A leak detection method for natural gas pipelines based on time-domain statistical features

In view of the amplitude-value statistical feature of pipeline dynamic pressure signals in time domain, an extraction method of these signals based on time-domain statistical features was proposed which can dynamically detect a gas leak of pipelines. Through the normalized frequency distribution histogram (k-P diagram) of each frame signal, the probability bandwidth feature of amplitude values distributed near the average can be extracted from dynamic pressure signals in time domain, which is independent of the amplitude value and waveform of signals, and also unaffected by the frequency offset of signals, so it has a good adaptability. Taking the statistical feature vector of dynamic pressure signals in normal working conditions whose dimension is reduced by PCA as a target sample, a PCA-SVDD diagnostic model was established that can offer reliable detection of a pipeline leak. The results of off-line detection of historical data and a continuous on-line test showed that this method can detect a leak of gas pipelines effectively.