Data Reliability Analysis during Containment Leakage Test Based on Statistical Software R

The most important part in the calculation of the containment leakage is to perform the linear regression on time for a series of data measured at different times. The significance test of the regression and residual analysis are the substantial means to evaluate the test results. This paper analyzes the data of the containment test during the commissioning and startup phase of a power plant based on the statistical software R, and explores the regression diagnosis before the leakage calculation by examining the independence, normality and heteroscedasticity of the regression model and the elimination of extreme sample points impact on the reliability of the result. Through the regression diagnosis on the examples, it was found that in the samples which leakage rate is calculated, there may be problems that affect the regression results and then the affect the final results, such as autocorrelation, non-normality and heteroscedasticity. Therefore, the validity of the data shall be evaluated by the regression diagnostic methods, while calculating the leakage rate, and the final results shall be corrected by appropriate methods for samples that fail the test.