| Natural gas will become one of the main clean energy sources in China due to its less pollution in the process of exploitation,and its combustion products can be completely burned without dust.With the laying of natural gas pipeline in China and the construction of city supporting facilities,the problems of seasonal contradiction between supply and demand,the imperfect infrastructure of transmission and distribution pipe network need to be solved,the gas regulator is one of the most critical equipment in the transmission and distribution system.In this paper,through the research of predictive management system of pressure regulator,we hope to improve the management efficiency of gas pipeline network and strengthen the security of gas pipeline network.According to the working principle of automatic pressure regulation of gas regulator,the predictive maintenance management system of gas regulator based on big data is designed.The RTU pressure collector is installed in the regional pressure regulating cabinet or pressure regulating box of urban residents,commerce and industry,the outlet pressure and other parameters of 50 pressure regulators in Suzhou were collected.By comparing three different feature sample extraction methods,the deep neural network learning model using sparse self encoder feature extraction is selected to process the outlet pressure parameters in advance.The state parameters such as outlet pressure collected by data collector or IOT pressure sensor are processed in advance.Then the processed low dimensional outlet pressure data and corresponding eigenvalues are input into the outlet pressure stable point prediction model to obtain the corresponding outlet pressure stable value.Finally,the characteristic value and the corresponding stable value of outlet pressure are input into the fault diagnosis model based on SVM classifier to judge whether the fault is or not and the fault type of the gas regulator,so as to realize the fault early warning and alarm of the gas regulator.The results show that:(1)In the regulator fault analysis engine,the neural network algorithm combined with sparse self coding feature extraction and deep learning is adopted.(2)The core of the fault analysis engine of the pressure regulator is the diagnosis model,which forms the gas consumption characteristic map and fault characteristic map based on the learning of expert experience and parameter characteristics.(3)Prediction of stable value of outlet pressure In the test model,the relative error between the steady value of outlet pressure and the actual steady value of some pressure regulators is in the range of-1.36%-4.17%,and the smaller the relative error is,the more reliable the model is;in the fault detection model of pressure regulators,the accuracy of the four fault types is more than 90%.(4)The fluctuation of daily outlet pressure and actual pressure regulation accuracy of the fault regulator is higher than that of the normal operation regulator.The results show that the range is larger,and the difference between the maximum and minimum daily outlet pressure is larger;the pressure regulator with high closing level fault is higher than the actual closing level in normal operation,and the curve fluctuation is larger;the proportion of excessive pressure/over closing level fault reaches 42.42%,which is the most prone fault in the current gas regulator.(5)The preventive maintenance and management function platform of the pressure regulator can give early warning for 5 cases,i.e.high closing pressure,outlet pressure deviation,instantaneous under pressure,shut-off valve failure and vent valve failure;and give alarm for 3 cases,i.e.excessive pressure/closing level exceeding standard,excessive pressure/closing set outlet pressure high and internal leakage.The predictive management system realizes the remote supervision of the regulator,improves the maintenance efficiency of the regulator,and reduces the risk of transmission and distribution network failure caused by the regulator. |
PDF Full Text Request