Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Forecasting severe nuclear accidents

Thesis/Dissertation ·
OSTI ID:5559634

The occurrence of severe accidents in commercial nuclear power plants is forecast. A Random Escalation Model (REM) is proposed that uses accident data of both high and low levels of severity to forecast the most-severe ones, which are viewed as escalations of low-severity accidents. Information from safety studies is coherently combined with the data using Bayesian probability theory. REM forecasts are more accurate in the sense that their uncertainties are less than those that only include accidents of the same severity. The special structure of the REM includes other forecasting models such as those derived from the Event-Tree model and the NRC models for accident consequences. By simple manipulation of the influence diagrams of REMs we gain considerable insights and derive two results: A Separable Updating Theorem that allows the parameters of a general REM to be updated in a simple form, and a Level Absorption Theorem that shows that the REM structure is preserved under change in number and classification of levels. Methods of the probabilistic risk analysis adopted by the U.S. Nuclear Regulatory Commission are shown to be asymptotic cases of a consequence model that has the REM structure.

Research Organization:
California Univ., Berkeley, CA (USA)
OSTI ID:
5559634
Country of Publication:
United States
Language:
English