No abstract

Abstract-This work investigates the use of Artificial Neural Networks (ANN) for radiation dose prediction due to a nuclear power plant (NPP) accident with radioactive material release. The main objective is to avoid necessity of using complex timeconsuming simulators during the emergency. Training, test and production data sets have been generated by realistic simulations on the precise atmospheric dispersion system used in CNAAA Brazilian NPP. Considering a hypothetical Lost of Coolant Accident (LOCA), several ANN architectures have been trained with a wide range of atmospheric conditions in order to predict spatial effective doses. As a result, a Backpropagation Multilayer Perceptron (MLP) with 5 layers demonstrated to achieve the best generalization, reaching a correlation factor of 0.990 for the validation dataset. On the other hand the GRNN reached a correlation factor slightly worse (0.986) but very faster.

No abstract