成品油管道泄漏次声波监测及信号处理方法  

Study on infrasonic leakage monitoring and signal processing for product oil pipeline

作　　者：尹渊博 李玉星[1] 杨文 芦澍 张晨 刘翠伟[1] 杨凯 王武昌[1] YIN Yuanbo;LI Yuxing;YANG Wen;LU Shu;ZHANG Chen;LIU Cuiwei;YANG Kai;WANG Wuchang(College of Pipeline and Civil Engineering,China University of Petroleum(East China)/Shandong Key Laboratory of Oil&Gas Storage and Transportation Safety;PipeChina South China Pipeline Co.Ltd.;China Petroleum Pipeline Bureau)

机构地区：[1]中国石油大学(华东)储运与建筑工程学院·山东省油气储运安全重点实验室 [2]国家管网集团华南分公司 [3]中国石油天然气管道局有限公司

出　　处：《油气储运》2024年第8期905-915,共11页Oil & Gas Storage and Transportation

基　　金：国家自然科学基金项目“基于气-液-声耦合机理的液体管道泄漏动态压力波传播规律研究”,52274066;国家重点研发计划“油气管网安全运维的大数据分析理论、算法及应用”,2021YFA1000104;广东省重点领域研发计划“油气储运重大基础设施灾害防御关键技术及装备研发与示范”,2019B111102001。

摘　　要：【目的】随着油气输送管道总里程不断增加,泄漏监测已成为保证管道安全平稳运行的关键技术之一。次声波监测因其灵敏度高、定位精度高、维护费用低等诸多优点备受关注,但其在成品油管道中的工程应用有待探讨。【方法】基于次声波监测基本原理,自主搭建了液体管道泄漏监测实验装置,分析了不同的泄漏孔尺寸、管道压力以及泄漏点距离工况下,次声波传感器采集信号的特征。分析了db系小波基与sym系小波基1~9层小波变换的信号处理效果;利用信号的15个时域特征与4个频域特征参与随机森林分类模型建模,以ROC(Receiver Operating Characteristic)曲线下的面积(Area Under Curve,AUC)作为目标函数对模型参数进行优化,并采用基于WT-RF(Wavelet Transform-Random Forest)的方法对实验数据进行信号处理与分类。【结果】将新建方法应用于国家管网集团华南分公司某成品油输送管道发现,经过sym2小波基8层分解处理后的次声波信号在时频域具有明显可识别特征,随机森林识别模型结合定位信息可实现生产管道中泄漏工况的误报率、漏报率均为0;在91 km的成品油管道监测区间,定位误差800 m左右,稳定泄漏速率0.0016 m^(3)/s,最小可检测泄漏速率为0.00046 m^(3)/s。【结论】次声波泄漏监测技术在成品油管道测试中获得了较好的效果,其误报率、漏报率均极低,且定位误差小,相关研究成果可为该技术在成品油管道的应用提供技术支持与参考。[Objective]With the total length of oil and gas transmission pipelines increasing due to booming development,pipeline leakage monitoring has emerged as one of the critical technologies to ensure the safe and stable operation of these pipelines.Infrasonic monitoring has garnered significant attention due to its high sensitivity,high positioning accuracy,and low maintenance costs.However,its engineering application in product oil pipelines requires further discussion.[Methods]Based on the basic principle of infrasonic monitoring,an experimental setup for liquid pipeline leakage monitoring was independently constructed,aimed to analyze the characteristics of signals acquired by infrasonic sensors across different leak hole sizes,pipe pressures,and distances from these sensors to the leak points.The signal processing effects of wavelet transforms at 1–9 layers on the db and sym wavelet bases were analyzed.Subsequently,a random forest classification model was established,incorporating fifteen time-domain features and four frequency-domain features of the signals.The model parameters were optimized,using the Area Under Curve(AUC)of the Receiver Operating Characteristic(ROC)curve as an objective function.Furthermore,the experimental data were processed and classified,utilizing the method based on Wavelet Transform-Random Forest(WT-RF).[Results]The proposed approach was applied to a product oil transmission pipeline section of PipeChina South China Pipeline Co.Ltd.,resulting in the following findings.Following an 8-layer decomposition on the sym2 wavelet basis,the infrasound signals exhibited distinct recognizable characteristics in both the time and frequency domains.The random forest identification model,supported by positioning information,showcased a zero false alarm rate and missing alarm rate under leakage conditions of the production pipeline.At a 91 km monitoring interval along the product oil pipeline,the positioning error was about 800 m,facilitating reliable monitoring up to a leak rate of 0.0016 m^(3)/s,wit

关 键 词：成品油管道 泄漏 次声波 监测 小波分析 随机森林 最小可检测泄漏量 

分 类 号：TE88[石油与天然气工程—油气储运工程]