Development of convolutional neural network model in predicting SLB scram worth for safety analysis

Abstract

In this study, we developed an artificial neural network to predict SLB scram worth during accident analysis of reload core design of OPR1000 nuclear power plant. Currently, the most important step in a reload core design is to search cycle loading patterns. However, time is limited and short due to a fixed design period. In addition, due to the complicated accident analysis procedure, the load pattern is selected without checking important safety factors during the load pattern selection stage. Among the various safety factors, SLB accidents are the shortest margin. To improve such a problem, it is necessary to have the ability to quickly predict the SLB scram worth when selecting a loading model. In this research, we applied a convolution neural network for predicting the two-dimensional core power distribution, which was a previous study, and developed an artificial neural network that predicts SLB scram worth which is the next step. The data for learning was generated using code developed by KNF. A random loading pattern was produced for three cases to confirm the train data sensitivity, and the artificial neural network was trained with supervised learning. As a result, the SLB scram worth Ave. Relative Error and Max. Relative Error were predicted to be 0.398% and 3.056% for the 1 case. For 2 case, the SLB scram worth Ave. Relative Error and Max. Relative Error were predicted to be 0.353% and 3.254%. For 3 case, the SLB scram worth Ave. Relative Error and Max. Relative Error were predicted to be 0.307% and 3.579%. The developed artificial neural network can quickly check the safety factor when optimizing the loading pattern and shorten the additional feedback time due to dissatisfaction with the safety analysis. This can be of great help to designers. Not only that, it has the potential to expand into other rods worth-related accident analyses in the future, which can help simplify reload core design.