Development of Efficient Global Sensitivity Analysis Method for BEPU in Nuclear Power Plant

The best estimate plus uncertainty (BEPU) analysis is recommended by IAEA for safety analysis of nuclear power plant accidents, and two important parts are included in the BEPU analysis, that is, the uncertainty analysis and the sensitivity analysis. Among them, the sensitivity analysis mainly aims at evaluating the influences of the inputs on the target outputs, and ranking the importance of the input parameters. There are many sensitivity analysis methods available for the analysis of nuclear power plant transients, and the local sensitivity analysis based on linear or monotonous hypothesis is often adopted, which is not applicable for complex nuclear power systems. Meanwhile, the global sensitivity analysis method can hardly be adopted in nuclear power systems since expensive computational cost requires to obtain reliable sensitivity measures. For the purpose of developing a global sensitivity analysis method with relatively low computational cost and high accuracy for the BEPU analysis of the nuclear power system, the optimization study was carried out in the research to decrease the computational cost of the momentindependent method. First of all, the fivepoints Gaussian quadrature scheme was adopted to replace the integral calculation of the momentindependent sensitivity measures. Then, the high dimension model representation technique was utilized to reduce the dimension of the input space, so that the sensitivity analysis can be carried out based on the assumption of input parameter independence. The Pearson system was then utilized to estimate the cumulative distribution function of the target output. The momentindependent sensitivity measure of each input parameter can finally be determined based on the moment estimation algorithm. The total computational cost of proposed efficient momentindependent global sensitivity analysis method is very low, and the required number of code calculations is only about five times the number of the input parameters. In order to validate the proposed method, one numerical example and two typical engineering tests were utilized to evaluate the applicability and reliability of the efficient global sensitivity analysis method in the simulation of nonlinear or nonmonotonic systems, and satisfying results were obtained in these test cases. Finally, the proposed technique was applied to the sensitivity analysis of an accident transient of the nuclear power system, namely the LOFT (lossoffluid test) large break loss of coolant accident. In order to further validate the accuracy of the proposed method in the simulation of real nuclear power system, more than one million code calculations were carried out to estimate the realistic momentindependent sensitivity measures. Results show that the proposed method can accurately identify the important parameters in the accident scenario, as well to rank the importance of the parameters.