Development of a neural simulator for research reactor dynamics

A research reactor simulator was designed and developed by neural networks model. This simulator can predict the reactor power and temperatures (fuel, clad and coolant) in normal and accident condition considering reactivity feedbacks. The main advantage of this method, as compared with custom calculational methods (simulation with PARET and RELAP) is real time simulation without the need for much skilled or experienced setting. The response of benchmark reactor core predicted by neural simulator was compared to that obtained from PARET code and close agreement was observed.