长输管道信号降噪及工况识别方法研究  

Noise reduction of signal and condition recognition of long-distance pipeline

作　　者：余东亮[1] 张来斌[1] 梁伟[1] 叶迎春[1] 王朝晖[1] 

机构地区：[1]中国石油大学机电工程学院,北京102249

出　　处：《石油学报》2009年第6期937-941,共5页Acta Petrolei Sinica

基　　金：国家高技术研究发展计划(863)项目(2008AA06Z209);中国石油天然气集团公司创新基金(2006-A类);教育部新世纪优秀人才支持计划项目(NCET-05-0110)联合资助

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、工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,增强样本空间划分能力,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。The weak negative pressure wave signals mixed in the powerful background signals and noise were often filtered out from the signal of long-distance pipeline. In order to resolve the problem, the local projective algorithm combined with the wavelet packet analysis technology was used to reduce the noise of signal. First of all, the sampling series were reconstructed using the delay coordinates. The background noise signal and the negative pressure wave signal of the high dimensional phase space were separated into different sub-spaces. The reconstruction of sub-spaces could reduce the strong background signal and get the negative pressure wave signal with a small amount of random noise. Then, wavelet packet analysis technology was used to extract the weak feature signal and retain the turning-point information. Field tests indicated that the combined method could substantially improve the value of signal-to-noise ratio. Because of the frequent regulation of long-distance pipeline and the complicated operation conditions, it was quite difficult to classify the similar signals, which could cause false alarm and failure alarm. The double-weight neural network was used to recognize the signals of various conditions. The construction of various closed hypersurfaces could get the optimal coverage for the sample space, enhance the classification capability of the sample space and improve the recognition effect. The field tests showed that this method had higher ability than the BP network and RBF network in the condition recognition.

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 

分 类 号：TE973.6[石油与天然气工程—石油机械设备]