Development of an artificial neural network model for on-line thermal margin estimation of a nuclear reactor core

Abstract

One of the key safety parameters related to thermal margin in a Pressurized Water Reactor (PWR) core, is Departure from Nucleate Boiling Ratio (DNBR), which is to be assessed and continuously monitored during dperation via either an analog or a digital monitoring system. The digital monitoring system, in general, allows more thermal margin than the analog system through the on-line computation of DNBR using the measured parameters as inputs to a simplified, fast running computer code. The purpose of this thesis is to develop an advanced method for on-line DNBR estimation by introducing an atrificial neural network model for best-estimation of DNBR at the given reactor operating conditions. the neural network model, consisting of three layers with five operating parameters in the input layer, provides real-time prediction accuracy of DNBR by training the network against the detailed simulation results for various operating conditions. The overall training procedure is developed to learn the characteristics of DNBR behaviour in the reactor core. First, a set of random combination of input variables is generated by Latin Hypercube Sampling technique peformed on a wide range of input parameters. Second, the target values of DNBR to be referenced for training are calculated using a detailed simulation code, COBRA-IV. Third, the optimized training input data are selected. Then, training is performed using an Error Back Propagation algorithm. After completion of trainig, the network is tested on the examining data set in order to investigate the generalization capability of the network responses for the steady state operating condition as well as for the transient situations where DNB is of a primary concern. The test results show that the values of DNBR predicted by the neural network are maintained at a high level of accuracy for the steady state condition, and are in good agreements with the transient situation, although slightly conservative as compared to those p...