Scientists at the National Research Centre Kurchatov Institute in Russia have developed a neural network that analyses thermophysical parameters to simulate severe accidents at nuclear power plants.
Nuclear safety analysis of NPPs requires the creation of mathematical models of reactor cores for various operating modes. The most difficult task is modelling of severe design accidents. In this case, the core is destroyed or melted, and the coolant density is redistributed. Under such conditions, it is necessary to accurately calculate how the neutron multiplication factor of particles that cause a chain reaction of fission of uranium nuclei will change.
Nuclear safety analysis must take into account actual physical processes with a minimum number of assumptions. For this purpose, certified neutron physics programs are used that correctly model the spatial-energy distribution of neutrons based on data that describes the state of the system as fully as possible. However, the more accurate the model, the more computing resources it requires. This makes the task of analysing severe accidents practically unsolvable without modern technologies.
The Kurchatov Institute has developed a technique using an artificial neural network (ANN), which makes it possible to conduct a realistic analysis of nuclear safety using elements of artificial intelligence at all stages of serious accidents, where the number of possible states of the system amounts to tens of thousands. The peculiarity of the neural network is that it can accumulate “experience” and extrapolate it to identify nuclear-hazardous conditions that may have previously been missed. This makes it possible to more accurately predict the parameters of nuclear hazardous situations and take measures to manage non-design accidents. The use of ANN significantly reduces calculation time.
“The analysis of nuclear safety in non-design accidents is based on the methodology of basic states,” said Alexander Glazkov, a researcher at the department for analysing the nuclear safety of nuclear energy facilities at the Kurchatov Institute. “Its essence is to find basic states in which the geometry of the system remains unchanged, but thermophysical parameters can change within established limits. The model we developed is intended for a basic state in which fuel assemblies have not yet collapsed.”
He continued: “At the same time, we found that creating and training one universal ‘large’ ANN model that would work for all basic states requires a huge amount of resources. Therefore, it was decided to develop separate models for the corresponding basic states. As part of the continuation of this work, ANN models have already been tested and debugged for other basic states at various stages of destruction of the reactor core.”
