Researchof Nuclear Power Plant Fault Diagnosis Based On Data Driven

The fault diagnosis of nuclear power plant is always the focus in this field.It is directly related to the safe operation of nuclear plant.The goal of fault diagnosis aims to help the operator to find the cause of malfunction andprevent the fault develop into the nuclear accident and threaten the pubic.With the development of information,the fault diagnosis based on data driven is the new focus.As complex system,the nuclear power plant has hundreds of subsystems and thousands of facilities.The real-time monitoring system has twenty thousands of monitoring parameters.It is very difficult to build up the accurate models in this huge database.But the fault diagnosis based on data driven can deal with the data information of process without the accurate model.As the important part of data driven fault diagnosis,SVM(support vector machine)based on the structure risk minimizing principle and can work out the dimensional problems.In addition,it needs a few samples.So it has excellent value of theoretical study and application.This article studies the nuclear power plant fault diagnosis system by SVM and other diagnosis methods based on data driven.A variety systems of nuclear plant exist many nonlinear relationship,but the principal component analysis lacks of ability to deal with nonlinear relationship data in the abnormal monitoring.So this article uses the kernel principal component analysis(KPCA)to boost the ability of principal component analysis.The KPCA projects the data from input space to feature space through a nonlinear mapping function and extracts the principal information.This can increase the linear separability of samples.And we calculate the T~2 and SPE statics to monitor the malfunctions whether happened.It's always a difficult problem to gain the samples when diagnosis with the nuclear plant.Because the nuclear power plant equipment has high reliability and the data samples can only be gained by simulation.So we use the SVM to diagnosis and predict the trend in research.The SVM has high accuracy in classification of small data problems.Besides,SVM has high ability to deal with nonlinear data and has strong generalization ability.So SVM can solve the problems of the insufficient of sample data.In this paper,we study the design of nuclear power plant fault diagnosis system by SVM and some data driven approaches.This article uses the CPR1000 simulator of Hongyanhe to simulate the six kinds of accidents to do the jobs of the nuclear power plant abnormal monitoring,fault diagnosis and prediction.The results of diagnosis and prediction is good.