Airport apron operation risk assessment method based on complex network

Abstract

The invention discloses an airport apron operation risk assessment method based on a complex network, and belongs to the technical field of risk control. The method comprises the following steps: analyzing an airport apron safety report by adopting an event chain analysis method to construct a complex network; determining a risk static index of each node; simulating a dynamic propagation process of risks in the complex network based on an SIRS model, and determining a risk dynamic index of each node; by taking the dynamic and static indexes of each node as characteristics, classifying the nodes by adopting a spectral clustering algorithm, respectively calculating an average risk comprehensive value of each cluster node, and determining a risk level of each cluster node based on a quantile; and acquiring risk events of the airport apron at a certain moment, determining a risk comprehensive value and a risk level of each risk event, and performing hierarchical and classified management on the risk events. According to the method, a multi-dimensional risk assessment index system is constructed to assess the risk propagation capability and blocking capability of the nodes, classification and grading of the risk nodes are realized, and scientific decision support can be provided for airport managers.

Description

Technical Field

[0001] The present invention relates to the technical field of risk control, and in particular to an airport apron operation risk assessment method based on a complex network. Background Art

[0002] Airport aprons are an important part of the aviation transportation system. Their operation requires the coordination of multiple personnel, equipment and processes, which is highly complex and risky. Traditional apron safety management methods often focus on statistical analysis of individual risk events, such as accident frequency, accident cause classification, etc. However, this static analysis is difficult to reveal the correlation and propagation path between risk events, especially in complex scenarios with multiple factors coupled, and has great limitations.

[0003] In recent years, complex network theory has been gradually applied to the field of aviation risk management. For example, through network analysis, the risk propagation behavior of pilots in flight missions is studied; by constructing interactive networks, the impact of air traffic controllers' operational behaviors on overall operational safety is analyzed. However, most of these studies focus on dynamic interaction analysis at the pilot and controller levels, and few studies apply complex network theory to ground operations, especially risk propagation and management of airport apron operations.

[0004] Airport apron operation risks are unique: there are many types of personnel and equipment involved, the operation scenarios are highly dynamic, and multiple risk factors (such as equipment failure, human error, and environmental factors) interact with each other. These characteristics determine that it is difficult to accurately assess the risks in apron operations by relying solely on static indicators. Summary of the invention

[0005] In order to solve the above problems, the present invention provides an airport apron operation risk assessment method based on complex networks, which adopts a multi-dimensional risk assessment indicator that integrates static and dynamic features to improve the comprehensiveness and accuracy of risk assessment.

[0006] In order to achieve the above object, the solution provided by the present invention is as follows: A method for airport apron operation risk assessment based on complex networks includes the following steps: S1. Analyze the apron safety report using the event chain analysis method and construct a complex network; the complex network uses events as nodes and establishes directed edges based on the dependency relationship between events, and the node with zero out-degree value is the top event node; S2. Based on the complex network, determine the risk static indicators of each node. The risk static indicators include accident neighbor risk rate and goal-oriented centrality. The accident neighbor risk rate of a node is the ratio of the proportion of top event nodes in the out-degree neighbor nodes of the node to the average proportion of top event nodes in the out-degree neighbor nodes of all nodes in the entire complex network. It is used to evaluate the direct impact of the node on the occurrence of the accident, that is, to measure the possibility of the node directly causing the accident. It is a local indicator; goal-oriented centrality is used to measure the node i On the slave node s Top event node t The importance of being an intermediary node in all shortest paths is a global indicator used to identify key nodes in the network that have important indirect effects on the occurrence of accidents; The calculation formula of goal-oriented centrality is as follows: ; In the formula, C og ( i ) is a node i The goal-oriented centrality of ; Г is the set of top event nodes; П is the set of non-top event nodes, that is, the set of all nodes in the network except the top event nodes; d st ( i ) indicates a slave node s Along a directed edge to a node t All shortest paths passing through the node i The number of paths, d st Represents a slave node s Along directed edges to nodes t The total number of all shortest paths; S3. Based on the SIRS (susceptible-infected-recovered-re-susceptible) model, the dynamic propagation process of risks in the complex network is simulated to determine the risk dynamic indicators of each node. The risk dynamic indicators include the average peak infection ratio and the average peak infection period, which are used to quantify the risk propagation ability of each node and evaluate the impact range and propagation speed of each node on the entire network. The state of the node in the SIRS model evolves in the order of susceptible state-infected state-recovered state, among which the node in the infected state evolves according to the infection rate. β Infect the susceptible nodes to the infected state, and then c Recover to the recovery state, the nodes in the recovery state are susceptible again m Transformed into a susceptible state again, leading to the re-transmission of risk; S4. Using the static and dynamic indicators of each node in the complex network as features, the spectral clustering algorithm is used to classify the nodes, and the average comprehensive risk value of each cluster node is calculated respectively, and the risk level of each cluster node is determined based on the quantile; The average risk composite value is calculated as follows: ; ; SC ( i ) is the i The average comprehensive risk value of cluster nodes; N i For the i A collection of cluster nodes; SCI ( j ) is a node j The greater the risk comprehensive value, the greater the risk; S k ( j ) is a node j No. k Normalized eigenvalues ​​of the features, where the normalized eigenvalue of the average peak infection period is obtained by reverse minimum-maximum normalization, and other normalized eigenvalues ​​are obtained by minimum-maximum normalization; S5. Obtain the risk events of the airport apron at a certain moment, determine the comprehensive risk value and risk level of each risk event, give priority to risk events with high comprehensive risk values, and adopt corresponding management strategies for each risk event according to its risk level.

[0007] In step S1, the risk events recorded in the airport apron operation are converted into a unified directed complex network through event chain analysis, so as to fully reflect the causal relationship and risk propagation path between the risk events. As a specific implementation of the present invention, step S1 includes: S11. Extract all risky events in a certain apron safety report and determine the subnetwork of event development in the apron safety report based on the event chain analysis method. In the subnetwork, events are nodes, and each event is divided into three categories: initial event, development event and top event according to its position in the event chain. Among them, the initial event is the starting point of the event chain, the development event is the propagation link of the event chain, and the top event is the end point of the event chain. The node corresponding to the top event is the top event node, and it does not have directed edges pointing to other nodes; in the subnetwork, each node is connected by a directed edge, and the direction of the directed edge is determined according to the dependency relationship between events; S12. Repeat step S11 to establish a sub-network for each apron safety report, and then link and integrate all sub-networks to form a complex network.

[0008] As a specific implementation of the present invention, the calculation formula of the accident neighbor risk rate is as follows: ; ; ; ; In the formula, APR( i ) is a node i The accident neighbor risk rate; P nb ( i ) is a node i The probability that an accident is directly caused by the out-degree neighbor node; is the average probability of an accident being directly caused by the out-degree neighbor nodes of all nodes in the complex network; N is the total number of nodes in the complex network; d ( u ,Г) is the indicator node u The indicator function belonging to the top event node set Г; Nb( i ) is a node i The set of out-degree neighbor nodes; N( i ) is Nb( i )The total number of nodes in As a specific implementation of the present invention, obtaining the average peak infection ratio and the average peak infection period of each node includes the following steps: S31, initializing parameters and determining the infection rate set; S32, selecting a non-top event node from the complex network, setting the selected node to an infected state, and setting the remaining nodes to a susceptible state; S33, selecting an infection rate from the infection rate set and substituting it into the SIRS model, repeatedly simulating the virus propagation process of the selected node, counting the peak infection ratio and the time corresponding to the peak infection ratio in the entire complex network during each virus propagation process, and then calculating the average values ​​as the first peak infection ratio and the first infection cycle of the selected node under the infection rate; S34, repeating step S33, traversing all infection rates in the infection rate set, obtaining the first peak infection ratio and the first infection period of the selected node under different infection rates, and then calculating the average values ​​respectively as the average peak infection ratio and the average peak infection period of the selected node; S35. Repeat steps S32-S34 to traverse all non-top event nodes in the complex network.

[0009] In step S5, corresponding management strategies adopted for different risk levels can be formulated as needed, such as adopting different monitoring frequencies for different risk levels, or taking targeted management measures in combination with various dynamic indicators and static indicators.

[0010] Beneficial effects: The present invention constructs a multi-dimensional risk assessment index system to evaluate the risk propagation and blocking capabilities of nodes, realizes the classification and grading of risk nodes, and can provide scientific decision-making support for airport managers.

[0011] The present invention selects the SIRS model to simulate the dynamic transmission process of risks in apron operations. This choice is based on the particularity of aviation operations. Unlike general operations, relevant units in the aviation field usually use specific risk management strategies to suppress potential risks and ensure the normal operation of the system. However, as time goes by, the effect of risk management may gradually weaken, causing some nodes to be exposed to risks again. In the SIRS model of the present invention, nodes in the recovery state are re-susceptible at a higher rate. m Transforming into a susceptible state again can better reflect the risk transmission characteristics in apron operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a schematic diagram of a complex network in an embodiment of the present invention; Figure 2 It is a two-dimensional visualization of high-dimensional data after dimensionality reduction. DETAILED DESCRIPTION

[0013] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0014] A method for airport apron operation risk assessment based on complex networks includes the following steps: S1. Use event chain analysis method to analyze the apron safety report and construct a complex network; the complex network uses events as nodes and establishes directed edges according to the dependencies between events. The node with zero directed edges pointing to other nodes is the top event node, which specifically includes the following steps: S11. Extract all risky events in a certain apron safety report and determine the subnetwork of event development in the apron safety report based on the event chain analysis method. In the subnetwork, events are nodes, and each event is divided into three categories: initial event, development event and top event according to its position in the event chain. Among them, the initial event is the starting point of the event chain, the development event is the propagation link of the event chain, the top event is the end point of the event chain, and the node corresponding to the top event is the top event node. In this embodiment, the initial event and the development event in the subnetwork are risk factors for the occurrence of the top event. These risk factors are divided into four categories: personnel, equipment, environment and management. Each category includes several specific events. For details, refer to Table 1; in the subnetwork, each node is connected by a directed edge, and the direction of the directed edge is determined according to the dependency relationship between events. There are two types of dependency relationships, namely causal dependency and conditional dependency. Among them, causal dependency refers to one event directly causing another event, and there is a clear unidirectional relationship. Conditional dependency means that one event creates conditions for the occurrence of another event, but does not directly cause it.

[0015] S12. Repeat step S11 to establish a sub-network for each apron safety report, and then link and integrate all sub-networks to form a complex network.

[0016] Table 1 Risk node statistics

[0017] The complex network constructed in this embodiment is as follows Figure 1 As shown, it contains 82 nodes and 244 directed edges, which comprehensively reflects the risk propagation path between risk events.

[0018] S2. Based on the complex network, determine the risk static indicators of each node, the risk static indicators include accident neighbor risk rate and goal-oriented centrality, the accident neighbor risk rate is the ratio of the proportion of top event nodes in the out-degree neighbor nodes of the node to the average proportion of top event nodes in the out-degree neighbor nodes of all nodes in the entire complex network; The calculation formula of goal-oriented centrality is as follows: ; In the formula, C og ( i ) is a node i The goal-oriented centrality of ; Г is the set of top event nodes; П is the set of non-top event nodes, that is, the set of all nodes in the network except the top event nodes; d st ( i ) indicates a slave node s Along directed edges to nodes t All shortest paths passing through the node i The number of paths, d st Represents a slave node s Along directed edges to nodes t The total number of all shortest paths; The calculation formula of the accident neighbor risk rate is as follows: ; ; ; ; In the formula, APR( i ) is a node i The accident neighbor risk rate, P nb ( i ) is a node i The probability that an accident is directly caused by the out-degree neighbor node; is the average probability of an accident being directly caused by the out-degree neighbor nodes of all nodes in the complex network; N is the total number of nodes in the complex network; d ( u ,Г) is the indicator node u The indicator function belonging to the top event node set Г; Nb( i ) is a node i The set of out-degree neighbor nodes; N( i ) is Nb( i )The total number of nodes in S3. Based on the SIRS (susceptible-infected-recovered-re-susceptible) model, the dynamic propagation process of risks in the complex network is simulated to determine the risk dynamic indicators of each node, wherein the risk dynamic indicators include the average peak infection ratio and the average peak infection period, which are used to quantify the risk propagation capability of each node and evaluate the impact range and propagation speed of each node on the entire network; The state of the node in the SIRS model evolves in the order of susceptible state-infected state-recovered state, where the node is in one of the susceptible state, infected state, and recovered state. The susceptible state is a healthy state that is easily infected, the infected state is an infected state, and the recovered state is a healthy state that recovers from the infected state to a healthy state that is not easily infected. The nodes in the infected state evolve according to the infection rate. β Infect the susceptible nodes to the infected state, and then c Recover to the recovery state, the nodes in the recovery state are susceptible again m Transformed into a susceptible state again, leading to the re-transmission of risk; The dynamics of the SIRS model can be described by the following set of equations: ; ; In the formula, S ( t ), I ( t ), R ( t ) are time t The proportion of nodes in susceptible state, infected state and recovered state at that time.

[0019] Obtaining the average peak infection ratio and average peak infection period of each node includes the following steps: S31, initialize parameters, determine the infection rate set B = { β j | j ∈[1, k ]}, k is the total number of infection rates in set B; S32, selecting a non-top event node from the complex network, placing the selected node in an infected state, and placing the remaining nodes in a susceptible state; S33, selecting an infection ratio from the infection rate set and substituting it into the SIRS model, repeatedly simulating the virus propagation process of the selected node, counting the peak infection ratio and the time corresponding to the peak infection ratio in the entire complex network during each virus propagation process, and then calculating the average values ​​as the first peak infection ratio and the first infection cycle of the selected node under the infection rate; ; ; ; ; In the formula, I peam,m ( v i , β j ) is a node v i The infection rate β j Next m The peak infection rate during the entire simulation at the time of the simulation; T is the total time; max(*) is the maximum value function; I m ( t ; v i , β j ) is a node v i The infection rate βj Next m The time of simulation t The infection ratio of the entire complex network at that time; argmax(*) is the return function, which returns the time when the function takes the maximum value; T peam,m ( v i , β j )for I m ( t ; v i , β j ) corresponding to the time; For Node v i The infection rate β j The first peak infection ratio under v i The infection rate β j The total number of simulations; For Node v i The infection rate β j The first infection cycle under S34, repeating step S33, traversing all infection rates in the infection rate set, obtaining the first peak infection ratio and the first infection period of the selected node under different infection rates, and then calculating the average values ​​respectively as the average peak infection ratio and the average peak infection period of the selected node; ; ; In the formula, For Node v i The average peak infection rate of For Node v i The average infection period S35, repeating steps S32-S34 to traverse all non-top event nodes in the complex network; S4. Using the static and dynamic indicators of each node in the complex network as features, the spectral clustering algorithm is used to classify the nodes, and the average comprehensive risk value of each cluster node is calculated respectively, and the risk level of each cluster node is determined based on the quantile; The average risk composite value is calculated as follows: ; ; SC ( i ) is the i The average comprehensive risk value of cluster nodes; N i For the i A collection of cluster nodes; SCI ( j ) is a node j The greater the risk comprehensive value, the greater the risk; S k ( j ) is a node j No. k Normalized eigenvalues ​​of the features, where the normalized eigenvalue of the average peak infection period is obtained by reverse minimum and maximum (min-max) normalization, and other normalized eigenvalues ​​are obtained by minimum and maximum (min-max) normalization.

[0020] In order to compare the adaptability of various clustering algorithms, this embodiment compares multiple clustering algorithms through the Calinski-Harabasz index (CH index for short). The comparison results are shown in Table 2. It can be seen from Table 2 that the spectral clustering algorithm has the best effect.

[0021] Table 2 CH index of three clustering models

[0022] The nodes are divided into four categories using the spectral clustering algorithm. In order to more intuitively display the spectral clustering results and analyze the distribution of different risk nodes in the dimensionality reduction space, this embodiment uses the Uniform Manifold Approximation and Projection (UMAP) to reduce the dimensionality of high-dimensional data. Figure 2 shown. Figure 2 The distribution of four types of risk nodes in two-dimensional space after high-dimensional data dimensionality reduction is shown, which can more clearly observe the distribution characteristics of the four types of risk nodes in the reduced dimensional space. The average value of risk indicators and the number of nodes of each type are shown in Table 3.

[0023] Table 3 Average values ​​of various node risk indicators and number of nodes

[0024] The comprehensive risk value of each node is shown in Table 4, and the average comprehensive risk value of each cluster node is shown in Table 5.

[0025] Table 4 Comprehensive risk value of each node

[0026] Table 5 Average comprehensive risk value of each cluster node

[0027] In this embodiment, the risk level of each cluster node is divided as follows: taking 75% and 25% as quantiles, if the average comprehensive risk value is greater than or equal to the comprehensive risk value corresponding to the 75% quantile, it is high risk; if the average comprehensive risk value is less than the comprehensive risk value corresponding to the 25% quantile, it is low risk; otherwise, it is medium risk.

[0028] According to the above classification rules, the comprehensive risk value corresponding to the 75% quantile is 0.41, and the comprehensive risk value corresponding to the 25% quantile is 0.17. Therefore, Cluster 1 is high risk, Cluster 2 and Cluster 3 are medium risk, and Cluster 4 is low risk.

[0029] S5. Obtain the risk events of the airport apron at a certain moment, determine the comprehensive risk value and risk level of each risk event, give priority to risk events with high comprehensive risk values, and adopt corresponding management strategies for each risk event according to its risk level.

[0030] For low-risk clusters, only basic monitoring and optimized resource allocation are required to avoid unnecessary management cost investment. For medium and high risk clusters, static and dynamic indicators are combined to sort them and match the corresponding management strategies: the cluster with the highest accident neighbor risk rate has the highest possibility of accident occurrence, and real-time monitoring and strict operation specifications are needed to prevent individual violations from directly causing accidents. Management measures include strengthening on-site inspections, increasing safety training, and establishing a violation accountability mechanism; the cluster with the highest goal-oriented centrality plays a key intermediary role in the accident transmission chain, and it is necessary to optimize process management and improve the stability of key nodes to prevent the entire system from being affected by individual node errors. Possible measures include multi-level decision review, optimization of cross-departmental coordination mechanism, and ensuring accurate information transmission; the cluster with the highest average peak infection ratio has the greatest impact on accident transmission, and a transmission isolation mechanism needs to be established to prevent the spread of risks. Possible measures include optimizing information transmission methods, adjusting processes to isolate key nodes with high transmission risks, and establishing a rapid response mechanism; the cluster with the lowest average peak infection period means that the risk spreads rapidly in a short period of time, and rapid intervention and optimization of emergency response processes are required. The management focus is to shorten the emergency response time and ensure that sudden risks can be discovered and handled in a timely manner.

[0031] For this embodiment, combined with Table 3 and Table 5, it can be seen that cluster 4 is low risk and only requires basic monitoring and optimized resource allocation. The other three clusters are medium and high risk. The four indicators are ranked respectively. Cluster 1 has the highest average accident neighbor risk rate and the shortest average peak infection period. Management measures include strengthening on-site inspections, increasing safety training, establishing a mechanism for accountability for violations, and optimizing emergency response processes to shorten emergency response time; Cluster 2 has the highest goal-oriented centrality, and management measures include multi-level decision-making review and optimization of cross-departmental coordination mechanisms to ensure accurate information transmission; Cluster 3 has the highest average peak infection ratio, and management measures include optimizing information transmission methods, adjusting processes to isolate key nodes with high transmission risks, and establishing a rapid response mechanism.

[0032] The above are only preferred specific implementations of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed in the embodiments of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for airport apron operation risk assessment based on complex networks, characterized in that: The steps include: S1. Analyze the apron safety report using the event chain analysis method and construct a complex network; the complex network uses events as nodes and establishes directed edges based on the dependency relationship between events, and the node with zero out-degree value is the top event node; S2. Based on the complex network, determine the risk static index of each node; the risk static index includes the accident neighbor risk rate and the goal-oriented centrality. The accident neighbor risk rate of a node is the ratio of the proportion of top event nodes in the out-degree neighbor nodes of the node to the average value of the proportion of top event nodes in the out-degree neighbor nodes of all nodes in the entire complex network; The goal-oriented centrality is calculated as follows: ； In the formula, C og ( i ) is a node i goal-oriented centrality; Г is the set of top event nodes; П is the set of non-top event nodes; δ st ( i ) indicates a slave node s Along a directed edge to a node t All shortest paths passing through the node i The number of paths; δ st Represents a slave node s Along a directed edge to a node t The total number of all shortest paths; S3. Based on the SIRS model, the dynamic propagation process of risk in the complex network is simulated to determine the risk dynamic indicators of each node; the risk dynamic indicators include the average peak infection ratio and the average peak infection period. The state of the node in the SIRS model evolves in the order of susceptible state-infected state-recovered state, among which the node in the infected state evolves according to the infection rate. β Infect the susceptible nodes to the infected state, and then γ Recover to the recovery state, the nodes in the recovery state are susceptible again μ Transformed into susceptible state again; S4, using the static index and dynamic index of each node in the complex network as features, classifying the nodes using a spectral clustering algorithm, calculating the average comprehensive risk value of each cluster node respectively, and determining the risk level of each cluster node based on the quantile; The calculation formula of the average risk composite value is as follows: ； ； SC ( i ) is the i The average comprehensive risk value of cluster nodes; N i For the i A collection of cluster nodes; SCI ( j ) is a node j The greater the risk comprehensive value, the greater the risk; S k ( j ) is a node j No. k Normalized eigenvalues ​​of the features, where the normalized eigenvalue of the average peak infection period is obtained by reverse minimum-maximum normalization, and other normalized eigenvalues ​​are obtained by minimum-maximum normalization; S5. Obtain the risk events of the airport apron at a certain moment, determine the comprehensive risk value and risk level of each risk event, give priority to risk events with high comprehensive risk values, and adopt corresponding management strategies for each risk event according to its risk level.

2. According to the complex network-based airport apron operation risk assessment method of claim 1, it is characterized in that: Step S1 includes: S11, extract all risky events in a certain apron safety report and determine the sub-network of event development in the apron safety report based on the event chain analysis method; in the sub-network, events are used as nodes, and each event is divided into three categories: initial event, development event and top event according to the position of the event in the event chain, wherein the initial event is the starting point of the event chain, the development event is the propagation link of the event chain, and the top event is the end point of the event chain. Each node is connected by a directed edge, and the direction of the directed edge is determined according to the dependency relationship between the events; S12. Repeat step S11 to establish a sub-network for each apron safety report, and then link and integrate all sub-networks to form a complex network.

3. According to the complex network-based airport apron operation risk assessment method of claim 1, it is characterized in that: The calculation formula of the accident neighbor risk rate is as follows: ； ； ； ； In the formula, APR( i ) is a node i The accident neighbor risk rate; P nb ( i ) is a node i The probability that an accident is directly caused by the out-degree neighbor node; is the average probability of an accident being directly caused by the out-degree neighbor nodes of all nodes in the complex network; N is the total number of nodes in the complex network; δ ( u ,Г) is the indicator node u The indicator function belonging to the top event node set Г; Nb( i ) is a node i The set of out-degree neighbor nodes; N( i ) is Nb( i )The total number of nodes in .

4. According to the complex network-based airport apron operation risk assessment method of claim 1, it is characterized in that: Obtaining the average peak infection ratio and average peak infection period of each node includes the following steps: S31, initializing parameters and determining the infection rate set; S32, selecting a non-top event node from the complex network, setting the selected node to an infected state, and setting the remaining nodes to a susceptible state; S33, selecting an infection rate from the infection rate set and substituting it into the SIRS model, repeatedly simulating the virus propagation process of the selected node, counting the peak infection ratio and the time corresponding to the peak infection ratio in the entire complex network during each virus propagation process, and then calculating the average values ​​as the first peak infection ratio and the first infection period of the selected node under the infection rate; S34, repeating step S33, traversing all infection rates in the infection rate set, obtaining the first peak infection ratio and the first infection period of the selected node under different infection rates, and then calculating the average values ​​respectively as the average peak infection ratio and the average peak infection period of the selected node; S35. Repeat steps S32-S34 to traverse all non-top event nodes in the complex network to obtain the average peak infection ratio and average peak infection period of each node.

Images