A hybrid algorithm of TCN-iTransformer for aircraft aerodynamic parameter estimation based on dual attention mechanism

Abstract

Based on the dual attention mechanism, the paper proposes a new hybrid algorithm of TCN-iTransformer for aircraft aerodynamic parameter estimation. This algorithm adopts a new iTransformer architecture, which applies Self-Attention and Feed-Forward Neural Network (FFNN) in inverted dimensions, and enhances the feature extraction by the prepositive placing an improved Temporal Convolutional Network (TCN) based on the Frequency-Enhanced Channel Attention Mechanism (FECAM). To effectively suppress the noise in flight data, the result of Variational Mode Decomposition (VMD) based on the rime optimization algorithm (RIME) is used as the supplementary input for TCN-iTransformer. To verify the effectiveness of this algorithm, the paper applies the flight test data of a typical propeller aircraft to train and validate the model. It conducts a comparative analysis of its estimation effect with that of Transformer and Long Short-Term Memory (LSTM) under the same circumstances. Results show that the proposed new hybrid algorithm is practical and effective, which performs well in terms of both efficiency and accuracy. To further demonstrate its superiority and necessity, comparisons with the SOTA and ablation studies were conducted, which validated its optimal performance and the necessity of its proposal.