Recurrent Neural Network Model for Signal Prediction in a LOCA Event

Abstract

In this study, an artificial neural network (ANN) model with the Elman recurrent structure is developed to predict thermo-hydraulic signals in a loss of coolant accident (LOCA). In the prediction procedure, a few previous samples of a signal at the time of the accident are fed to the ANN and the output value of next time steps are estimated by the network output. The model is used to predict signals after a LOCA in a water-water energetic reactor (VVER). Trained networks then are utilized to estimate signals after a LOCA in a pressurized water reactor (PWR). Networks are trained with the data obtained from the benchmark RELAP5/MOD 3.2 simulation of the event. Results show that Elman recurrent neural network model can predict thermo-hydraulic signals at the LOCA event in the pressurized water reactor well. This fast alternative method could guarantee the safe and reliable exploration of nuclear power plants at an accident time. This tool is especially useful for damaged signals.

Keywords: Safety, loss of coolant accident, artificial neural network, Elman recurrent network.