MF-TimeSformer: A Multi-Frequency Feature Enhancement Network for Predicting Unsteady Flow Field of Plane Cascades

The prediction of unsteady flow field in plane cascades is of great significance for analysing compressor instability. A rapid and accurate prediction method of unsteady flow field is a challenge due to the high-dimensional and nonlinear dynamic behavior. In view of the two shortcomings of the existing spatio-temporal predictive network models, namely, the inability to capture the multi-frequency characteristics and massive consumption of computing resources, we designed the MF-TimeSformer model, which uses wavelet transform and ConvLSTM to capture the spatio-temporal multi-frequency characteristics in the flow field, and uses the filled prediction tokens to achieve the prediction function. We design a number of experiments to verify the effectiveness and generalization of the model and explore the role of filled prediction tokens and multi-frequency feature fusion in the prediction of spatio-temporal sequence. The results show that MF-TimeSformer has a good effect on the prediction of the unsteady flow field of plane cascades, and the structure of each part of the model plays an extremely important role.