Dynamic PRA: An Overview of New Algorithms to Generate, Analyze and Visualize Data, ANS Winter Meeting.

摘要

State of the art PRA methods, i.e. Dynamic PRA (DPRA) methodologies (1), largely employ system simulator codes to accurately model system dynamics. Typically, these system simulator codes (e.g., RELAP5 (2)) are coupled with other codes (e.g., ADAPT (3), RAVEN (4) that monitor and control the simulation. The latter codes, in particular, introduce both deterministic (e.g., system control logic, operating procedures) and stochastic (e.g., component failures, variable uncertainties) elements into the simulation. A typical DPRA analysis is performed by: 1. Sampling values of a set of parameters from the uncertainty space of interest 2. Simulating the system behavior for that specific set of parameter values 3. Analyzing the set of simulation runs 4. Visualizing the correlations between parameter values and simulation outcome. Step 1 is typically performed by randomly sampling from a given distribution (i.e., Monte-Carlo) or selecting such parameter values as inputs from the user (i.e., Dynamic Event Tree (3), DET). In Step 2, a simulation run is performed using the values sampled in Step 1). These values typically affect the timing and sequencing of events that occur during the simulation. The objective of Step 3 is to identify the correlations between timing and sequencing of events with simulation outcomes (such as maximum core temperature). In a classical PRA (event-tree/fault-tree based) environment, such analysis is performed by observing and ranking the minimal cut sets that contribute to a Top Event (e.g., core damage). In a DPRA environment, however, data generated is more heterogeneous since it consists of both: Temporal profiles of state variables Timing of specific events. The visual exploration of such data is a new research topic and it is especially relevant when uncertainty quantification is performed on many parameters for complex systems such as nuclear power plants.