Devon Energy’s Kathy Ball, speaking at the 2017 SAS Global Forum in Orlando, opined that the Industrial Internet of Things (IIoT) will radically change how humans, machines and infrastructure operate. While disruptive, the IIoT is to create new value streams including real time automated decisions and reactions that will ‘massively improve’ operational efficiencies. SAS, augmented with an open source software ecosystem for data science (Anaconda, R, Scala), provides a comprehensive analytical toolset. Ball showed how Devon trained a neural network to interpret downhole video logs and detect water inflow. Handling the large video dataset meant deploying Hadoop/deep learning tools ‘not present in the market.’ Pattern recognition was applied to a frame-by-frame breakdown of the video stream to spot changes in text annotation and imagery. The system automatically determined water inflow determine depths, time and orientation. Ball also touched on text analytics with deep neural networks (using snazzy D3.js data-driven documents) and on pattern recognition in microseismic data. Paper SAS1341-2017.
Predicting equipment failure is a poster child of the analytics movement. SAS Institute’s Adriaan Van Horenbeek showed how to do it using R along with SAS asset performance analytics. Vibration signals are a key indicator of asset health. Spectral analysis is used to identify failure modes, currently an offline, manual process. Using a fast Fourier transform, this can be done in real time, comparing frequency domain fingerprints with NASA’s prognostics data repository. Paper SAS527-2017.
Vipin Gupta showed how Petronas forecasts production in the age of oil and gas big data. Production forecasts based on data analytics, aka technical potential (TP) forecasts, are better able to capture the patterns created by past behavior of wells and reservoirs. TP is more than the ‘monotonous,’ straight lines of conventional decline curve analysis. The shift from DCA to TP involved standardizing definitions and a ‘systematic and focused review’ of processes that previously were not transparent. TP combines science and the statistical art, Petronas applies ARIMA and UCM to process raw time series data from Schlumberger’s Oil Field Manager. Pareto analysis also ran. Benchmark results showed that 80% of the fields had a sub 20% error in the estimate. Paper SAS0563-2017.
SAS Institute’s Shahrzad Azizzadeh presented work performed by an unnamed Middle Eastern oil products company that has used SAS OptModel to optimize its 230 million liters/day distribution network and avoid bottlenecks. To handle large numbers of potential failure scenarios, a Benders decomposition algorithm was implemented. Paper SAS0681-2017.
More presentations on the SAS Forum homepage.
© Oil IT Journal - all rights reserved.