基于机器学习的冻土区融沉变形管段识别方法  

Identification of Thaw Settlement Deformation Pipe Section in Frozen Soil Area Based on Machine Learning

作　　者：刘燊 刘啸奔 李睿 李博 陈朋超 张宏[1] Liu Shen;Liu Xiaoben;Li Rui;Li Bo;Chen Pengchao;Zhang Hong(National Engineering Laboratory for Pipeline Safety/MOE Key Laboratory of Petroleum Engineering/Beijing Key Laboratory of Urban Oil and Gas Distribution Technology,China University of Petroleum(Beijing;PipeChina North Pipeline Co.Ltd.;PipeChina Shenyang Pipeline Inspection Center)

机构地区：[1]中国石油大学(北京)油气管道输送安全国家工程实验室/石油工程教育部重点实验室/城市油气输配技术北京市重点实验室 [2]国家管网集团北方管道有限责任公司 [3]国家管网沈阳管道检测中心

出　　处：《石油机械》2022年第3期106-114,共9页China Petroleum Machinery

基　　金：国家自然科学基金项目“逆断层作用下X80管道屈曲演化与韧性破损机理研究”(52004314);北京市自然科学基金项目“时变温压载荷作用下大口径直埋热水管道-土体耦合机制与失效机理研究”(8214053);新疆自治区天山青年计划项目“复杂载荷作用下高钢级管道韧性断裂与后屈曲失效行为”(2019Q088);中国石油大学(北京)青年拔尖人才科研基金项目“断层作用下高强钢管道失效机理与可靠性评价”(2462018YJRC019);中国石油大学(北京)科研基金项目“基于大数据的天然气管网智能运行与控制研究”(2462020YXZZ045)。

摘　　要：目前工业界采用人工识别的方法,对整条管线的惯性检测单元(IMU)应变检测数据进行逐段识别的做法存在耗时多、识别效率不高以及判断标准不一致等问题。鉴于此,通过建立机器学习模型,提出了弯曲变形危险管段智能识别方法,实现了对冻土区融沉变形管段的智能识别。首先统计了漠大一线冻土区管线中弯曲应变值超过0.125%的管段,包括弯头段、凹陷段和融沉导致的弯曲变形段等,使用1阶数字低通滤波法降低IMU应变检测数据中的噪声干扰,然后结合几何/漏磁检测数据截取IMU应变检测数据中不同管段类型的样本数据,从中提取了11种典型数据特征值,利用主成分分析法对11种特征值进行降维处理,最后建立决策树和随机森林模型进行识别分类。研究结果表明,不同管段类型的长度特征是影响模型分类效果的重要因素,在测试集中决策树模型出现了过拟合,识别准确率大幅下降,随机森林模型识别准确率达到了90%以上。该识别方法为管线完整性评价提供了技术基础。At present,the pipeline industry mainly adopts the artificial identification method to identify the IMU strain detection data of the whole pipeline step by step,which results in many problems such as time consuming,low identification efficiency and inconsistent judgment standards.In the paper,by means of building machine learning model,the intelligent identification of thaw settlement deformation pipe section in frozen soil area was realized.Firstly,statistics was conducted to obtain the pipe section with bending strain value exceeding 0.125%in the frozen soil area of the Moda 1 pipeline,including the elbow section,the depressed section and the bending deformation section resulted from thaw settlement,and the first-order digital lowpass filtering method was used to reduce the noise interference in IMU strain detection data.Then,combined with the geometric/magnetic flux leakage detection data,the sample data of different pipe section types in IMU strain detection data were cut out,from which 11 typical data characteristic values were extracted,and the principal component analysis method was used to descending dimension of the 11 characteristic values.And finally,decision tree and random forest model were built to conduct identification and classification.The study results show that the length feature of different pipe section types is an important factor affecting the classification effect of the model;in the test set,over-fitting occurs in the decision tree model,and the identification accuracy of random forest model exceeds 90%;and this method provides a technical basis for the evaluation of pipeline integrity.

关 键 词：IMU 弯曲应变 冻胀融沉 机器学习 智能识别 

分 类 号：TE832[石油与天然气工程—油气储运工程]