Spatiotemporal Anomaly Detection Applied to Flow Measurement Points in Natural Gas Production Plants

Abstract

In an oil production unit, the volume of production is measured by Measurement Points distributed throughout the process plant. The distribution of these points is designed for the operational monitoring of the plant, and ensure that all produced fluid is measured. In 2000, ANP and INMETRO published the Technical Regulation of Measurement (TRM), establishing requirements for measurement systems in the production units. After TRM, ensurance of the measured value of oil and gas became not only legal but also operational issue. To facilitate the monitoring of problems in Flow Measurement Points (FMPs), this work proposes a method based on Dynamic Bayesian Networks for Detection of Anomalies in values reported by FMPs. The approach explored the relationship among the volumes reported by FMPs at an instant of time, and the temporal relationship of the values of a Measurement Point. Two experiments were carried out with the proposed method using real data from a production plant. The first one aimed at evaluating the impact of variating parameters of the method on the predicted values, reported by a FMP. The second experiment aimed at verifing the effectiveness of the modeling in detecting anomalies, using the parametrization that obtained the best performance in the first experiment. The approach was promising and was able to identify most of the anomalies present in the used data set.