A SVR-BN Based Early Warning Method For Gas Turbines Of Natural Gas Long-distance Pipelines

Gas turbine is the core power equipment for long-distance natural gas pipeline,it often fails and triggers self-protection stop system because of its complex structure,many parts and work load,which will affect the stability of the main line of gas transmission and cause serious economic losses.At present,the SCADA system is widely used in the state monitoring of gas turbine units,which contains a large number of process parameter data.It has been found that many emergency shutdown failures have process parameters signs.Because the data of process parameters can be predicted,if we use the data to carry out fault early warning research,the downtime rate of gas turbine units will be reduced greatly.In this paper,the gas turbine unit is regared as the research object,the fault warning method is carried out by using the SCADA system parameter data.The research of fault early warning methods mainly includes the research of process parameter prediction methods,the research of fault diagnosis methods and the establishment of fault early warning models.(1)In order to carry out the process parameter prediction and fault diagnosis method,the process of gas turbine auxiliary system is analyzed.The FMEA method is used to analyze the possible failure of the components of the auxiliary system and the influence on the system.Combined with the analysis result of FMEA,the fault database of gas turbine auxiliary system is established.(2)SVR is applied to predict the process parameters.Firstrly,the process parameters in the SCADA system is analyzed,the process parameters that are suitable for prediction is found.Then,in order to improve the performance of SVR prediction,particle swarm optimization and chaos theory are applied to optimize the selection of SVR parameters.Finally,the chaotic signal generated by Chen mapping is used to test the performance of the optimization method,and the optimization method is proved to be of great superiority.(3)Bayesian network method is selected for system fault diagnosis.Firstrly,appropriate mapping rules is established and map the fault tree to Bayesian networks.And then,in order to solve the problem that the condition probability of the complex Bayesian network is not easy to obtain,combining the fusion theory with the fuzzy theory,a method of calculating the conditional probability of the Bayesian network is proposed based on the knowledge of expert judgment.At last,the reasoning method of Bayesian network is briefly introduced.(4)Establishment and application of fault early warning model.Firstiy,the concept of deviation degree and the moving window method are put forward.A fault early warning model is established by combining SVR prediction method with BN fault diagnosis method.Then the fault early warning model is applied to the auxiliary system of gas turbine,the early warning model of fuel gas system and lubricating oil system are established respectively.At last,the practicality and accuracy of the established fault warning model are verified.