AAPG European Region, 3rd Hydrocarbon Geothermal Cross Over Technology Workshop

Datapages, Inc.Print this page

Minimizing Geothermal Exploration Costs Using Machine Learning As A Tool To Drive Deep Geothermal Exploration


With the increasing amount of complex and huge dataset available in the geoscience industry, there is a need of techniques that are able to extract meaning from those datasets. Machine and deep learning are emerging techniques that have proven their efficacity in automatizing various tasks such as geomechanical properties characterization (Keynejad et al. 2017); automatic fault interpretation (Araya-Polo et al, 2017, Bugge et al. 2018, Guitton et al, 2017), geophysical data inversion (Araya-Polo et al, 2018, Smith et al., 2010) and lithofacies classification (Hall et al. 2016). In this work we propose two machine learning techniques that can be used to guide the geologist’s interpretation and accelerate their work, hence minimizing the costs. Firstly, we propose some machine learning approach for the automatic lithofacies classification using well logs, that have been applied to the recently drilled (2018) and successful GEo-01 geothermal exploration well in Satigny, Geneva. If successful, this data-driven approach has several benefits, among them the most useful will be a standardization, overs different wells, of the lithofacies interpretation, as well as the its quickness. In fact, theoretically, as long as the well logs are available, a quick lithofacies log could be made in real time and can be validated by an experienced geologist. Then, a clustering approach integrating 4 different seismic attributes (Energy, Semblance, Similarity and RMS statistic) is proposed to guide the seismic interpretation. This approach allows combining each attribute and grouping them in seismic facies that have the same attributes characteristics (Figure 1). From this result is pretty clear that the seismic section could be separated into two different zones: the left part shows a predominant class (yellow), associated to Mesozoic units. However, the right part of the section is less consistent, even if the yellow class seems to be in continuity with the left part.