终端区4D飞行轨迹预测与冲突预警

为提升终端区飞行轨迹预测精度，实现航空器短时冲突预警，建立一种基于孪生支持向量回归的终端区4D飞行轨迹预测模型。对历史飞行轨迹应用重采样算法，降低轨迹数据规模；利用墨卡托投影将轨迹点经度、纬度与高度化为x-y-z坐标，采用孪生支持向量回归算法学习预测模型，实现短时航空器飞行轨迹动态预测；计算两架航空器水平、垂直距离，建立航空器冲突预警指示函数；对孪生支持向量回归算法进行超参数灵敏度分析，分析各超参数对模型预测效果的影响。根据机场真实数据进行仿真实验，证明：基于孪生支持向量回归的4D飞行轨迹预测模型能够准确捕捉航空器运动趋势，且泛化能力强；所提模型x-y-z坐标预测均方根误差是BP神经网络预测结果的32%，35%和61%，单次预测计算用时减少约0.13 s。

4D Flight Trajectory Prediction and Conflict Warning in Terminal Area

In order to improve the accuracy of flight trajectory prediction in the terminal area and achieve the goal of short-term conflict warning between two aircraft, a 4D flight trajectory prediction methodology in the terminal area based on twin support vector regression was established. Firstly, the resampling algorithm was applied to the raw flight trajectory to reduce the scale of trajectory dataset. Mercator projection was used to convert the longitude, latitude and height of trajectory point into x-y-z coordinates. Then, the twin support vector regression approach was contributed to construct the prediction model to achieve the goal of aircraft flight trajectory dynamic prediction in short-term. By calculating the horizontal and vertical distance between two aircraft, an aircraft conflict warning indicator function was established. The influence of each hyperparameter on the prediction effect was analyzed through hyperparameter sensitivity analysis. Simulation experiments based on history trajectory in an airport prove that the 4D flight trajectory prediction model based on twin support vector regression can accurately catch the trend of aircraft movement and have robust generalization ability. The root mean square error of x-y-z coordinates of the proposed model is 32%, 35% and 61% of the back-propagation neural network, and the calculation time for a single prediction is reduced by about 0.13 s.