基于飞行数据的着陆偏跑道参数预测

Prediction of runway excursion landing parameters based on flight data

摘要:

偏跑道事件是一类频发且危险的民航安全事件，如何降低该事件发生概率是民航运行过程关注的焦点之一。首先基于真实的飞行数据集，根据飞行手册与中国民航行业标准选出有偏出跑道风险的阳性样本与无偏出跑道风险的阴性样本，同时确定关键操作节点；然后对其进行重复测量实验，获得影响偏跑道事件的重要参数及各个参数的时序特征；最后使用灰色关联分析筛选参数并基于LSTM 神经网络，训练对各个关键操作节点的重要参数以及偏跑距离的预测模型。结果表明：模型的MAE 均小于或等于1.2，精度较高。未来可根据提前训练好的模型与进近过程实时数据，实现对偏跑道事件的在线提示功能。

Abstract:

The occurrence of runway excursion is a frequent and dangerous safety issue in civil aviation, and reducing the probability of such incidents is a key focus in aviation operations.?This study first selects positive samples with the risk of runway excursion and negative samples without the risk from real flight data and determines crucial time point based on flight manuals and China Civil Aviation Industry Standard.Subsequently, repeated measure experiments to these samples are conducted to obtain important parameters affecting runway excursion and the temporal characteristics of these parameters.Finally, grey relational analysis is used to select parameters,and LSTM neural networks and these selected parameters are used to train models to predict the important parameters and off runway center line distance of all crucial time points .The experiments show that the MAE of each model is consistently less than or equal to 1.2, indicating high accuracy. In the future, the pre-trained models can be used with real-time approach data to achieve online warning for runway excursion.