Research On Acoustic Signal Abnormal Detection Method Based On Cluster Analysis

As one of the five transportation modes,pipeline transportation has the characteristics of high efficiency and convenience.It is the main transportation mode of natural gas,oil and other energy sources.Due to the continuity of pipeline transportation,if the pipeline leaks in the process of transportation,the medium in the pipeline will continue to drain,resulting in environmental pollution and even major safety accidents,with incalculable consequences.Therefore,pipeline leakage detection is of great significance and necessity.At present,the main research direction of pipeline leakage detection is modeling diagnosis abnormal signal detection.In this method,the off-line training model is extracted from the acoustic signal samples to realize the mapping relationship between the features and the fault,so as to judge whether the pipeline leaks.However,the establishment of the model is usually affected by the selection of training samples,so the detection sensitivity and accuracy of the model are related to the samples selected during training.In addition,the modeling diagnosis method takes the whole frame acoustic signal as the detection object for detection.Although the detection can be used to determine whether a frame acoustic signal is abnormal,the position,amplitude,quantity and other information of the abnormal interval signal in the acoustic signal can not be obtained to distinguish the leakage signal and interference signal,so there may be leakage and false alarm.In this paper,the idea of the whole acoustic signal detection is changed,the interval division feature extraction of a frame of acoustic signal is carried out,and the clustering model free method is used to analyze the feature,extract the interval of the abnormal feature,so as to realize the abnormal detection of acoustic signal.This paper first introduces the concept of clustering,similarity measurement and clustering method.Then,the simulation acoustic signal is taken as the research object to extract the interval division features,and the hierarchical clustering method suitable for the distribution shape of the features and the similarity between the single connected classes are selected for clustering analysis.The single connection distance of the extracted features is analyzed.The threshold value of the single connection distance is set to get the specific clustering.The normal clusters in the clustering results are removed,and the interval sub signals of the remaining abnormal clusters are extracted as the abnormal interval sub signals.The method of determining the threshold value of single connection distance based on statistics is studied,and the shortcomings of the method are analyzed.On this basis,an adaptive method of determining the threshold value of single distance is proposed.Compared with the statistical method,the distance threshold determined by the adaptive method can be adjusted adaptively according to the features of the current acoustic signal to be detected,which has better adaptability and flexibility.The acoustic signal data collected by the field simulation leakage and the data collected by the valve internal leakage monitoring experimental platform under the laboratory environment are used for testing.The test results show that the acoustic signal anomaly detection method studied in this paper can detect and extract the abnormal interval sub signals in a frame of acoustic signal,and at the same time,the number,amplitude and characteristics of abnormal sub signals can be obtained Compared with the method of modeling diagnosis to detect the whole frame signal,this method can further mine the information of outliers,which can provide a good research foundation for the following research of leak location,reducing false alarm and so on.In addition,the clustering based outlier detection method also has the advantage of no modeling,so it does not have the disadvantage of being affected by the sample signal characteristics.