Using Bayesian Belief Networks to Predict Radioactive Releases after a Severe Accident in a BWR (PSAM-0144)

Abstract

The paper deals with an application within the STERPS project (Source Term Indicator Based on Plant Status) which is part of the 5th Euratom Framework Program. The project aims at developing a computer based tool for the rapid and early diagnosis of plant status and for the prediction of environmental releases. The approach is based on a probabilistic plant model using the Bayesian Belief Network (BBN) methodology. A BBN models relations between variables which are relevant to some problem. In this way, meaningful results can be obtained in spite of incomplete or uncertain information. In the STERPS application, the outcome is typically a number of possible plant states ranked according to probability, each with an associated environmental source term. The paper describes the development of a BBN for the Swedish boiling water reactor (BWR) Oskarshamn 3. The analysis used the generic BBN software Netica (developed by Norsys Inc.), with the user interface SPRINT, which was developed within the project for handling of the BBN. The user interface includes a set of questions and background information, which are used in order to gain information about crucial plant parameters during the course of a severe accident. SPRINT also includes graphical presentation of analysis results, both in terms of node probabilities and as characteristics for radioactive releases (amount, composition, and timing). The customization to the Oskarshamn 3 nuclear power plant included identification of key plant parameters for inclusion in the BBN, through a systematic consideration of fission product transport and retention phenomena in plant. Plant systems for mitigation of accidents as well as implemented severe accident management strategies at the plant were also considered. The functionality and practicability of the SPRINT software is being demonstrated in the recently started EU project EURANOS (European approach to nuclear and radiological emergency management and rehabilitation strategies). For the Oskarshamn 3 plant, SPRINT has been tested in connection with an emergency exercise at Oskarshamn 3. Preliminary results and conclusions show that the project has successfully demonstrated the suitability of the BBN technique for modeling the complex conditions after a severe accident in a nuclear power plant. The user interface is simple, and after some adaptation SPRINT will be suitable for use in plant technical support centers and at national emergency centers. The plant BBN models will also be a very useful tool for training and education. The prediction capabilities of the resulting models can be efficiently verified using results from plant PSA quantification and from accident analysis codes. In conclusion, the described technique has proved to be a very promising prediction tool for plant status diagnosis and estimation of the source term in severe accident situations.