Last year we reported a shift in PNEC’s focus from data management per se to a ‘broader, holistic rendering of the business-data-IT triangle.’ The 19th Pennwell/PNEC data integration conference held in Houston earlier this year consolidated the transformation, under the oversight of GeoComputing’s Joel Allard, chairman of the PNEC advisory board.
ConocoPhillips’ Richard Barclay’s keynote offered a ‘drill down into analytics,’ tracing the history of artificial intelligence from the development of neural nets, expert systems and genetic algorithms in the mid to late 20th Century. Around 1980 things stalled with the onset of the ‘AI winter*’ as computers were not up to the task. Things changed with IBM’s Deep Blue chess playing computer, the DARPA driving grand challenge, Watson, Siri and now AlphaGo. Hardware and software performance has now caught up and also, ‘all of human knowledge’ is available over the internet to train expert systems. In oil and gas there is much public domain information available that can be ‘scraped’ from websites and PDFs. Today, any job that entails analyzing data is a potential target for AI. ‘Your company is full of people doing these jobs.’ Barclay divvies up the problem set into ‘easy modeling,’ where the truth is measurable and amenable to AI and other subjective questions such as a well top. ‘Data thinking’ goes beyond traditional physics models. Companies that rely only on physical models are leaving money on the table. Hess’ work on frac fluid and proppant is the poster child for data-driven approach. In the 19th Century, the steam engine brought about the industrial revolution, putting a lot of horses out of work. Analytics will likely change some things for the better but it will also eliminate jobs. In the Q&A it emerged that AI has yet to impact the upstream. ConocoPhillips has tried IBM Watson on well data, but found that it relied on background information which was not available. Analytics teams are plagued by ‘traditional’ data management issues and spend ‘80-90% of their time on the data problem. Questioned as to whether data driven models will replace the simulator, Barclay replied that today, AI struggles with new data and how generalizable its models really are. We are in a transition but AI will overtake a lot of these physical models. It would be nice to experiment more, but in the current climate there is not a lot of R&D dollars to try wild things.
Hakan Sarbanoglu (Chevron) offered a more nuanced take on upstream information management. The challenges of data diversity, long life cycle, real time, cross silo were well known before the arrival of ‘big data’ a couple of years back. Core applications keep subject matter experts happy, but they are distributed around the place and tend to have non system of record storage. They are tied together with behind the scenes data integration and point-to-point ‘back door’ interfaces. More data is moving to the cloud. But this brings its own problems like harmonizing multiple applications before the move. ETL and other tools don’t work in the cloud with its new paradigm of APIs and managed services. Hadoop, at 10 years, is old fashioned. Today we have the data lake which is ‘a lot like the old data warehouse!’ Chevron is piloting various access methods to the data lake, leveraging a logical data warehouse architecture, a canonical taxonomy and cross-silo master data. Echoing Barclay, Sarbanolglu concluded that the current low oil price is limiting deployment of the exciting new technology. Still, now is a good time to refresh your reference architecture and leap forward when things pick up.
Kuwait Oil Co.’s Khawar Qureshey with help from Eudoxus Systems has built a field data quality and optimization system (Fdqos) around Fico’s mathematical programming technology. The system helps KOC satisfy data quality requirements and maintain peak production by optimization across wells, gathering centers and export terminal. Data from KOC’s Schlumberger Finder data base is optimized in a PostgreSQL optimization data base along with Fico optimization modeler 4.4.0. Fico’s Xpress Insight rapid application development environment is used to produce web-based end-user tools. French software house Artelys provides support to the Fdqos team.
Jay Hollingsworth provided an update on developments in Energistics’ standards line-up which Oil IT Journal readers should be familiar with. Beiting Zhu-Colas (Geosiris) presented a recent development in the standards space, a public domain tool to explore and validate data in Energistics’ Resqml V2 format. Resqml is used to exchange reservoir models between different software vendors’ tools. Resqml uses Energistics’ packaging convention (EPC) to bundle standards-based reservoir data into an XML file with bulk data stored in HDF5. The Resqml Explorer checks data against with the official schema and adds configurable business rules. Geosiris builds on the open standards theme with an implementation that leverages the Eclipse Foundation’s modeling framework and ‘Zest’ viewer to view the EPC package content as an interactive graph. Eclipse’s object constraint language is used for business rule validation.
EnergyIQ’s Steve Cooper dropped something of a standards bombshell with his proposal for new ‘data objects’ as a foundation for effective data management. Data objects are not new to the upstream, but are currently ‘hidden in applications.’ To support interoperability we need to abandon ‘point to point’ solutions. A data object should contain metadata, attributes, quality and governance information along with accepted values and ranges and an audit history. ‘We are not yet able to deliver this degree of granularity yet.’ The EnergyIQ team is working with a consortium of operators and vendors on attribute definition. Initial focus is on the well hierarchy, leveraging PPDM’s ‘what is a well’ work. The aim is for an ‘implementation-agnostic’ standard. ‘Folks like to talk about JSON, XML. We don’t want to get dragged down into the reeds on this.’ Ultimately, object definitions will be transferred to a standards organization., Cooper was joined by Matt Huber who, notwithstanding the technology ‘agnostic’ claim, described an open source environment comprising restful web services, NiFi, Kafka and NoSQL. ‘Finance and healthcare are already doing this.’ Huber showed a data exchange manager widget for transfer between Geographics and Open Works along with services for data validation and coordinate transform. The standards community reacted vigorously to this encroachment into its bailiwick. Cooper argued, ‘We have no illusions. Previous attempts have gone nowhere. Agreement is hard. But look at WIAW, there is hope.’ For Energistics all this is ‘reinventing the wheel.’ More cross-examination from Oil IT Journal ascertained that the initiative a) has no name, b) is not close to PPDM and c) that names of consortium partners are not available. Not a great start for the ‘open’ initiative?
EnergySys’ Peter Black has reviewed ten years worth of digital oilfield writings. Of the 57 papers, ‘two were quite interesting!’ His quest for ‘definition and recipes’ for digital oil met with disappointment and led him to wonder if the digital oilfield existed at all. Most papers came from suppliers or vendors and of the large oils, BP and Shell dominated. Some digital oilfield tools and techniques are not really being used. Others are ‘useful but not transformational.’ Was CERA’s DOF push a mistake? Black thinks it’s better to focus on an ‘efficient and productive oil field.’ Black disses big data as ‘a solution in search of a problem’ although the cloud does get his seal of approval as ‘most important and impactful.’ Integration is easier in the cloud as witnessed by Zapier where there are over 500 cloud-based apps that can be linked together to create your own workflows. The digital oilfield was a bad idea because it put technology before the business. In the Q&A some skepticism was expressed as to the role of the cloud as an enabler of interoperability. Already there are competing cloud platforms in the oil and gas space. ‘Big oil and gas operators are hostages to proprietary clouds. Until this problem is solved you’ll never capture and use data on your own terms.’
Noah’s Fred Kunzinger presented the results of an E&P information management maturity benchmark study conducted for Repsol Exploration. Kunzinger observed that ‘industry is strange, it partners with its worst enemies in joint ventures. Execs all think that everybody else is ahead of them.’ In fact there are ‘pockets of brilliance and terror in every company.’ All have ‘hybrid’ IM organizational models, with 1/3 reporting to IT, 1/3 to the business and 1/3 to a technical services group. Data and information management often equates to geotechs on a mostly low to mid-level career path. While one of the 11 companies studied is doing enterprise level IM planning, most are ‘tactical’ and project based. Automation is ‘not really there’ and there is a lot of ‘gimme the data and I’ll do it myself!’ Many users only trust what they control. This is ‘kind of sad,’ and one in the eye for developers of corporate repositories. Businesses do not seem to care whether or not their local information architecture ties into the enterprise! Only a couple had formal governance in place, others may do it on a case by case/project basis, depending on individuals’ influence and preference. Data ownership fared better. Kunzinger despairs of exaggerated vendor claims, but yes, some 30-70% of time is still spent accessing the data. Some have full time data hunter gatherers. Data is not shared across assets. There are pockets of what passes as ‘analytics’ but this is actually reporting. Many have big data initiatives as in ‘we have Hadoop and we are trying to figure out what to do with it!’ A lot of dabbling is going on. On the final summary report card most score around 3 out of 5. There is however a consistent desire for improvement. Changing the culture is tied to the degree of executive involvement and support. The ROI of data management is hard to sell. ‘We all need to work on this.’ A tentative plot of IM maturity and corporate return on capital shows a good correlation.
David Johnson is not a big data dabbler! Petrolink’s cloud-based infrastructure drives drillers’ efficiency and scalability in particular with a flagship deployment for a middle east NOC. Petrolink’s technology targets larger oils and NOCs with big legacy databases. The NOC selected the solution to distribute a country-wide, 7.5 terabyte real time database. The ability to offer triple geographical redundancy and replication was a key consideration. Petrolink has also introduced process-orientated data quality, security and ownership and an alignment of terminology. The client’s main aim was cost reduction which is hard when data management is considered as a ‘cost of doing business.’ The trick is to figure the cost of ‘corporate non-productive time.’ The new solution represents a move forward from the sneakernet era, enabling analytics and avoiding mis-calibrated and unused sensors.
Patricia Herrera Torres told how back in 2004 an audit found that, Shell’s geoscientists spent 50% of their time on searching and gathering data. This was down to the absence of technical data management personnel. Since then, Shell has defined new roles and positions for instance, a project data manager understands IT and knows about data management principles, subsurface data types, and possesses soft skills. While the discipline is now established, it has a long way to go before it is embedded across Shell. One reason is that it is hard to attract graduates or to dislodge staff from other disciplines.
EP Energy’s Chris Josefy suggested a renewed focus on solving the ‘small data challenge.’ Josefy takes inspiration from the work of the Small Data Group which aims for ‘timely, meaningful insights organized and packaged to be accessible, understandable and actionable for everyday tasks.’ The suspicion is that the big data movement has left people behind in the quest for some machine learning future. For EP Energy this boiled down to using the PPDM ‘what is a well’ pamphlet to align API numbers across geoscience, production and operations data sources. SAP Business Objects were then used to connect to the company’s different data sources into dashboard a.k.a. an ‘analytical workbench.’ Josefy describes the approach as ‘the somewhat unified theory of people doing the right thing.’
Jess Kozman (CLTech Consulting) introduced the reference information model (RIM) a UML-based approach to analyzing the business. Kozman has used the RIM to make a graphical map of the CEO’s mind, a holistic, big picture of the business. The methodology combines Energistics’ earlier work on a business process reference model and PPDM’s what is a well. CLTech has used this to help find the answer to questions like ‘what are our finding, development and lifting costs.’ The approach allows different stakeholder who may be looking at data in different systems to understand each other and reach agreement. In one instance this involved protracted analysis of exactly where to place the data custody division between subsurface and production. The exercise helps with workflow and data mapping and supports business decisions and change management.
More from PNEC.
* Wikipedia has it that there have been several AI winters and springs since then.
© Oil IT Journal - all rights reserved.