期刊
ENERGY REPORTS
卷 7, 期 -, 页码 7640-7647出版社
ELSEVIER
DOI: 10.1016/j.egyr.2021.10.093
关键词
Artificial intelligence; Artificial neural network; Failure prediction; Gas pipeline; Pattern recognition; Supervised learning
In this study, gas pipeline incidents from 2002 to 2020 in the US were analyzed to predict different types of failures using Artificial Neural Networks and Support Vector Machine. The study identified that the Medium Gaussian SVM integrated with ANOVA and Holdout cross-validation performed better than other algorithms with 74.8% accuracy.
Gas pipelines are often subjected to various kinds of damages such as corrosion, welding failure, and excavation damages, due to harsh environmental conditions. The failure in gas pipelines may lead to catastrophic damages like human life loss, economic loss, etc. Predicting pipeline health is of critical importance to avoid these damages. In this study, 875 incidents are extracted from US DOT PHMSA from 2002 to 2020. For each of the incident, different parameters such as Age, NPS, Wall Thickness, Material, Operating Pressure, Location, and Area is analyzed. Two supervised learning techniques Artificial Neural Networks and Support Vector Machine are used to predict and classify different natural gas pipeline failures i.e. Corrosion, Pipeline Material, or Weld Failure and Excavation Damage by using actual pipeline incident data. One-Way ANOVA F-test is used to select the important features of the input dataset. The supervised models (Backpropagation Neural Network and SVM) are trained and tested on the input data. The performance of the models is assessed based on accurate predictions made by the trained models on the testing dataset. It is observed that Medium Gaussian SVM integrated with ANOVA (and Holdout cross-validation) performs better than other algorithms and yields 74.8% accuracy. (C) 2021 Published by Elsevier Ltd.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据