Answers are grounded in this paper and cite their sources.
In this work, metamodels capable of reproducing the thermal-hydraulic behavior of subsystems of the reactor coolant system (RCS) during the early phase of a Station Blackout (SBO) leading to a severe accident—i.e., up to the point of fission product release in the RCS—are proposed as a first step toward the development of accelerated surrogate models for severe accident simulations. The work is performed in the frame of the EU-funded project ASSAS, which aims at developing a basic-principles severe accident simulator for a generic 1300 MW pressurized water reactor (PWR), by replacing ASTEC (severe accident code developed by ASNR) models with Machine Learning (ML) surrogate models. The work presented here focuses on the thermal-hydraulic CESAR module for the primary and secondary circuits. A suitable training dataset must be generated. A base case scenario is considered, involving an SBO without any safety measures until the onset of fission product release. From this base case, various calculations are performed by depressurizing remotely the primary circuit at different times, followed by the recovery of emergency water injection, also at different times. From these ASTEC calculations only the variables needed for the surrogate model development are extracted. These include state variables for each control volume within the RCS domain, boundary conditions, and additional variables that may provide information about the overall evolution of the accident and could be useful for ML. The surrogate models are expected to compute each time step while also accounting for user decisions interactively during the accident simulation. The ML method considered here is based on Recurrent Neural Networks, which are commonly used for time series. The developed LSTM-based surrogate models achieved a mean absolute error (MAE) on the order of 10 −3 on an independent test dataset. Temperatures and steam pressures were predicted with the highest accuracy, whereas void fractions and hydrogen partial pressures exhibited larger errors. Among the four metamodels, the steam generator secondary side showed the most robust predictive performance, while the primary side was more sensitive to rare events. Larger error metrics were especially found during changes in operational state.
No takes yet. Share an insight, caveat, or question.
D'Alessandro et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: