A hybrid EMD-GRU model for pressure prediction in air cyclone centrifugal classifiers

Abstract

Predicting the pressure in air cyclone centrifugal classifiers is important for analyzing the flow field and improving classification performance, but traditional methods struggle due to the presence of noise in the pressure signals. In this study, a model combining empirical mode decomposition (EMD) and gate recurrent unit (GRU) is proposed and genetic algorithm (GA) is used to further increase the prediction accuracy of the model. The air pressure at the feeding port, fine particle discharge port and coarse particle discharge port of the air classifier are selected for prediction and the intrinsic mode functions (IMFs) with high correlation are selected for the denoising effect using EMD and the Pearson correlation coefficient (PCC). Signal denoising facilitates better feature extraction and simplifies neural network models. The results show that the best prediction among five models is achieved by the EMD-GRU model, with a root mean square error (RMSE) of 0.0549, 0.0177, and 0.0203 for the three ports. In addition, the effects of different parameters on the classification efficiency of the air classifier are investigated. The results reveal that the air classifier can achieve the best classification effect when the rotational frequency is 10.83 Hz, the feeding rate is 0.4 kg/s and the inclination angle is −4°. This study introduces a new idea for pressure prediction and flow field simulation in air classifiers and provides a new reference for optimizing the classification performance of air cyclone centrifugal classifiers.