Comparison of back propagation network and radial basis function network in Departure from Nucleate Boiling Ratio (DNBR) calculation

Abstract

Abstract Since estimating the minimum departure from nucleate boiling ratio (MDNBR) requires complex calculations, an alternative method has always been considered. One of these methods is neural network. In this study, the Back Propagation Neural network (BPN) and Radial Basis Function Neural network (RBFN) are introduced and compared in order to estimate MDNBR of the VVER-1000 light water reactor. In these networks, the MDNBR were predicted with the inputs including core mass flux, core inlet temperature, pressure, reactor power level and position of the control rods. To obtain the data required to design these neural networks, an externally coupledcode was developed and its ability to estimate the thermo-hydraulic parameters of the VVER-1000 reactor was compared with other numerical solutions of this benchmark and the Final Safety Analysis Report (FSAR). After ensuring the accuracy of this coupled-code, MDNBR was calculated for 272 different conditions of reactor operating, and it was used to design BPN and RBFN. Comparison of these two neural networks revealed that when the output SMEs of the two systems were approximately the same, the training process in RBFN was much faster than in BPN and the maximum network error in RBFN was less than in BPN.