Decision making under uncertainty by trustworthy dynamic bayesian networks for severe accident management in nuclear power plants

Severe Accident Management (SAM) of a Nuclear Power Plant (NPP) relies on a set of actions to mitigate the consequences of severe accidents, and recover its safe and stable state. Dynamic Bayesian Networks (DBNs) can support decision-making during accident progression and, thus, serve as Accident Management Support Tools (AMSTs). In this work, we propose a methodological framework for quantifying, in real-time, the uncertainty of the output of a DBN-based AMST to enable trustworthy decision-making with regards to the selection of the best action to mitigate the developing accident scenario. The proposed methodology is exemplified on a Loss of Coolant Accident (LOCA) in a WWER-1000 nuclear reactor. Results show that accounting for the uncertainty of the output of the DBN enables a reliable and robust selection of the proper mitigative actions to avoid severe consequences, ultimately strengthening the support for safe accident management decisions.