Predicting Leaks in Natural Gas Distribution Networks Using Generalized Linear Models

摘要

Leakages in gas pipelines are associated with significant safety, economic, and environmental concerns. In 2014, there were 113 gas distribution pipeline incidents across the United States. The annual loss rate from all downstream components of the natural gas (NG) system in the Boston region, MA, was estimated to be around $90 million worth of NG fuel in 2015. Environmental concerns further support this effort given that methane (the major component of NG) is a potent greenhouse gas. To address these issues, numerous research studies have developed sensing technologies and hydraulic simulations for detecting leaks in pipeline networks. However, most of these methods lack integration of the detected leaks and infrastructure properties to provide feedback and support risk-informed decisions about a given pipeline network. Through analysis of a citywide pipeline leakage and maintenance data set for a mid-size city in the United States, authors have developed a statistical model relating pipeline infrastructure properties and detected leaks on segments over multiple periods of surveys. A calibrated generalized linear model provides the expected number of leaks for pipeline segments, based on factors such as pipe age, material, and diameter. This model reveals information about (1) influential network characteristics contributing to leakages, and (2) hot-spot leak regions. The identified influential factors and hotspot leak regions can utilize decisions on optimal sensor placement or leak surveying and hence improve maintenance activities.