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

作者：Yu, Dongliang^*; Zhang, Laibin; Liang, Wei; Ye, Yingchun; Wang, Zhaohui

来源：Acta Petrolei Sinica, 2009, 30(6): 937-941.

• Condition recognition
• Double-weight neural networks
• High dimensional phase space
• Local projective
• Long distance pipelines
• Negative pressure waves
• Operation conditions
• Wavelet Packet Analysis

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.

• 出版日期2009
• 单位中国石油大学（华东）