2020 Oil and Gas Machine Learning Symposium

The Advertas/Geophysical Insights-backed online event hears Paradise use cases from RockServ and Idemitsu Norge. Southwest research Institute’s SLED, smart leak detection. AgileDD’s open source Tabio toolkit for data retrieval from scanned documents. Yokogawa’s ‘intelligent’ GOSP gas oil separation plant. IBM’s Production Optimum Advisor.

Ali Bakr (RockServ) presented a multi-disciplinary workflow for unsupervised machine learning based interpretation of a 3D seismic survey across the Shell-operated Sr Field, offshore Nile Delta. Conventional mapped with Hampson Russel spectral decomposition was followed by ML using Geophysical Insights’ Paradise software. Attributes were ranked with principal component analysis and classified using Paradise’s self-organizing map. The approach confirmed the main targets and located a ‘significant other channel’ that was undetected by conventional spectral decomposition.

Sharareh Manouchehri (Idemitsu Norge) also presented a Paradise case history on the Norwegian North Sea Wisting field. Wisting is a very shallow target that has been surveyed with an ‘ultra high resolution’ PCable survey. This uses short-offsets only, so AVO was not an option. Instead, Paradise was used to perform PCA/SOM on the 1 ms data. The result was a ‘clear improvement in reservoir characterization as compared to traditional quantitative interpretation’.

Heath Spidle from SWRI, the San Antonio based Southwest Research Institute has developed ‘SLED’, SWRI’s smart leak detection system. SLED provides ‘automated, unmanned detection and quantification of fugitive methane emissions. SLED provides an early indication of an unexpected emission. Drone-based sensor data requires special attention as the motion and viewpoint of the camera defy static pattern recognition algorithms. SWRI’s ‘powerful deep learning algorithms’ offer low false positives and pinpoint ‘otherwise invisible’ emissions. 96% precision and 2% false positive rates are reported. Tools of the trade include the FLIR MWIR infra red camera and the NVIDIA Tegra embedded/mobile GPU. SLED/M is the methane detection variant. Another system, SLED/C, detects crude oil leaks. More from SWRI.

Amit Juneja (Agile Data Decisions) presented ‘Tabio’, an open-source toolkit for detecting and extracting tabular data from scanned documents. Tabio uses machine learning to locate tables and figure text alignment, spacing and numbers and letters. Tabio is released as open source under the MIT license ‘for the benefit of the data management community’. Tabio development was sponsored by Total, Technip, Saipem, Schlumberger, Subsea7 and IFPen.

Mustafa Al-Naser (Yokogawa and King Fahd University of Petroleum and Minerals (KFUPM)) presented an ‘intelligent approach to GOSP (gas-oil separation) and enhanced oil recovery’. GOSP plants are traditionally operated at fixed conditions, ignoring ambient temperature variations, leading to lower recovery. The project set out to optimize setpoints at the high and low pressure production traps. A GOSP dynamic simulator was built by Yokogawa’s OmegaLand unit to investigate optimal settings for different ambient temperatures. ‘Artificial intelligence techniques’ determine the optimum pressure required to maximize production.

Crystal Lui (IBM Canada) presented the Prediction-Optimization Framework, an AI based tool developed for a major Canadian oil sands operator (Lui was previously with Suncor). The solution uses a system of systems approach that spans data silos across the Syncrude process flagging potential process upsets before they happen. Scenario generation integrates mass balance and machine learning to generate production plans based on dynamic events and operational objectives.

More from the Oil & Gas Machine Learning Symposium.

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.