Understanding the subsurface is key to deliver reliable well construction and efficient production operations. Embracing digitalization in subsurface helps to improve accuracy, reduce risks, and accelerate cycle time. Crew change, availability of high-precision data sets and increased pressures on margins imply that human interpreters need to be supplemented by automation and machine learning (ML) driven insights through digital techniques.
ML solutions can accelerate interpretation to modeling to reservoir engineering workflows by optimizing first break picking in seismic processing, improving fault detection, stratigraphic interpretation in seismic and reconstructing logs and detect outliers in wellbore. Scalable machine learning requires reliable data products, but operators are fraught with data wrangling challenges across sources without lineage or context. Domain users cannot collaborate well with data scientists further impeding ML models from moving from innovation to production.
New wellbore and seismic data can be aggregated across vendors, data stores and contextualized to reliable data products by automated DataOps pipelines. Domain experts can understand these data products and collaboratively work with data scientists on an intuitive ML workbench democratizing the ML craft and providing first principal guard-rails. An MLOps pipeline can manage model versions and continuously deliver qualified ML models into elastic compute-clusters for reliable result prediction on new datasets.
Such a digital system can account for ML models drift recalibration and regional localization ensuring the solution remains operational and reliable over time. Reliable data products through DataOps pipelines feeding contextualized information to ML models deployed and operationalized using MLOps in the cloud, result in efficient and intelligent solutions that optimize subsurface processing, interpretation, and modeling workflows.
Energy companies face ever increasing challenges such as closing the demand/supply gaps, recovering cash from projects quickly while minimizing risks, mitigating emissions to meet climate change targets, and dealing with complex hybrid energy systems. Current approaches to understanding and optimizing the assets therefore soon become unscalable and insufficient. Gone are the days of easy oil and crew change impacting asset team competency; complexity of subsurface data.; volumes and newer subsurface data formats drive the need to rethink our approaches to data and the need for assisted/accelerated data-driven workflows.
