 Uhr

  • Geothermal energy exploration, a sustainable and renewable energy resource, is critical for reducing reliance on fossil fuels and mitigating climate change. Geothermal exploration and development require high-cost and high-risk decisions based on very complex and highly uncertain data. Recent advances in Machine Learning and Data Science offer a wide range of techniques that can help us tackle these challenges. Here we present two examples of how these techniques can help us make better decisions when exploring geothermal energy.

    Our first example shows how Data Science can help us in the optimal selection of geothermal project sites. For optimal selection, we first need to rate potential project sites (i.e. prospects) against each other. For this purpose, we propose a data-driven rating scheme, which aims to capture the primary factors determining the success of geothermal projects in the Upper Rhine Graben (URG): (1) resource, (2) risk and (3) cost. Each of these factors depends on second and third level parameters (e.g. risk - exploration risk - well control), which are rooted in measurable physical parameters (e.g. distance to the nearest well). This hierarchical scheme enables the estimation of key factors for each prospect, allowing comparisons with existing geothermal projects and facilitating informed selection of future project sites.

    Our second example employs Machine Learning to help us identify tectonic faults that we can target in the reservoir. Machine learning algorithms are increasingly employed to process and analyze vast amounts of geophysical data. For instance, supervised deep learning, a technique where models are trained to recognize a specific part of the data based on a large number of labelled training examples, has proven very successful in many areas of Earth Science, such as seismic interpretation, seismology and remote sensing. Here we show how to apply this technique to detect tectonic faults in 3D seismic data robustly and quickly, a critical task for precise well-planning in the fractured reservoirs of the URG.

    These examples illustrate the potential of Machine Learning and Data Science in overcoming challenges in prospect and target selection during geothermal exploration.