Uncertainty Analysis Algorithm Research Based On Surrogate Model

Uncertainty analysis and sensitivity analysis are necessary steps for model design optimization increasingly.They aim to quantify the influence factors of input variables and obtain the varies of response function under the influences of uncertainty input parameters.In this paper,the phenomenon of unprotected loss of flow accident in molten salt reactor is taken as the research object.The sampling algorithm and meta-model algorithm are used to analyze the uncertainty and sensitivity of transient safety parameters based on five key,parameters.In this paper,five kinds of uncertainty analysis methods are studied and applied to the safety accidents of molten salt reactors.The practical application of uncertainty and sensitivity analysis is carried out,especially the global sensitivity analysis is calculated by various algorithms.The sampling algorithm mainly includes Latin hypercube sampling and sobol sequence sampling.The meta-modeling algorithms include polynomial chaos expansion method,stochastic collocation method and arbitrary polynomial chaos expansion algorithm.In the application process,the coupling process of DAKOTA/SALib and GenFlow was presented,and the arbitrary polynomial chaos expansion algorithm program was developed.The thermal parameters in the molten salt reactor were quantitatively analyzed and studied.The results show that the peak temperature depends strongly on PUI input.Input and output temperatures of CATH only rely on PUI and the other four parameters show almost the small effect.There are significant interaction effects among inputs for the Y6.Furthermore,FLP,FLD has the significant influence for DHX.The impact of VOT is minimal among all the parameters.The meta-modeling based methods need fewer samples than sampling-based method and show similar results.The polynomial chaos expansion and SC method have significant advantage in computing time,as the sensitivity analysis results are obtained from few samples.The estimation errors of SC2,PCE2,APC method are relatively large than other meta-modeling methods.The sampling method may converge slowly when only one parameter has a significant effect.If the simulation model has an unsmooth surface and jump discontinuities in the model,there could be errors for meta-based and sampling-based method.For this kind of model,the standard deviation and coefficients of variation are much larger in sensitivity indices.The large standard deviation and coefficient of variation mean the global sensitivity analysis results obtain with low confidence and significant uncertainty.Through the research in this paper,the sensitivity analysis and uncertainty analysis of nuclear safety analysis program are mastered.The analysis and development of arbitrary polynomial chaos expansion algorithm are carried out,and application in the molten salt reactor model of the uncertainty analysis algorithm based on surrogate model is completed.