ECOD算法在飞机不稳定进近检测中的应用  

Application of ECOD algorithm for detection of aircraft unstable approach

作　　者：卢晓光[1] 许忠睿 张喆[1] 文贵宏 LU Xiaoguang;XU Zhongrui;ZHANG Zhe;WEN Guihong(Tianjin Key Laboratory for Advanced Signal Processing,Civil Aviation University of China,Tianjin 300300,China;Shandong Airlines Co.,Ltd.,Jinan 250107,China)

机构地区：[1]中国民航大学天津市智能信号与图像处理重点实验室,天津300300 [2]山东航空股份有限公司,济南250107

出　　处：《安全与环境学报》2024年第5期1872-1878,共7页Journal of Safety and Environment

基　　金：民航局安全能力建设基金项目([2023]276号)。

摘　　要：在飞机进近和着陆阶段,一旦发生不稳定进近就可能导致航空事故发生,因此终端空域内不稳定进近检测是航空器运行监控领域的热点研究问题。针对终端区的不稳定进近检测,利用OpenSky提供的开源航空器监视数据提出了基于数据驱动的检测方法。从能量管理的角度入手,构建基于无监督异常检测(Empirical-Cumulative-Distribution-based Outlier Detection,ECOD)算法的不稳定进近检测模型,并结合主成分分析(Principal Components Analysis,PCA),获取了飞机能量状态的异常评分进而实现检测。复飞事件检测的验证分析结果表明检测模型在准确率与效率方面具有优势,模型可实现实时部署与在线更新。Once an unstable approach occurs during one of the approach and landing phases,it may lead to an incident or accident.Therefore,the detection of unstable approaches in terminal areas has attracted a lot of research interest in the field of flight operation status monitoring.Aiming at the detection algorithms for unstable approaches,a data-driven method for unstable approaches detection is proposed using OpenSky s open-source aviation surveillance data set.The unstable approaches detection model is based on the Empirical-Cumulative-Distribution-based Outlier Detection(ECOD)unsupervised anomaly detection algorithm from the perspective of energy management.ECOD computes an outlier score of each data point by aggregating estimated tail probabilities across dimensions,abnormal scores are then obtained for the detection of unstable approaches.The distance of the aircraft from the runway,altitude,ground speed,vertical rate,and total energy are used as features to form a training dataset that includes 5-dimensional features.After dimensionality reduction by PCA,the detection model is obtained by training using the ECOD algorithm.The probability of an unstable approach increases with higher anomaly scores.According to the unstable approach criterion,the labeled data are acquired and the ECOD model is compared with two other popular data-driven models,iForest and HDSCAN.The ECOD model achieves a detection accuracy of 0.73,a breakthrough in the accuracy of unstable approach detection.Aircraft energy state monitoring in conjunction with the ECOD model contributes to energy management and unstable approach monitoring during aircraft approaches and landings based on previously established energy safety boundaries and identified air traffic flow.Go-around data is fed to this model,the results demonstrate the accuracy and efficiency of the proposed model.

关 键 词：安全工程 飞行安全 不稳定进近 数据驱动 无监督异常检测(ECOD) 异常评分 

分 类 号：X949[环境科学与工程—安全科学] TP311[自动化与计算机技术—计算机软件与理论]