A Framework Based Isolation Forest for Detecting Anomalies in Natural Gas Production

Abstract

AbstractNatural gas is one of the main fossil fuels, and it is widely used in residential and industrial applications. The demand for natural gas is constantly increasing. However, due to the complex and diverse production environment for gas production, abnormal events that occur during the production of natural gas wells will reduce the gas production of gas wells with sufficient gas reservoirs. At present, detecting abnormal event in gas production mainly relies on engineers according to their own experience. This method is unreliable and requires a lot of manpower. In this paper, the first unsupervised framework for detecting anomalies in natural gas production is proposed. In this framework, a novel data convention method using a time window is proposed to enable the capture of the contextual anomaly. Besides, a low time-complexity and a small memory-requirement method called Isolation Forest is used to build a detector. Moreover, the maximum information coefficient (MIC) based feature selection mechanism reduces the high dimension caused by data convention in order to solve the increasing complexity of natural gas data sets. We apply our framework to several real natural gas well production data set labeled manually. Observations show that this framework increases the accuracy of the detection in the actual gas well production.KeywordsNatural gasAnomaly detectionIsolation forest