HMM-based Pipeline Safety Monitoring DAS Signal Timing Information Mining And Identification Method

With the rapid development and extensive applications of distributed acoustic sensing(DAS)to long distance pipeline safety monitoring,it is still challenging to find a very efficient way to achieve highly correct recognition and really deep understanding of physical events sensed in a wide dynamic environment,as the vibration signals usually exhibit non-linear and non-stationary characteristics caused by the complicated environments.In this thesis,a Hidden Markov Model(HMM)based pipeline security monitoring DAS signal timing information mining and identification method is proposed.The method combines the two-fold information of feature vector and feature vector time series evolution relationship.Compared with the traditional recognition algorithm based on static features or feature vectors of signals in a certain time period,it has higher event recognition rate.At the same time,the hidden state sequence of the event signal is output,which is convenient for mining the evolution law of various events.The experimental results with real field test data show that the average recognition accuracy of this work is as high as 98.2% for frequently encountered five typical events along buried pipelines,which is superior to traditional machine learning methods such as DT,RF,SVM,BN,etc.At the same time,the proposed method has high timeliness and meets the requirements of real-time online use.The specific work is as follows:(1)Analyze the progress of signal processing technology at home and abroad in the field of pipeline safety monitoring.For the problem that the existing method has low recognition rate due to neglecting signal timing evolution information,this paper is based on the principle of DAS pipeline monitoring based on Φ-OTDR technology and the dynamic timing identification of HMM.Excellent performance,proposed HMM-based pipeline safety monitoring DAS signal timing mining and identification method.(2)According to the randomness and instability of the one-dimensional time signal of each spatial point of DAS,a multi-analytic domain feature extraction method including time domain,frequency domain,transform domain and model parameters is proposed,and the value feature selection is proposed.And the LDA feature dimension reduction algorithm tests the validity of the feature set and visualizes the test results.(3)HMM timing mining and identification methods including five kinds of typical event signals including background noise,artificial excavation,mechanical excavation,traffic interference and factory interference are proposed.Aiming at the parameter initialization method of HMM system,a parameter initialization method based on Kmeans algorithm is proposed.The multi-dimensional observation sequence is obtained after long-term signal feature extraction.The solution of estimating the HMM observation matrix by GMM is proposed.The sample is obtained based on Viterbi algorithm for the first time.At the same time of the label,the optimal hidden state sequence is output,and the time evolution of the signal is mined.(4)Use field data to verify the effectiveness of the HMM pipeline safety monitoring method.Compared with the evaluation indexes of five traditional classifiers and HMM classifiers,the test results show that HMM has higher recognition rate than the other five models,and the total recognition rate can reach 98%.At the same time,HMM outputs various event hidden states.The sequence is a good description of the evolution of different event signals.Finally,the timeliness of the HMM method is tested.The results show that the HMM online processing speed can reach 125 samples/second,which meets the real-time requirements of the system.