A dynamic neural network aggregation model for transient diagnosis in nuclear power plants

Abstract

A dynamic neural network aggregation (DNNA) model was proposed for transient detection, classification and prediction in nuclear power plants. Artificial neural networks (ANNs) have been widely used for surveillance, diagnosis and operation of nuclear power plants and their components. Most studies use a single general purpose neural networks for fault diagnostics with limited reliability and accuracy. The proposed system in this study uses a two level classifier architecture with a DNNA model instead of the conventional single general purpose neural network for fault diagnosis. Transients' type, severity and location were individually obtained by assigning neural networks for different purposes. The model gave satisfactory performance in the system tests and proved to be a better method from comparison. Few previous diagnostic systems focus on the prediction of transients' severity. The proposed system can provide more accurate numerical values other than qualitative approximation for transient's severity.