基于神经网络的管道环焊缝失效预测及灵敏性分析  

Failure Prediction and Sensitivity Analysis of Pipeline Girth Weld Based on Neural Network

作　　者：齐峰 淦邦 成涛林 马卫锋 姚添 王珂 QI Feng;GAN Bang;CHENG Taolin;MA Weifeng;YAO Tian;WANG Ke(PipeChina Network Corporation Eastern Oil Storage and Transportation Co.Ltd,Xuzhou Jiangsu 220005,China;CNPC Tubular Goods Research Institute,Xi'an 710061,China)

机构地区：[1]国家管网集团东部原油储运有限公司,江苏徐州220005 [2]中国石油集团工程材料研究院有限公司,西安710061

出　　处：《西南大学学报（自然科学版）》2024年第3期159-167,共9页Journal of Southwest University(Natural Science Edition)

基　　金：国家自然科学基金项目(51904332);陕西省青年科技新星项目(2021KJXX-65);国家管网集团揭榜挂帅项目(WZXGL202107);陕西省重点研发计划项目(2020GY-179)。

摘　　要：对油气管道焊缝所处风险等级进行准确的预测是保证管道安全运行必不可少的环节,本研究首先在分析影响焊缝失效主要因素的基础上,构造了适用于管道环焊缝失效预测的3层神经网络模型.其次,针对传统灵敏度分析方法难以综合考虑各因素对焊缝失效影响程度的问题,从理论层面分析了基于神经网络的失效预测灵敏度分析方法,并将其嵌入到所研发的神经网络中.最后,针对高、中和低风险的环焊缝数据量严重不平衡的现状,提出一种双嵌套整体相关度最小的训练样本选择算法,可以在训练中较好地解决这一问题.对721个焊口实际数据进行预测试验发现,本研究提出的失效预测神经网络模型可行且有效,高、中风险识别率达100%,低风险识别率达98.8%.The failure of the girth weld of the oil and gas pipeline could make the pipeline run in a dangerous state,so it is necessary to predict the risk level of the oil and gas pipeline efficiently and accurately.Based on the main parameters of pipeline failure prediction,the risk levels of pipeline girth weld were divided into high,medium and low risk levels,and a neural network model suitable for failure prediction of pipeline girth welds was developed.Because the traditional sensitivity analysis method is difficult to comprehensively consider the interaction of various production factors of girth welds,the sensitivity analysis method suitable for failure prediction of girth welds was theoretically analyzed,and embedded in the developed neural network.In actual production,there is a serious imbalance in the amount of data of girth welds with high,medium and low risks.Therefore,a double-nested training sample selection algorithm with minimal overall correlation was proposed to solve the problem of unbalanced number of risk samples in training neural networks.The results show that the proposed girth weld failure prediction neural network model is feasible and effective.The recognition rate of high and medium risk was 100%,and the recognition rate of low risk was 98.8%.

关 键 词：环焊缝 失效预测 神经网络 灵敏性分析 样本选择 

分 类 号：TE88[石油与天然气工程—油气储运工程] TP391.4[自动化与计算机技术—计算机应用技术]