면적속도 기반 DTW를 이용한 이탈 항적 탐지

Abstract

In the field of aviation safety, data analysis research is being conducted to detect the precursors of aviation accidents. A deviation trajectory, which means the trajectory of an aircraft deviating from its normal trajectory, is one of the precursor signs of an aviation accident that occurs in an en-route section. The en-route section means segment of flight from the termination point of a departure procedure to the origination point of an arrival procedure. trajectory data is track data meaning positional information according to time. A general method for calculating the dissimilarity of trajectory data having different numbers of sample points constituting the trajectory is dynamic time wrapping (DTW). However, area-based dissimilarity from normal tracks is more useful than distance-based dissimilarity for deviation trajectory detection. Therefore, in this study, ‘area velocity-based DTW (hereafter referred to as ADTW)' was developed to calculate the value corresponding to the area. Dissimilarity was calculated by DTW and ADTW methods using ADS-B (automatic dependent surveillance-broadcast) trajectory data, and then deviation tracks were determined using clustering-based outlier detection technology. For clustering, the k-medoids method, PAM (partitioning around medoids), was used. As a result of calculating the average silhouette, it was classified into two clusters, and the deviation track was detected using the dissimilarity from the representative track of each cluster. Detection criteria were constructed using quantiles. When evaluated using the average silhouette width of clusters, ADTW performed better than DTW.