Trajectory prediction(TP)is critical for enhancing flight safety and operational reliability in small to medium-sized private and corporate aircraft,which involve complex multiple inputs and multiple outputs.While exi...Trajectory prediction(TP)is critical for enhancing flight safety and operational reliability in small to medium-sized private and corporate aircraft,which involve complex multiple inputs and multiple outputs.While existing TP methods primarily focus on extracting coupled features,they often neglect the independent features of individual outputs,leading to unsatisfactory learning performance.To address this limitation,this paper proposes a multitask learning-based TP method using bi-directional long short-term memory(Bi-LSTM),consisting of two key components:1)a multi-source feature fusion part that automatically extracts and integrates coupled evolutionary features across flight modes,and 2)a multitask learning part that mines independent change characteristics of each output.Firstly,the trajectory sequences are categorized into short,medium,and long-period flight modes to better capture temporal dependencies.The coupled characteristics in every flight mode are automatically excavated and integrated via the Bi-LSTM and fully connected network in the multi-source feature fusion part.Secondly,the fusion output is sent into the multitask learning part and every task has a customized model to learn independent evolutionary features of each output.Experimental results on real-world flight trajectories demonstrate the superiority of the proposed method in both one-step and multi-step prediction scenarios,highlighting its ability to leverage both coupled and independent features of flight trajectories.展开更多
文摘Trajectory prediction(TP)is critical for enhancing flight safety and operational reliability in small to medium-sized private and corporate aircraft,which involve complex multiple inputs and multiple outputs.While existing TP methods primarily focus on extracting coupled features,they often neglect the independent features of individual outputs,leading to unsatisfactory learning performance.To address this limitation,this paper proposes a multitask learning-based TP method using bi-directional long short-term memory(Bi-LSTM),consisting of two key components:1)a multi-source feature fusion part that automatically extracts and integrates coupled evolutionary features across flight modes,and 2)a multitask learning part that mines independent change characteristics of each output.Firstly,the trajectory sequences are categorized into short,medium,and long-period flight modes to better capture temporal dependencies.The coupled characteristics in every flight mode are automatically excavated and integrated via the Bi-LSTM and fully connected network in the multi-source feature fusion part.Secondly,the fusion output is sent into the multitask learning part and every task has a customized model to learn independent evolutionary features of each output.Experimental results on real-world flight trajectories demonstrate the superiority of the proposed method in both one-step and multi-step prediction scenarios,highlighting its ability to leverage both coupled and independent features of flight trajectories.
