Sea area airport runway toughness evaluation and improvement method based on graph neural network

Abstract

The invention relates to the technical field of airport runway toughness evaluation, in particular to a sea area airport runway toughness evaluation and improvement method based on a graph neural network, and the method comprises the steps: calculating the fault mode similarity and fault influence degree between components, taking a combined component as a logic node, and constructing an initial topological graph in combination with a spatial distance; fusing the marine environment anomaly probability and the real-time environment deviation, updating the initial topological graph, and generating an optimized topological graph; and on the basis of the optimized topological graph, historical monitoring data and historical sea area environment data, a pre-training graph neural network model is adopted to obtain a toughness index of each logic node so as to screen risk logic nodes, and a toughness improvement measure is performed in a targeted manner. According to the method, the dynamic topological graph fused with the environmental characteristics is constructed, so that the precision and reliability of toughness evaluation of the sea area airport runway are improved, and continuous and safe operation of the runway is ensured.

Description

Technical Field

[0001] This application relates to the field of airport runway resilience assessment technology, specifically to a method for assessing and improving the resilience of marine airport runways based on graph neural networks. Background Technology

[0002] Graph Neural Networks (GNNs) are deep learning models specifically designed for processing graph-structured data. Their core lies in deeply mining the relationships between entities, enabling feature learning and accurate analysis of non-Euclidean data, effectively compensating for the modeling shortcomings of traditional neural networks (such as CNNs and RNNs) when dealing with irregular data. Coastal airport runways, constructed using marine space, adapted to the unique marine environment, and employing differentiated technologies to ensure structural stability and safety, are crucial aviation infrastructures for coastal cities to alleviate land resource constraints and expand air transport capacity.

[0003] While relying on marine space effectively saves land resources and improves takeoff and landing safety, offshore airport runways inevitably face severe challenges from the complex marine environment, including tides, corrosion, and waves. This makes accurate assessment and enhancement of their resilience particularly crucial. Existing resilience assessment methods mainly rely on "structural monitoring + environmental monitoring" and intelligent maintenance and repair, but they still have significant limitations: On the one hand, existing methods often neglect the global correlation between components, resulting in the inability to capture the chain reactions triggered by anomalies, making preventive maintenance lack comprehensiveness; on the other hand, the analysis of multi-source environmental data often remains at a simple superposition level, severing the deep correlation and complementary mechanisms between environmental factors and runway failure events, leading to a lack of assessment dimensions and severely restricting the accuracy and reliability of offshore airport runway resilience assessment. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method for assessing and improving the resilience of marine airport runways based on graph neural networks, thereby resolving existing issues.

[0005] The method for assessing and improving the resilience of marine airport runways based on graph neural networks in this application adopts the following technical solution: One embodiment of this application provides a method for assessing and improving the resilience of marine airport runways based on graph neural networks. The method includes the following steps: Collect historical fault records and multi-source monitoring data of various components on the airport runway in the sea area, as well as marine environmental data; Based on historical fault records, the number of the same fault type and the time of occurrence of the faults among different components are counted to calculate the similarity of fault modes and the degree of fault impact among different components. Then, components that meet the conditions are merged into a logical node. The spatial distance between any two logical nodes and the degree of fault impact are fused to construct the initial topology graph at the current moment. Based on historical fault records and historical marine environment data, the probability of logical nodes failing when the marine environment is abnormal is statistically analyzed to calculate the influence strength of the marine environment on the edges in the initial topology graph. Combined with the deviation of the marine environment data at the current moment from the average level of historical marine environment data, the final weight of the edge in the graph neural network is calculated, and then the initial topology graph is updated to generate the optimized topology graph at the current moment. The vector composed of multi-source monitoring data and marine environmental data of all components under each logical node in the optimized topology map is input into the pre-trained graph neural network model to obtain the resilience index of each logical node. Based on the resilience index, risk logic nodes are screened, and the resilience of components under the risk logic nodes is improved.

[0006] Preferably, the calculation of the similarity of failure modes and the degree of failure impact among different components includes: From the historical fault records, extract the historical fault types of component i and component j, count the number of the same fault types and the total number of fault types of the two, and record the proportion of the number of the same fault types of the two to the total number of fault types of the two as the fault mode similarity between component i and component j. From the historical failure records, we count the number of times component j fails after component i fails, and the total number of times component i fails. We calculate the proportion of the former to the latter, which is denoted as the failure impact degree of component i on component j. This is used to characterize the probability that component j will fail after component i fails.

[0007] Preferably, merging components that meet the conditions into a single logical node includes: Based on the fault mode similarity and the fault impact, the fault correlation degree from component i to component j and the fault correlation degree from component j to component i are calculated respectively. If the normalized value of the minimum fault correlation degree between component i and component j is greater than the preset correlation threshold, and the spatial straight-line distance between component i and component j is less than the preset clustering radius, then component i and component j are merged into a single logical node. Otherwise, they are not merged. All components are traversed, and components that meet the conditions are merged.

[0008] Preferably, the fault correlation degree between any two components is positively correlated with the fault mode similarity and the fault impact degree, respectively.

[0009] Preferably, the initial topology graph at the current moment includes: Each component in the m-th logical node is combined with all components in the n-th logical node to form a component pair, and the mean of the fault impact of all component pairs is calculated; the ratio of the mean of the fault impact between the m-th logical node and the n-th logical node to the spatial distance is normalized and denoted as the edge connection coefficient between the m-th logical node and the n-th logical node. If the edge connection coefficient between the m-th logical node and the n-th logical node is greater than or equal to the preset connection threshold, then a connection is established between the m-th logical node and the n-th logical node; otherwise, no connection is established. Traverse all logical nodes to obtain the initial topology graph.

[0010] Preferably, the calculation of the influence intensity of the marine environment on the edges in the initial topology graph includes: Based on historical fault records and historical marine environment data, the total number of times the k-th marine environment data exceeds its corresponding preset normal standard value and the total number of historical faults of all components in the two logical nodes on the v-th edge in the initial topology graph when the k-th marine environment data exceeds the preset normal standard value are respectively recorded as the first statistical value and the second statistical value. Calculate the average value of the amount by which the environmental data of the kth marine area exceeds the preset normal standard value, and denote it as the average anomaly intensity of the environmental data of the kth marine area. By combining the first statistical value, the second statistical value, and the average anomaly intensity, the influence intensity of the k-th marine environmental data on the v-th edge is calculated.

[0011] Preferably, the calculation of the final weights of the edges in the graph neural network includes: Marine environmental data that exceeds the preset normal standard value is recorded as marine environmental data exceeding the standard, and the historical environmental impact weight of the v-th edge is calculated. , its value is the weighted sum of the influence intensity of all types of marine environmental data exceeding the standard on the v-th edge, and the weight is the proportion of the total failure degree of all components in the two logical nodes of the v-th edge under each type of marine environmental data exceeding the standard to the sum of the total failure degree of all components in the two logical nodes of the v-th edge under all types of marine environmental data exceeding the standard. Calculate the deviation rate of the current marine environmental data relative to historical marine environmental data. And the deviation rate of the multi-source monitoring data of each of the two logical nodes on the v-th edge relative to the historical multi-source monitoring data. , ; Calculate the consistency of the influence of the marine environment on the v-th edge. The influence of consistency is respectively the same as the deviation rate. and Differences, deviation rates and The differences are negatively correlated; Based on historical environmental impact weights Impact on consistency and deviation rate and Calculate the final weight of the v-th edge.

[0012] Preferably, the influence consistency The calculation formula is: In the formula, norm() represents the normalization function.

[0013] Preferably, updating the initial topology graph includes: If the final weight of edge v in the initial topology graph is less than the preset structure construction threshold, then edge v is removed from the initial topology graph; otherwise, edge v is retained. All edges in the initial topology graph are traversed, and the initial topology graph is updated.

[0014] Preferably, the risk logic node is a logic node in the optimized topology graph whose resilience index is less than a preset resilience threshold.

[0015] This application has at least the following beneficial effects: The purpose of this invention is to analyze the correlation and complementarity between components and between components and environmental factors, thereby obtaining the mutual influence between different components and the proportion of environmental factors in the process of mutual influence, thereby improving the accuracy of the assessment of the toughness of marine airport runways and thus improving the effectiveness of the improvement process.

[0016] The specific scenario targeted by this invention is: This application constructs an initial topology graph that combines functional logic and physical characteristics by integrating historical fault records and spatial topological relationships. Specifically, it aggregates components into homogeneous logical nodes by utilizing fault mode similarity and correlation, and establishes sparse connections between nodes by combining spatial distance and fault impact. This effectively solves the pain point of "close but not related, related but not connected" in traditional methods, eliminates noise interference and redundant edges, and provides a high-fidelity topology foundation for subsequent dynamic correction by introducing marine environmental factors and high-precision resilience assessment of graph neural networks. Furthermore, this application dynamically calculates the final weights of graph neural network edges by integrating historical fault data and real-time environmental deviations, and updates the initial topology graph based on a preset pruning threshold to generate an optimized topology graph at the current moment. This method quantifies the real-time modulation effect of marine environmental factors on the node association strength. By eliminating invalid edges that are weakly affected by environmental interference or have lost their association, it effectively solves the problem that the initial topology graph structure cannot reflect the real working conditions. Thus, while retaining the core risk transmission path, it removes noise interference, providing a high-fidelity graph structure foundation that can adapt to dynamic changes in the marine environment for high-precision resilience assessment of graph neural networks. Furthermore, this application optimizes the topology graph and inputs the node feature vectors containing multi-source monitoring data and marine environmental data into the graph neural network model. Through the model's feature learning and aggregation mechanism, the resilience index of each logical node is output. This method makes full use of the environmental modulation information and component correlation contained in the dynamic graph structure, overcomes the limitations of single data evaluation, and realizes the real-time and accurate quantification of the resilience status of marine airport runway components under different environmental conditions. This improves the accuracy and reliability of marine airport runway resilience assessment and provides core data support for subsequent risk identification and resilience improvement. Ultimately, this application accurately screens out risk logic nodes in a sub-healthy state based on high-standard preset resilience thresholds. By deeply analyzing the coupling relationship between node types and marine environmental data, it identifies the dominant environmental interference factors, then matches and executes targeted resilience enhancement plans, and conducts closed-loop review and iterative optimization after remediation until the resilience index of risk nodes reaches the standard. This moves the risk prevention and control checkpoint forward, achieving precise governance and proactive defense of vulnerable links in marine airport runways, ensuring their continuous safe operation in complex marine environments. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the steps of a method for assessing and improving the resilience of marine airport runways based on graph neural networks, provided in one embodiment of this application. Figure 2 This is a flowchart of a component merging process provided in one embodiment of this application. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the method for assessing and improving the resilience of marine airport runways based on graph neural networks proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0021] The following description, in conjunction with the accompanying drawings, details the specific scheme of the graph neural network-based method for assessing and improving the resilience of marine airport runways provided in this application.

[0022] This application provides an embodiment of a graph neural network-based method for assessing and improving the resilience of marine airport runways. Specifically, the method is described below. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps: Step S1: Collect historical fault records and multi-source monitoring data of various components on the airport runway in the sea area, as well as marine environmental data.

[0023] First, for the components of the marine airport runway, historical fault records of the current marine airport runway are extracted by reading the inspection and maintenance logs. Next, multi-source monitoring data of various components on the marine airport runway are collected in real time using sensors installed at their corresponding locations. These components are categorized into several types, including in this embodiment: pavement layer, base layer, pile foundation, revetment structure, drainage system, etc. The process of acquiring multi-source monitoring data specifically includes: 1. Embedded fiber optic sensor and laser flatness detector: installed on the paving layer to collect surface stress, crack width and flatness of the paving layer in real time.

[0024] 2. Vibrating wire displacement sensor and ground-penetrating radar: installed on the base layer to collect data on the vertical displacement, compaction degree and moisture content of the base layer in real time.

[0025] 3. Built-in strain gauge and ultrasonic flaw detector: installed in the pile foundation, used to collect the axial stress, horizontal displacement and corrosion depth of the pile foundation in real time.

[0026] 4. Earth pressure cell, piezometer, and total station: installed on the revetment structure to collect real-time data on lateral earth pressure, top displacement, and seepage flow of the revetment structure.

[0027] 5. Electromagnetic flow meter and liquid level sensor: installed in the drainage system to collect real-time data on the flow velocity and water depth in the drainage system.

[0028] Furthermore, marine environmental data is collected from meteorological stations and marine environmental sensors deployed around the runway. In this embodiment, this includes seawater salinity, wave period, and water level change. Implementers may also set the types and quantities of marine environmental data to be collected according to specific circumstances. This embodiment does not impose any special restrictions.

[0029] Finally, the data obtained above is preprocessed by outlier removal and missing value completion. The outlier removal and missing value completion processes are well-known techniques, and the specific implementation process will not be described in detail.

[0030] Step S2: Based on historical fault records, count the number of the same fault type and the time of occurrence of the fault among different components to calculate the similarity of fault modes and the degree of fault impact among different components, and then merge the components that meet the conditions into a logical node; fuse the spatial distance between any two logical nodes and the degree of fault impact to construct the initial topology graph at the current moment.

[0031] While spatially uniformly partitioning airport runways in a sea area can effectively divide the physical components of the runway initially, this method relies solely on geographical location, neglecting the consistency of failure modes and deep functional connections between components. This spatial distance-based partitioning method can easily lead to the forced connection of components with vastly different failure types or unrelated functions when constructing the graph structure, introducing noisy connections and reducing the accuracy of subsequent resilience assessments. Furthermore, if edges are constructed solely based on spatial proximity, components that are not geographically adjacent but are functionally tightly coupled may be overlooked, resulting in the graph structure lacking crucial connection paths and ultimately leading to incomplete and inaccurate resilience assessment results.

[0032] Therefore, in order to construct a topology graph that includes both spatial topological relationships and functional logical relationships, this embodiment first uses historical fault records to count the number of the same fault type and the time of occurrence of faults among different components, in order to calculate the fault mode similarity and fault impact between different components, and then merges the components that meet the conditions into a single logical node; then, it integrates the spatial distance between any two logical nodes and the fault impact to construct the initial topology graph at the current moment, as detailed below: First, based on historical fault records, the number of identical fault types and the time of occurrence of faults among different components are statistically analyzed to calculate the similarity of fault modes and the degree of fault impact among different components. Specifically: In this embodiment, taking any component i and component j as an example, the following calculation is performed: extract the historical fault types of component i and component j from the historical fault records, count the number of the same fault types of the two and the total number of the two fault types, and record the proportion of the number of the same fault types of the two in the total number of the two fault types as the fault mode similarity between component i and component j. From the historical fault records, we count the number of times component j fails after component i fails, and the total number of times component i fails. We calculate the proportion of the former to the latter, which is denoted as the fault impact degree of component i on component j.

[0033] Furthermore, based on the similarity of failure modes and the degree of failure impact, components that meet the conditions are merged into a single logical node. Specifically: Based on the fault mode similarity and the fault impact, the fault correlation degree between any two components is calculated. Specifically, the fault correlation degree between any two components is positively correlated with the fault mode similarity and the fault impact.

[0034] It should be understood that a positive correlation means that the dependent variable increases as the independent variable increases, and the dependent variable decreases as the independent variable decreases. The specific relationship can be additive or multiplicative, etc., and is determined by the actual application. This application does not impose any special restrictions.

[0035] Preferably, as an implementation method, in this embodiment, the fault mode similarity between component i and component j is multiplied by the fault influence of component i on component j, and the result is used as the fault correlation degree from component i to component j. Similarly, the fault mode similarity between component j and component i is multiplied by the fault influence of component j on component i, and the result is used as the fault correlation degree from component j to component i. That is, the fault correlation degree between any two components includes two values.

[0036] If the normalized value of the minimum fault correlation degree between component i and component j is greater than the preset correlation threshold, and the spatial straight-line distance between component i and component j is less than the preset clustering radius, then component i and component j are merged into a single logical node. Otherwise, they are not merged. All components are traversed, and components that meet the conditions are merged.

[0037] It should be noted that in this embodiment, the preset clustering radius is 50 meters, which aims to ensure that the merged logical nodes maintain spatial connectivity and avoid incorrectly aggregating components that are too far apart in physical distance and cannot generate direct mechanical transmission, thereby ensuring the accuracy of the spatial features of the subsequent topology graph. In practical applications, as other implementation methods, implementers can also set their own values ​​according to specific circumstances. This embodiment does not impose any special restrictions.

[0038] Preferably, the flowchart of the component merging process provided in this embodiment is as follows: Figure 2 As shown.

[0039] In this embodiment, the preset association threshold is set to 0.8. This is based on a strict screening requirement for the tightness of functional coupling between components. The aim is to aggregate components that are physically independent but highly coordinated in fault logic into the same logical node. Only when two components are highly similar in fault mode and have a significant fault propagation relationship in time series will they be judged to belong to the same functional block. This value can effectively avoid erroneously merging components with occasional or weak correlation, thereby ensuring that the generated logical nodes have clear physical meaning and fault consistency. This lays a solid node foundation for the subsequent construction of a high-precision graph neural network model. In practical applications, as other implementation methods, implementers can also set their own values ​​according to specific circumstances. This embodiment does not impose any special restrictions.

[0040] It should be noted that there are many commonly used normalization methods. In this embodiment, the maximum and minimum value normalization method is used to normalize the fault correlation degree to the range of [0,1]. In actual application, as other implementation methods, implementers can also use other normalization methods according to specific circumstances. This embodiment does not impose any special restrictions on the selection of normalization methods.

[0041] The process of normalizing data using the maximum-minimum normalization method is a well-known technique, so the process of normalizing the multiplication result using it will not be described in detail.

[0042] It should be noted that, unless otherwise specified, all normalization in this embodiment uses the maximum-minimum value normalization method.

[0043] Furthermore, this embodiment constructs an initial topology graph at the current moment by integrating the spatial distance between any two logical nodes and the fault impact degree. Specifically: In this embodiment, each component in the m-th logical node and all components in the n-th logical node are combined into a component pair, and the average fault impact of all component pairs is calculated; the ratio of the average fault impact between the m-th logical node and the n-th logical node to the spatial distance is normalized and recorded as the edge connection coefficient between the m-th logical node and the n-th logical node. If the edge connection coefficient between the m-th logical node and the n-th logical node is greater than or equal to the preset connection threshold, then a connection is established between the m-th logical node and the n-th logical node; otherwise, no connection is established. Traverse all logical nodes to obtain the initial topology graph.

[0044] It should be noted that the preset connection threshold of 0.8 in this embodiment is used to eliminate noisy connections and retain the most decisive key association paths when constructing the initial topology graph, ensuring the sparsity and effectiveness of the graph structure. Since the edge connection coefficient combines spatial proximity and functional correlation, the high threshold of 0.8 means that only those pairs of nodes that are spatially close and have strong functional interactions will establish edge connections. This effectively filters out weak correlation interference caused by proximity but unrelated functions, focusing on the core links that can truly determine the overall resilience transmission of the airport runway in the sea area. This prevents the graph structure from being too complex and redundant, and improves the efficiency of subsequent graph neural network processing and the accuracy of feature extraction. In practical applications, as other implementation methods, implementers can also set their own values ​​according to specific circumstances. This embodiment does not impose any special restrictions.

[0045] To further clarify, the calculation process for the spatial distance between the m-th logical node and the n-th logical node is as follows: each component in the m-th logical node is combined with all components in the n-th logical node to form a component pair, the distance between any component pair is calculated, and the average of the distances of all component pairs is taken as the spatial distance between the m-th logical node and the n-th logical node.

[0046] Based on the entire construction process of the initial topology graph described above, it can be understood that this embodiment does not construct a static physical topology based solely on geographical location, but rather an intelligent dynamic knowledge graph that deeply integrates historical fault statistics and functional logical relationships. This construction method is chosen because there are complex fault coupling effects among the components of a sea-area airport runway. Simple physical proximity does not necessarily imply risk transmission, while components physically far apart may have fatal synergistic effects on specific functional chains. First, components are merged into logical nodes using fault mode similarity and fault impact, achieving a granular upgrade from "points" to "clusters," ensuring homogeneity within logical nodes. Then, edges are constructed by integrating the spatial distance and fault impact of logical nodes. This embodiment simultaneously considers the spatial continuity of the structure and the risk transmission of functions, thus forming an initial topology graph that reflects both the physical skeleton of the runway and reveals the evolutionary patterns of potential risks. This construction method solves the pain point of "near but unrelated, related but not connected" in traditional graph construction methods, providing a high-fidelity initial structural foundation for subsequent dynamic correction by introducing environmental factors.

[0047] Thus, this embodiment constructs an initial topology graph that combines functional logic and physical characteristics by integrating historical fault records and spatial topological relationships. Specifically, it aggregates components into homogeneous logical nodes by utilizing fault mode similarity and correlation, and establishes sparse connections between nodes by combining spatial distance and fault impact. This effectively solves the pain point of "near but not related, related but not connected" in traditional methods, eliminates noise interference and redundant edges, and provides a high-fidelity topology foundation for subsequent dynamic correction by introducing marine environmental factors and high-precision resilience assessment of graph neural networks.

[0048] Step S3: Based on historical fault records and historical marine environment data, calculate the probability of logical nodes failing when the marine environment is abnormal, in order to calculate the influence strength of the marine environment on the edges in the initial topology graph. Combined with the deviation of the marine environment data at the current moment from the average level of historical marine environment data, calculate the final weight of the edge in the graph neural network, and then update the initial topology graph to generate the optimized topology graph at the current moment.

[0049] After determining the initial topology map of the marine airport runway, the construction of the initial topology map is mainly based on the historical failure statistics of the components, which reflects the static functions and spatial relationships between the components. It does not fully consider the dynamic interference of the marine environment on the state of the components. The core difference between marine airport runways and land airport runways is that marine airport runways are exposed to the complex marine environment for a long time and are significantly affected by dynamic factors such as tides, corrosion, wind and waves. If the real-time modulation effect of these environmental factors on the strength of the relationship between components and nodes is ignored, and the static map is used directly for evaluation, the toughness assessment results will not accurately reflect the real working conditions.

[0050] Therefore, this embodiment needs to further analyze marine environmental data and quantify the impact of environmental factors on logical nodes. Specifically, based on historical fault records and historical marine environmental data, this embodiment calculates the probability of logical nodes failing when the marine environment is abnormal, in order to calculate the influence strength of the marine environment on edges in the initial topology graph. Combined with the deviation of the current marine environmental data from the average level of historical marine environmental data, the final weight of the edge in the graph neural network is calculated, thereby updating the initial topology graph and generating the optimized topology graph at the current moment. The specific process is as follows: First, this embodiment, based on historical fault records and historical marine environment data, statistically analyzes the probability of logical nodes failing when the marine environment is abnormal, in order to calculate the intensity of the marine environment's influence on edges in the initial topology graph. Specifically: Based on historical fault records and historical marine environment data, the total number of times the k-th marine environment data exceeds its corresponding preset normal standard value and the total number of historical faults of all components in the two logical nodes on the v-th edge in the initial topology graph when the k-th marine environment data exceeds the preset normal standard value are respectively recorded as the first statistical value and the second statistical value. Calculate the average value of the amount by which the environmental data of the kth marine area exceeds the preset normal standard value, and denote it as the average anomaly intensity of the environmental data of the kth marine area. By combining the first statistical value, the second statistical value, and the average anomaly intensity, the influence intensity of the k-th marine environmental data on the v-th edge is calculated. Specifically: the influence intensity of the k-th marine environmental data on the v-th edge. The expression is: In the formula, This represents the total number of historical failures of all components within the two logical nodes on the v-th edge of the initial topology graph when the k-th marine environmental data exceeds the preset normal standard value; that is, the second statistical value. This represents the total number of times the k-th type of marine environmental data exceeds its corresponding preset normal standard value, i.e., the first statistical value; represents the average anomaly intensity of the k-th type of marine environmental data; norm() represents the normalization function; This indicates a preset constant greater than 0, used to prevent the denominator from being 0. The value of is 0.01 in this embodiment. Under the premise of ensuring that the denominator is not 0 and does not excessively affect the calculation result, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0051] Among them, the preset normal standard values ​​of marine environmental data can be obtained from the technical specifications for corrosion protection of port engineering structures in the sea area where the airport runway is located.

[0052] Based on the concept of impact strength, it can be understood that impact strength is a core indicator that quantifies the destructive ability of a specific marine environmental factor (such as salinity and waves) to two logical nodes connected by an edge in a topological graph. It reflects the depth of the causal relationship between environmental anomalies and node failures, and is used to characterize the extent to which marine environmental factors dominate the risk transmission weight of that edge. The calculation of impact strength is influenced by three key factors: first, the frequency of simultaneous failures of nodes at both ends of the edge when there is an environmental anomaly (i.e., the ratio of the second statistical value to the first statistical value); second, the total frequency of marine environmental anomalies; and third, the average anomaly intensity of marine environmental data exceeding normal standards. Specifically, the higher the frequency of simultaneous failures of logical nodes due to a particular type of marine environmental anomaly... The greater the average magnitude of marine environmental exceedances, i.e., the greater the average anomaly intensity, the higher the calculated impact intensity value. This reflects the extremely strong destructive and stressful nature of the current marine environmental factors on the components within the logical node. In subsequent assessments, high impact intensity will cause edges to gain greater weight in the graph neural network, thus making the model pay more attention to this path that is severely threatened by the environment. Conversely, if the frequency of node failures or the magnitude of exceedances is low when the current marine environmental anomaly occurs, the impact intensity value will be lower. This reflects that the impact of the current marine environmental factors on the edges is weak or sporadic, and the components are mainly affected by their own aging or other factors. The environmental weight of the edges in the resilience assessment is reduced accordingly, thereby avoiding the model from misjudging risks that are not dominated by the environment.

[0053] Furthermore, this embodiment calculates the final weights of edges in the graph neural network based on the strength of the influence of the marine environment on the edges in the initial topology graph, and in combination with the degree of deviation of the current marine environment data from the average level of historical marine environment data. Specifically: In this embodiment, marine environmental data that exceeds the preset normal standard value is recorded as substandard marine environmental data, and the historical environmental impact weight of the v-th edge is calculated. , where is the weighted sum of the influence intensity of all types of marine environmental data exceeding the standard on the v-th edge. The weight is the proportion of the total fault degree of all components within the two logical nodes of the v-th edge under each type of marine environmental data exceeding the standard to the sum of the total fault degrees of all components within the two logical nodes of the v-th edge under all types of marine environmental data exceeding the standard. The calculation method for the fault degree of each component under the r-th type of marine environmental data exceeding the standard is as follows: calculate the ratio of the real-time multi-source monitoring data of each component to the corresponding preset standard value, denoted as the monitoring ratio; take the average of the monitoring ratios of all types of monitoring data as the fault degree of each component; and take the sum of the fault degrees of all components as the total fault degree of the component under the r-th type of marine environmental data exceeding the standard. For example, if the standard value of the crack width in the multi-source monitoring data of component b under the r-th type of marine environmental data exceeding the standard is 0.2 mm, then calculate the ratio of the actual monitored crack width to the standard value of the crack width as the monitoring ratio of the crack width; and take the average of the monitoring ratios of all types of multi-source monitoring data as the fault degree of each component. The preset standard values ​​for monitoring data can also be obtained from the technical specifications for corrosion protection of port engineering structures in the sea area where the airport runway is located.

[0054] Calculate the deviation rate of the current marine environmental data relative to historical marine environmental data. And the deviation rate of the multi-source monitoring data of each of the two logical nodes on the v-th edge relative to the historical multi-source monitoring data. , Specifically, the deviation of the k-th marine environmental data from its historical average level is taken as the deviation of the k-th marine environmental data. The average deviation of all marine environmental data is taken as the deviation rate of the current marine environmental data relative to historical marine environmental data. The deviation of any monitoring data point for each component from its historical average level is also calculated and recorded as the deviation of that monitoring data point. The average deviation of all monitoring data points for each component is recorded as the deviation characteristic value of that component. The average deviation characteristic value of all nodes under each logical node is taken as the deviation rate of each logical node. .

[0055] It should be noted that when the facility is newly built or the historical monitoring data accumulation time is less than the preset time threshold, the design standard value of the sea area airport is used as the historical average value to participate in the deviation calculation. The preset time threshold is 3 years. In actual application, as another implementation method, the implementer can also set it according to the specific situation. This embodiment does not impose any special restrictions.

[0056] Calculate the consistency of the influence of the marine environment on the v-th edge. The influence of consistency is respectively the same as the deviation rate. and Differences, deviation rates and The differences are negatively correlated; It should be understood that a negative correlation means that the dependent variable decreases as the independent variable increases, and the dependent variable increases as the independent variable decreases. It can be a subtraction relationship or a division relationship, depending on the actual application.

[0057] Preferably, in this embodiment, the consistency of the influence of the vth edge is... The calculation formula is: In the formula, norm() represents the normalization function.

[0058] Furthermore, based on the weight of historical environmental impact... Impact on consistency and deviation rate and Calculate the final weight of the v-th edge, specifically: The final weight of the v-th edge in this embodiment The expression is: , where norm() represents the normalization function, which iterates through all edges in the initial topology graph to obtain the final weight of each edge.

[0059] Based on the final weight, it can be understood that the final weight is a quantitative assessment of the importance and activity of each edge in the graph structure at the current moment, after comprehensively considering historical lessons learned and real-time state changes. It reflects the effectiveness of edges in participating in resilience assessment under specific marine environmental conditions and is used to characterize whether edges should be retained and used for information transmission in the graph neural network. The calculation of the final weight is mainly affected by three factors: the weight of historical environmental influence, the consistency of influence, and the difference in deviation rate between logical nodes. Specifically: when the weight of historical environmental influence is large, it indicates that the long-term destructive power of the marine environment is strong; when the consistency of influence is high, it indicates that the marine environment is causing node anomalies in real time; and when the difference in deviation rate between logical nodes is high... A lower degree indicates strong real-time resonance or synchronous deterioration among logical nodes, resulting in a higher final weight value. This reflects that the edge is in an active state of "high risk, strong correlation, and real-time damage," representing a vulnerable and critical link in the current maritime airport runway. It must be retained in the optimized topology graph and given high attention to ensure that the graph neural network can capture the core risks. Conversely, if the historical environment has a smaller impact weight, lower consistency, and greater difference in deviation rate among logical nodes, the final weight value will be lower. This reflects that the edge is in a state of "low risk, failure, or isolation" in the current environment, contributing negligibly to the resilience assessment. Therefore, it will be pruned and removed when updating the topology graph.

[0060] Furthermore, based on the final weights of the edges in the graph neural network, the initial topology graph is updated to generate the optimized topology graph at the current time step. Specifically: If the final weight of edge v in the initial topology graph is less than the preset structure construction threshold, then edge v is removed from the initial topology graph; otherwise, edge v is retained. Traverse all edges in the initial topology graph, update the initial topology graph, and denote the updated topology graph as the optimized topology graph.

[0061] It should be noted that the preset structural pruning threshold is set manually. In this embodiment, the preset structural pruning threshold is 0.4. This is to establish a "suppressing falsehoods and retaining the true" filtering mechanism during the dynamic simplification of the graph structure. The aim is to eliminate invalid edges that are weakly affected by the environment or have lost their correlation, while retaining potential weakly correlated paths. This value avoids over-pruning caused by an excessively high threshold, preventing the omission of hidden risks by eliminating some potential correlated links that may be activated under extreme conditions. It also prevents the graph structure from becoming too dense due to an excessively low threshold, avoiding a large number of irrelevant weak connections that interfere with the feature extraction of the graph neural network. Therefore, the threshold setting of 0.4 achieves appropriate pruning of the graph structure, ensuring both computational efficiency and maintaining the comprehensiveness and robustness of the system risk assessment.

[0062] Thus, this embodiment dynamically calculates the final weights of graph neural network edges by integrating historical fault data and real-time environmental deviations, and updates the initial topology graph based on a preset pruning threshold to generate an optimized topology graph for the current moment. This method quantifies the real-time modulation effect of marine environmental factors on the node association strength. By eliminating invalid edges that are weakly affected by environmental interference or have lost their association, it effectively solves the problem that the initial topology graph structure cannot reflect the real working conditions. This removes noise interference while retaining the core risk transmission path, providing a high-fidelity graph structure foundation that can adapt to dynamic changes in the marine environment for high-precision resilience assessment of graph neural networks.

[0063] Step S4: Input the vector composed of the optimized topology map and the multi-source monitoring data of all components under each logical node and marine environmental data into the pre-trained graph neural network model to obtain the resilience index of each logical node.

[0064] After obtaining the optimized topology map of the current sea area airport runway, based on the graph structure and combined with the graph neural network model, the vector composed of multi-source monitoring data and marine environmental data of all components under each logical node in the optimized topology map and the optimized topology map is input into the pre-trained graph neural network model to obtain the resilience index of each logical node, so as to achieve accurate assessment of the resilience of the sea area airport runway components. The specific process is as follows: the vector composed of multi-source monitoring data and marine environmental data of all components under each logical node in the optimized topology map and the optimized topology map are input into the pre-trained graph neural network model (GNN), and the resilience index of each logical node in the optimized topology map at the current time is output.

[0065] To further clarify, the training process of the pre-trained graph neural network model is as follows: First, historical multi-source monitoring data, historical marine environmental data, and corresponding topological maps are used as input samples, and a training dataset is constructed by combining the resilience index generated from historical expert ratings as a supervision label. Then, the samples are input into the graph neural network model for forward propagation to obtain the predicted resilience index. By calculating the loss function error between the predicted resilience index and the true resilience index generated from historical expert ratings, the backpropagation algorithm is used to iteratively update the graph aggregation weights and parameters inside the model until the error converges, thereby enabling the model to grasp the mapping law from environmental disturbances and component association features to the resilience index.

[0066] The specific process of pre-training graph neural networks is a well-known technique and will not be elaborated further.

[0067] In this embodiment, the topology graph and node feature vectors containing multi-source monitoring data and marine environmental data are input into the graph neural network model. Through the model's feature learning and aggregation mechanism, the resilience index of each logical node is output. This method makes full use of the environmental modulation information and component correlation contained in the dynamic graph structure, overcomes the limitations of single data evaluation, and realizes real-time and accurate quantification of the resilience status of marine airport runway components under different environmental conditions. That is, it improves the accuracy and reliability of marine airport runway resilience assessment and provides core data support for subsequent risk identification and resilience improvement.

[0068] Step S5: Based on the resilience index, screen risk logic nodes and improve the resilience of components under the risk logic nodes.

[0069] First, risk logic nodes are screened based on the resilience index, specifically: Logic nodes in the optimized topology graph whose resilience index is less than the preset resilience threshold are designated as risk logic nodes. It should be noted that the preset resilience threshold in this embodiment is 0.9. The preset resilience threshold of 0.9 in this embodiment is based on the extremely high safety and reliability requirements of marine airport runways as critical transportation infrastructure. A risk control strategy of "better to be strict than lenient" is adopted. The resilience threshold of 0.9 means that only when the logical node shows extremely small deviations in structural integrity and environmental adaptability is it considered safe and acceptable. Early warning and improvement measures are triggered when the risk is still in its infancy, thereby moving the safety defense line forward, preventing losses due to minor issues, and ensuring that the marine airport runway can maintain absolute safety margin and operational resilience under any operating conditions.

[0070] Furthermore, the types of risk logic nodes are obtained, and their coupling relationship with marine environmental data is analyzed to determine the dominant marine environmental influencing factors. For example, pile foundation risk corresponds to high salinity, and pavement layer risk corresponds to high water level. Further, based on the combination of "risk logic node type + dominant marine environmental factor influence", the preset toughness improvement measures library is queried to obtain the corresponding toughness improvement solutions. For example, for pile foundations under high salinity, measures such as adding anti-corrosion coatings and pile foundation reinforcement are adopted, and for pavement layers under high water level, measures such as crack repair and drainage system expansion are adopted.

[0071] Furthermore, after the resilience of the risk logic node is remedied, the resilience index of the risk logic node is recalculated. The risk logic node is then judged to meet the standard according to the risk logic node screening method in step S4. If it does not meet the standard, the same resilience enhancement measures are added until the resilience meets the standard.

[0072] Thus, this embodiment accurately screens out risk logic nodes in a sub-healthy state based on a high-standard preset resilience threshold. By deeply analyzing the coupling relationship between node types and marine environmental data, it identifies the dominant environmental interference factors, matches and executes targeted resilience enhancement schemes, and conducts closed-loop review and iterative optimization after remediation until the resilience index of the risk nodes reaches the standard. This moves the risk prevention and control checkpoint forward, achieving precise governance and proactive defense of vulnerable links in marine airport runways, and ensuring their continuous safe operation in complex marine environments.

[0073] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0074] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0075] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them; modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions of some of the technical features, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for assessing and improving the resilience of marine airport runways based on graph neural networks, characterized in that, The method includes the following steps: Collect historical fault records and multi-source monitoring data of various components on the airport runway in the sea area, as well as marine environmental data; Based on historical fault records, the number of the same fault type and the time of occurrence of the faults among different components are counted to calculate the similarity of fault modes and the degree of fault impact among different components. Then, components that meet the conditions are merged into a logical node. The spatial distance between any two logical nodes and the degree of fault impact are fused to construct the initial topology graph at the current moment. Based on historical fault records and historical marine environment data, the probability of logical nodes failing when the marine environment is abnormal is statistically analyzed to calculate the influence strength of the marine environment on the edges in the initial topology graph. Combined with the deviation of the marine environment data at the current moment from the average level of historical marine environment data, the final weight of the edge in the graph neural network is calculated, and then the initial topology graph is updated to generate the optimized topology graph at the current moment. The vector composed of multi-source monitoring data and marine environmental data of all components under each logical node in the optimized topology map is input into the pre-trained graph neural network model to obtain the resilience index of each logical node. Based on the resilience index, risk logic nodes are screened, and the resilience of components under the risk logic nodes is improved.

2. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The calculation of the similarity of failure modes and the degree of failure impact among different components includes: From the historical fault records, extract the historical fault types of component i and component j, count the number of the same fault types and the total number of fault types of the two, and record the proportion of the number of the same fault types of the two to the total number of fault types of the two as the fault mode similarity between component i and component j. From the historical failure records, we count the number of times component j fails after component i fails, and the total number of times component i fails. We calculate the proportion of the former to the latter, which is denoted as the failure impact degree of component i on component j. This is used to characterize the probability that component j will fail after component i fails.

3. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The step of merging components that meet the conditions into a single logical node includes: Based on the fault mode similarity and the fault impact, the fault correlation degree from component i to component j and the fault correlation degree from component j to component i are calculated respectively. If the normalized value of the minimum fault correlation degree between component i and component j is greater than the preset correlation threshold, and the spatial straight-line distance between component i and component j is less than the preset clustering radius, then component i and component j are merged into a single logical node. Otherwise, they are not merged. All components are traversed, and components that meet the conditions are merged.

4. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 3, characterized in that, The fault correlation between any two components is positively correlated with the fault mode similarity and the fault impact, respectively.

5. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The initial topology graph at the current moment includes: Each component in the m-th logical node is combined with all components in the n-th logical node to form a component pair, and the mean of the fault impact of all component pairs is calculated; the ratio of the mean of the fault impact between the m-th logical node and the n-th logical node to the spatial distance is normalized and denoted as the edge connection coefficient between the m-th logical node and the n-th logical node. If the edge connection coefficient between the m-th logical node and the n-th logical node is greater than or equal to the preset connection threshold, then a connection is established between the m-th logical node and the n-th logical node; otherwise, no connection is established. Traverse all logical nodes to obtain the initial topology graph.

6. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The calculation of the influence intensity of the marine environment on the edges in the initial topology graph includes: Based on historical fault records and historical marine environment data, the total number of times the k-th marine environment data exceeds its corresponding preset normal standard value and the total number of historical faults of all components in the two logical nodes on the v-th edge in the initial topology graph when the k-th marine environment data exceeds the preset normal standard value are respectively recorded as the first statistical value and the second statistical value. Calculate the average value of the amount by which the environmental data of the kth marine area exceeds the preset normal standard value, and denote it as the average anomaly intensity of the environmental data of the kth marine area. By combining the first statistical value, the second statistical value, and the average anomaly intensity, the influence intensity of the k-th marine environmental data on the v-th edge is calculated.

7. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 6, characterized in that, The calculation of the final weights of the edges in the graph neural network includes: Marine environmental data that exceeds the preset normal standard value is recorded as marine environmental data exceeding the standard, and the historical environmental impact weight of the v-th edge is calculated. , its value is the weighted sum of the influence intensity of all types of marine environmental data exceeding the standard on the v-th edge, and the weight is the proportion of the total failure degree of all components in the two logical nodes of the v-th edge under each type of marine environmental data exceeding the standard to the sum of the total failure degree of all components in the two logical nodes of the v-th edge under all types of marine environmental data exceeding the standard. Calculate the deviation rate of the current marine environmental data relative to historical marine environmental data. And the deviation rate of the multi-source monitoring data of each of the two logical nodes on the v-th edge relative to the historical multi-source monitoring data. , ; Calculate the consistency of the influence of the marine environment on the v-th edge. The influence of consistency is the same as the deviation rate. and Differences, deviation rates and The differences are negatively correlated; Based on historical environmental impact weights Impact on consistency and deviation rate and Calculate the final weight of the v-th edge.

8. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 7, characterized in that, The impact on consistency The calculation formula is: In the formula, norm() represents the normalization function.

9. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The update of the initial topology graph includes: If the final weight of edge v in the initial topology graph is less than the preset structure construction threshold, then edge v is removed from the initial topology graph; otherwise, edge v is retained. All edges in the initial topology graph are traversed, and the initial topology graph is updated.

10. The method for assessing and improving the resilience of marine airport runways based on graph neural networks as described in claim 1, characterized in that, The risk logic node is a logic node in the optimized topology graph whose resilience index is less than a preset resilience threshold.

Images