抄録
Probabilistic Risk Assessment (PRA) is an important tool for evaluating risk in nuclear power plants. Dynamic PRA (DPRA) is an extension of traditional PRA methods that account for dynamic and phenomenological effects associated with time-dependent complex dynamic systems. Here we investigated dynamic event trees and optimization per focus on identifying the highest probability of system failure. Using a Branch-and-Bound (B&B) algorithm that relies on and develops bounding functions to prune or delete branches that will not yield the optimal solution (i.e., clad failure). The approach also used novel LENDIT (length, energy, number, distribution, information, time) metrics and S2R2 (state, systems, resources, response) sets to support an expert-based approach that is linked to constraints per use of the B&B algorithm. Results to date indicate that this approach is effective in reducing simulation time and thus mitigating the state explosion of thermal-hydraulic states. The work demonstrated the ability to evaluate uncertainty such that a risk-informed, quantitative PIRT (phenomena identification ranking table) is generated. Quantitative PIRT (QPIRT) can be used to improve models and identify validation needs with respect to risk. Two case studies using reference PWR and BWR plant configurations under SBO were evaluated. The implementation of the B&B algorithm yielded a significant reduction (>50%) in simulation costs. QPIRT ranking for the PWR showed that lower fidelity models combined with system redundancy produces adequate results with respect to risk. For BWR SBO, the modeling uncertainty does not present a challenge with respect to risk. Recovery of the SBO requires either restoration of the AC power or activation of firewater injection combined with operator action to depressurize the system through the automatic depressurization system. The timing of both automatic and manual safety system actuation is indeed critical to the outcome reactor state. Reactor power and firewater injection capacity provided the highest degree of correlation to model success.
