An online operator support tool for severe accident management in nuclear power plants using dynamic event trees and deep learning.

Operating staffs of a nuclear power plant (NPP) are responsible for returning the NPP to a stable state and alert authorities if there is the potential for offsite radiological consequences following an accident. An operator support tool (OST) using deep learning techniques and trained by data from dynamic probabilistic safety/risk assessment (DPSA/DPRA) is proposed to assist the NPP personnel in decision-making. The DPSA/DPRA methodology employs time-dependent branching conditions based on the evolving state of the NPP and accounts for complex hardware/process/software/human interactions to predict possible outcomes of the initiating event. A large number of scenarios generated from the DPSA/DPRA performed for a pressurized water reactor station blackout as a function of time were used to train the OST to predict possible offsite dose outcomes at 2-mile and 10-mile site boundaries for emergency response planning. The results show that the OST can predict offsite dose levels with greater than 90% accuracy. [ABSTRACT FROM AUTHOR]