The influence of model uncertainty is most pronounced in areas of limited knowledge and large uncertainties like severe accident (SA) calculations. Lack of a systematic methodology for this purpose makes this assessment difficult. This paper describes the treatment of model uncertainty in SA analysis for nuclear power plants, which is an area that has had limited past research. This paper aims at a systematic subject assessment. By review of available approaches, a methodology is structured to deal with alternative modeling options in SA code structure. The proposed methodology comprises three phases: the probability of each model is estimated (phase 1), the input uncertainty is quantified (phase 2), and the Bayesian model averaging technique is utilized to integrate the calculations of alternative models into the SA code (phase 3). Through this process, the degree of belief is quantified for the performance of alternative code models. The methodology evaluates available information and data from experiments and code predictions. The application of the proposed methodology is demonstrated on fission product release models for the LP-FP-2 SA experiment of the LOFT (Loss-of-Fluid Test) facility.
