Machine Learning Method for Validation of the Computational Model of Nonstationary Xenon Processes in the VVER Reactor Based on the Algorithm of Separation of Variables

In the paper, we present a method for validating the KORSAR/GP software package in terms of a mathematical model of nonstationary xenon processes in a VVER reactor that is based on the separation of spatial and temporal variables. The data obtained from various high-power VVER installations in experiments to study the spatial distribution of energy release under nonstationary reactor poisoning conditions caused by the action of various regulators are used. The model is based on the classification of means of affecting reactivity by the type of variable in energy release, which undergoes changes significant for the process as a result of this impact. Nonstationary xenon poisoning processes, which involve control rods of the control and protection systems and water exchange operations with a change in the concentration of boric acid, as well as both of the listed methods, both in the presence of a change in the neutron power of the reactor and when maintaining its constant value, are considered. A machine learning method on the basis of regression analysis making it possible to estimate the error in calculating the parameters of the energy release field under conditions of spatial, temporal, and spatiotemporal feedback of the xenon concentration and regulators is developed. On the basis of the processed experimental data, a training array, which is used for machine learning of this model, is formed. As a result of the developed algorithm, an error estimate for the model of the computing code with allowance for the partial impact of various means of changing the reactivity in a given calculation is made.