Prediction of Pressure Fluctuation of Double-Suction Centrifugal Pump in Different Cavitation Conditions based on EMD-GRU

Abstract The double-suction centrifugal pumps play a crucial role in various applications such as in irrigation-drainage projects, water diversion projects, and urban water supply and drainage projects. The pressure fluctuation is one of the crucial parameters for assessing the operational stability of pump. The pressure fluctuation exhibits distinct characteristics in various cavitation conditions. A pressure fluctuation prediction method based on Empirical Mode Decomposition (EMD) and Gated Recurrent Unit (GRU) is proposed in this paper. The model enables accurate forecasting of pressure fluctuation signals when pump operates at different cavitation conditions. The data were obtained from experimental measurements of a double-suction centrifugal pump. The original signal was decomposed into multiple sub-signals with varying energy levels using the EMD method. The neural network of GRU was employed for training and data prediction on individual sub-signals. The predicted results of each sub-signal were combined to form the final pressure fluctuation signals. The method based on EMD-GRU proves effective in predicting pressure fluctuation in three different cavitation conditions. The mean absolute error (MAE) in time-domain characteristics is below 6%, and the prediction error for the main frequency is below 1%. The prediction of low-frequency characteristics still requires improvement.