A Predictive Aircraft Trajectory Prediction Method Based on Transformer Encoder and LSTM
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
维普
万方数据
为了解决飞机目标机动数据集缺失的问题,文章利用运动学建模生成了丰富的轨迹数据集,为网络训练提供了必要的数据支持.针对现阶段轨迹预测运动学模型建立困难及时序预测方法难以提取时空特征的问题,提出了一种结合Transformer编码器和长短期记忆网络(Long Short Term Memory,LSTM)的飞机目标轨迹预测方法,即Transformer-Encoder-LSTM模型.新模型可同时提供LSTM和Transformer编码器模块的补充历史信息和基于注意力的信息表示,提高了模型能力.通过与一些经典神经网络模型进行对比分析,发现在数据集上,新方法的平均位移误差减小到 0.22,显著优于CNN-LSTM-Attention模型的 0.35.相比其他网络,该算法能够提取复杂轨迹中的隐藏特征,在面对飞机连续转弯、大机动转弯的复杂轨迹时,能够保证模型的鲁棒性,提升了对于复杂轨迹预测的准确性.
In order to solve the problem of missing aircraft target maneuver data sets,this paper uses kinematic modeling to generate a rich trajectory data set,which provides necessary data support for network training.In order to solve the problem that it is difficult to establish a kinematic model for trajectory prediction at the current stage and that it is difficult to extract spatiotemporal features with the time series prediction method,an aircraft target trajectory prediction method that combines the Transformer encoder and the Long Short Term Memory network(LSTM)is proposed.It can provide supplementary historical information and attention-based information representation provided by LSTM and Transformer modules at the same time,improving model capabilities.Through comparative analysis with some classic neural network models on the data set,it is found that the average displacement error of this method is reduced to 0.22,which is significantly better than 0.35 of the CNN-LSTM-Attention model.Compared with other networks,this algorithm can extract hidden features in complex trajectories.When facing complex aircraft trajectories with continuous turns and large maneuvers,it can ensure the robustness of the model and improve the accuracy of prediction of complex trajectories.
NETL
NSTL
维普
万方数据
