Treatment of epistemic uncertainties with non-probabilistic approaches in applications of Probabilistic Risk Assessment
Abstract
In Nuclear Power Plants, Probabilistic Risk Assessment (PRA) insights contribute to achieve a safe design and operation. In that context, decision making process must be robust and uncertainties must be taken into account. In the current approach of PRA, parameter uncertainty due to a lack of knowledge is generally represented by a log-normal distribution. However, resorting to a specific probability distribution for PRA parameter uncertainty representation seems to be questionable and could lead to ambiguous results in decision making in some cases. More recently, other mathematical frameworks have been proposed to represent parameter uncertainty in a more appropriate manner. In this paper, we will discuss some important issues in parameter uncertainty representation and study two non-probabilistic frameworks which have been recently proposed to be used in risk assessment: a hybrid (i.e. probabilistic-possibilistic) framework and the Dempster-Shafer theory of evidence. The Dempster-Shafer theory of evidence framework is recognized in this paper as the most attractive framework with regard to hybrid framework and traditional probability framework in the context of PRA. A unified Dempster-Shafer representation for parameter uncertainty is proposed to deal with current issues and this representation is studied in some PRA practical examples.