Dynamic Risk Assessment through Automatic Accident Sequence Generation using Optimized Simulations of Nuclear Power Plants

Abstract

Probabilistic safety assessment (PSA), as currently used to evaluate nuclear power plant (NPP) safety, is presented as a comprehensive and systematic method to analyze the types of accidents that can occur at NPPs, the likelihood of these accidents occurring, and the impact of the accidents by quantifying them in a probabilistic manner. To assess NPP safety, PSA assumes specific scenarios and analyzes safety for the representative scenarios using conservative assumptions for complex cases. Although event trees (ETs) are used in PSA to model accident scenarios represented by binary failure and success based on conservative success criteria for safety systems and operator actions, ETs have limited capacity for analyzing dynamic characteristics such as the stochastic processes, operator response times, and other time-dependent factors. Recently, dynamic PSA has been performed to evaluate realistic risk assessment considering dynamic accident scenarios using dynamic event trees (DETs), which have been commonly used in dynamic PSA methods. This is expected to clarify the differences between the current static-based PSA and dynamic PSA reflecting dynamic characteristics. In a DET, branching occurs at user-specified dynamic conditions, such as operator response times and timings of safety system operations. While this leads to a more realistic and mechanistically consistent analysis of the system under consideration, the number of branches may become extremely large when the branching is generated based on various factors. This results in poor visibility of the accident sequences due to its extensive size and also causes an increase in the number of simulations required for estimating risk. Furthermore, quantifying risk in dynamic PSA remains challenging because adequate operator response models that can provide a branch probability according to the timing of operator action in dynamic scenarios have not yet been addressed. To overcome these challenges associated with dynamic PSA, this paper proposes a novel dynamic PSA framework, a dynamic risk assessment through automatic accident sequence generation using optimized simulations of nuclear power plant (DRAGON). DRAGON consists of three modules: Optimized simulation, automatic accident sequence generation, and dynamic risk assessment. First, optimized simulation module employes a simulation optimization algorithm, which is an iterative process to identify limit surface (i.e., the boundary between success and failure scenarios in uncertain scenario domain), focusing simulations on the scenarios near this boundary. Second, automatic accident sequence generation module analyzes the optimized simulation results and provides the branching points using the alpha shape method, which can adjust the level of detail using an alpha parameter, and generates dynamic accident sequences by controlling the number of branching points to manage sequence complexity with maximized scenario coverage. Last, dynamic risk assessment module evaluates the time-dependent human reliabilities using time available as dynamic features and human reliability analysis data. Dynamic risk is then quantified by assigning the occurrence probabilities to each branch of dynamic accident sequences, that is generated from automatic accident sequence generation module, using time-dependent human reliability models. To demonstrate the application of the proposed framework to dynamic PSA, two case studies were conducted considering loss of coolant accident (LOCA) and station blackout (SBO). The results showed that DRAGON can effectively reduce the required number of simulations and provide dynamic accident sequences with controllable complexity. Additionally, the risk was evaluated more realistically than static PSA. Based on these findings, DRAGON is expected to support flexible decision-making concerning dynamic accident sequences. Furthermore, for the practical validation and application of the proposed framework, a platform was developed and implemented to automate the DRAGON workflow.