Pressure pulsations intelligent prediction model for load rejection of pumped storage power station based on data augmentation and one-dimensional convolutional neural network

Extreme pressure pulsations during the load rejection transitions will pose a threat to the safety of pumped storage power stations (PSPs). Fast and accurately predicting pressure pulsations in extreme working conditions is essential for power plant safety. Therefore, a pressure pulsations intelligent prediction model based on data enhancement and one-dimensional convolutional neural network (1D-CNN) is proposed. Firstly, a data augmentation method for enhancing pressure pulsations data is proposed to solve the scarcity of load rejecting data. The water hammer pressure (WP) and pulsing pressure are then extracted from the enhanced data using the Savitzky-Golay filter and sent into the 1D-CNN for training. After that, a fine simulation model of the PSP is built utilizing the method of characteristics to calculate the simulation WP for the condition that needs to be predicted. Finally, the calculated WP is input into the trained 1D-CNN model to predict the corresponding pressure pulsations. The effectiveness of the suggested pressure pulsations intelligent prediction model is confirmed using measured transition data from a PSP in China. The findings demonstrate that there are only 2 m and 1.01 m prediction errors for the maximum pressure at the volute inlet and minimum pressure at the draft tube inlet, respectively. • A pressure pulsations prediction model based on VMR and CNN is proposed. • A data augmentation method named variational mode reconstruction is presented. • A refined model of the pumped storage power station is constructed utilizing MOC. • The effectiveness of the proposed method is verified using an actual PSP.