Success Criteria Analysis using Deep Neural Network and Monte Carlo Dropout for Dynamic Probabilistic Safety Assessment

Heo, Yunyeong; Bae, Junyong; Jo, Wooseok; Lee, Seung Jun

Permalink

https://hdl.handle.net/2142/121804 Copy

Description

Title

Success Criteria Analysis using Deep Neural Network and Monte Carlo Dropout for Dynamic Probabilistic Safety Assessment

Author(s)

• Heo, Yunyeong
• Bae, Junyong
• Jo, Wooseok
• Lee, Seung Jun

Issue Date

2023

Keyword(s)

• Dynamic probabilistic safety assessment
• Deep neural network
• Monte Carlo dropout
• Success criteria
• Simulation

Abstract

Dynamic probabilistic safety assessment (PSA) performs analysis and simulation of a large number of scenario sequences, unlike legacy PSA. When performing safety assessments that reflect dynamic characteristics, success criteria such as allowable time are of great importance. However, numerous scenarios have different success criteria. When a lot of variables that affect the state of a power plant change, success criteria can also change significantly. Therefore, when performing dynamic PSA and interpreting the results, if we try to assign meaning to each scenario (simulation calculation), we must find success criteria for each scenario. On the other hand, if we try to assign meaning by merging scenarios, we need to derive the range (bound) of allowable time. Therefore, we must find success criteria for a large number of scenarios that are affected by numerous variables. Thus, our research team proposes a study to explore success criteria for dynamic PSA. This study utilizes an optimization algorithm called Deep-SAILs. By using Monte-Carlo dropout and deep-learning, this algorithm efficiently finds the limit surface to determine success criteria for numerous scenarios. To do this, we need to perform scenario analysis that is suitable for the purpose of conducting dynamic PSA. During this process, we extract variables that require analysis and have a significant impact, set their bounds, and find the limit surface called success criteria. Moreover, the criteria can be any parameter in the simulation system code such like peak cladding temperature, pressurizer pressure. To efficiently find the limit surface, we connect system codes like MAAP with Deep-SAILs to perform simulation and inference model calculations. This research aims to perform more complete safety assessments by deriving scenario-specific success criteria that sufficiently reflect dynamic meaning, thereby reducing the number of simulations needed to interpret the results of dynamic PS

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121804

Owning Collections