Research On Intelligent Monitoring Of Leaks Based On Signal Processing

As the service time of pipelines extends,the complexity of the pipe network increases,leading to the pipeline leakage threatens for pipeline safety.After the pipeline leaks,a quick leakage detection and location is of great importance to solve the problem.In recent years,the construction of “smart pipeline” has become a new trend in China’s pipeline development.In order to account with this trend,the networked,digitized and automated methods for leakage detection have become the research focus of oil and gas storage and transportation.This article conducts research on the basis of negative pressure wave signals.First,a pressure anomaly alarm and expert analysis system is established.This system is programmed in Microsoft Visual Studio 2010 based on the C ++ language.The V8 engine is used to integrate with the SCADA system.The point information in the PI database is scanned and calculated.According to mathematical modeling,determine the change of point information corresponding to the working conditions of leakage,starting and stopping pumps,and tank cutting,and determine the specific valve number and the point ID in the SCADA system according to the on-site process flow chart.After that,it was amended according to the experience of the on-site dispatcher,and the prepared algorithm rules were configured into the system.The pressure abnormality alarm and expert analysis system scanned and calculated the monitoring points according to the algorithm rules.In order to improve accuracy and reduce errors,based on pressure anomaly alarms and expert analysis systems,apply machine learning methods to monitor and locate negative pressure wave signals,and use indoor experimental loops to collect pressure signals of different operating conditions.Based on this,an improved local mean algorithm is used to denoise the pressure signals.For the pressure signal after de-noising,eight features are extracted,the optimal feature combination is found for the random forest and probabilistic neural network model,a detection model for pipeline leakage conditions is established,and the location of the leakage point is studied.The research results show that the analysis of the causes of abnormal pressure fluctuations under different working conditions collected in the experimental loop shows that the pressure signals of different pipeline working conditions are separable.During the pipeline leakage signal processing phase,the corresponding optimal interpolation method may be different for different signals.During the feature extraction phase of the pipeline leak signal,the classification effects of energy,energy entropy,kurtosis,kurtosis entropy,mean square error,mean square error entropy,sample entropy,and fuzzy entropy on the different working conditions of the experimental loop are analyzed.Good energy entropy and mean square error entropy are discarded.During the intelligent identification of pipeline leak conditions,two learning models were applied,namely the probabilistic neural network and the random forest model.The parameters of the probabilistic neural network were selected using the improved Drosophila optimization algorithm,and the two learning models were established Corresponding optimal feature library.In the leak point localization phase,compared with the leak point localization based on two sensors,the localization of the leak pressure based on three sensors can avoid the calculation of the negative pressure wave velocity,which has the advantages of small calculation amount and small error.The pressure signal of each sensor is used to locate the leakage point based on the three-sensor positioning principle.After analysis,a rough time delay estimation is first performed through a related algorithm,and an accurate time delay estimation is performed using an adaptive algorithm,thereby improving the positioning accuracy.