| In recent years,with the rapid development of the world’s civil aviation,air traffic flow has increased rapidly,and airspace restrictions and flight delays have become increasingly prominent.One of the challenges facing civil aviation today is how to fully use valuable airspace resources and manage air traffic more effectively.Trajectory based operation(TBO)is a more accurate,safer,and efficient air traffic management model,and trajectory prediction is one of the critical technologies of TBO,which is an effective measure for future airspace management under high traffic,high density,and short interval conditions,and can significantly improve the utilization of airspace resources.Therefore,in the face of the complex airspace environment,exploring a more accurate trajectory prediction method is the focus of future civil aviation development.Aiming at the problems of inadequate extraction of trajectory features and difficulty in overcoming the short-term memory of time series in existing trajectory prediction,a trajectory prediction model based on a convolutional neural network-bidirectional long short-term memory(CNN-BiLSTM)network combined with dual attention and genetic algorithm(GA)optimization is proposed.The feature attention module constructed by CNN combined with the attention mechanism has the advantage of extracting essential features from the input model.Then the temporal attention module formed by introducing the attention mechanism at the output of BiLSTM can capture long-term dependencies in the time series and highlight crucial historical information,and GA was used to optimize the hyperparameters of the model to achieve the best performance.Finally,a multifaceted comparison with other typical time-series prediction models based on real flight data verifies that the prediction model based on hyperparameter optimization and a dual attention mechanism has significant advantages in terms of prediction accuracy and applicability.Based on the above study,a lightweight dual self-Attention temporal convolutionbidirectional gated recurrent unit(DSA-TCN-BiGRU)trajectory prediction model is proposed to improve multi-step prediction accuracy and reduce operational complexity.In this model,the TCN combined with the self-attention mechanism provides highly stable training,high parallelism,and flexible perceptual domains.At the same time,BiGRU has the advantages of fewer parameters and fast convergence.The Bayesian optimization method is used to optimize the model by entirely using the prior function evaluation information,simplifying the model structure while improving the prediction performance.To verify the superior performance of the model,ablation experiments based on several flight datasets from Beijing Capital-Shanghai Hongqiao are conducted.The results show that the TCN-BiGRU model based on the dual selfattention mechanism performs best and is generally consistent with the actual flight trajectory.As a result,the improved prediction model is more accurate and robust,providing a decision basis for trajectory based operation. |
