中国安全科学学报 ›› 2020, Vol. 30 ›› Issue (6): 84-91.doi: 10.16265/j.cnki.issn1003-3033.2020.06.013

• 安全工程技术 • 上一篇    下一篇

基于DBN的飞机着陆弹跳预测方法

贾博 工程师, 孙延进 二级飞行员, 张贵明 二级飞行员   

  1. 东航技术应用研发中心有限公司,上海 201707
  • 收稿日期:2020-03-16 修回日期:2020-05-14 出版日期:2020-06-28 发布日期:2021-01-28
  • 作者简介:贾博(1987—),男,甘肃兰州人,博士,工程师,主要从事民航大数据分析、模式识别和机器学习等方面的工作。E-mail:icesea137@163.com。
  • 基金资助:
    国家自然科学基金资助(U1933125)。

A prediction method for bounced landing of aircraft based on DBN

JIA Bo, SUN Yanjin, ZHANG Guiming   

  1. China Eastern Technology Application R&D Center Co., Ltd., Shanghai 201707, China
  • Received:2020-03-16 Revised:2020-05-14 Online:2020-06-28 Published:2021-01-28

摘要: 飞机着陆弹跳是航班运行中经常发生的不正常事件,为掌握着陆弹跳的诱因并有效预防该类事件发生,首先,提出一种基于深度置信网络(DBN)的飞机着陆弹跳预测方法;其次,通过航空运行数据评估事件与着陆机场环境条件的相关性,以东航实际着陆弹跳事件为例,探究着陆时油门杆位置与弹跳的变化趋势;然后,分别讨论触地前飞行员操纵、飞机状态和飞机的不稳定进近对着陆弹跳的影响;最后,以不同信息的组合作为输入训练模型,对比预测精度,找到最优模型。研究结果表明: 基于DBN的方法适合利用大量航班数据预测着陆弹跳事件;当网络输入包含机场环境条件、飞机油门杆位置等着陆弹跳直接影响因素,以及不稳定进近等非直接影响因素时,该预测模型能够较准确地预测着陆弹跳事件,其预测准确度可达到94.78%。

关键词: 深度置信网络(DBN), 着陆弹跳, 快速存储记录器(QAR), 航空大数据, 不安全事件

Abstract: In order to grasp causes of bounced landing of aircraft which is a frequently occurring issue during flight operation, and effectively prevent such incidents, a prediction method for bounced landing based on DBN was proposed. Secondly, correlation between incidents and landing airports' environment was evaluated by using aviation data, and with an actual incident of China Eastern Airlines as an example, changing trend of bounced landing along with throttle stick position at touchdown was explored. Then, impacts of pilot control, aircraft status and unstable approach on incidents were discussed. Finally, different combinations of information were used as inputs to train the model, and their prediction accuracy was compared to find optimal one. The results show that DBN-based method is suitable for predicting bounced landing by utilizing flight data. When network input includes direct influencing factors such as airports' environment, throttle lever position, as well as indirect ones like unstable approach, this model can accurately predict accidents with a prediction accuracy as high as 94.78%.

Key words: deep belief network (DBN), bounced landing, quick access recorder (QAR), big data of aviation, unsafe incident

中图分类号: