(A) study on the advanced methods for on-line signal processing by using artificial intelligence in nuclear power plants

Abstract

In order to assist the operators at the transient states of a nuclear power plant, the automation of signal processing is needed. This study has the objective to process the signals from a nuclear power plant for the purpose of advising the operator. To meet this objective, in this study, two kinds of on-line signal processing system based on AI techniques are developed for the nuclear power plant application with on-line signals. First, an artificial neural network for signal prediction is developed for the adequate countermoves at transient states. The steam generator water level is adopted as the example and the outputs of a simulation program for the dynamics of steam generator combined with noises are used as the training patterns. For the training of the artificial neural network, the modified backpropagation algorithm is proposed for escaping quickly from local minima. The modified algorithm is different from the ordinary backpropagation algorithm in the aspect that the training rate coefficient is reqeatedly adjusted randomly and taken when the training is improved. This trial has an effect to search for an adequate magnitude of a training rate coefficient. The comparison result shows that the modified algorithm enables the neural network to be trained more quickly. The simulation result shows that the outputs of the artificial neural network are not sensitive to noises. Using the artificial neural networks proposed in this thesis, the operators can predict the next status of a plant and can take actions to maintain the stability of plant. Second, the multi sensor integration system has been developed for the identification of transient states. The developed system is divided into two parts; pre-processors and a fusion part. An artificial neural network is adopted in the fusion part to include the knowledge about the identification and to make a decision of the transient state. The developed pre-processors play a role of classifying the trend types of...