Specialists from the NRC "Kurchatov Institute" have developed a neural network model for nuclear safety analysis during severe beyond-design-basis accidents at nuclear power plants. The development is intended for modeling situations accompanied by the destruction and melting of the reactor core with loss of coolant.
Modeling such accidents requires iterating through tens of thousands of possible system states, which makes the task extremely resource-intensive. Traditional analysis methods, based on detailed physical simulations and calculations, take hours or even days, making them difficult to apply for operational response.
The new development by scientists significantly speeds up this process. The neural network analyzes thermophysical parameters and is capable of accumulating "experience," extrapolating it to identify nuclear-hazardous states that might have been previously missed. This makes it possible to more accurately predict the parameters of critical situations and take timely measures to manage the accident.
As explained by Alexander Glazkov, a researcher at the center, the peculiarity of the methodology lies in creating separate neural network models for each stage of accident development, rather than a single universal model, which allowed for a reduction in required computational resources.