Transfer learning for oil and gas wells: unlocking the collective potential of production data from multiple oil fields

The soaring field of artificial intelligence (AI) is disrupting all industries across the globe, and the oil and gas industry is no exception. This project proposes a completely new approach to virtual flow metering based on machine learning algorithms that learn from production data. A virtual flow meter helps production teams estimate the production from each well, a crucial task in operations. Our underlying idea is to combine and leverage production data from different wells and assets to increase the dataset and enable cross-learning, ultimately to achieve better performance at less cost. Since data is the fuel of any successful AI application, Solution Seeker's existing database of production data from more than 1000 wells is a unique and crucial enabler for the project. The innovation unlocks the collective potential of production data gathered around the world, to deliver flow rate estimates with higher accuracy and much less effort than traditional virtual flow metering systems. Accurate flow rates have the potential to unlock more than 190 BNOK in increased oil recovery on the Norwegian Continental Shelf and 12000 BNOK globally. This is a huge opportunity to create an AI success story in oil and gas operations, and foster greater industry collaboration since such data sharing across operators through a third party creates a win-win situation with better performance for all. The primary objective of this project is to develop a prototype virtual flow meter based on cross-learning and test it in live operations together with our partners. The most critical R&D challenges and related activities of the project are: dataset standardization across assets; creating new model architectures for cross-well learning; and developing online calibration strategies to keep models up to date. Solving these challenges through the planned project activities will lead to a highly automated virtual flow meter that learns from continuous streams of data from multiple fields. The project has successfully produced research results and innovations directly applicable in practice and has significantly improved the data-driven virtual flow meter technology. The results have been documented in three research papers, of which one is already reviewed and published in a Level 2 journal. Topics addressed by these papers are model uncertainty, calibration, scaling, and monitoring. Additionally, a whitepaper has been written that documents the technology for a broader audience. The project has also yielded two technology results, the first is an improved method for the preparation and management of training data from multiple assets, and the second is an improved method for updating the data-driven virtual flow meter. These research results are directly beneficial to real-time applications of data-driven virtual flow meters.

Prosjektet har vært en stor suksess for Solution Seeker, og resultatene støtter direkte oppunder et produkt basert på datadrevne strømningsmålinger, og er med på å muliggjøre driften av et slikt produkt. Både OKEA og Lundin (nå AkerBP) bruker resultater fra prosjektet i daglig drift av deres offshore olje- og gassfelter som var med i prosjektet, og det er planlagt å ta i bruk teknologien på ytterligere felter fremover. Den datadrevne modellen, som er et nevralt nettverk, leveres nå som en sanntidsbasert kunstig intelligens-tjeneste til hele seks operatører av offshore olje- og gassfelt, og vi er i ferd med å sette den opp for ytterligere to kunder. Totalt har prosjektresultatene potensialet til å påvirke til sammen nesten 2000 brønner fordelt over Asia, Europa, Nord- og Sør-Amerika med Solution Seekers nåværende kundebase. For de olje- og gassfeltene vi leverer tjeneste til som har mulighet til å sammenligne ytelsene av våre datadrevne modeller med multifasemeter eller fysikkbaserte modeller, observerer vi at vår tjeneste leverer god ytelse. Vi ser også at vi er mye mer skalerbare i måten vi leverer på, og at vi kan levere en tjeneste som er vedlikeholdsfri sett fra kundens perspektiv. Dette oppleves som en stor forbedring ovenfor de tradisjonelle produktene på markedet, og er et stort salgsargument mot fremtidige kunder, og åpner opp for å installere strømningsmålinger på felter hvor dette ellers ville vært lite aktuelt. Et verdidrivende aspekt med denne modellen, utover det å levere sanntids strømningsestimater, er muligheten den gir til å optimere olje- og gassproduksjonen og på den måten utnytte reservoaret og utbygget kapasitet til sitt fulle potensial. Dette gjelder både dag-til-dag optimering av gjennomstrømningen igjennom olje- og gassnetverket, men også hvordan reservoaret utnyttes optimalt over hele levetiden. Modellen er også egnet for andre utforskende offlline studier og analyser. Dette gjør at prosjektresultatene har potensielle ringvirkninger utover sanntidsestimering. I det videre arbeidet bygger vi nå flere applikasjoner rundt modellen for å kunne bidra i en større del av verdikjeden. Vi ser også flere muligheter for å forbedre og videreutvikle metoden, samt generalisere den til nye anvendelser. Vi har også begynt å bruke samme modelleringsteknologi mot andre bransjer, blant annet fiskeoppdrett, hvor vi har et stort prosjekt sammen med Salmon Evolution.

The soaring field of artificial intelligence (AI) is disrupting all industries across the globe, and the oil and gas industry is no exception. This project proposes a completely new approach to virtual flow metering based on machine learning algorithms that learn from production data. A virtual flow meter helps production teams estimate the production from each well, a crucial task in operations. Our underlying idea is to combine and leverage production data from different wells, assets and oil companies to increase the dataset and enable cross-learning, ultimately to achieve better performance at less cost. Since data is the fuel of any successful AI application, Solution Seeker’s existing database of production data from more than 1000 wells is a unique and crucial enabler for the project. The innovation unlocks the collective potential of production data gathered around the world, to deliver flow rate estimates with higher accuracy and much less effort than traditional virtual flow metering systems. Accurate flow rates have the potential to unlock more than 190 BNOK in increased oil recovery on the Norwegian Continental Shelf and 12 000 BNOK globally. This is a huge opportunity to create a lacking AI success story in oil and gas operations, and foster greater industry collaboration since such data sharing across operators through a third party creates a win-win situation with better performance for all. The primary objective of this project is to develop a prototype virtual flow meter based on cross-learning and test it in live operations together with our participating partners. The most critical R&D challenges that the project will face are related to: dataset standardization across assets; creating new model architectures for cross-well learning; and, developing online calibration strategies to keep models up to date. Solving these challenges will lead to a highly automated virtual flow meter that learns from continuous streams of data from multiple fields.