Modeling and measuring the effects of imprecision in accident management

Abstract

This paper presents two approaches for evaluating the uncertainties inherent in accident management strategies. Current PRA methodology uses expert opinion in the assessment of rare event probabilities. The problem is that these probabilities may be difficult to estimate even though reasonable engineering judgement is applied. This occurs because expert opinion under incomplete knowledge and limited data is inherently imprecise. In this case, the concept of uncertainty about a probability value is both intuitively appealing and potentially useful. This analysis considers accident management as a decision problem (i.e. “applying a strategy” vs. “do nothing”) and uses an influence diagram. Then, the analysis applies two approaches to evaluating imprecise node probabilities in the influence diagram: “a fuzzy probability” and “an interval-valued subjective probability”. For the propagation of subjective probabilities, the analysis uses a Monte-Carlo simulation approach. In case of fuzzy probabilities, fuzzy logic is applied to propagate them. We believe that these approaches can allow us to better understand uncertainties associated with severe accident management strategies, because they provide additional information regarding the implications of using imprecise input data.