(A) study on the uncertainty propagation of design and operational parameters of nuclear power plant

Abstract

For the best estimation of the uncertainties propagated from the design and operational parameters of the nuclear power plant, various methods of uncertainty analysis are developed. In Part I, to evaluate the influence of the uncertainties of design parameters, two methods which can estimate both the propagated uncertainty and sensitivity simultaneously are developed by combining and improving conventional methods. First, a two-step method using the Response Surface Method and the Monte Carlo Simulation Technique is developed. This method samples in multiple parameter space by well-chosen experimental design technique and then obtains response surface polynomial by a statistical regression technique. From this response surface, the uncertainty distribution of the output variable is obtained by Monte Carlo simulation. The other method is a sequential method which uses the Fourier Amplitude Sensitivity Test and the Stepwise Regression Technique. This method samples in the uncertainty space of input parameters according to a preselected integer frequency set. The uncertainty of output variable and the influence of each input parameter are then evaluated by Fourier Amplitudes, and finally a reduced nonlinear regression model is constructed by the Modified Stepwise Regression Technique. These methods are applied to the thermal margin analysis based on DNBR (Department from Nucleate Boiling Ratio) and LOCA (Loss Of Coolant Accident), which are the most important factors in deciding the operating margin of the nuclear power plant of pressurized water reactors, and applied also to risk assessment of the high level radioactive waste repository system, which has highly nonlinear behavior. These applications are used to compare the characteristics of each method. In Part II, to evaluate the influence of the uncertainties of operational parameters in on-line environment, an improved algorithm which can estimate state and noise statistics simultaneously is developed with Ad...