摘要
利用光纤作为传感器的管道安全预警系统可以对油气管道破坏性事件进行在线分析与智能识别。针对由光纤采集的振动信号,采用粗细提取结合加希尔伯特变换的方法,准确地将破坏性事件的振动信号从海量的原始数据中提取出来,经过小波降噪处理获得高信噪比的纯净信号,之后采用小波包、MFCC、时域能量窗口等多种方法全面地提取出信号的特征,对特征向量做主成分分析,从而获得独立、正交的无冗余特征向量,最后利用多级分类器对信号特征作综合识别分析以确定破坏性事件的类型,并引入了增量学习机制。该系统在真实现场的运行结果表明,不仅对破坏性事件的识别率达到了94.35%,满足了实际应用需求,而且能够满足实时性的要求。
Pipeline security warning systems with fiber optic sensors can identify and classify the online signals of devastating events intelligently. For the vibration signals collected by the optical fiber, this method uses rough and fine segmentation and Hilbert transform method to accurately extract the vibration signals of destructive events from the mass of raw data, and obtains pure signal through wavelet denoising method, and then extracts the features of the signals using wavelet packet, MFCC, the energy window in time. After a principal component analysis, these feature vectors are transformed into independent, orthogonal and non-redundant feature vectors. Finally, the type of destructive events can be determined by the use of multi-class classifier and the comprehensive identification analysis of the signal characteristics, and the incremental learning mechanism is introduced into the method. The result of the system running on site shows that the recognition rate of destructive events is up to 94.35%, which can not only meet practical needs, but also meet real-time requirements.
出处
《石油规划设计》
2017年第2期38-42,53,共6页
Petroleum Planning & Engineering
