Using advanced machine learning techniques to improve equipment health monitoring in NPPs

To improve the condition monitoring and maintenance in nuclear power plants (NPPs), we propose a hybrid condition prediction method using wavelet decomposition (WD), variational mode decomposition (VMD) and gated recurrent unit (GRU). Firstly, WD is employed to decompose the time series data into 2 subsequences of high and low frequencies respectively, enabling a better grasp of the dynamic characteristics within the original signal. Then, the VMD performs a secondary decomposition on the 2 subsequences to form multiple intrinsic mode functions (IMFs). By doing so, the complexity of the time series signal can be reduced, and this facilitates the accurate prediction of the original signal. Finally, the GRU is used to predict each IMF component, and the prediction results of the original signal are obtained by reconstructing the predictions of IMFs. The performance of the proposed method is validated by using the time series signals from reactor coolant pumps (RCPs) of a NPP, and comparisons with other two traditional single methods and two hybrid methods highlight the advantages of the proposed hybrid prediction method.