Fault Diagnosis Of Condenser Based On Principal Component Analysis&RBF Neural Network

Condenser is an important auxiliary equipment of thermal power plant, its performancedirectly affects the economic benefit and safety of the power plant，the study on faultdiagnosis methods of condenser has a very important theoretical significance andengineering value.This paper mainly use the history fault diagnosis data of condenser in Shajiao C powerplant’SIS system to generate fault diagnosis and fault verification data source data for RBFneural network,after the principal analysis of data to remove redundant data informationand then form the fault diagnosis data samples of RBF neural network. After determiningthe structure of RBF neural network, train the RBF neural network through numericalcalculation software Matlab and built fault pattern recognition RBF network. And we canconsider the accuracy of model through fault verification data.In this paper, I analyze theimpact of noise on the principal component analysis method, by means of adding differentSNR data to the fault diagnosis signal and then comparing changes of accumulatedvariance contribution rate at the same number of principal component. As we know, there isrelationship between the data is the premise of using principal component analysis method,analysing association between the symptom parameters through the correlation coefficientbetween them. This paper also uses the original training sample and the sample by principalcomponent analysis optimization respectively to train RBF neural network, and considerthe performance of based on principal component and RBF network.By means of the diagnosis examples in Shajiao C power plant’s condenser, the dataredundancy problem can be solved effectively when there is a linear correlation betweenthe data. And if the number of principal component is appropriate,the accuracy ofclassification of radial basis function neural network is also very high.And because of thedata reduction, there is a certain degree of the computing time and computing speed to be improved.