Research On Leakage Detection Method Of Refined Oil Pipeline Network Based On Deep Belief Network

With the development of China's economy,the demand for refined oil is increasing day by day.However,with the damage of pipelines damage and the destruction of man-made factors,pipeline leakage accidents happen frequently,which cause serious environmental pollution and economic losses.The traditional one-line leakage detection method can not meet the requirements of industrial field for the accuracy and accuracy of leakage detection.One of the main reasons is that the traditional centralized leakage detection system has a simple structure and can not satisfy the high-time detection of pipeline network data.Another factor is that the accuracy of leakage detection is seriously interfered by the on-site process,leading to high false positive rate.Based on the theory of distributed fault detection,this thesis puts forward an anomaly detection method based on time series.Combined with the characteristics of pipeline network,the thesis uses the depth belief network classifier model to classify pipeline network anomaly data,accurately find leakage and eliminate false alarms.The main contents include the following three aspects:Firstly,combined with the distributed fault detection model,the thesis puts forward a distribution management,centralized monitoring of the pipe network leakage detection system framework,which is mainly divided into remote control center and distributed detection nodes.The OPC standard unified interface is used to collect and store field data in the system framework.The real-time data flow model is employed to design the pipeline network data scheduling scheme,and this thesis uses multi-threaded technology to handle multiple tasks and then improve the efficiency of entire detection system.Secondly,the thesis presents a temporal sequence pattern method based on sequence temporal error.Compared with previous methods,the method not only can be applied to different data characteristics environment,but also has strong adaptability.Based on the pattern density anomaly detection method,several common data is selected as research object,which proved the generalization of this method.The simulation results show that the proposed method is effective in the anomaly detection of pipeline network,especially for the detection of small leakage anomalies.Thirdly,the thesis analyzes the reasons for the abnormal condition of pipeline network,and classifies abnormal conditions of working conditions.The thesis introduces the relevant methods of depth learning and uses deep belief network(DBN)classifier model to identify the type of the pipeline network.The abnormal conditions of the pipeline network are identified by inputting the equipment information of each monitoring station.The simulation results are verified by the real industry data.The results validate the effectiveness of the DBN model in pipeline network leakage detection and discuss the impact of DBN parameters on the recognition effect.