Energy Conference Network’s Oil & Gas machine learning conference took place in Houston earlier this year and heard from AI/ML aficionados in various contexts. A post show poll of attendees indicated current areas of interest in this community. Most showed interest in predictive maintenance, with less interest in natural language processing. With regard to obstacles to AI/ML deployment, the main cause was a ‘lack of a clear ROI/business case’. Having said that, attendees considered themselves to be in the vanguard of innovation and believe AI and automation to be the ‘main accelerator for IoT and digitization’.
Brian Emmerson from Pason Systems’ unit Verdazo Analytics presented on the use of machine learning in a geosciences context. A joint study between Verdazo and GLJ Petroleum Consultants investigated the performance of hydraulic fracturing operations at the Spirit River formation of Western Canada. Pason’s analytics platform, which is ‘used by over 120 companies in North America’, was applied to a large data set of production, fracking parameters and geological data. ML tools of choice included ICE and PDP, aka individual conditional expectation and partial dependence plots. The technique investigates a multi variate data space of formation tops, porosity from cores, pressure data, along with copious information completions and frac technology. The ICE/PDP addresses AI model ‘interpretability’ and provides an understandable entry point to the big data analysi, finding, for instance, that for the same proppant intensity, a tighter stage spacing yields a greater impact i.e. more gas. Emmerson concluded ‘choose your model parameters wisely, more is not necessarily better’. It is also important to include economics and the dollar value of a feature. The ICE/PDE approach shows how and why well performance varies. More on the Spirit River study from the Verdazo website.
Harshit Sharma from Lux Research described the ‘host of startups’ now offering solutions to three key downstream oil and gas challenges viz: emission reduction, feedstock optimization, and margin optimization. In the emissions space, Lux has spotlighted two startups offering remote GHG/methane monitoring. GHGSat offers its ‘Claire’ microsatellites with on-board imaging sensors that create spectral footprints of different trace gases. GHGSat is backed by Boeing and Schlumberger and was trialled last year in the Permian Basin. Another of Lux’ spotlights is Satelytics, a provider of analytics for aerial imagery data. The Satelytics platform has been used to detect methane and liquid hydrocarbon leaks in the Eagle Ford basin, identifying leaks as low as 10 ppm, without supervision. Backers include BP and Phillips66. In the robotics space, Lux sees Inuktun’s robotic pipeline crawler vehicles as promising. Inuktun’s robots perform visual inspections in confined and hazardous environments like inside pipes using amphibious pan, tilt, and zoom cameras, operating on a tether cable up to 7,000 ft long. The system has been used on BP’s BP Thunder Horse GoM platform. BP and Dow/DuPont are backers. On the AI/software front, Maana and Seeq got a plug from Lux. Sharma concluded that the downstream sector will need to act fast with impending fuel regulations and that automation and robotics will be part of the solution.
Kelly Rose described work performed at the US National Energy Technology Laboratory (NETL) investigating the likelihood of fluid and/or gas migration through the subsurface. SIMPA, Spatially integrated multivariate probabilistic assessment is a science and data-driven, fuzzy logic approach that is said to improve prediction of subsurface leakage and flow from fractures, wellbores and other pathways. SIMPA computes the magnitude and extent of natural and anthropogenic subsurface pathways and can help with site selection, risk analysis and production optimization. Rose went on to present the NETL’s work on subsurface trend analysis (STA) that seeks to constrain estimates of subsurface property values with a combination of geologic knowledge and spatio-temporal statistical methods. STA has analyzed data from 53,000 wells, some 500,000 US offshore datasets and 30 TB of data, to define offshore provinces with a common geological history. The work is to inform subsurface pressure prediction for oil and gas exploration and carbon dioxide sequestration. More from NETL. NETL is now working to incorporate more ML into its the spatial analysis of subsurface properties and on a natural language processing NLP function which collects information from relevant publications. SIMPA V.1 is available for download from the NETL.
Nathan Yeager (Machina Automation) described trends in intelligent (office) automation, notably ‘robotic process automation’ that ‘emulates human interactions with a digital system in the execution of a business process’. Yeager warned that few companies are currently deploying RPA but that the RPA ecosystem is evolving rapidly. Companies to watch include UiPath, Enate and ABBYY. Basic RPA can be augmented with machine learning with tools such as Skymind.
Simon Mantle presented BHGE’s use of machine learning and AI to solve complex problems in both upstream and downstream oil and gas. BHGE is working on digital twin for drilling optimization using deep learning. The idea is to leverage drilling log data to identify formations, combining historical data in the area with logging while drilling (LWD) data on gamma ray logs, neutron density and vibration. Working from training data, formations are identified using and unsupervised clustering algorithm. This data is used to determination the optimal drilling rate of penetration using a ‘global optimization algorithm’. A similar digital twin approach is used to model offshore platforms or refineries as ‘systems of systems’ in a network/graph approach. All is rolled up into BHGE’s AI Pipeline for model building. The Pipeline leverages TensorFlow, CubeFlow, Nvidia NGC containers and GCP Elastic Compute. More from BHGE.
More from Energy Conference Network.
© Oil IT Journal - all rights reserved.