An Isolation Forest-based system for drilling problem identification in geothermal wells

Geothermal energy systems often require drilling to multi-kilometre depths to access and extract heat. This is an expensive process, and the risk of unforeseen drilling problems can significantly affect the long-term economic sustainability of such projects. Moreover, the low availability of drilling data from geothermal development impedes efforts to improve drilling processes and mitigate this risk. Here, we demonstrate that publicly accessible legacy data from hydrocarbon exploration and development drilling can be used to develop an effective, generalisable system for the early identification of drilling problems. The system, based on the Isolation Forest algorithm, identifies problems that interrupted drilling at a rate significantly greater than random chance (detection rate 29% compared to 8.6% baseline, p =0.012), even when applied to new measurements in new geological contexts. Whilst there are significant limitations of such a system, which we discuss, our work highlights the utility of legacy data holdings in facing contemporary challenges.