Study On Intelligent Condition Monitoring And Fault Diagnosis Methods For Nuclear Power Plant

During the operation of nuclear power plant,the sources of abnormal events are complex and diverse,so it is difficult for operators to directly detect abnormal phenomena and diagnose faults.Especially,when the abnormal phenomena are not obvious and the development process is slow,fault information will be more difficult to be captured.This situation is unfavorable to the safe operation of NPP.If a serious accident occurs due to failure to timely detect the fault,it will have a serious negative impact.In view of these conditions,it is necessary to perform condition monitoring and fault diagnosis for nuclear power units.At present,the traditional system-level condition monitoring and fault diagnosis methods for nuclear power plant need to be improved in accuracy and other aspects.At the same time,monitoring and diagnosis technology combined with intelligent methods have great advantages in many aspects.As a result of the above reasons,an intelligent condition monitoring system based on iForest and an intelligent fault diagnosis system based on the stacked autoencoder of nuclear power plant are constructed in this paper.We also analyze the operation data of the nuclear power plant,and make in-depth research on several intelligent methodsFirstly,the nuclear power plant system is selected as the research object in this paper,and the intelligent condition monitoring system based on iForest and the intelligent fault diagnosis system based on the stacked autoencoder is designed.By building the flow chart of the system,each part of the condition monitoring and fault diagnosis can be displayed intuitively.The intelligent condition monitoring system based on iForest and the intelligent fault diagnosis processes are composed of offline model training and online condition monitoring.They specifically include four main steps:data preprocessing,feature extraction,model building,and real-time condition monitoring or fault diagnosis.Secondly,the method introduction and simulation test of iForest-based intelligent condition monitoring method are carried out.Since the quality of feature extraction is very important for the effect of condition monitoring,this paper focuses on the feature extraction method based on sparse autoencoder before the condition monitoring simulation test.In this paper,the operational data under 100%full-power condition and the operational data in both 90%and 100%full-power conditions are selected for simulation test.In each simulation test result,the visualization results of feature extraction and various evaluation indicators of condition monitoring are respectively displayed.In order to compare with the traditional methods and illustrate that the iForest-based intelligent state monitoring method has better performance,We also selects the single-class support vector machine(OCSVM)and the local outlier factor(LOF)to make the simulation test.The data used in the tests of these two methods is the same as the data in the iForest method.The results show that the evaluation criteria of the iForest method are superior to the traditional methods.The iForest method has obvious advantages.Finally,the method introduction and simulation test are carried out for the intelligent fault diagnosis method based on stacked autoencoder.The stacked autoencoder method uses the idea of deep learning,so it can process complex data.The neural network model used in this paper has a four-layer network structure.The sparse autoencoder can extract the features of the data well.The softmax classifier can clearly and effectively classify the fault types during the operation of the nuclear power plant.The model can achieve the purpose of diagnosing the operational data status of the nuclear power plant.The simulation test results show that the intelligent fault diagnosis method based on stacked autoencoder can judge several fault types well and perform well in fault diagnosis accuracy.