摘要
高原机场运行环境复杂,进近阶段易引发近地告警,因此,研究高原机场近地告警关键致因和预测。首先,提取环境、飞机状态、操纵、航迹4个方面的参数指标,作为致因因素初步范围,通过计算互信息值识别参数间的相似特征,删除冗余参数,引入FP-Growth关联规则挖掘算法识别关键致因。为解决航班样本数据高维度问题与预测过程黑箱特性,将卷积神经网络(Convolutional Neural Network, CNN)和决策树算法融合,建立近地告警预测模型,使用美国阿斯彭机场快速存取记录器(Quick Access Record, QAR)数据,对近地告警关键致因和预测模型进行了验证。得到6个与近地告警相关的关联规则,均为多指标组合关键致因;采用数据增强后的航班数据对预测模型进行了验证,预测结果表明模型可实现基于初始下降运行参数预测近地告警,平均准确率达到86.1%,并与其他预测模型进行了对比,在准确率、召回率、F1分数、虚警率(False Alarm Rate, FAR)和临界成功指数(Critical Success Index, CSI)方面均显示有更好的效果。
To ensure the safe operation of aircraft during the final descent phase,this paper investigates the key causes and predictive factors associated with Ground Proximity Warnings at plateau airports.First,parameters are extracted from four aspects—environment,aircraft state,aircraft control,and trajectory—to establish a preliminary set of causal factors.The similarity characteristics among these parameters are assessed by calculating mutual information values,which allows for the elimination of redundant parameters.Additionally,the FP-Growth association rule analysis algorithm is employed to identify key causes.Finally,to tackle the high dimensionality of flight sample data and the black box nature of machine learning prediction processes,a fusion of Convolutional Neural Networks(CNN)and decision tree algorithms is utilized to develop a ground proximity warning prediction model.The key causes that trigger ground proximity warnings,along with the prediction model,were validated using Quick Access Record(QAR)data from Aspen Airport,a plateau airport in the United States.The research identified six association rules related to ground proximity warnings:FATA 1 and VRTG 1—Pattern 1;W_h 1 and IVV 1—Pattern 1;FATA 1 and SPLG 1—Pattern 2A;VRTG 1 and SPLG 1—Pattern 2B;IVV 1 and CW 1—Pattern 4;and W_h 1 and ELEV 0—Pattern 5.All of these rules represent key causes involving combinations of multiple indicators.The prediction algorithm was validated using flight data that had been subject to data augmentation.The prediction results demonstrate that the combined model can effectively predict ground proximity warnings based on the operational parameters during the initial descent phase,achieving an average accuracy of 86.1%.The analysis and verification of randomly selected samples predicted for each mode revealed that the positive samples identified by the model indeed corresponded to instances where Ground Proximity Warnings were triggered in actual flights.In comparison to other prediction models,the model presented in this paper demonstrates superior performance in terms of accuracy,recall,F 1 score,False Alarm Rate(FAR),and Critical Success Index(CSI),further validating its reliability and effectiveness.
作者
齐雁楠
赵嘉星
吴祚禹
赵宇
QI Yannan;ZHAO Jiaxing;WU Zuoyu;ZHAO Yu(College of Air Traffic Management,Civil Aviation University of China,Tianjin 300300,China;Operation Supervisory Center of CAAC,Beijing 100710,China;Air China,Beijing 100621,China)
出处
《安全与环境学报》
北大核心
2025年第6期2061-2071,共11页
Journal of Safety and Environment
基金
国家重点研发计划项目(2022YFB4300904)
国家自然科学基金委员会与中国民用航空局联合资助项目(U1633124)。
关键词
安全工程
高原机场
近地告警
关联规则挖掘
卷积神经网络-决策树
safety engineering
plateau airport
ground proximity warning
association rule mining
convolutional neural network-decision tree
