基于聚类分析的航空器滑行过点时间预测

Prediction of Aircraft Taxiing Estimated Time of Arrival Based on Cluster Analysis

为了预测航空器滑行预计到达时间(Estimated time of arrival,ETA),减少场面冲突,提高机场运行效率,本文使用卡尔曼滤波算法对场面历史轨迹数据进行预处理。为了衡量轨迹样本间的距离,综合三类特征用于机场场面历史轨迹数据聚类。特征包含航空器滑行时段和场面航空器数量,以及参考动态时间规整(Dynamic time warping,DTW)算法提取的轨迹差异度特征。将两个样本特征的欧式距离作为样本间的相似度量;基于均差最大原则确定初始聚类中心,使用K-means算法对样本进行聚类,根据待规划航空器的所处时段和场面航空器数量选择匹配度最高的类簇,将其聚类中心样本的轨迹序列和塔台规划的静态路径相结合预测航空器滑行ETA。通过将实际轨迹数据与预测的滑行ETA进行对比分析,证明了本文预测航空器滑行ETA的准确性。

To predict aircraft taxiing estimated time of arrival(ETA),reduce conflicts,and improve the efficiency of the airport,this paper uses the Kalman filter algorithm to preprocess historical trajectory data of an airport. In order to measure the distance between trajectory samples, three kinds of features are synthesized for clustering of airport historical trajectory data,including the aircraft taxiing period and the number of surface aircraft. In addition,we refer to the dynamic time warp(DTW)algorithm to extract the difference features of the trajectory. The Euclidean distance of two sample features is taken as their similarity.Initial clustering center is determined based on the divided difference maximum principle. The K-means algorithm is used to cluster samples,and the cluster with the highest matching degree is selected according to the time period of the aircraft to be planned and the number of aircraft operating on the surface. The trajectory sequence of the cluster’s center sample and the static path planned by the tower are combined to predict the aircraft taxiing ETA. By comparing and analyzing the actual trajectory data with the predicted taxiing ETA,the good performance of predicting the aircraft taxiing ETA in this paper is verified.