An analysis of machine learning for the safety justification of VVER reactors

Abstract

Contemporary approaches to the safety justification of nuclear power plants are characterized by constantly growing requirements for the volume and complexity of a computational analysis. Machine learning provides the ability to analyze large volumes of calculations. The present paper examines the methodology of a multivariate computational safety analysis based on machine learning. The approach proposed in the work will significantly reduce the time spent on labor-intensive calculations of safety justification. The selected learning model is presented with particular attention paid for its properties optimal for the solution of the considered problem. The ways of solving the problems arising during the preparation of a learning sample and model learning are described; the results of the model application for one of the typical problems including the safety justification of VVER reactors during a postulated design-basis accident using the conservative approach are provided. In conclusion, issues requiring further research, as well as the prospects for the development of the proposed methodology are indicated.