Prediction of steam generator liquid level under main steam line break accident based on wavelet decomposition combined with deep learning

Abstract

Liquid level monitoring is essential for maintaining the safe operation of nuclear power circuits. During a Main Steam Line Break (MSLB) accident, significant fluctuations in the liquid level within the steam generator pose challenges for traditional measurement methods, which often fail to accurately capture the true liquid level. This study conducted experiments of MSLB accidents under controlled conditions, with parameters including heating power ranging from 8 to 16 kW, break pressures from 0.05 to 0.1 MPa, and relative break sizes between 20 % and 100 %. In selected conditions, rolling motions were introduced to simulate marine environments. Wavelet decomposition was utilized to extract features at varying frequency levels, and deep learning models were employed to predict each component. The proposed approach achieved a prediction accuracy of 88.3 %, outperforming direct predictions from raw data with improvements of 21.9 % in Mean Squared Error (MSE), 12.3 % in Mean Absolute Error (MAE), and 10.0 % in the coefficient of determination (R²). The detail component cD1 was found to have the most significant impact on overall prediction accuracy, highlighting it as a key parameter for further optimization. Furthermore, the use of wavelet-decomposed data significantly reduced computational complexity, enhancing time efficiency. These results demonstrate the effectiveness of the proposed method in improving prediction accuracy and operational efficiency, offering valuable support for the safe management of nuclear power systems during MSLB accidents.