Leak Detection For Gas Pipeline Based On Acoustic Data-driven

With the rapid development of the domestic economy, increasing demand for natural gas,there are more and more onshore pipelines and pipeline transportation task has become increasingly important. However, the pipeline leak has occurred due to pipeline corrosion,natural destruction, vandalism and other reasons. Pipeline leak will not only cause environmental pollution, economic losses, will lead to more serious injuries and other accidents. Therefore, for the protection of the safe operation of the pipeline properly,pipeline leak detection becomes particularly important. The acoustic method is a method with high sensitivity, accurate positioning and so on used in this article is the last ten years.A data-driven system for acoustic method was designed in this paper with the help of the high-pressure gas pipeline system designed on laboratory. The test is applied to accomplish the acquiring of the acoustic signals, Then using high-quality time-frequency analysis effect of generalized S transform analyze the acoustic signals and de-noising process, analyze and extract the signal leakage and interference signals in time domain and includes some of the characteristics of wavelet domain, provided the basis for the judgment leakage. Selected the leakage and interference signals commonly used statistical characteristics based on the genetic algorithm and got the most favorable conditions characterized by a combination of classification, and finally established muti-class classifiers based on support vector machine(SVM), completed the leakage and interference classification signal, reducing the false alarm rate leak detection.By the study of leak detection based on acoustic data-driven, the main conclusions of this paper are: ① leak acoustic signals characteristics on time and frequency domain significantly different from the normal operation of the pipeline signals, you can set a statistical feature threshold leakage judgment; ②The S transform can achieve good resultson the time-frequency analysis since the leakage signals have most noise on high-frequency an significantly leakage characteristics on low-frequency. ③Not all features are suitable for statistical signal classification, feature-based genetic algorithm can select a more appropriate classification applies to feature the best feature subset; ④The support vector machines can be used to distinguish leakage and interference signals,and the accurate identification of signal leakage rate can reached 92%.