Comparative Study on Vibration Prediction Methods of Reactor Internals Based on Neutron Noise Characteristic Frequency Time-Series Signal

The vibration status of reactor internals is directly related to the operational safety and the maintenance node of nuclear power plant. Therefore, it is important to analyze and predict the vibration of these internals. This paper proposes a method for predicting the vibration of reactor internals based on the time-series signals of neutron noise characteristic frequency bands. The method, from two perspectives of single-cycle and double-cycle, utilizes statistical learning and machine learning methods for prediction, and an experiment was conducted using neutron noise signals collected from a nuclear power plant. The results show that, in terms of analysis methods, the processing of characteristic frequency band time-series signals can effectively utilize the temporal information in the signals. In terms of prediction methods, statistical learning models achieve higher accuracy for single-cycle prediction while machine learning models achieve higher accuracy for double-cycle prediction.Therefore, the combination of characteristic frequency band time-series signal analysis methods and appropriate prediction models can provide guidance for the prediction and determination of maintenance nodes in nuclear power plants.