Fault Diagnosis Method Based On Sample Generation And Interval Markov Feature In Pipeline

Energy plays an essential role in the economy development in a country,and pipeline is an important way to ensure energy transportation.Now,with the in-service time increase of pipeline,different damage caused by external forces and natural corrosion lead to pipeline condition deteriorate year by year and potential danger growing sharply.In case of leakage,there will be serious economic losses and environmental pollution.Therefore,it cannot be ignored to maintain operation safety.Under the complex oil pipe network structure,the signal attenuation caused by the pressure wavelength long-distance propagation and the frequent valve and pump adjustment lead to false alarm and miss alarm in the identification of leakage.In addition,recognition accuracy of small leakage is not high due to insufficient signal acquisition accuracy and external interference.Therefore,how to identify weak leakage faults and working condition adjustment faults under complex working conditions to improve the anomaly detection accuracy and reduce the false alarm rate and false alarm rate is an urgent problem to be solved for pipeline operation safety.Aiming at the recognition difficult problems of weak pipeline leakage fault and working condition adjusting fault under complex working condition in existing pipeline leakage detection method,a fault diagnosis method based on sample generation and interval Markov feature extraction in pipeline is proposed from the aspect of small samples generalization and deep feature extraction,in combination with artificial intelligent identification.Specific researches are as follows:1.A small sample generation method based on ELM is proposed.Firstly,fault samples are generated from the historical data,which are used to establish the training set by up sampling and down sampling,Then ELM model is trained.Secondly,the training weights are obtained and the output is weighted.Finally,the weighted output and the weight are reversed to derive the birth sample.2.An interval Markov fault feature extraction method is proposed.In order to further search for the features that can more effectively present the time series changes and overcome the dependence of traditional Markov features on the state of single point,the Markov features are improved based on the traditional Markov chain,in which the state of single point is replaced by interval state and the method of state division is improved by introducing the concept of quantile to improve the accuracy of fault feature extraction.3.Based on sample generation method and internal Markov feature,a pipeline leakage fault diagnosis algorithm is designed,in which SVM model is utilized to classify the different fault class.Based on the recognition results,a homologous signal similarity matching method is designed based on interval Markov features combined with wavelet anomaly detection to locate the faults.Experimental results demonstrate the effectiveness of the proposed method.In addition,we discuss the influence of parameters on the results.