Drilling the Perfect Geothermal Well: an International Research Coordination Network for Geothermal Drilling Optimization Supported by Deep ML and Cloud Based Data Aggregation


Adam SCHULTZ, Pradeep ASHOK, Alain BONNEVILLE, Daniel BOUR, Rolando CARBONARI, Trenton CLADOUHOS, Geoffrey GARRISON, Roland HORNE, Gunnar Kaldal, Susan PETTY, Robert RALLO, Carsten SORLIE, Dang TON, Matt UDDENBURG, Eric VAN OORT, Leandra WEYDT

Key Words:

drilling optimization, database, machine learning, cost reduction


Stanford Geothermal Workshop







Paper Number:


File Size:

1798 KB

View File:


The US Department of Energy Geothermal Technologies Office supported EDGE Consortium has developed a sustainable database of complex, heterogeneous geothermal drilling data from 113 wells in various geologic settings in the US and Iceland, and an expert system using machine learning tools to guide geothermal drillers to substantially reduce the cost of geothermal wells. Best practices were developed for ingestion of heterogenous data containing missing and erroneous data. The resultant cloud-hosted FAIR data repository based on CKAN, compatible and interoperable with existing geothermal data repositories was designed, built, and stored by PNNL while assuring that data provided remains proprietary to the provider. Teams at Oregon State University, Stanford University, and the University of Texas at Austin used modern machine learning techniques to develop methods for predicting cost, ROP, bit wear, and other important parameters for variety of environments. Data models have been developed to predict: (a) Rate of Penetration (ROP), (b) Non-productive Time (NPT), (c) the occurrence of ‘Problems’, and (d) drilling costs. Data analytics and modeling approaches used/developed included: (a) Self-Organizing Maps (SOM), (b) Random Forest Classifier (RFC), (c) Natural Language Processing (NLP) of driller’s comments, (d) Probabilistic models, and (e) Process mining. For instance, the most significant variables controlling the ROP were found to be the pumping rate, depth, and the diameter of the hole. Using these and other parameters, a model with an R2=0.77 was achieved. These results were obtained analyzing wells primarily drilled in ingenious basement rock. More work is needed to determine how to best incorporate lithological complexity seen in sedimentary basins. A robust Bayesian Network cost model has been designed which is applicable to a variety of geological settings and well designs suitable for a wide range of scenarios. High level analysis using this model shows that the primary driver of cost is rig time, followed by casing, drilling mud, and cement. Work on these and related analytical tools continues. We invite geothermal developers to become EDGE expert system users, and for those interested in tailoring the system to their specific requirements, by providing additional drilling and contextual data additional training of the expert system will improve the system organically. Such tools will help developers to streamline well design and operations to reduce costs significantly. A subscrption system is being launched with a tiered structure to encourage users to become data providers.

ec2-3-230-143-213.compute-1.amazonaws.com, you have accessed 0 records today.

Press the Back button in your browser, or search again.

Copyright 2023, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.

Attend the nwxt Stanford Geothermal Workshop, click here for details.

Accessed by: ec2-3-230-143-213.compute-1.amazonaws.com (
Accessed: Friday 19th of April 2024 02:34:14 AM