长输管道信号降噪及工况识别方法研究

《长输管道信号降噪及工况识别方法研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、第3()卷第6期石油学报Vo1．30No．62009年11月ACTAPETROLEISINICANov．2009文章编号：0253—2697(2()09)06—0937—05长输管道信号降噪及工况识别方法研究余东亮张来斌梁伟叶迎春王朝晖(中国石油大学机电工程学院北京102249)摘要：鉴于长输管道中微弱负压波信号往往淹没在强背景信号及噪声中而被过滤掉等特点，应用局部投影降噪法结合小波包分析技术进行信号降噪。首先，将采样序列通过相重构，在高维的相空间上将背景噪声信号以及负压波特征信号分解到不同的子空间上，利用子空间的重构，分离出强背景信号及含有少量随机噪声的负压波特征信号，然后对

2、这两类信号分别进行小波包分析，提取弱特征信号，保留拐点信息。试验表明，两者结合能大幅提高信号的信噪比。此外，油气管道调节频繁、_T-况复杂多变也给相似信号准确识别带来困难，易造成误报警和漏报警。应用双权值神经网络识别各种工况信号，通过构造多种封闭超曲面完成对样本空间的最佳覆盖，增强样本空问划分能力，提高识别效果。该方法具有较强的工况识别能力，识别效果优于BP网络和RBF网络。关键词：长输管道；信号降噪；工况识别；局部投影法；双权值神经网络中图分类号：TE973．6文献标识码：ANoisereductionofsignalandconditionrecognitionoflong

3、-distancepipelineYUDongliangZHANGLaibinLIANGWeiYEYingchunWANGZhaohui(FacultyofMechanicalandElectronicEngineering，ChinaUniversityofPetroleum，Beijing102249，China)Abstract：Theweaknegativepressurewavesignalsmixedinthepowerfulbackgroundsignalsandnoisewereoftenfilteredoutfromthesignaloflong—distan

4、cepipeline．Inordertoresolvetheproblem，thelocalprojectivealgorithmcombinedwiththewaveletpacketanalysistechnologywasusedtoreducethenoiseofsigna1．Firstofall，thesamplingserieswerereconstructedusingthedelaycoordi—nates．Thebackgroundnoisesignalandthenegativepressurewavesignalofthehighdimensionalph

5、asespacewereseparatedintodifferentsub—spaces．Thereconstructionofsub-spacescouldreducethestrongbackgroundsignalandgetthenegativepressurewavesignalwithasmallamountofrandomnoise．Then，waveletpacketanalysistechnologywasusedtoextracttheweakfeaturesignalandretaintheturning—pointinformation．Fieldtes

6、tsindicatedthatthecombinedmethodcouldsubstantiallyimprovethevalueofsig—nal—to—noiseratio．Becauseofthefrequentregulationoflong—distancepipelineandthecomplicatedoperationconditions，itwasquitedifficulttoclassifythesimilarsignals，whichcouldcausefalsealarmandfailurealarm．Thedouble-weightneuralnet

7、workwasusedtorecognizethesignalsofvariousconditions．Theconstructionofvariousclosedhypersurfacescouldgettheoptimalcoverageforthesamplespace，enhancetheclassificationcapabilityofthesamplespaceandimprovetherecognitioneffect．Thefieldtestsshowedthatthism