Control rod position reconstruction based on K-Nearest Neighbor Method

Abstract

Regression method based on Radial Basis Function Neural Network (RBFNN) has been applied to reconstruct the control rod position from the fixed in-core neutron detector signals, and the raw reconstructed rod position which has a decimal part should be rounded into integer. In this paper, K-Nearest Neighbor (K-NN) classification method is utilized to obtain the reconstructed rod position in the form of integer directly. Neutronics code SMART is used to simulate in-core neutron detector signals for different rod position, and noises are added to the simulated signals to generate training data set and test data set. The simulation results show that the K-NN method can reconstruct the control rod position accurately, and modification strategy based on calibration factor is used to improve the rod position monitoring accuracy when there are mismatches between actual physical factors and modeled physical factors.