| 基于在线更新LSTM网络的短期4D航迹预测算法 |
| |
| 引用本文： | 石庆研,王文青,韩萍.基于在线更新LSTM网络的短期4D航迹预测算法[J].信号处理,2021,37(1):66-74. |
| |
| 作者姓名： | 石庆研 王文青 韩萍 |
| |
| 作者单位： | 中国民航大学智能信号与图像处理天津市重点实验室 |
| |
| 基金项目： | 国家重点研发计划（2016YFB0502405）；中央高校（3122014C004） |
| |
| 摘 要： | 航班飞行过程中一些因素会对当前飞行轨迹产生影响，从而导致实时航迹与历史航迹相比有一定的差异，使得仅基于历史航迹数据的航迹预测模型的预测性能变差。为解决该问题，提出了一种基于在线更新长短期记忆（Long Short-Term Memory，LSTM）网络的短期4D航迹预测算法，该算法由基于历史航迹数据的预测模型初始化参数训练和基于实时航迹数据的预测模型参数在线更新两部分构成。首先建立基于LSTM神经网络的航迹预测模型，使用历史航迹数据进行训练并保存训练完成的预测模型参数，然后使用实时航迹数据对航迹预测模型进行在线训练并微调参数，使用在线更新参数后的预测模型实现4D航迹短期预测，以期达到提升预测准确度的目的。利用实际航迹数据对算法的性能进行验证，结果表明新方法能够考虑实时飞行过程中各因素对航迹产生的影响，有效提升经度、纬度、高度和时间的预测准确度，并具有良好的泛化能力。
|
| 关 键 词： | 4D航迹预测 在线更新 长短期记忆网络 循环神经网络 |
| 收稿时间： | 2020-10-12 |
Short-term 4D Trajectory Prediction Algorithm Based on Online-updating LSTM Network |
| |
| Affiliation: | Tianjin Key Lab for Advanced Signal Processing, Civil Aviation University of China |
| |
| Abstract: | Some factors in the flight process will have an impact on the current trajectory. There is a certain difference between the real-time trajectory and the historical trajectory, and the prediction performance of the trajectory prediction model based on historical trajectory data becomes worse. To solve this problem, a short-term 4D trajectory prediction algorithm based on online-updating long short-term memory (LSTM) is proposed. The prediction algorithm is composed of two parts: the initial parameters training of the prediction model based on historical trajectory data and the parameters online-updating for the prediction model based on real-time trajectory data. The trajectory prediction model is established through the LSTM neural network first. The historical trajectory data are used to train the model and the trained parameters of the model are saved. Then, the real-time trajectory data are used to retrain and fine-tune the parameters of the trajectory prediction model. The online-updating prediction model is used to predict the short-term 4D trajectory data, so as to achieve the purpose of improving the prediction accuracy. The actual trajectory data are used to verify the performance of the algorithm. Experimental results show that the new prediction model with a good generalization ability can take into account the influence of various factors on the trajectory during the real-time flight process and improve the prediction accuracy of longitude, latitude, height, and time effectively. |
| |
| Keywords: | |
|
| 点击此处可从《信号处理》浏览原始摘要信息 |
|
点击此处可从《信号处理》下载全文 |
