With a moderate- to low-oil-price environment being the new normal, improving process efficiency, thereby leading to hydrocarbon recovery at reduced costs, is becoming the need of the hour. The oil and gas industry generates vast amounts of data that, if properly leveraged, can generate insights that lead to recovering hydrocarbons with reduced costs, better safety records, lower costs associated with equipment downtime, and reduced environmental footprint. Data analytics and machine-learning techniques offer tremendous potential in leveraging the data.
An analysis of papers in OnePetro from 2014 to 2020 illustrates the steep increase in the number of machine-learning-related papers year after year. The analysis also reveals reservoir characterization, formation evaluation, and drilling as domains that have seen the highest number of papers on the application of machine-learning techniques. Reservoir characterization in particular is a field that has seen an explosion of papers on machine learning, with the use of convolutional neural networks for fault detection, seismic imaging and inversion, and the use of classical machine-learning algorithms such as random forests for lithofacies classification.
Formation evaluation is another area that has gained a lot of traction with applications such as the use of classical machine-learning techniques such as support vector regression to predict rock mechanical properties and the use of deep-learning techniques such as long short-term memory to predict synthetic logs in unconventional reservoirs.
Drilling is another domain where a tremendous amount of work has been done with papers on optimizing drilling parameters using techniques such as genetic algorithms, using automated machine-learning frameworks for bit dull grade prediction, and application of natural language processing for stuck-pipe prevention and reduction of nonproductive time.
As the application of machine learning toward solving various problems in the upstream oil and gas industry proliferates, explainable artificial intelligence or machine-learning interpretability becomes critical for data scientists and business decision-makers alike. Data scientists need the ability to explain machine-learning models to executives and stakeholders to verify hypotheses and build trust in the models. One of the three highlighted papers used Shapley additive explanations, which is a game-theory-based approach to explain machine-learning outputs, to provide a layer of interpretability to their machine-learning model for identification of identification of geomechanical facies along horizontal wells.
A cautionary note: While there is significant promise in applying these techniques, there remain many challenges in capitalizing on the data—lack of common data models in the industry, data silos, data stored in on-premises resources, slow migration of data to the cloud, legacy databases and systems, lack of digitization of older/legacy reports, well logs, and lack of standardization in data-collection methodologies across different facilities and geomarkets, to name a few.
I would like to invite readers to review the selection of papers to get an idea of various applications in the upstream oil and gas space where machine-learning methods have been leveraged. The highlighted papers cover the topics of fatigue damage of marine risers and well performance optimization and identification of frackable, brittle, and producible rock along horizontal wells using drilling data.
This Month’s Technical Papers
Machine-Learning Work Flow Identifies Brittle, Fracable, Producible Rock Using Drilling Data
Machine-Learning Techniques Assist Data-Driven Well-Performance Optimization
Digital-Twin Approach Predicts Fatigue Damage of Marine Risers
Recommended Additional Reading
SPE 201597 Improved Robustness in Long-Term Pressure-Data Analysis Using Wavelets and Deep Learning by Dante Orta Alemán, Stanford University, et al.
SPE 202379 A Network Data Analytics Approach to Assessing Reservoir Uncertainty and Identification of Characteristic Reservoir Models by Eugene Tan, the University of Western Australia, et al.
OTC 30936 Data-Driven Performance Optimization in Section Milling by Shantanu Neema, Chevron, et al.
Yagna Oruganti, SPE, is a senior data scientist with Microsoft in Houston. At Microsoft, her area of focus is artificial-intelligence and machine-learning applications for the energy industry, with a specific focus on sustainability. During the past 11 years, Oruganti has held various positions as research scientist, reservoir engineer, and data scientist. Her work experience includes 7 years at Baker Hughes, where she focused on reservoir simulations for unconventional shale plays and on machine learning for various subsurface applications. Oruganti has authored or coauthored more than 14 technical publications in the areas of reservoir engineering, carbon sequestration, and data analytics and machine learning in the oil and gas industry. She holds a bachelor’s degree in chemical engineering from the Indian Institute of Technology Madras and a master’s degree in petroleum engineering from The University of Texas at Austin. She is a member of the JPT Editorial Review Committee and serves on the SPE Data Science and Engineering Analytics Advisory Committee. Oruganti can be reached at yagna.oruganti@microsoft.com.