| The real-time monitoring of gas pipeline leakage is one of the necessary guarantees of natural gas energy transportation safety,but the long-distance transportation pipeline network is mostly buried underground which 's background environment is complex and fickle,so acquiring these weak pipeline leakage signal needs laid high-sensitive and anti-interference sensor equipments and precise classification recognition methods for a better monitoring performance.At the same time,some different parameters such as gas supply pressure and gap size can also lead to different changes of how many pipeline gas leak,so identification the leakage size also has a direct and significant role to know the ponderance of the leakage accident's influence.This paper perceives and obtains the acoustic signals that carrying these leakage information based on the optical fiber F-P acoustic sensing technology which has many advantages,such as good sensitivity,high positioning accuracy,short test time,strong ability of adapting,long life of system and so on.And puts forward a method to extract feature-set which coalesced the analysis results of time-domain and frequency-domain from the audio characteristics of acoustic signal.And then using assiociation analysis method for mining those strong association rules between the signal features and one target event,and build a classifier based on these association rules,and finally realized the pipeline leakage detection under complicated background noise and leakage size recognition,better solve the application problem of the above.The main work of this paper includes the following aspects:First,analyze the research background,present situation,demand of pipeline leakage detection,and the mechanism of pipeline leakage acoustic signal's production and optical fiber F-P acoustic sensing technology's perception.Finally,form the overall framework about gas pipeline leakage detection method proposed by this paper based on the principle of data mining.Second,research the feature extraction methods based on characteristics of fiber optic acoustic sensor signal and extract MFCC features from the angle of frequency-domain,extract AR model parameters features form the angle of time-domain respectively to form a feature-set.And then studing the methods of feature selection based on variance and feature binarization based on FCM.Third,research on the design method of association rules classifier of pipeline leakage detection.The first step is to complete the frequent itemsets mining and association rules generation and triming between features and the target event based on Apriori algorithm,in which four important evaluation index of association analysis include support,confidence,Kulc coefficients and unbalance factor was introduced.Finally,complete the construction of association rules classifier based on these strong association rules and other important parameters.Last,build a gas pipeline leakage test experiment platform and establish a leakage events database which contains 12 categories of events.Using ten fold cross validation and classifier evaluation index to complete the performance evaluation,comparison and analysis of proposed leakage detection method in this paper based on the database.The experimental results show that the method proposed by this paper can reach to 99.53%of a comprehensive evaluation index called F1-measure in leakage under complex environment detection performance,and own 98.14%of F1-measure on average for 6 types of target event about the recognition of leakage size. |
PDF Full Text Request