Gas Leakage Detection Based On Approximate Entropy And Improved K - Means

Natural gas is widely used because of its clean and efficient characteristics, and the transportation mode is generally used in pipeline transportation. In the transport process, such as the occurrence of pipeline leak, which affects not only the normal transportation of natural gas and people’s daily life, but also caused great economic losses, affecting the ecological environment, and even cause casualties. Therefore, how to detect pipeline leakage quickly and accurately is extremely important. In this paper, the theory of signal processing and pattern recognition is used to carry out the following research on pipeline leak detection:1, A comprehensive experimental platform for gas pipeline leak detection is built. The experimental platform can be used to realize the on-line monitoring and experimental simulation of the gas pipe line and branch pipe under different pressure and different leakage aperture.2, A signal feature extraction method based on the combination of approximate entropy and wavelet transform is proposed. Firstly, the approximate entropy is used to select the appropriate acoustic emission signal for further analysis, then the details of the original signal are extracted based on wavelet decomposition, in the meantime, the signals are reconstructed according to the obtained details. Finally, six general characteristics, such as variance, peak in the frequency domain, coefficient of kurtosis, skewness, rising count and ringing counts, are extracted according to the reconstructed signals. In order to remove redundant features, Relief algorithm and correlation algorithm are exploited to select the peak in the frequency domain and ringing counts for analysis. Experimental results show that the proposed method not only reduces the amount of data processing, but also highlights the main features of the signal.3, In order to achieve the judgment of the non-supervision leakage, the two signal feature sets are processed by the improved K means clustering method, so that the leakage signals and non-signals can be distinguished, So as to achieve the purpose of accurate judgment. An improved initial cluster center selection methodis proposed. The initial cluster centers are determined based on the density of the data, the isolated points are removed based on that. Experimental results show that, compared with the original clustering method, the improved k-means clustering method can removal the influence caused by the isolated points and manual selection of the cluster numbers, and can automatically determine the number of cluster numbers, achieved a better goal for the detection.