Prediction of Pressurized Water Reactor Core Design Parameters Using Artificial Neural Network for Loading Pattern Optimization

The loading pattern optimization for reactor core is to find the most economical loading pattern of all loading patterns, satisfying the safety restriction requirements. If the reactor core design parameters for all loading patterns could be computed with numerical analysis codes, the optimal loading pattern could be obtained. However, it is limited because of the huge computation time. Therefore, various methods such as Simulated Annealing (SA) algorithm [1] have been studied to effectively perform calculations for finding the optimal loading pattern. If these design parameters such as the peaking factor and the cycle length could be calculated faster than the numerical analysis codes, the optimal loading pattern could be found faster. In this study, reactor core design parameter prediction algorithms using an Artificial Neural Network (ANN) has been developed to replace the numerical analysis codes in the loading pattern optimization. A deep learning algorithm was used to improve the accuracy of predictions because it can solve more complex and nonlinear problems using multiple hidden layers for feature extraction and transformation. In addition, a system has been developed that automatically generates training data for predicting the reactor core design parameters using data from the Westinghouse 2-loop plant. In this paper, the peaking factor and the cycle length according to the loading patterns were predicted using the deep learning algorithms and automatically generated training data.