The Method Of Relational Support Vector Regression And Its Small Sample Failure Prediction Of Steam Generator Parts In Nuclear Power Industry

The steam generator is an important equipment of nuclear power plant to performing energy exchange between first loop and second loop.Its parts lead to all sorts of defects easily due to the terrible working condition in the process of operation with the risk of bringing out radioactive substance.So the reliability of steam generator occupies an essential position in various nuclear power parts.The failure of steam generator unit is a typical small sample reliability problem due to the particularity of nuclear power industry.It’s difficult to collect a large amount of data to performing prediction.Therefore,the paper put forward the relational support vector regression(r-SVR)methods based on the grey relational analysis method and support vector regression(SVR)method.And put it into failure prediction of small sample failure of steam generator parts.The main contributions of the thesis are as follows:1.The small sample data mining method of grey relational analysis in the reliability analysis and the data prediction method of SVR are studied systematically.Grey relational analysis changes the data weight according to the correlation between multi index factors and key factors to mine the hidden information;SVR can train the model with given data and perform prediction with it.But the training model obtained with small sample is instability and inaccurate.2.Put forward r-SVR,which combines the advantages of the grey correlation analysis to calculate the correlation degree between the factors and optimize data weights with the advantages of SVR,which can perform prediction to improve the accuracy of the prediction ability of model.Cross validation method is introduced to optimize the parameters of r-SVR kernel function,which avoids the artificial selection and makes the model more stable.In order to make the prediction result more convincing,the RVM,PSVR and SVR are introduced at the same time.Finally,small sample prediction software based on r-SVR is developed with MATLAB GUI platform.It can achieve r-SVR,SVR,PSVR,RVM and grey correlation degree calculation.3 The r-SVR is used to predict the material properties of plunger pump suction drainage valve which is an important part of the steam generator.And the relationship between the 4 parameters of extrusion temperature,extrusion speed,quenching method and failure condition and the tensile property of the material is studied.The results compared with SVR show that the prediction performance of r-SVR is improved about 50.0%;Comparing with the PLS-BPNN algorithm,the prediction performance is improved 36.2%.4.r-SVR is used to predict the explosion pressure of the steam generator heat transfer tubes.The relationship between the three parameters of defect length,defect wrap angle,and depth of the defect with the blasting pressure is studied.The average percentage of deviation is merely 1.17%,which is improved than SVR(8.4%),PSVR(9.8%)and RVM(11.3%)significantly.The accuracy of the prediction is very reliable.5.The heat transfer tube was tested by Instron 8872 electro hydraulic servo tension compression fatigue test machine.And the r-SVR was used to predict the fatigue life of the heat transfer tube with the test data.The average percentage deviations between the predicted results and actual life expectancy is 8.3%,over the tolerance range slightly,but still within the acceptable range.The predicted results have a certain reference value for the plugging operation of actual steam generator tube.The r-SVR method is recommended in the later small sample forecasting.