Optimising the drilling process for geothermal wells using legacy oil field data and machine learning

Deep geothermal heat represents a massive opportunity to provide low-carbon district heating for towns and cities. Space heating represents a large percentage of total energy use in Northern Europe; nearly 40% of all UK energy use (BEIS, 2022) is for heating, predominantly from natural gas. Global pressures on the international gas market and the urgent need to decarbonise the heating system to deliver NetZero highlight the need for identifying renewable heat sources to replace gas.

However, finding reliable high temperatures requires drilling to several-kilometres depth. Achieving sustainable heat supply, without depletion, means that wells must intersect deep permeable strata which are impossible to detect from the surface. Well prognosis is therefore heavily reliant on data from legacy drilling. Drilling is always an expensive process and any operational issues can impose significant additional costs, as rigs capable of drilling such boreholes have rental rates of many €1000s per day. Even when the drilling is completed, financial returns are slow and reaching profit takes years. Therefore, reassuring investors requires de-risking such projects through mitigating avoidable additional costs.

Digital data from wells penetrating many kilometres are needed for understanding subsurface processes. Only small numbers of deep geothermal wells have been drilled, so the best alternatives are legacy hydrocarbon exploration boreholes; these are good analogies for geothermal wells as they rely on permeability at depth. Such legacy hydrocarbon data are increasingly openly available through National Data Repositories (NDR) and/or Geological Survey Organisations. 

The EU Horizon programme funded OptiDrill project (101006964) is using legacy well data to optimise the drilling process, by linking drilling parameters with petrophysical data to understand the constraints upon the drilling processes. This will allow causes of interruptions to drilling and unnecessary down-time to be assessed and hopefully eliminated.

NDR archives have been trawled for modern drilling and logging data that admits optimal analysis. An Isolation Forest machine-learning algorithm was used to analyse Measurement-While-Drilling derived Rate-of-Penetration data and geophysical log data, identifying zones of anomalous responses quickly and without supervision. Examination of newly available daily drilling reports (DDR) data, from the NDR, allows these anomalous responses to be associated with breaks in drilling operations and their causes to be understood. This allows both refinement of the anomaly-detection algorithm for the identification of drilling problems, and differentiation between problems caused by drilling or geological issues and those caused by operational and logistical difficulties (e.g. procurement delays). Where drilling issues are identified these can be used to develop remediation strategies for future wells drilled in similar conditions, through revised drilling programmes and optimised well designs that minimise avoidable drilling operations such as unnecessary round trips etc.