In her keynote address, Maria Juul (Norwegian Petroleum Directorate) explained that the ‘digital transformation’ is now a central part of Norwegian national strategy. The transformation includes the internet of things, big data and technology, with the ‘theme of data’ running throughout the ‘data-driven’ transformation. The downturn has put the Norwegian continental shelf under pressure. But the digital transformation can counterbalance the low oil price through ‘IT, standards and automation.’ Machine learning and robotics will ‘transform the way we work.’ But ML is ‘also about people.’ Statistics Norway, the national agency has predicted that AI and robotics will see ‘one in three jobs disappear.’ For Juul, these technological advances bring opportunities for innovation and new jobs, so long as new skills and proficiencies are learned. Key among these are IT security, HSE and data quality. The future is ‘both complex and simplified.’ IT security is a priority in NPD’s 2016-2020 strategy plan.
Mark Lantz (IBM) entertained the audience with a super-geeky presentation of the future of tape. Worldwide data growth is running at 42% annually and storage is a race between disk and tape. Hitherto $/GB storage costs have been driven by disk technology but as the 2015 INSIC tape storage roadmap showed, disk drive scaling is slowing while tape is forecast to raise the storage bar at a steady 33%/year for another decade. 80% of data is inactive and can be stored on tape. Tape is energy efficient, secure with a long media life. A 2015 investigation by the Clipper Group of the total cost of ownership, over a nine-year period, of a petabyte archive growing at 55%/year found a 6.7x cost advantage of LTO over disk. Could it be that tape’s evolution will slow down too? Seemingly not. While disk is approaching the ‘superparamagnetic limit’ due to grain size, tape’s much lower areal density can continue to scale log-linearly for maybe 20 years. INSIC is forecasting 250TB cartridges by 2025. Lantz concluded that today, ‘tape is good for big data and its cost advantage over disk will continue to grow.’
Kine Johanne Aardal gave an assured presentation of her startup’s (Robotic Resource Insight R2I) work on creating business value through data integration, analytics and visualizations. The online demo shows R2I’s Microsoft PowerBI development accessing public NPD data via a compelling GUI.
Li Dawei (a.k.a David Lee) outlined PetroChina’s use of data mining algorithms in petroleum. PetroChina has 1.6 million employees and operates the largest oilfield in China with 100,000 wells, and 70 large IT systems. Moreover, PetroChina has 2 petabytes of data ‘whose deep value has not yet been realized.’ Enter data mining as a way of leveraging this huge dark dataset. In China, data mining PHds peaked in 2012. The subject is now considered ‘mature.’ All that is needed is to select the best algorithm between ‘classification, regression, clustering, estimation, prediction and association.’ Dawei has tested the approach on the C&C Reservoirs global oil and gas field database, looking for key factors that influence recovery. It turns out that ANN (artificial neural network) and SVC (support vector network) regression techniques are both ‘applicable’ with ANN best for recovery factor regression and SVC the tool of choice for classification.
Duncan Irving, now a ‘think big’ consultant chez Teradata agreed that data science has made it into the upstream. While data-driven decision making has a long history in business-at-large (Irving cited Joe Lyons’ deployment of the LEO computer at his eponymous teashops in 1951). Despite BP’s multi-million dollar investment in NASA AI spin-out Beyond Limits, in general, oil and gas has missed out on the big data movement because of culture, skillsets, platforms and, until recently, the lack of an economic driver. The culture issue is well illustrated in geophysics which turned its back on IT twenty years ago and built its own HPC infrastructure. This is now ‘quite impenetrable to generic IT.’ But this can be fixed through improved data management. Companies have a choice, either keep storing data as we done for years or ‘add a few smarts’ to data management and make it (and the data managers) more useful to the business. Enter business-focused data management and the chief data officer, an ‘emerging’ C-level role in oil and gas. There needs to be a shift from ‘custodianship’ (creating walls, avoiding change) to ‘steward’ (sharing, teaching). Stewards can mentor and support ‘citizen data management’ and ‘do better than storing everything in PowerPoint.’
David Holmes Dell/EMC also thinks that, access to data should not give a competitive advantage. In Norway, NPD has promoted open access to data for decades via Diskos. So how does IT confer a competitive edge? Perhaps it is the applications and how we use them that are transformational. But deploying applications is costly in terms of infrastructure and they are expensive to deliver and maintain. The cloud promises to change this but there is an assumption that apps will take care of persistence and availability. Unfortunately, today’s ‘legacy, pre-cloud’ data management apps (Schlumberger’s Petrel Studio, Landmark’s Open Works) just assume that such infrastructure is already there. Cloud-native means a different set of assumptions and a shift to a ‘software-defined’ environment.
Holmes advocates a ‘multi-cloud’ environment with cloud-native new apps and infrastructure-as-a-service for legacy. ‘Look at how much you spend just keeping the lights on, most IT budgets are spent on keeping things alive.’ Holmes advocates a cloud-first strategy, either on or off premises, or one of the private cloud providers, like Dell’s own VirtueStream. Some large ERP-type apps may require specialist skills to migrate, probably with an abstraction layer to separate apps from virtualized hardware. Today all E&P software vendors are racing to be cloud native. Holmes, citing Gartner, sees a return of in-house app development. ‘75% of business apps will be built not bought in 2020.’ The movement will also see the rise of open source software. It used to need a huge team to do a ‘hello world’ app. Today teams can put together their own apps and avoid vendor lock-in. While it may not be possible to write a conventional reservoir flow simulator in a couple of weeks, ‘machine learning will write you one!’ Will a ‘no physics’ methodology be OK for the SEC? Probably not? Will it be useful? Definitely. The tools of the trade, the Hadoop ecosystem, may be terrifying, but you can always use a packaged solution like the Dell-backed Cloud-Foundry, the fastest path to innovation and a virtuous circle of development. In the Q&A, Holmes admitted that the risk of competing cloud ‘platforms’ in the future was real.
New Digital Business’ Jonathan Jenkins provided a progress report on the Subsurface Applications Benchmark, a joint venture with Aupec. The SAB kicked off in 2011 and has studied the marketplace for six years, providing detailed trends for subsurface application and data tool usage. Schlumberger Petrel ‘absolutely dominates’ the subsurface, per company and per user. Since the downturn there has been growth in cheaper tools like Open dTect. Landmark’s DecisionSpace ‘is not dead’ and is starting to creep up the charts, ‘offering some competition for Petrel.’ For seismic interpretation the top three are Petrel, Kingdom and Geoprobe.
SeisWorks ‘is still there’ and Open dTect is ‘coming up strong.’ For mapping and visualization, ArcGIS and Petrosys are equal at the top spot. For static geomodelling its Petrel, RMS, Skua/Gocad with DecisionSpace making some headway. Reservoir Engineering is still dominated by Schlumberger either with Eclipse or increasingly, by Petrel RE. Landmark continues to dominate drilling. Despite the best efforts at portfolio rationalization, the number of tools in a workflow remains stable. In data management, Petrel Studio ‘does not appear to be getting huge traction,’ the Petrel reference project and OpenWorks ‘dominate’ and ‘spatial databases are no longer dominated by ESRI.’ BlueCube and Landmark Earth are now launching as cloud ready and EnergySys is ‘designed for the cloud’ and available in a ‘pay by barrel’ license. Open dTect with all bells and whistles is available for $200/day. Some majors still do in-house development. Both Total and Shell prefer to build in the face of expensive vendor tools. But their peers still buy software. Inter-tool compatibility is much better today than before.
Nina Reiersgård and Per Kåre Foss described how the new cloud-based Statoil data platform is being built. The drivers for the SDP were a) the complex and non-scalable legacy infrastructure, b) more real time data, c) the difficulty of finding stuff in the 30 petabyte archive and d) the advent of novel data sources like drones, the robot ‘snake’ and satellite imagery. 2015 saw the kick-off Statoil’s future IT project and the search for a new (additional) data platform. The idea was for connectivity to the legacy platform from a ‘one stop’ data platform in the cloud, supporting analytics and external collaboration. A bold, multi-cloud strategy means that Amazon is used for infrastructure and HPC, the data platform is on Microsoft Azure and some subsurface and future HPC on Google. The SDP spans data storage across multiple databases and API connections to SAP and other enterprise tools. Legacy systems without an API are replicated to the SDP. The cloud ingests both streaming and batch data from the IoT. Predictive maintenance is currently the main app, but Statoil is working on subsurface analytics. All data in the cloud is encrypted so it is ‘actually more secure than legacy.’ Loading processes are all automated and only seven people run the platform. The recently announced Statoil digital center of excellence was enabled by the SDP. The company is ‘recruiting heavily,’ both internally and externally.
Teradata’s Jane McConnell provided a peek into the future of subsurface data management in the ‘managed data lake.’ It is clear that a lot is happening in IT with analytics, unstructured data and hackathons. Data use is changing so data management needs to change too. A data lake groups original format data in a ‘collection of storage instances.’ These need modelling and management, ‘otherwise you will have a data swamp.’ Whereas in the old world data was loaded manually, in the new world of the data lake, data is picked up from a directory and automatically ingested into the lake through predefined pipelines. Enter Kylo* (a recent Teradata acquisition) as a data lake management platform. Kylo manages enterprise class data lakes in Hadoop and Spark. Teradata has developed a pipeline for LAS (log) data using an Apache NiFi template. Regarding seismic data McConnell observed, ‘standards are good but it would be better nice if they did not assume we were writing to 9 track tape!’ In a another nod to Norway’s regulators, McConnell also stated that, ‘the ability to access data should not confer a competitive advantage.’
Halliburton’s Ashwani Dev recalled the disruption of Amazon, Uber, AirBNB and others to ponder on the impact of all this on the upstream. Well completions ‘create 12 terabytes of data/day’ which is amenable to big data analytics. BDA is coming in right now in the form of open source tools like R and Python, but the industry is reluctant and slow to adopt. Dev observed that despite fifty years of stuck pipe research on OnePetro, ‘we are still nowhere!’ But if we manage to use all the data, leveraging the emerging technologies we can expect actionable insights. Dev’s open technology slide was very busy with maybe 50 or so ‘open source emerging technologies’ making up the big data ecosystem. One current use case is reservoir petrophysical property prediction, now ‘90% accurate.’ An advisory tool has been developed to rationalize unused functions in software by monitoring mouse clicks. ‘We (or maybe you) are paying for software that we don’t use.’ Dev wound up with a plug for Halliburton’s OpenEarth community backed by RedHat, Energistics, Total, CGG, Shell, Devon, Statoil and IHS which is set to drive the open source model à la Linux.
Magnus Svensson of Norway’s Epim joint industry body traced the tortuous history of Norway’s production reporting. Back in 2006, the first version of Epim’s daily production report was released. This has been through several evolutions, most recently the Epim reporting hub, a common data sharing platform that connects field production data to operations and the regulator. Reflecting on weaknesses in the current systems, after fourteen years of effort, Svensson wondered, ‘what went wrong?’ Some technology components (read the semantic web) did not to fulfill their early promise. Also production reporting is complex with different daily, monthly, partner, asset and yearly reports along with updates and reconciliations. Such issues are particularly relevant as we move into the new world of big data, digitalization and the data lake. Text input can be valuable but is hard to capture, especially when in ‘offshore language.’ PDF still rules, ‘this is a challenge, we cant get rid of it.’ Even a ‘trivial’ daily production report is not trivial at all!
Visit the ECIM home page here.
* Kylo looks like a GUI-driven reinvention of the Unix Shell!
© Oil IT Journal - all rights reserved.