Nominated for The Prize
This project has been nominated for the Exploration Innovation Prize 2023.
Read more about the prize and be sure to vote for your favourite nominee by May 10th.
Over the last decade, the oil and gas industry has been scrambling to translate the progress within machine learning into tangible and valuable new ways of working with subsurface data. This effort has led to countless new proof of concepts, tools, and new features in existing software.
Moreover, the large operators have all developed innovation labs in the hopes of harnessing their internal capabilities to achieve competitive advantage in strategic important applications. However, the oil and gas industry has struggled to bridge the gap between R&D and scalable daily use for internally developed machine learning capabilities.
Aker BP has recently made a significant breakthrough.
They have developed an innovative and game-changing way to close the gap between the development of machine learning models and their efficient use by domain specialists in core exploration and reservoir development workflows.
The company has used an approach to machine learning (ML) similar to software development. By utilizing the implemented connection from the model host environment into legacy subsurface software tools through Python plug-ins, they can enable ML models into everyday use. The main goal has been to allow specialists to utilize machine learning models efficiently when delivering core tasks.
Aker BP’s technical innovation and organizational transformation have been driven by a combination of three principles:
First, approach machine learning operations (MLOps) similarly to software development (DevOps). This involved design, model development, and operations phases, with each phase providing input in an iterative feedback loop.
MLOps implementation enables version control, continuous integration and delivery, monitoring and alerting, and model hosting and model serving.
Second, «meet the experts” at their “traditional” legacy tools such as Petrel and Tech Log to efficiently obtain results and avoid forcing the users to use new digital tools.
Aker BP’s facilitation of seamless ingestion of machine learning-generated subsurface exploration data sets such as missed pay intervals, shows, and lithology predictions into existing legacy tools, enables cross-combining machine learning data sets with traditional subsurface evaluations.
Third, empowering cultural change by bringing the development process closer to the subsurface community with the early involvement of the domain specialists. Involving subsurface domain experts includes iterative feedback loops from the end users and gives them ownership of the development process.
This has fundamentally helped Aker BP to build trust and transparency in the new dataset and models generated, as well as to enhance the scientific core of the machine learning models. This approach has also resulted in the quality and relevance of the machine learning models to geoscientists.
The integrated organizational (cultural) and technological changes at Aker BP have allowed specialists to reduce the time used on standard subsurface workflows. Using practices based on MLOps, Aker BP has provided the in-house subsurface community with fast access to the brand-new set of tools and data that complement already established scientific practices.
Thanks to the machine learning models developed in-house, geoscientists can now reconstruct missing well logs, perform lithology predictions, do shale volume calculations, get a map view of potentially missed pay intervals, or get an unbiased assessment of log quality with fewer resources than traditionally used. With the successful integration of the machine learning models into the exploration teams, they can now process vast amounts of data more efficiently, achieve better data-driven interpretations and allow the knowledge of geologists and geophysicists to be used more optimally.
Moreover, they are given the tools to generate consistent quick-look interpretations in seconds rather than months, providing significant acceleration in data-driven processes like prospect screening in exploration and missed pay evaluations. The tools are already in use in various teams in the organization and are gradually being adopted as part of standard in-house work practices.
In conclusion, Aker BP’s solution comprises the combination of the pragmatic MLOps implementation, optimal IT infrastructure as well as empowering a cultural shift in the use of the new datasets, which unlocks more value from the data in subsurface characterization workflows.
The intention has never been to replace human expertise, but rather to let the machines do what they do best – processing and consistently screening vast amounts of data, thus freeing up time for the geoscientists to concentrate on subsurface evaluations. This is an excellent example of a digital front-runner that enables how to unlock the true potential of digitalization through organizational and technological transformation.
Aker BP was awarded the Insight Award by the Norwegian Computer Society in November 2022 for implementing infrastructure for automated machine learning systems that facilitate new ways of working in exploration.