There have already been several oil price crises, which confronted the industry with major challenges and hurdles. With the COVID-19 pandemic, the industry is now experiencing an unprecedented situation: On the one hand, the so-called supply shock causes a sudden reduction or increase in the range of goods or services; on the other hand, there is a sudden increase (or decrease), temporary demand for goods or services (demand shock) – this creates a massive imbalance. Oil prices were already falling before COVID-19, and this trend was fueled in particular by the global commitment to making a transition to renewable and sustainable energy sources. According to expert estimates, the oil demand should peak in 2025 – and not, as previously assumed, in 2040.
European oil companies such as BP, Shell, Total, Equinor, and Eni are ahead of the curve when shifting towards renewable and alternative energies. However, change brings numerous challenges; for example, processes and cash flows are becoming more complex – and this is when the need for investment is very high.
For this reason, the oil and gas industry has to rely on innovation and digital transformation: Reducing production, processing, and transport costs through the transformation of the value chain and operations is the right way to position itself for the future – this is where artificial intelligence comes in (AI) plays a special role.
The Data At A Glance
Traditionally, oil and gas companies are geared towards efficiency and cost optimization. As a result, they often work in silos without a single view of data and applications. This structure makes addressing larger cross-functional machine learning (ML) problems difficult. So far, the use of AI in the industry has mostly been limited to localized, punctual solutions that have no major influence on other applications. Currently, only around 29 percent of energy companies think their AI implementations are working satisfactorily.
An example: A common ML use case is improving drilling efficiency and reducing non-productive time (NPT). The crux of the matter: Most algorithms only deal with very specific aspects – for example, the torque on the mast motor with stuck pipes – instead of taking a comprehensive approach that takes data from different drilling rigs into account.
Cross-Functional Approach And Data Quality
While oil and gas companies have huge amounts of data, much of it is in documents, reports, and scanned assets. Therefore, cross-referencing, duplicating, transforming, and merging cross-functional data sets is still a major challenge. Unattended, large language models such as BERT and GPT3 can make sensible use of unstructured organizational data with the help of ML models and provide a remedy here.
Oil and gas companies are increasingly using integrated data lakes managed by a cross-functional governance team, making cross-domain data accessible. This includes unstructured, transactional, real-time, IoT, or pyrotechnical data. New standards such as Open Subsurface Data Universe (OSDU) also support companies in consolidating data processes. Companies can already make significant progress by forecasting the relevant operating parameters for oil production – this is made possible by integrating operational data into ML-based analyzes.
Machine Learning Models Merge With Physics And Engineering
Oil and gas operations are critical to safety: For this reason, all decisions based on ML models should be explainable and comply with physics and engineering principles. For AI-based operations to enable decision-making, the coordination of ML models with physics-based simulation models is crucial. This is accompanied by the increased use of a different programming model that expands data-supported AI with physics, for example.
“Physics informed Neural Networks” – This is a deep learning framework that enables the synergetic combination of a mathematical model and data specifically applicable to problems with nonlinear partial differential equations. (Raissi, Perdikaris, & Karniadakis, 2019)
AI Feynman – A recursive multidimensional symbolic regression algorithm that combines neural network adaptation with physically inspired techniques. This can help find a symbolic expression that matches the data trained on the neural networks. (Udrescu & Tegmark, 2020)
These technologies are still in the middle of development. But it is already visible that the trend is developing towards the explainability of the ML models used in decision-making. This also provides insights to minimize unplanned downtime, limit bottlenecks in processes/workflows, and reduce security incidents.
The Look Into The Crystal Ball
European oil and gas companies are on a delicate path in making the long-term transition from fossil fuels to an uncertain, renewable-dominated future. The COVID-19 pandemic has created its own set of challenges. As the world moves towards a new normal with lower oil demand, more alliances will emerge in the future – at the same time, the pressure to increase process efficiency is growing.
Cross-functional silos will merge in the future, and the focus will shift towards integrated data and governance. For example, drilling efficiency improves significantly through so-called integrated reservoir modeling and geological and geophysical interpretation models. These models are driven by real-time data from drilling and production wells and past drilling reports. In addition, there will be new engineering workflows that are also controlled by reservoir modeling and geotechnical influences.
Equally likely is the introduction of targeted automation in wells, in which the system dynamically calibrates and plans the logistics, including personnel, equipment, and consumables.
The role of AI in the oil and gas industry will certainly be in the foreground in the future – however, the industry expects more scope and functionality from the technology. The ability to integrate physical and engineering technologies with data-driven analytics, ML, and automation is groundbreaking for oil and gas companies.