Simulation Optimization Framework for Dynamic Probabilistic Safety Assessment of Nuclear Power Plants

Abstract

Probabilistic safety assessment (PSA) is a useful and comprehensive method to quantify the risk of complex systems such as nuclear power plants (NPPs). PSA based on event trees and fault trees provides not only numerical risk but also an integrated understanding of risk and safety issues. Due to such advantages, thus, PSA has been widely used in the risk assessment of NPPs in recent decades. Current PSA is a static approach by nature, PSA has limitations to consider time-dependent scenarios and dynamic interactions in the risk assessment. For that reason, dynamic PSA has been introduced and studied to complement static-based PSA by considering the dynamic characteristics in the accident sequences. In contrast to static-based PSA, dynamic PSA has been introduced as a complementary methodology that considers dynamic scenarios with time-dependent interactions between the system and human operations in physical modeling such as thermal-hydraulic simulation for risk assessment. Dynamic PSA does not require conservative assumptions such as sequence grouping, success criteria, mission time, etc., moreover, it increases the realism by considering the above assumptions in dynamic scenarios directly in the risk assessment. Hence, dynamic PSA has many advantages such as finding previously unidentified dynamic risk contributors, estimating the risk for the dynamic scenarios realistically, and focusing on the behavior of physical systems or human actions in the risk assessment. However, the various research on dynamic PSA has a common challenge in that the number of dynamic scenarios to be analyzed increases exponentially. Thus, simulating all possible scenarios, taking into account the dynamic interactions between random failures of hardware, failure times of events, recovery actions by operators, and uncertainties in the full-scope system simulation model, is highly impractical. An approach is therefore necessary to manage the number of simulations for performing dynamic PSA efficiently. The objective of this paper is to propose an efficient framework to reduce the massive number of simulations for realistic risk assessment while providing both qualitative and quantitative risk insight like the static-based PSA. The proposed simulation optimization framework uses an optimization algorithm to reduce, as much as reasonably achievable, the large number of dynamic scenarios to be evaluated in dynamic PSA. This framework includes a total of 6 steps. In the first and second steps, an initiating event for risk assessment is selected, and the event sequences and system failure regarding the initiating event are analyzed, respectively. In the third step, possible dynamic scenarios are generated for performing dynamic PSA. The next step is optimizing the number of simulations via an optimization algorithm. The optimization algorithm is developed with two options in consideration of efficiency and coverage (also called accuracy in this work): 1) relatively low efficiency and high coverage, and 2) high efficiency and reasonable coverage. Also, various optimizing factors such as depth, simulation cost, conditional core damage probability, and coverage can be utilized to optimize the number of simulations in order to assess the appropriate level of the risk. In the fifth and last steps, it is possible to quantify the risk, develop the dynamic event trees, and visualize the results such as the core damage domain based on the results from the optimization algorithm. To demonstrate the application of the proposed framework to dynamic PSA, two case studies were conducted considering loss of coolant accidents (LOCAs). The first and second case studies consider small- and large- LOCAs, respectively. In the case studies, simulations were performed based on TH models that physically modeled for Zion NPP, and a total of six steps of the framework and the results of each step are covered. The application allowed the simulation optimization framework to evaluate the risk by optimizing the number of simulations according to two options. In the case of option 1, the simulation cost was 46% and 54% at resolution level 3 of small and large LOCAs, respectively, and in both cases, 100% coverage was achieved. In the case of option 2, the simulation cost was 13% and 26%, respectively, under the same conditions, and in both cases, coverage of more than 99% was achieved. In another case study considering only dynamic scenarios in small LOCA, 100% coverage was achieved with about 48% simulation in option 1, and more than 96% coverage was achieved in option 2 with 13% simulation. Through case studies, it has been shown that CCDP can be quantitatively evaluated like static PSA or dynamic ET can be developed, and it has also been established that it is possible to provide qualitative and quantitative insight by analyzing dynamic scenarios with high efficiency in simulations.