Dynamic Liquid Level Prediction of Steam Generator under Main Steam Pipe Rupture Accidents

To enhance the real-time and accurate monitoring of liquid levels in steam generators during main steam pipe rupture accidents, thereby ensuring the safe operation of nuclear power systems, a dynamic liquid level prediction method is proposed. First, experiments simulating main steam pipe rupture conditions are conducted using a scaled model of the AP1000 steam generator. This involves the integration of electric ball valve control and high-speed camera image recognition to collect data on liquid levels and key thermal parameters. Next, a liquid level time series dataset is constructed, followed by wavelet decomposition and correlation analysis to examine the time-frequency characteristics of the liquid level itself and its relationship with thermal parameters. Finally, a deep learning liquid level prediction model based on Informer and DLinear is established to perform a comparative analysis of the prediction results. The results indicate that the DLinear model outperforms the Informer model in terms of prediction accuracy and model robustness, accurately reflecting the characteristics of severe liquid level fluctuations and demonstrating its suitability and advantages in handling long-term sequence dependency issues. The DLinear model improves the mean squared error, mean absolute error. and coefficient of determination by 24.9% 16.0%, and 9.3%, respectively, compared to the Informer model. It achieves a prediction accuracy of 81.5% within a ±5 mm error range. capturing detailed changes in liquid levels while exhibiting stronger robustness and generalization ability. This study verifies the efficiency and engineering application potential of the DLinear model in liquid level prediction tasks, providing technical support for accident warnings and intelligent monitoring in nuclear power plants.