Research On Fault Diagnosis Of Nuclear Power Plants Based On GrC-SDG

With the deterioration of the ecological environment and the increasing demand for electricity,nuclear energy is attracting more and more attention as the most potential clean energy.However,due to the serious potential radioactive hazards of nuclear power plants,the safety issue has seriously hampered the rapid and efficient development of the nuclear power industry.In order to ensure the safe and stable operation of nuclear power plants,it is necessary to monitor the condition of the nuclear power plant systems.After the anomalies found,fault diagnosis should be conducted as soon as possible to help the operators understand the fault information and take effective measures to avoid serious consequences in time.In view of the Fuqing NPP Unit 2 and the existing methods of fault diagnosis,the key technologies of condition monitoring,fault diagnosis and severity evaluation for nuclear power plants are studied.A new fault diagnosis method for nuclear power plants is presented and a corresponding fault diagnosis system is developed in this paper.The main research work is as follows:(1)The threshold method combined with Qualitative Trend Analysis(QTA)is used to monitor the operation parameters of nuclear power plants and improve the sensitivity of condition monitoring,so as to realize the early detection of abnormal parameters.(2)The basic methods of Signal Directed Graph(SDG)are studied and the SDG fault diagnosis process is improved by decision table.Then,the SDG fault diagnosis method based on diagnostic rules is developed,to avoid repetitive search inference of SDG mode,obtain the path of fault propagation,and improve the speed of diagnosis.The SDG model of nuclear power plants is established to clearly express the impact of the relationship between the operating parameters and its correctness is testified by the simulator.(3)The attribute reduction algorithm and similarity reasoning algorithm of Granular Computing(GrC)are studied.The attribute reduction algorithm is applied to reduce the decision table.In this way,the complexity of decision tables can be reduced and the efficiency of data processing will be improved by deleting unnecessary condition attributes in the case of assuring classification ability.The similarity reasoning algorithm is used to ensure the correctness of the decision results.Then,a fault diagnosis method based on GrC-SDG is formed.(4)Three major deep learning models of deep neural network are studied and the Deep Belief Network(DBN)model suitable for nuclear power plants is selected.The mapping relation between the evaluation parameters and the severity of fault is established by DBN model,so as to estimate the severity of the fault.(5)The C#4.0 language is used to develop a fault diagnosis system for nuclear power plants,integrating state monitoring,fault diagnosis and severity evaluation.The tests and analysis are carried out by using the simulator of Fuqing NPP Unit 2.The test results show that the proposed method of condition monitoring,fault diagnosis and degree evaluation can timely and accurately detect system anomalies,identify the types of faults and obtain more accurate estimation results,and certify the feasibility of the method.At the same time,the effectiveness of the system is certified.This paper lays foundation for the engineering application and further research.