基于多维特征的终端区异常轨迹实时检测

Research on real-time detection of abnormal trajectory in terminal area based on multidimensional features

为辅助管制员评估终端区航空器的运行风险,降低终端区内不安全事故的发生率,提出了一种基于轨迹信息熵和飞行距离的异常轨迹检测模型。首先,运用基于时间比的自顶向下算法将轨迹分段,并用基于密度的聚类算法找出中心轨迹,利用中心轨迹计算轨迹簇的信息熵;其次,在水平轨迹和垂直剖面两个维度进行异常轨迹的检测,综合两个维度的检测结果得到异常轨迹最佳判别阈值;最后,将飞行轨迹预测模型应用于异常轨迹检测模型中进行异常轨迹的实时检测,并运用对比试验证明异常轨迹实时检测模型的有效性。结果表明,该模型检测结果与一线实际运行情况基本一致。

To assist the controller in assessing the operational risks of aircraft and reduce the incidence of unsafe accidents in the terminal area,real-time detection of abnormal trajectories of aircraft in the terminal area is carried out so that the aircraft with potential safety hazards in the terminal area can be identified in time and accurately,and then alert.The controller pays attention to ensure the safe operation of the aircraft in the terminal area.This paper proposes an abnormal trajectory detection model based on trajectory information entropy and flight distance and combines the flight trajectory prediction model with this model to achieve real-time detection of abnormal trajectories.First,this paper uses the top-down algorithm based on time ratio to segment the trajectory,uses the density-based clustering algorithm to find the center trajectory,and uses the center trajectory to calculate the information entropy of the trajectory clustering;secondly,in the horizontal trajectory and the vertical profile above,the abnormal trajectories are detected in two dimensions respectively.combining the two-dimensional detection results,the best discrimination threshold of abnormal trajectories is obtained.Finally,the flight trajectory prediction model is applied to the abnormal trajectory detection model to detect the abnormal trajectory in real-time.In the flight trajectory prediction,when time,latitude,longitude,and altitude are considered,the predicted error is 397.48 m;when time,latitude,longitude,and energy altitude are considered,the predicted error is 647.29 m,which meets the flight trajectory requirements.Besides,the prediction time-span is 3 minutes,and the required time is 2.734 s,which meets the requirements of long-term flight trajectory prediction.The effectiveness of the model for real-time detection of abnormal trajectories is verified by comparative experiments.Experimental results show that the effective detection rate of this model is 84.62%,and the false alarm rate is 81.48%.