工程科学学报 Chinese Journal of Engineering 基于有向权值网络的航班运行风险传播与控制 王岩韬杨志远刘锟谢春生 Flight operation risk propagation and control based on a directional-weighted complex network WANG Yan-tao,YANG Zhi-yuan,LIU Kun,XIE Chun-sheng 引用本文: 王岩韬,杨志远,刘锟,谢春生.基于有向权值网络的航班运行风险传播与控制.工程科学学报,2022,44(1):114-121.doi: 10.13374j.issn2095-9389.2020.06.15.002 WANG Yan-tao,YANG Zhi-yuan,LIU Kun,XIE Chun-sheng.Flight operation risk propagation and control based on a directional- weighted complex network[J].Chinese Journal of Engineering,2022,44(1):114-121.doi:10.13374/j.issn2095- 9389.2020.06.15.002 在线阅读View online:https::/doi.org10.13374.issn2095-9389.2020.06.15.002 您可能感兴趣的其他文章 Articles you may be interested in 基于多变量混沌时间序列的航班运行风险预测模型 Flight operation risk prediction model based on the multivariate chaotic time series 工程科学学报.2020,42(12:1664 https:/1doi.org/10.13374.issn2095-9389.2019.12.09.002 混沌人工鱼群的鲁棒保性能控制权值矩阵优化方法 A weighting matrix optimization method for robust guaranteed cost control based on chaos artificial fish swarm algorithm 工程科学学报.2018.40(4:500 https:/1doi.org10.13374j.issn2095-9389.2018.04.014 含有自校正模型的加权多模型自适应控制 Weighted multiple model adaptive control with self-tuning model 工程科学学报.2018.40(11:1389htps:/doi.org10.13374.issn2095-9389.2018.11.013 基于集成神经网络的剩余寿命预测 Remaining useful life prediction based on an integrated neural network 工程科学学报.2020,42(10:1372htps:/ldoi.org10.13374.issn2095-9389.2019.10.10.005 剪切浓密床层孔隙网络模型与导水通道演化 Pore network model of tailings thickener bed and water drainage channel evolution under the shearing effect 工程科学学报.2019,41(8:987htps:/oi.org10.13374.issn2095-9389.2019.08.004 能量均衡的间断连接无线网络数据转发策略 Data forwarding strategy for wireless network with intermittent connectivity based on energy equilibrium 工程科学学报.2017,396):962htps:/1doi.org/10.13374j.issn2095-9389.2017.06.020
基于有向权值网络的航班运行风险传播与控制 王岩韬 杨志远 刘锟 谢春生 Flight operation risk propagation and control based on a directional-weighted complex network WANG Yan-tao, YANG Zhi-yuan, LIU Kun, XIE Chun-sheng 引用本文: 王岩韬, 杨志远, 刘锟, 谢春生. 基于有向权值网络的航班运行风险传播与控制[J]. 工程科学学报, 2022, 44(1): 114-121. doi: 10.13374/j.issn2095-9389.2020.06.15.002 WANG Yan-tao, YANG Zhi-yuan, LIU Kun, XIE Chun-sheng. Flight operation risk propagation and control based on a directionalweighted complex network[J]. Chinese Journal of Engineering, 2022, 44(1): 114-121. doi: 10.13374/j.issn2095- 9389.2020.06.15.002 在线阅读 View online: https://doi.org/10.13374/j.issn2095-9389.2020.06.15.002 您可能感兴趣的其他文章 Articles you may be interested in 基于多变量混沌时间序列的航班运行风险预测模型 Flight operation risk prediction model based on the multivariate chaotic time series 工程科学学报. 2020, 42(12): 1664 https://doi.org/10.13374/j.issn2095-9389.2019.12.09.002 混沌人工鱼群的鲁棒保性能控制权值矩阵优化方法 A weighting matrix optimization method for robust guaranteed cost control based on chaos artificial fish swarm algorithm 工程科学学报. 2018, 40(4): 500 https://doi.org/10.13374/j.issn2095-9389.2018.04.014 含有自校正模型的加权多模型自适应控制 Weighted multiple model adaptive control with self-tuning model 工程科学学报. 2018, 40(11): 1389 https://doi.org/10.13374/j.issn2095-9389.2018.11.013 基于集成神经网络的剩余寿命预测 Remaining useful life prediction based on an integrated neural network 工程科学学报. 2020, 42(10): 1372 https://doi.org/10.13374/j.issn2095-9389.2019.10.10.005 剪切浓密床层孔隙网络模型与导水通道演化 Pore network model of tailings thickener bed and water drainage channel evolution under the shearing effect 工程科学学报. 2019, 41(8): 987 https://doi.org/10.13374/j.issn2095-9389.2019.08.004 能量均衡的间断连接无线网络数据转发策略 Data forwarding strategy for wireless network with intermittent connectivity based on energy equilibrium 工程科学学报. 2017, 39(6): 962 https://doi.org/10.13374/j.issn2095-9389.2017.06.020
工程科学学报.第44卷.第1期:114-121.2022年1月 Chinese Journal of Engineering,Vol.44,No.1:114-121,January 2022 https://doi.org/10.13374/j.issn2095-9389.2020.06.15.002;http://cje.ustb.edu.cn 基于有向权值网络的航班运行风险传播与控制 王岩韬四,杨志远,刘锟,谢春生 中国民航大学航空公司人工智能民航局重点实验室,天津300300 ☒通信作者,E-mail:caucwyt@126.com 摘要为了分析航班运行风险传播过程,进而有效控制保障飞行安全,基于复杂网络理论,首先参照民航局咨询通告选取 机组、航空器、运行环境共29个终端因素作为网络节点,统计民航安全监察记录,根据事件中节点关系,构建无向网络:统计 前后节点间的作用关系和发生概率,提出一种有向带权的航班运行风险网络:然后,引入改进感染率和改进恢复率概念,构建 一种适用于航班运行风险传播分析的改进SIR(Susceptible-infected-recovered)模型;定义感染起始范围,最后采取多参数控制 方式,大规模计算该有向带权网络的传播和控制过程.结果表明:有向网的平均最短路径为1.788,属于小世界网络:参照使用民 航常规管控措施,有向网节点感染下降幅度可达到37.4%:对入度值排序前3或前4的节点控制后,感染节点峰值下降率高达50.6% 和58.1%.网络传播抑制明显.结果证实:在该航班运行风险有向带权网络中,按人度值控制节点对抑制风险传播最为有效 关键词航空运输:航班运行风险:复杂网络:有向带权网络:改进SR模型 分类号N945.24:U8:V355.2 Flight operation risk propagation and control based on a directional-weighted complex network WANG Yan-tao,YANG Zhi-yuan,LIU Kun,XIE Chun-sheng Airlines Artificial Intelligence Key Laboratory of Civil Aviation Administration,Civil Aviation University of China,Tianjin 300300,China Corresponding author,E-mail:caucwyt@126.com ABSTRACT The flight operation risk is equal to the occurrence probability multiplied by the severity of the consequences.Flight operation risks include many types,forms,and numbers,and they frequently change with conditions.In the face of this complex system, through principle analysis,the risk formation mechanism research,and the spreading process,a scientific risk management and control method can be constructed.Based on the risk management technology,an informative and automated management control system can be developed and applied.The overall safety level of flight operations will be effectively improved.To analyze and study the flight operations risk propagation and then effectively control flight safety based on the complex network theory,29 terminal factors were selected as network nodes according to the Civil Aviation Administration's advisory notice,initially including the flight cabin crew,civil aviation aircraft,and operating environment.Civil aviation safety monitoring records from 2009 to 2014 were counted,and an undirected network was constructed based on node relationships.The relationships and occurrence probability between the nodes were counted,and a directed and weighted network was constructed.The concepts of improved infection rate and improved recovery rate were introduced,and an improved susceptible-infected-recovered(SIR)model suitable for flight operation risks was proposed.Finally,the initial infection range was clearly defined,and a multi-parameter control method was adopted.For directed networks,large-scale propagation and control simulations were calculated.The results indicate that the average shortest path of the directed network was 1.788,which belonged to the small-world network.The directed network infection node decreased to 37.4%with conventional control measures.After controlling top three or four nodes of the entry degree value sequence,the infected nodes peak drop rate was the biggest, 收稿日期:2020-06-15 基金项目:国家自然科学基金资助项目(01933103)
基于有向权值网络的航班运行风险传播与控制 王岩韬苣,杨志远,刘 锟,谢春生 中国民航大学航空公司人工智能民航局重点实验室,天津 300300 苣通信作者, E-mail:caucwyt@126.com 摘 要 为了分析航班运行风险传播过程,进而有效控制保障飞行安全,基于复杂网络理论,首先参照民航局咨询通告选取 机组、航空器、运行环境共 29 个终端因素作为网络节点,统计民航安全监察记录,根据事件中节点关系,构建无向网络;统计 前后节点间的作用关系和发生概率,提出一种有向带权的航班运行风险网络;然后,引入改进感染率和改进恢复率概念,构建 一种适用于航班运行风险传播分析的改进 SIR(Susceptible-infected-recovered) 模型;定义感染起始范围,最后采取多参数控制 方式,大规模计算该有向带权网络的传播和控制过程. 结果表明:有向网的平均最短路径为 1.788,属于小世界网络;参照使用民 航常规管控措施,有向网节点感染下降幅度可达到 37.4%;对入度值排序前 3 或前 4 的节点控制后,感染节点峰值下降率高达 50.6% 和 58.1%,网络传播抑制明显. 结果证实:在该航班运行风险有向带权网络中,按入度值控制节点对抑制风险传播最为有效. 关键词 航空运输;航班运行风险;复杂网络;有向带权网络;改进 SIR 模型 分类号 N945.24;U8;V355.2 Flight operation risk propagation and control based on a directional-weighted complex network WANG Yan-tao苣 ,YANG Zhi-yuan,LIU Kun,XIE Chun-sheng Airlines Artificial Intelligence Key Laboratory of Civil Aviation Administration, Civil Aviation University of China, Tianjin 300300, China 苣 Corresponding author, E-mail: caucwyt@126.com ABSTRACT The flight operation risk is equal to the occurrence probability multiplied by the severity of the consequences. Flight operation risks include many types, forms, and numbers, and they frequently change with conditions. In the face of this complex system, through principle analysis, the risk formation mechanism research, and the spreading process, a scientific risk management and control method can be constructed. Based on the risk management technology, an informative and automated management control system can be developed and applied. The overall safety level of flight operations will be effectively improved. To analyze and study the flight operations risk propagation and then effectively control flight safety based on the complex network theory, 29 terminal factors were selected as network nodes according to the Civil Aviation Administration’s advisory notice, initially including the flight cabin crew, civil aviation aircraft, and operating environment. Civil aviation safety monitoring records from 2009 to 2014 were counted, and an undirected network was constructed based on node relationships. The relationships and occurrence probability between the nodes were counted, and a directed and weighted network was constructed. The concepts of improved infection rate and improved recovery rate were introduced, and an improved susceptible-infected-recovered (SIR) model suitable for flight operation risks was proposed. Finally, the initial infection range was clearly defined, and a multi-parameter control method was adopted. For directed networks, large-scale propagation and control simulations were calculated. The results indicate that the average shortest path of the directed network was 1.788, which belonged to the small-world network. The directed network infection node decreased to 37.4% with conventional control measures. After controlling top three or four nodes of the entry degree value sequence, the infected nodes peak drop rate was the biggest, 收稿日期: 2020−06−15 基金项目: 国家自然科学基金资助项目(U1933103) 工程科学学报,第 44 卷,第 1 期:114−121,2022 年 1 月 Chinese Journal of Engineering, Vol. 44, No. 1: 114−121, January 2022 https://doi.org/10.13374/j.issn2095-9389.2020.06.15.002; http://cje.ustb.edu.cn
王岩韬等:基于有向权值网络的航班运行风险传播与控制 115 as high as 50.6%/58.1%,the risk spread in the network was significantly suppressed.The results confirm that controlling nodes based on the entry degree value is the most effective method to suppress risk propagation in the directed and weighted network KEY WORDS air transportation:flight operation risk;complex network;directional weighted network;improved SIR model 航班运行风险是发生概率与后果严重程度的 进一步分析和总结发现:①节点间存在单向和双 乘积,其影响种类、形式、数目繁多,且随条件不 向作用关系:②不同节点间传播的可能性和作用强度 断改变.面对此复杂系统,通过对风险形成机理、 差异较大:③以往研究中传播初始点随机选择,是有 传播过程的原理性分析和研究,从而构建科学的 悖于真实情况的.这三点问题需要改进和解决 风险管控方法,是提高安全水平的必要措施.国外 鉴于以上,依据民航规章选取风险要素,统计 民航在安全管理研究多集中于风险评估,2000年 安全监察数据用以改造无向网络,首次构建了航 左右已有实用性成果P引.国内关于民航风险的研 班运行风险的有向加权网络;然后,引入改进感染 究成果最早见于1999年,孙瑞山教授刊基于系统 率和改进恢复率概念,提出一种改进SIR模型:定 工程构建了航空公司安全评估理论和方法,而后 义传染起始范围,综合安全管理方法和网络参数 于2013年研发了飞行安全风险评估工具问.王岩 分析,制定多种控制方案,并进行大量仿真计算: 韬团队于2014年以航班运行多因素分析为基 最后,以风险传播感染数量作为评价标准,旨在得 础,提出了一套具有普遍性、实用性的风险体系, 到具备理论支撑和实践价值的风险管控措施 并用于致险因素推断m后续,采用机器学习多算 1复杂网络节点选取 法协作方式,于2019年将评估精度提升至95%8: 并且实现了对起飞、巡航、降落等不同飞行阶段 网络是指由节点和连接节点的边组成,构建 的短时预测,尝试了一周内的短期预测0-☒而 航班运行风险网络,就是将运行中的风险因素作 后,张继凯,针对不同风险等级提出航班差异化 为节点,因素之间的相互关系作为连线.鉴于民航 控制策略,并于深圳航空公司实践验证.至此,风 局咨询通告AC-121-FS-2015-125于2015年在全国 险分析与量化评估等应用技术问题已得到较好解 实施至今,已有54家运输航空公司实践多年且验 决,而风险产生条件、传播过程、控制消散方法等 证有效,因此选取其作为风险分析体系,具体包 机理性问题尚缺乏研究 括机组、航空器和运行环境3方面共5级风险因 在上述研究和方案中,存在一种通用假设,即 素.其中,终端因素可与运行数据一一对应,可作 风险因素间关系呈现树状结构.然而,运行风险实 为后文建网数据依据,因此将第5级的终端因素 则与气象条件、机组资质、机场设施、障碍物、飞 作为网络节点,如表1.其中,节点编号1~13隶属 机维修甚至政策保障等多因素相关,而各因素间 机组风险类,节点14~26隶属运行环境风险类, 通过逻辑关系互相作用,错综交汇最终形成网络 节点27~29隶属航空器风险类 结构.欧美对于民航领域中复杂网络理论已有一 以节点1为例:机组级别一般分为学员、二 定研究,具有代表性的成果包括:Belkoura等4通 副、一副、机长、教员和检查员.经由2009年至今 过航空网络拓扑结构,从机场、机型、航空公司三 的民航不安全事件分析可得,当机组级别差距过 方面分析了网络动态特性,并提出网络结构优化 大时交叉检查效果会显著降低,甚至出现过技术 方法;Voltes-Dorta等l构建了以机场为节点,航 级别高的正驾驶操作不符程序,副驾驶已发现却 线为边的航空网络;近年在国内,武喜萍等研 不敢提示的事例.因此,级别搭配不合理是潜在风 究了空中交通流量网络的静态特性、抗毁性和延 险点.而节点2:因现今主力机型的系统提示是全 误传播模型;王兴隆等刀与齐雁楠和高经东1基 英文的,国内主干和国际航路的空地通话也使用 于空管规则与网络层间逻辑关系,建立相依网络 英文,通话能力对安全影响明显:国际民航英语等 模型;吴明功等则采用独立主元分析在线识别 级低于3级的机组英语通话能力很弱,民航不安 空中交通复杂性;王岩韬团队20于2019年首次 全事件记录中发生过飞行员听不懂提醒的状况 将复杂网络应用于航班运行风险研究,使用SIR 2无向与有向网络构建 (Susceptible-infected-recovered)病毒传播模型分析 了风险传播阈值和变化规律,但其使用的无向网 2.1网络构建 络忽略了风险节点间实则较多的单向作用关系. 根据文献[7]中2009一2014年民航安全监察
as high as 50.6%/58.1%, the risk spread in the network was significantly suppressed. The results confirm that controlling nodes based on the entry degree value is the most effective method to suppress risk propagation in the directed and weighted network. KEY WORDS air transportation;flight operation risk;complex network;directional weighted network;improved SIR model 航班运行风险是发生概率与后果严重程度的 乘积[1] ,其影响种类、形式、数目繁多,且随条件不 断改变. 面对此复杂系统,通过对风险形成机理、 传播过程的原理性分析和研究,从而构建科学的 风险管控方法,是提高安全水平的必要措施. 国外 民航在安全管理研究多集中于风险评估,2000 年 左右已有实用性成果[2–3] . 国内关于民航风险的研 究成果最早见于 1999 年,孙瑞山教授[4] 基于系统 工程构建了航空公司安全评估理论和方法,而后 于 2013 年研发了飞行安全风险评估工具[5] . 王岩 韬团队[6] 于 2014 年以航班运行多因素分析为基 础,提出了一套具有普遍性、实用性的风险体系, 并用于致险因素推断[7] . 后续,采用机器学习多算 法协作方式,于 2019 年将评估精度提升至 95% [8] ; 并且实现了对起飞、巡航、降落等不同飞行阶段 的短时预测[9] ,尝试了一周内的短期预测[10–12] . 而 后,张继凯[13] 针对不同风险等级提出航班差异化 控制策略,并于深圳航空公司实践验证. 至此,风 险分析与量化评估等应用技术问题已得到较好解 决,而风险产生条件、传播过程、控制消散方法等 机理性问题尚缺乏研究. 在上述研究和方案中,存在一种通用假设,即 风险因素间关系呈现树状结构. 然而,运行风险实 则与气象条件、机组资质、机场设施、障碍物、飞 机维修甚至政策保障等多因素相关,而各因素间 通过逻辑关系互相作用,错综交汇最终形成网络 结构. 欧美对于民航领域中复杂网络理论已有一 定研究,具有代表性的成果包括:Belkoura 等[14] 通 过航空网络拓扑结构,从机场、机型、航空公司三 方面分析了网络动态特性,并提出网络结构优化 方法;Voltes-Dorta 等[15] 构建了以机场为节点,航 线为边的航空网络;近年在国内,武喜萍等[16] 研 究了空中交通流量网络的静态特性、抗毁性和延 误传播模型;王兴隆等[17] 与齐雁楠和高经东[18] 基 于空管规则与网络层间逻辑关系,建立相依网络 模型;吴明功等[19] 则采用独立主元分析在线识别 空中交通复杂性;王岩韬团队[20] 于 2019 年首次 将复杂网络应用于航班运行风险研究,使用 SIR (Susceptible-infected-recovered)病毒传播模型分析 了风险传播阈值和变化规律,但其使用的无向网 络忽略了风险节点间实则较多的单向作用关系. 进一步分析和总结发现:①节点间存在单向和双 向作用关系;②不同节点间传播的可能性和作用强度 差异较大;③以往研究中传播初始点随机选择,是有 悖于真实情况的. 这三点问题需要改进和解决. 鉴于以上,依据民航规章选取风险要素,统计 安全监察数据用以改造无向网络,首次构建了航 班运行风险的有向加权网络;然后,引入改进感染 率和改进恢复率概念,提出一种改进 SIR 模型;定 义传染起始范围,综合安全管理方法和网络参数 分析,制定多种控制方案,并进行大量仿真计算; 最后,以风险传播感染数量作为评价标准,旨在得 到具备理论支撑和实践价值的风险管控措施. 1 复杂网络节点选取 网络是指由节点和连接节点的边组成,构建 航班运行风险网络,就是将运行中的风险因素作 为节点,因素之间的相互关系作为连线. 鉴于民航 局咨询通告 AC-121-FS-2015-125 于 2015 年在全国 实施至今,已有 54 家运输航空公司实践多年且验 证有效[21] ,因此选取其作为风险分析体系,具体包 括机组、航空器和运行环境 3 方面共 5 级风险因 素. 其中,终端因素可与运行数据一一对应,可作 为后文建网数据依据,因此将第 5 级的终端因素 作为网络节点,如表 1. 其中,节点编号 1~13 隶属 机组风险类,节点 14~26 隶属运行环境风险类, 节点 27~29 隶属航空器风险类. 以节点 1 为例:机组级别一般分为学员、二 副、一副、机长、教员和检查员. 经由 2009 年至今 的民航不安全事件分析可得,当机组级别差距过 大时交叉检查效果会显著降低,甚至出现过技术 级别高的正驾驶操作不符程序,副驾驶已发现却 不敢提示的事例. 因此,级别搭配不合理是潜在风 险点. 而节点 2:因现今主力机型的系统提示是全 英文的,国内主干和国际航路的空地通话也使用 英文,通话能力对安全影响明显;国际民航英语等 级低于 3 级的机组英语通话能力很弱,民航不安 全事件记录中发生过飞行员听不懂提醒的状况. 2 无向与有向网络构建 2.1 网络构建 根据文献 [7] 中 2009—2014 年民航安全监察 王岩韬等: 基于有向权值网络的航班运行风险传播与控制 · 115 ·
116 工程科学学报,第44卷,第1期 表1风险网络节点 Table 1 Risk network node Node type Node Node name Node Node name number Node type number 1 Crew qualification level matching 14 Temporary air diversion 2 Crew English level 15 Controller's radiotelephone 3 Crew collaboration communication level 4 Crew technical characteristics Large areas of thunderstorms,moderate 16 or severe icy areas,and turbulence Captain flight experience in the airway 6 Captain's familiarity with the airport 17 Rain,snow,fog,and other Crew risk 7 Copilot flying experience weather in airport factors 8 Copilot's familiarity with the airport 18 Runway friction effect Operational environment 9 Transient fatigue risk factors 19 Airport landing standards 10 Cumulative fatigue 20 Flight procedure complexity 11 Special passenger pressure 21 Approach terrain and obstacles Q Flight inspection 22 Airport equipment and facilities status 13 Change route before takeoff 23 Runway length and slope 27 Landing approach involves 品 Airport temporary restriction notice equipment failure Aircraft risk factors 28 Aircraft failure rate 25 Destination airport busyness 29 Navigation database encoding 26 Alternate airport busyness 中特殊事件、工作差错及不安全事件总计25401 性,因此予以清除.关于节点28.当前民航主力机 起的统计资料,当事件记录中显示节点间存在作 型故障率极低,如最为临界的一台发动机故障发 用关系时,即进行连线.举例,如描述中出现了 生概率仅为106.鉴于统计数据中没有体现出影 “因空中临时改航,使得机组操纵时间拉长,… 响,也同样删除.由此,构建该无向网络对应的邻 短期疲劳程度加重”,这种情况则认为节点14和 接矩阵N,建立网络如图1所示.限于数据年限, 9这两个因素间是相关的,进行连线:反之,没有相 未包含737MAX数据.而737MAX因设计缺陷,存 关记录,就不连线四连线后发现,节点28和 在操纵失控状态,本质上是不满足适航标准的,是 29是网络外的孤立节点.分析可知,现今运行规 没有资格进入运行范畴的.此外,鉴于数据未反映 章已将节点29作为起飞基本条件之一,即航班飞 出发动机故障率的影响,则使用其发生概率106 行均满足此条件,失去了风险产生和传播的可能 作为无影响区间判断的阈值 00 1010 1 1000 0 00000000 00 0 0 00 00000 0 0 000000000 00 0 0 0 0 0 0 0 00 0 0 0 0 0 00 0 0 0
中特殊事件、工作差错及不安全事件总计 25401 起的统计资料,当事件记录中显示节点间存在作 用关系时,即进行连线. 举例,如描述中出现了 “因…空中临时改航,使得机组操纵时间拉长,… 短期疲劳程度加重”,这种情况则认为节点 14 和 9 这两个因素间是相关的,进行连线;反之,没有相 关记录 ,就不连线 [22] . 连线后发现 ,节 点 28 和 29 是网络外的孤立节点. 分析可知,现今运行规 章已将节点 29 作为起飞基本条件之一,即航班飞 行均满足此条件,失去了风险产生和传播的可能 性,因此予以清除. 关于节点 28,当前民航主力机 型故障率极低,如最为临界的一台发动机故障发 生概率仅为 10–6 . 鉴于统计数据中没有体现出影 响,也同样删除. 由此,构建该无向网络对应的邻 接矩阵 N,建立网络如图 1 所示. 限于数据年限, 未包含 737MAX 数据. 而 737MAX 因设计缺陷,存 在操纵失控状态,本质上是不满足适航标准的,是 没有资格进入运行范畴的. 此外,鉴于数据未反映 出发动机故障率的影响,则使用其发生概率 10–6 作为无影响区间判断的阈值. N = 0 0 1 1 1 0 1 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 1 1 1 1 1 1 1 0 0 1 1 1 1 0 1 1 1 1 1 0 0 1 1 1 0 1 0 1 1 1 1 1 1 1 0 0 0 0 1 1 0 1 1 0 0 1 0 0 0 1 1 0 1 1 0 1 1 0 1 1 1 0 0 1 0 0 1 0 1 1 1 1 1 0 0 1 1 1 0 1 1 1 0 1 0 1 1 0 0 0 0 0 1 0 0 1 1 0 0 1 0 0 0 0 0 0 1 0 1 0 1 0 1 1 0 0 1 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 1 1 1 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 1 0 0 0 0 1 1 1 0 1 0 1 1 0 1 0 0 0 1 1 1 1 1 0 0 0 0 0 0 1 0 0 1 0 1 0 1 0 1 1 0 0 1 1 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 1 0 0 1 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 1 1 0 1 1 1 0 1 1 0 1 0 0 1 1 0 0 0 1 0 0 1 1 1 0 1 0 1 1 0 0 0 0 0 1 1 1 0 1 1 0 1 1 0 1 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 1 0 1 0 0 0 1 1 0 0 0 0 0 1 1 0 0 0 0 1 1 0 1 1 1 0 0 1 1 1 1 1 1 1 0 1 1 1 0 0 1 1 1 0 1 0 1 1 0 1 1 1 0 1 0 1 1 1 0 0 1 0 0 0 0 0 1 1 0 0 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 1 1 1 1 1 1 1 0 0 1 1 0 0 1 1 1 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 1 1 0 1 0 1 1 0 0 1 1 1 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 1 0 1 1 0 1 1 0 0 1 0 0 1 0 1 0 0 0 0 1 1 1 1 1 1 1 1 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 1 1 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 1 1 0 1 1 1 1 0 0 1 0 0 1 0 1 1 1 1 1 0 0 1 1 0 0 1 1 0 0 1 0 1 1 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 0 1 1 1 1 1 0 1 1 1 表 1 风险网络节点 Table 1 Risk network node Node type Node number Node name Node type Node number Node name Crew risk factors 1 Crew qualification level matching Operational environment risk factors 14 Temporary air diversion 2 Crew English level 15 Controller’s radiotelephone communication level 3 Crew collaboration 4 Crew technical characteristics 16 Large areas of thunderstorms, moderate or severe icy areas, and turbulence 5 Captain flight experience in the airway 6 Captain’s familiarity with the airport 17 Rain, snow, fog, and other weather in airport 7 Copilot flying experience 8 Copilot’s familiarity with the airport 18 Runway friction effect 9 Transient fatigue 19 Airport landing standards 10 Cumulative fatigue 20 Flight procedure complexity 11 Special passenger pressure 21 Approach terrain and obstacles 12 Flight inspection 22 Airport equipment and facilities status 13 Change route before takeoff 23 Runway length and slope Aircraft risk factors 27 Landing approach involves equipment failure 24 Airport temporary restriction notice 28 Aircraft failure rate 25 Destination airport busyness 29 Navigation database encoding 26 Alternate airport busyness · 116 · 工程科学学报,第 44 卷,第 1 期
王岩韬等:基于有向权值网络的航班运行风险传播与控制 .117 因实际运行中风险因素间作用具有方向,且 不同节点间传播概率差别很大.因此,在无向网基 26 础上,构建有方向、带权值的网络,步骤为: (1)定义传播方向.如前述“.空中临时改航, 41 8 使得短期疲劳程度加重”,即设为节点14单向指 10 向节点9 0 (2)定义权值.在步骤(1)的基础上,统计不安 全事件中前一节点影响下一节点的频次,并转化 20 为单向影响发生概率。根据频次及对应概率区间, 定义有向网权值,具体如表2. (3)专家评议.来自国航、南航、东航等国内6家 6 航空公司,共18位飞行、机务、运控专家对步骤(1)和 27 (2)进行轮流评议,再集中讨论各组意见,最终决定删去 图1无向网络 与业务逻辑有悖的连线14条,定义权值取值区间如 Fig.1 Undirected network 表2.最终建立邻接矩阵M,构建有向网络如图2. 表2权值设置规则 Table 2 Weight setting rules Weight setting Probability of previous node affecting the next node Statistical frequency/probability 1 High probability Statistical frequency 100,occurrence probability c(3.94 x 103,1] 0.8 More likely [50,100,1.97×10-3,3.94×10-] 0.5 May occur [10,50:3.94×10,1.97×101 0.2 Low probability [1,10:3.94×10,3.94×10 0 Typically does not affect the next node Statistical frequency =0;[0,3.94 x 10] 度值,两者加和等于总度值.一般情况下,度值越 高,代表该节点对网络的影响越大 度分布是指不同度值的节点总数占网络总节 点的百分比,用于描述网络中不同度数的分布情 况.将网络N个节点中度值为a的节点数记为 Freq(,则计算度分布P)如式(I): 15 P(a)=Freq(a) (1) N (2)聚类系数 点的聚类系数(Clustering coefficient)反映了该 节点的一级近邻间的集团性质,近邻间联系越紧 12 密,节点的聚类系数越高.将节点i与相邻n个节 点间实际存在连边条数记为e,而e,与这些相邻节 图2有向带权网络 点间最大可能存在的连边条数的比值,称为节点 Fig.2 Directional weighted network i的聚类系数,记作C,如式(2): 2.2网络参数 2ei Ci=- (2) (1)度值及度分布 n(n-1) 在有向网络中,点的总度值(Degree value)是 (3)介数. 指与该节点相连边的总数.指向该节点的连边数 点的介数(Betweenness)是一个重要的全局几 量称为入度值,由该节点发出的连边数量称为出 何量参数.在有向网络中,将任意节点j到节点
因实际运行中风险因素间作用具有方向,且 不同节点间传播概率差别很大. 因此,在无向网基 础上,构建有方向、带权值的网络,步骤为: (1)定义传播方向. 如前述“…空中临时改航, 使得短期疲劳程度加重”,即设为节点 14 单向指 向节点 9. (2)定义权值. 在步骤(1)的基础上,统计不安 全事件中前一节点影响下一节点的频次,并转化 为单向影响发生概率. 根据频次及对应概率区间, 定义有向网权值,具体如表 2. (3)专家评议. 来自国航、南航、东航等国内 6 家 航空公司,共 18 位飞行、机务、运控专家对步骤(1)和 (2)进行轮流评议,再集中讨论各组意见,最终决定删去 与业务逻辑有悖的连线 14 条,定义权值取值区间如 表 2. 最终建立邻接矩阵 M,构建有向网络如图 2. 表 2 权值设置规则 Table 2 Weight setting rules Weight setting Probability of previous node affecting the next node Statistical frequency/probability 1 High probability Statistical frequency ≥ 100, occurrence probability ϵ (3.94 × 10−3,1] 0.8 More likely [50, 100; 1.97 × 10−3, 3.94 × 10−3] 0.5 May occur [10, 50; 3.94 × 10−4, 1.97 × 10−3] 0.2 Low probability [1, 10; 3.94 × 10−5, 3.94 × 10−4] 0 Typically does not affect the next node Statistical frequency = 0; [0, 3.94 × 10−5] 2.2 网络参数 (1)度值及度分布. 在有向网络中,点的总度值(Degree value)是 指与该节点相连边的总数. 指向该节点的连边数 量称为入度值,由该节点发出的连边数量称为出 度值,两者加和等于总度值. 一般情况下,度值越 高,代表该节点对网络的影响越大. 度分布是指不同度值的节点总数占网络总节 点的百分比,用于描述网络中不同度数的分布情 况. 将网络 N 个节点中度值为 α 的节点数记为 Freq(k),则计算度分布 P(k) 如式(1): P(α) = Freq(α) N (1) (2)聚类系数. 点的聚类系数(Clustering coefficient)反映了该 节点的一级近邻间的集团性质,近邻间联系越紧 密,节点的聚类系数越高. 将节点 i 与相邻 n 个节 点间实际存在连边条数记为 ei,而 ei 与这些相邻节 点间最大可能存在的连边条数的比值,称为节点 i 的聚类系数,记作 Ci,如式(2): Ci = 2ei n(n−1) (2) (3)介数. 点的介数(Betweenness)是一个重要的全局几 何量参数. 在有向网络中,将任意节点 j 到节点 25 13 2 15 26 12 24 17 18 8 23 4 1 27 5 7 3 20 6 21 22 11 14 10 9 16 19 图 1 无向网络 Fig.1 Undirected network 1 4 5 7 3 6 8 22 15 11 2 26 13 25 12 18 16 17 19 20 21 23 27 24 14 10 9 图 2 有向带权网络 Fig.2 Directional weighted network 王岩韬等: 基于有向权值网络的航班运行风险传播与控制 · 117 ·