Time series prediction of key parameters of pump-type machinery in nuclear power plants based on empirical wavelet transform-gate recurrent unit (EWT-GRU)

This study proposed a hybrid prediction model named EWT-GRU to address several issues in pump-type machinery fault prediction. These issues include low prediction accuracy of single models, insufficient extraction of time-correlated information, and challenges in multi-step prediction. The EWT-GRU model combines Empirical Wavelet Transform (EWT) and Gate Recurrent Unit (GRU) to overcome the low prediction accuracy problem of single models. In the multi-step prediction strategy, a sliding time window technique was employed for multistep recursive prediction to mitigate the accumulation of prediction errors. The proposed prediction model was applied to predict key parameters of main pumps, and case tests were conducted for both single-step and multi-step predictions. The results showed that the proposed hybrid prediction model has the lowest RMSE and MAPE values for two sequences under different prediction steps, which are controlled within 0.37 and 0.85 %, respectively. The model integrates the advantages of EWT, GRU, and sliding recursive prediction, which enables multi-step prediction and reduces error accumulation, providing a sound basis for predictive maintenance and control.