Research On Fault Diagnosis Technology Of Nuclear Power Station Based On Multi-class Support Vector Machine

Fault Diagnosis System plays an important role in ensuring the safety and economy ofnuclear reactor operation. At present, traditional threshold alarm method is mainly adopted inChina’s nuclear power stations, which can’t provide the fault’s reasons, trend and so on.Therefore, it is very meaningful to research the fault diagnosis of nuclear power station. Thesupport vector machine has the advantages that the good performance and the highergeneralization ability in solving classification problems based on limited sample informationand large dimension, and the fault samples of nuclear power station have also the samecharacteristics. Therefore, aiming to typical fault of nuclear power station’s primary loop andsecondary loop system, the nuclear power station’s fault diagnosis techniques and methodswhich are based on Multi-class support vector machine are researched and explored in thispaper in order to solve the difficulties and problems in the nuclear power station’s faultdiagnosis.Firstly, at the basis of existing research, the typical failure and phenomena of the nuclearpower station are analyzed and the fault samples are collected.Secondly, with the study of the statistical theory and support vector machine theory, amulti-classification algorithm based on the support vector machine is presented by comparingseveral multi-classifications and considering the characteristics of the nuclear power station’stypical failure, and its feasibility has been proved.Thirdly, through the research of main issues of the support vector machine, e.g. theselection of kernel function and nuclear parameters optimization, a support vector machinemulti-classification model is built in this paper by using multi-classification algorithm whichis presented in the last chapter. It is applied to the fault diagnosis of nuclear power station toachieve fault diagnosis based multi-class support vector machine, and compared with theresult of neural network classification.At last, the paper researches the forecasting technique of information granules based onthe necessity of forecasting the change of a parameter when the nuclear power station is underthe circumstance, and the technique is used in predicting the Stabilizer water level variationtendency of the SGTR which is as an example. With the experimental analysis of nuclear power station fault diagnosis system, thefeasibility and effectiveness, which the multi-class support vector machine is used in faultdiagnosis of nuclear power equipment is demonstrated.