Predicting low-probability/high-consequence events
PRAs often require quantification of the probabilities of various low-probability events, such as accident-initiating events and hardware-fault events. A two-stage Bayes/empirical Bayes data pooling procedure is presented for use in combining as many as five different types of relevant data. A Poisson model is assumed for the event in question. Empirical Bayes methods are used to determine the population variability curve, while Bayesian methods are used to specialize this curve to the specific event in question. The procedure is illustrated by an example in which we estimate the probability of failure of a hypothetical large dam based on (1) a deductive event-tree-type analysis of the probability, (2) historical US dam failure data, (3) the opinions of a committee of several dam experts, and (4) the operating history for the dam in question. A Stage-2 posterior distribution is produced which incorporates these data sources. Similar distributions are produced for various combinations of data types and used to assess the contribution of each data source.
- Research Organization:
- Los Alamos National Laboratory (LANL), Los Alamos, NM (United States)
- DOE Contract Number:
- W-7405-ENG-36
- OSTI ID:
- 5152370
- Report Number(s):
- LA-UR-82-1668; CONF-820656-2; ON: DE82018373
- Resource Relation:
- Conference: Workshop on low-probability/high consequence risk analysis, Arlington, VA, USA, 15 Jun 1982
- Country of Publication:
- United States
- Language:
- English
Similar Records
NONPARAMETRIC EMPIRICAL BAYES ESTIMATION THROUGH DECONVOLUTION IN PROBABILISTIC RISK ANALYSIS
Empirical Bayesian inference and model uncertainty