A Network Resilience Evaluation Method Based on Data Fusion and Adaptive Weight Update

Abstract

This invention discloses a network resilience evaluation method based on data fusion and adaptive weight update in the field of data processing technology. The method includes acquiring headquarters-branch office data of industries in a target region, classifying the headquarters-branch office data, fusing the classified data at the city level, preprocessing the fused data to obtain preprocessed data, setting corresponding KPIs based on the headquarters-branch office data, calculating the KPI change rate based on historical and newly added headquarters-branch office data, and calculating the updated coupling strength between headquarters-branch office data based on the KPI change rate. This invention constructs a hierarchical, multi-source spatiotemporal data industry network model, and combines it with an adaptive association strength update method and a spatiotemporal hierarchical weight adjustment mechanism to achieve real-time updates of headquarters-branch office coupling strength and node / edge capacity, thereby improving the accuracy of industry network resilience assessment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically a network resilience evaluation method based on data fusion and adaptive weight update. Background Technology

[0002] As cities expand and their functional layouts become increasingly sophisticated, the differentiated development of various urban networks provides varying degrees of technological support for urban development. Among these, industrial networks, as crucial hubs connecting economic activities within urban clusters, play a key role in maintaining regional economic stability and driving systemic growth. The resilience of industrial networks directly determines the recovery capacity and sustainable development potential of the urban cluster's economic system in the face of external shocks such as market fluctuations, natural disasters, or policy adjustments, as well as internal imbalances within headquarters or branches.

[0003] However, existing methods for studying network resilience are mostly based on macroscopic integrated analysis using fully integrated, static data, often neglecting the different structural and functional performances of multi-level nodes, such as headquarters and branches. This results in an insufficient characterization of the internal interaction processes of the industrial network, leading to a discrepancy between the accuracy of the network resilience assessment and the actual value. Meanwhile, empirical studies targeting urbanization processes and regional industrial organization characteristics have not yet incorporated multi-source, multi-temporal and spatial data fusion and adaptive weight update strategies into their evaluation frameworks, making it difficult to meet the needs for "precise" and "dynamic" monitoring and early warning of industrial network resilience. Summary of the Invention

[0004] The purpose of this invention is to provide a network resilience evaluation method based on data fusion and adaptive weight update. By building an industrial network model with hierarchical, multi-source spatiotemporal data, and combining it with an adaptive correlation strength update method and a spatiotemporal hierarchical weight adjustment mechanism, the method can realize real-time updates of the coupling strength between headquarters and branches and the capacity of nodes and connections, thereby improving the accuracy of industrial network resilience assessment.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A network resilience evaluation method based on data fusion and adaptive weight update includes: Acquire headquarters-branch data of industries in the target area, classify the headquarters-branch data, merge the classified data at the city level, and preprocess the merged data to obtain preprocessed data. The headquarters-branch data includes industry type data, latitude and longitude coordinates of the headquarters, latitude and longitude coordinates of the branches, number of enterprises of the same industry type, connection path between headquarters and branches, and enterprise operation data. Based on the headquarters-branch data, set corresponding KPIs (Key Performance Indicators), calculate the KPI change rate based on historical and new headquarters-branch data, and calculate the updated coupling strength between the headquarters-branch data based on the KPI change rate. Based on the latitude and longitude coordinates, the geographic layer data and external shock index of the industry are obtained. Based on the geographic layer data and external shock index, the coupling strength and the rate of change of the coupling strength are calculated. Based on the rate of change, the dynamic simulation parameters are adjusted. The dynamic simulation parameters include city node capacity, enterprise removal threshold, cascade failure ratio and network reconfiguration ratio. The number of headquarters and branches in each of the cities is counted, and the city information of the headquarters and branches is obtained to obtain city nodes and city pairs. The city nodes and city pairs are marked on the spatial map using the annotation method. Construct a multi-industry directed hierarchical network Gi(N, E), where Gi represents the network of each industry category, node N is the city node after integrating headquarters and branches, and edge E is the directed city pair. A data simulation method is used to perform multi-dimensional perturbation simulations on each directed hierarchical network Gi(N, E), and the first network resilience of the directed hierarchical network is calculated. ; For each directed hierarchical network Gi(N, E), a cascade failure simulation is performed to calculate the resilience of the second network. ; For each directed hierarchical network Gi(N, E), network reconfiguration simulation is performed to calculate the resilience of the third network. ; Output first network resilience Second network resilience and third network resilience Based on industry and geographical factors, resilience change curves were plotted.

[0006] As a further aspect of the present invention: the headquarters-branch office connection path is a directed line segment used to represent the spatial relationship between the headquarters and the branch office.

[0007] As a further aspect of the present invention: a city node is the city information of the headquarters structure and the city information of the branch offices, and a city pair is a combination of the city information of the headquarters and the branch offices corresponding to the headquarters.

[0008] Preferably, the text format for city nodes and city pairs is: city where the headquarters is located, city where the branch offices are located, number of enterprises.

[0009] As a further aspect of the present invention: in outputting the first network resilience Second network resilience and third network resilience Subsequently, resilience change curves are plotted by combining industrial and geographical factors to support multi-dimensional industrial collaborative development and risk prevention and control decisions.

[0010] As a further aspect of the present invention: the industry type data includes manufacturing data, construction data, general production service data, technology-intensive productive service data, and general service industry data.

[0011] As a further aspect of the present invention: the classification of the headquarters-branch data and the fusion of the classified data at the city level, followed by preprocessing of the fused data to obtain preprocessed data, includes: The headquarters-branch data is divided into operational data, information data, and geographic spatiotemporal data types to obtain categorized data; The categorized data is fused at the city level using a data fusion method; The city-level classification data is deduplicated using a data deduplication method to obtain deduplicated data. Based on the deduplicated data, interpolation is used to interpolate the missing temporal and spatial values ​​in the deduplicated data to obtain smooth data; Based on the smoothed data, the normalization method is used to clean the smoothed data to obtain cleaned data; Based on the cleaning data, the cleaning data is standardized using a data standardization method to obtain preprocessed data.

[0012] As a further aspect of the present invention: based on the KPI change rate, calculating the updated coupling strength between the headquarters-branch data includes: Define a coupling mapping function f(ΔKPI) between the KPI change rate and the headquarters-branch data; Based on the Bayesian smoothing algorithm, the formula is as follows: Calculate the updated coupling strength between the headquarters and branch office data, where, Indicates the updated coupling strength. Indicates the previous coupling strength. This indicates the instantaneous coupling strength compared to the previous value. This indicates the smoothness, which is determined by the industry type and geographic information data.

[0013] As a further aspect of the present invention: the construction of the multi-industry directed hierarchical network Gi(N, E) includes: The number of headquarters-branch offices in the city corresponding to each edge in E is used as the weight of the corresponding edge; Define the direction from the branch office to the headquarters as the direction of the edge.

[0014] As a further aspect of the present invention: the data simulation method is used to perform multi-dimensional perturbation simulation on each directed hierarchical network Gi(N,E), and the first network resilience of the directed hierarchical network is calculated. ,include: Remove nodes and their corresponding incoming and outgoing edges in a specific order to perturb the nodes. Remove each edge in a specific order, and determine whether the ratio of the removed incoming and outgoing edges of a node exceeds the coupling strength. ; When the ratio of the removal of an incoming edge to the outgoing edge of a node exceeds the coupling strength Then, it is determined that the node has also been removed and the perturbation of the opposite edge has been completed; Select a perturbation mode and perform the perturbation, wherein the perturbation mode includes a random perturbation mode and a deliberate perturbation mode; The maximum connected subgraph, the second largest subgraph, the network efficiency, and the average network degree of the directed hierarchical network are selected as key parameters for measuring network performance. The changes of these key parameters are recorded after each perturbation removal, and curves are plotted. Formula used: Calculate the first network resilience of a directed hierarchical network ,in, express The parameters characterize the network resilience. This represents the value of the t parameter, which includes the maximum subgraph, network efficiency, and average degree.

[0015] As a further aspect of the present invention: the selected perturbation mode includes: Select a random perturbation mode; Sort all directed hierarchical networks Gi(N, E) in a preset order, generate non-repeating random numbers, and remove directed hierarchical networks Gi(N, E) in sequence based on the random numbers until all directed hierarchical networks have been removed.

[0016] As a further aspect of the present invention: the selected perturbation mode includes: Select the intentional disturbance mode; Determine the sorting parameters for all directed hierarchical networks, and set the sorting order for all directed hierarchical networks based on the sorting parameters. The sorting parameters include betweenness centrality, proximity centrality, in-degree and out-degree, and number of headquarters. Before each deliberate perturbation mode begins, calculate any sorting parameter of all remaining directed hierarchical networks, select the first-ranked directed hierarchical network, and remove the first-ranked directed hierarchical network. Repeat this process until all directed hierarchical networks have been removed.

[0017] As a further aspect of the present invention: for each directed hierarchical network Gi(N, E), a cascade failure simulation is performed to calculate the resilience of the second network. ,include: The perturbation mode with the lowest network resilience value as the key parameter is selected as the baseline mode. Formula used: Calculate the cascade failure ratio, where CFR is the cascade failure percentage. It represents the number of existing headquarters at node k. Let k be the initial number of headquarters institutions, and select 95%, 90%, 85%, 80%, 75%, and 70% as typical cascading failure ratios. After each perturbation, determine whether all remaining nodes satisfy the condition. If the cascading failure rate is fixed, remove the corresponding node and repeat until the network is stable before proceeding with the next object perturbation. Continue this process until all objects are removed. Record the variation curves of key parameters, calculate the integral of the variation curves, and use the integral of the variation curves as the second network resilience characterized by the key parameters. .

[0018] As a further aspect of the present invention: for each directed hierarchical network Gi(N, E), network reconfiguration simulation is performed to calculate the third network resilience. ,include: Determine the network reconfiguration mode, which includes the nearest neighbor reconfiguration mode and the central reconfiguration mode; When the nearest neighbor reassignment mode is adopted, after each nearest neighbor directed hierarchical network is removed, the three closest directed hierarchical networks in the remaining directed hierarchical networks are selected, and the branches in the removed directed hierarchical networks are divided into three equal parts and reassigned to the corresponding directed hierarchical networks. When using the central redistribution mode, the formula is as follows: Constructing reconfiguration coefficients ,in, Let the centrality of node i be... The spatial distance between nodes k and i; Select the three nodes with the highest reconfiguration coefficients from node i, and reconfigure them according to the coefficient ratio; Formula used: Set the network reconfiguration ratio (RR), where... This represents the number of branches that were removed from node k. Let be the initial number of branches for node k, and select 20%, 30%, and 40% as typical reconfiguration ratios to represent the proportion of branches that are removed from the node. After each directed hierarchical network perturbation, determine whether all remaining nodes satisfy the condition. If a fixed reconfiguration ratio is used, the branch structure of the node that has been removed will be reconfigured according to the determined reconfiguration pattern. This process is repeated until the network is stable, and then the next perturbation is performed until all directed hierarchical networks have been removed. Record the variation curves of the key parameters, and calculate the integral to obtain the network resilience characterized by the key parameters. .

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, by constructing an industrial network model with hierarchical and multi-source spatiotemporal data, a dynamic multi-source spatiotemporal data fusion module can clean, interpolate, and standardize heterogeneous data such as business operations, information, and geography. At the same time, with the adaptive correlation strength update method and spatiotemporal hierarchical weight adjustment mechanism, the coupling strength of headquarters-branch institutions and the capacity of nodes and connections can be updated in real time. This not only improves the accuracy of industrial network resilience assessment, but also provides operable technical support for the coordinated development and risk prevention of urban agglomerations.

[0020] 2. In this invention, by using industry data from multiple city headquarters and branches, and by calculating the updated coupling strength between headquarters and branches based on the KPI change rate, it is possible to construct a multi-layer time-varying industry network to more precisely characterize the differences in structure and function between headquarters and branches, and effectively reduce estimation errors caused by static and fully integrated analysis, thereby significantly improving the accuracy of industry network resilience evaluation.

[0021] 3. In this invention, by introducing multiple disturbance modes on the basis of the traditional single disturbance mode and combining the two dynamic processes of cascading failure and network reconfiguration, the limitations of subjectively setting a single scenario are avoided. Among them, cascading failure simulation can reveal the chain destruction mechanism, and network reconfiguration simulation can reflect the system's self-repair capability. Through the combined effect of the two, the dynamic response of the industrial network under external shocks can be depicted more comprehensively and realistically.

[0022] 4. In this invention, by allocating smoothing coefficients through industry clustering or risk grading methods, the coupling strength within different clusters can maintain a stable inheritance of historical trends while responding promptly to changes in the latest KPIs. By setting an adaptive weight update mechanism, the "historical inertia" and "real-time sensitivity" of the data can be taken into account. It can be flexibly configured among different industry characteristics or risk levels to ensure the robustness and forward-looking nature of the assessment results.

[0023] 5. In this invention, by providing a modular and configurable resilience estimation framework, and with disturbance modes that can be freely selected or combined according to various scenarios such as market fluctuations, natural disasters, and data risks, and with cascading failure ratios and network reconfiguration ratios that can be parameterized according to actual business needs and system recovery methods, it is possible to promptly monitor information flow transmission efficiency, regional coordination, and the tightness of internal relationships. By selecting corresponding key parameters for index-based evaluation, it can flexibly adapt to various city clusters and industrial organizations, facilitating its promotion and application in industrial collaborative development planning, risk prevention and control, and decision support. Attached Figure Description

[0024] Figure 1 This is a flowchart of the industrial network resilience evaluation provided in the embodiments of the present invention; Figure 2 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example: Please see Figures 1-2 In this embodiment of the invention, a network resilience evaluation method based on data fusion and adaptive weight update includes the following steps: S1: Obtain headquarters-branch data of industries in the target area, classify the headquarters-branch data, merge the classified data at the city level, and preprocess the merged data to obtain preprocessed data; S2: Set corresponding KPIs based on headquarters-branch data, calculate the KPI change rate based on historical and new headquarters-branch data, and calculate the updated coupling strength between headquarters-branch data based on the KPI change rate. S3: Based on latitude and longitude coordinates, obtain the geographic layer data and external shock index of the industry; based on the geographic layer data and external shock index, calculate the coupling strength and the rate of change of the coupling strength; and adjust the dynamic simulation parameters based on the rate of change. S4: Count the number of headquarters and branches in each city, obtain the city information of the headquarters and branches, obtain city nodes and city pairs, and use the annotation method to mark the city nodes and city pairs on the spatial map. S5: Construct a multi-industry directed hierarchical network, where Gi represents the network of each industry category, node N is the city node after integrating headquarters and branches, and edge E is the directed city pair; S6: Use data simulation methods to perform multi-dimensional perturbation simulations on each directed hierarchical network and calculate the first network resilience of the directed hierarchical network. S7: Perform cascade failure simulation for each directed hierarchical network and calculate the resilience of the second network; S8: For each directed hierarchical network, perform network reconfiguration simulation to calculate the resilience of the third network; S9: Output the first network resilience, the second network resilience, and the third network resilience, and plot the resilience change curves based on industry and geographical factors.

[0027] In this embodiment, the industry type is manufacturing, the industry type data is manufacturing data, and a total of 3,390 data entries of headquarters-branch institutions were collected. These 3,390 data entries were then integrated into the basic dataset of 337 prefecture-level cities and municipalities directly under the central government according to the categories of operational data, information data, and geographic spatiotemporal data.

[0028] Preferably, after filtering out duplicate data based on the company coordinates and the connection path between headquarters and branch offices, and cleaning data with missing content, the above 3390 data entries are stored in shapefile format, and the geographic reference coordinate system is set to WGS1984.

[0029] Preferably, a city node is the city information where the headquarters is located and the city information where the branches are located, and a city pair is the combination of the city information where the headquarters and the corresponding branches are located.

[0030] Preferably, the text format for city nodes and city pairs is: city where the headquarters is located, city where the branch offices are located, number of enterprises.

[0031] Preferably, in outputting the first network resilience Second network resilience and third network resilience Subsequently, resilience change curves are plotted by combining industrial and geographical factors to support multi-dimensional industrial collaborative development and risk prevention and control decisions.

[0032] Preferably, step S1 includes: S11: Divide the headquarters-branch data into operational data, information data, and geographic spatiotemporal data types to obtain categorized data; S12: Use data fusion methods to integrate categorized data at the city level; S13: Use data deduplication methods to remove duplicate data from the city-level categorized data to obtain deduplicated data; S14: Based on the deduplicated data, the missing temporal and spatial values ​​in the deduplicated data are interpolated to obtain smooth data; S15: Based on the smoothed data, the normalization method is used to clean the smoothed data to obtain cleaned data; S16: Based on the cleaned data, the cleaned data is standardized using data standardization methods to obtain preprocessed data.

[0033] In this embodiment, step S1 is the acquisition and preprocessing of multi-source spatiotemporal data, specifically including: S1-1. Collect data on the headquarters and branches of the manufacturing industry within the research scope, mainly including industry type (manufacturing), city where the headquarters is located, latitude and longitude coordinates of the headquarters, city where the branches are located, latitude and longitude coordinates of the branch structure, number of enterprises, and connection path between headquarters and branches. A total of 3,390 initial data entries were collected. S1-2. The above-mentioned multi-source data are integrated into the basic dataset of 337 prefecture-level cities and municipalities directly under the central government according to the categories of business data, information data, and geographic spatiotemporal data. S1-3. After filtering out duplicate data based on the company coordinates and the connection path between headquarters and branch offices, and cleaning data with missing content, store the above data in shapefile format and set the geographic reference coordinate system to WGS1984. In this embodiment, step S2 is the adaptive coupling strength update, which specifically includes: S2-1, Known initial evaluation coupling strength =1, smoothing coefficient α=0.8, KPI change rate vector of this industry network Calculate ; S2-2, Calculate the updated coupling strength 0.915 In this embodiment, step S3 is the spatiotemporal layering weight adjustment, which specifically includes: S3-1. Based on the geographical stratification of the manufacturing network and the external shock index, the initial capacity of core cities (cities with a population of over 2 million) is set to 400, the initial capacity of ordinary cities (cities with a population between 500,000 and 2 million) is set to 100, and the initial capacity of peripheral cities (cities with a population of less than 500,000) is set to 20. The corresponding enterprise removal thresholds are set to 0.9, 0.8, and 0.7, respectively. S3-2. The internal coupling strength changes are determined based on industry type and external shocks, and the dynamic simulation parameters are adjusted in real time. For the simulation process of the manufacturing industry, the above standard parameters are selected, with the cascading failure ratio (CFR) set to 0.2 and the network reconfiguration ratio (RR) set to 0.3.

[0034] In this embodiment, step S4 is used to extract key nodes and city pairs based on the data processed above. The shapefile data is read using the Geopandas library in Python, and 315 city nodes and 1558 city pairs are identified.

[0035] In this embodiment, step S5 uses the NetworkX library in Python to build the network, with city nodes as network nodes, city pairs as network edges, branch structures to headquarters as connection directions, and the number of enterprises in the same city pair as network edge weights. The city coordinates are the spatial average coordinates of each headquarters and branch.

[0036] In this embodiment, step S6 performs multi-dimensional perturbation simulation based on the network architecture. The perturbations can be classified according to the object and the pattern, specifically including: S6-1. First, perform perturbation simulations on the nodes, simulating multiple perturbation modes. Start with random perturbation simulations, generating a random sequence [1, 315] from the city nodes according to their natural order. Determine the node number for each perturbation, and use the `nx.remove_nodes()` function of the NetworkX library to remove the corresponding nodes sequentially, recording the changes in key parameters (relative size of the largest connected subgraph, network efficiency, and average degree). Similarly, use the `nx.centrality()` function of NetworkX to sort the nodes by their betweenness centrality, proximity centrality, number of headquarters, and network in-degree, removing the corresponding nodes sequentially based on the sorting results, and recording the changes in key parameters. S6-2. Next, perturbation simulation is performed on the edges, with multiple perturbation modes simulated. First, random perturbation simulation is performed. The network edges are generated into a random sequence [1, 1558] according to their natural order, and the edge numbers for each perturbation are determined. The `remove_edges` function from the NetworkX library is used to remove the corresponding edges sequentially, recording the changes in key parameters. Similarly, edges are sorted according to their betweenness centrality, proximity centrality, and edge weight, and the corresponding edges are removed sequentially based on the sorting results. The removal ratio of incoming and outgoing edges at each node is compared with the current coupling strength. The size relationship is passed to the node removal process, and changes in key parameters are recorded; S6-3. Plot the variation curves of key parameters under each disturbance object and disturbance mode, and calculate the normalized integral as the manufacturing network resilience characterized by a certain parameter in this scenario. In this embodiment, step S7, based on multi-dimensional disturbance simulation, selects the disturbance type with the lowest resilience among the parameters of the manufacturing network—the disturbance targeting the intermediate centrality of city nodes—and performs cascading failure simulation, specifically including: S7-1. First, calculate the current values ​​of all parameters, including key resilience parameters (relative size of the largest connected subgraph, network efficiency, average degree) and the number of headquarters and branch structures for each city node. Then, use NetworkX's nx.betweennes_centrality() to calculate the weighted directed betweenness centrality of each node in the current network. Next, use NumPy's sorting function to sort them, select the node with the highest betweenness centrality, and use nx.remove_nodes() to remove the node, along with the edges associated with it. S7-2. After each removal of a target node, we calculate the number of headquarters in each remaining node and compare it with the number of initial headquarters stored in order of betweenness centrality. If the proportion is less than CFR, the node is removed in a cascade manner, and the associated edges are also removed. S7-3. Repeat the steps in S5-2 until none of the remaining nodes in the network are cascade failure nodes. Then repeat the next round of node removal and subsequent processes starting from S5-1 until all nodes are removed. S7-4. Based on the recorded key parameter values, plot the variation curves of the key resilience parameters, and obtain the manufacturing network resilience characterized by each parameter after normalization integration. In this embodiment, step S8, based on the cascading failure simulation, considers the dynamic recovery process of the industrial network after being disturbed and performs network reconfiguration simulation, specifically including: S8-1. First, simulate the disturbance and cascading failure of a single node according to S5-1, S5-2, and S5-3. S8-2. Calculate the number of remaining branches for each existing node. The difference between this number and the node's initial branch structure is the number of branches to be removed. Ratio this number to the initial branch structure number. If the ratio > 1, the number of branches to be removed is determined by the difference between the remaining number of branches and the initial branch structure number. If so, it is determined that the node needs to undergo network reconfiguration; S8-3. For distance reconfiguration, first find the network node that needs to be reconfigured and calculate its distance to all existing nodes. Select the three network nodes with the closest distance, transfer the branch that needs to be reconfigured to these three nodes on an average basis, and construct the corresponding connection edges. S8-4. For central reconfiguration, first find the network nodes that need to be reconfigured, and then calculate the reconfiguration of all existing nodes. and select from them The three network nodes have the largest values, and the branch structures that need to be reconfigured are arranged according to... After rounding down the value, it is transferred to the corresponding network node, and a corresponding edge is constructed. S8-5. Select different reconfiguration modes for different simulation needs. Here we choose distance reconfiguration. Based on S6-1 and S6-2, repeat S6-3 until all nodes that need to be reconfigured are reconfigured. Repeat the whole process of starting from S6-1 to remove nodes in the next round until all nodes are removed. S8-6. Based on the recorded key parameter values, plot the variation curves of the key resilience parameters, and obtain the manufacturing network resilience characterized by each parameter after normalization and integration. In this embodiment, step S9, based on the simulation process of the above three dimensions, can obtain the dynamic change process of the resilience of the manufacturing network.

[0037] Table 1 below shows the types and basic parameters of industrial networks.

[0038] Table 1 .

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A network resilience evaluation method based on data fusion and adaptive weight update, characterized in that, include: Acquire headquarters-branch data of industries in the target area, classify the headquarters-branch data, merge the classified data at the city level, and preprocess the merged data to obtain preprocessed data. The headquarters-branch data includes industry type data, latitude and longitude coordinates of the headquarters, latitude and longitude coordinates of the branches, number of enterprises of the same industry type, connection path between headquarters and branches, and enterprise operation data. KPIs are set based on the headquarters-branch data, the KPI change rate is calculated based on historical and new headquarters-branch data, and the updated coupling strength between the headquarters-branch data is calculated based on the KPI change rate. Based on the latitude and longitude coordinates, the geographic layer data and external shock index of the industry are obtained. Based on the geographic layer data and external shock index, the coupling strength and the rate of change of the coupling strength are calculated. Based on the rate of change, the dynamic simulation parameters are adjusted. The dynamic simulation parameters include city node capacity, enterprise removal threshold, cascade failure ratio and network reconfiguration ratio. The number of headquarters and branches in each of the cities is counted, and the city information of the headquarters and branches is obtained to obtain city nodes and city pairs. The city nodes and city pairs are marked on the spatial map using the annotation method. Construct a multi-industry directed hierarchical network Gi(N, E), where Gi represents the network of each industry category, node N is the city node after integrating headquarters and branches, and edge E is the directed city pair. We employ data simulation methods to perform multi-dimensional perturbation simulations on each directed hierarchical network and calculate the first network resilience of the directed hierarchical network. For each directed hierarchical network, a cascade failure simulation is performed to calculate the resilience of the second network. For each directed hierarchical network, network reconfiguration simulation is performed to calculate the resilience of the third network; Output the first network resilience, the second network resilience, and the third network resilience, and plot the resilience change curves based on industry and geographical factors.

2. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 1, characterized in that: The industry type data includes manufacturing data, construction data, general production service data, technology-intensive production service data, and general service industry data.

3. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 2, characterized in that: The process involves classifying the headquarters-branch data, merging the classified data at the city level, and preprocessing the merged data to obtain preprocessed data, including: The headquarters-branch data is divided into operational data, information data, and geographic spatiotemporal data types to obtain categorized data; The categorized data is fused at the city level using a data fusion method; The city-level classification data is deduplicated using a data deduplication method to obtain deduplicated data. Based on the deduplicated data, interpolation is used to interpolate the missing temporal and spatial values ​​in the deduplicated data to obtain smooth data; Based on the smoothed data, the normalization method is used to clean the smoothed data to obtain cleaned data; Based on the cleaning data, the cleaning data is standardized using a data standardization method to obtain preprocessed data.

4. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 3, characterized in that: Based on the KPI change rate, calculate the updated coupling strength between the headquarters and branch office data, including: Define a coupling mapping function f(ΔKPI) between the KPI change rate and the headquarters-branch data; Based on the Bayesian smoothing algorithm, the formula is as follows: Calculate the updated coupling strength between the headquarters and branch office data, where, Indicates the updated coupling strength. Indicates the previous coupling strength. This indicates the instantaneous coupling strength compared to the previous value. Indicates smoothness.

5. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 4, characterized in that: The construction of the multi-industry directed hierarchical network Gi(N, E) includes: The number of headquarters-branch offices in the city corresponding to each edge in E is used as the weight of the corresponding edge; Define the direction from the branch office to the headquarters as the direction of the edge.

6. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 5, characterized in that: The method employs data simulation to perform multi-dimensional perturbation simulations on each directed hierarchical network Gi(N, E) and calculates the first network resilience of the directed hierarchical network, including: Remove nodes and their corresponding incoming and outgoing edges in a specific order to perturb the nodes. Remove each edge in a specific order, and determine whether the ratio of the removed incoming and outgoing edges of a node exceeds the coupling strength. ; When the ratio of the removal of an incoming edge to the outgoing edge of a node exceeds the coupling strength Then, it is determined that the node has also been removed and the perturbation of the opposite edge has been completed; Select a perturbation mode and perform the perturbation, wherein the perturbation mode includes a random perturbation mode and a deliberate perturbation mode; The maximum connected subgraph, the second largest subgraph, the network efficiency, and the average network degree of the directed hierarchical network are selected as key parameters for measuring network performance. The changes of these key parameters are recorded after each perturbation removal, and curves are plotted. Formula used: Calculate the first network resilience of a directed hierarchical network, where, express The parameters characterize the network resilience. This represents the value of the t parameter, which includes the maximum subgraph, network efficiency, and average degree.

7. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 6, characterized in that: The selected perturbation mode includes: Select a random perturbation mode; All directed hierarchical networks are sorted in a preset order, and non-repeating random numbers are generated. Based on the random numbers, directed hierarchical networks are removed in sequence until all directed hierarchical networks have been removed.

8. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 6, characterized in that: The selected perturbation mode includes: Select the intentional disturbance mode; Determine the sorting parameters for all directed hierarchical networks, and set the sorting order for all directed hierarchical networks based on the sorting parameters. The sorting parameters include betweenness centrality, proximity centrality, in-degree and out-degree, and number of headquarters. Before each deliberate perturbation mode begins, calculate any sorting parameter of all remaining directed hierarchical networks, select the first-ranked directed hierarchical network, and remove the first-ranked directed hierarchical network. Repeat this process until all directed hierarchical networks have been removed.

9. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 8, characterized in that: The calculation of the second network resilience by performing cascade failure simulation on each directed hierarchical network Gi(N, E) includes: The perturbation mode with the lowest network resilience value as the key parameter is selected as the baseline mode. Formula used: Calculate the cascade failure ratio, where CFR is the cascade failure percentage. It represents the number of existing headquarters at node k. Let k be the initial number of headquarters institutions, and select 95%, 90%, 85%, 80%, 75%, and 70% as typical cascading failure ratios. After each perturbation, determine whether all remaining nodes satisfy the condition. If the cascading failure rate is fixed, remove the corresponding node and repeat until the network is stable before proceeding with the next object perturbation. Continue this process until all objects are removed. Record the variation curves of the key parameters, calculate the integral of the variation curves, and use the integral of the variation curves as the second network resilience characterized by the key parameters.

10. The network resilience evaluation method based on data fusion and adaptive weight update according to claim 9, characterized in that: The process of performing network reconfiguration simulations for each directed hierarchical network and calculating the third network resilience includes: Determine the network reconfiguration mode, which includes the nearest neighbor reconfiguration mode and the central reconfiguration mode; When the nearest neighbor reassignment mode is adopted, after each nearest neighbor directed hierarchical network is removed, the three closest directed hierarchical networks in the remaining directed hierarchical networks are selected, and the branches in the removed directed hierarchical networks are divided into three equal parts and reassigned to the corresponding directed hierarchical networks. When using the central redistribution mode, the formula is as follows: Constructing reconfiguration coefficients ,in, Let the centrality of node i be... The spatial distance between nodes k and i; Select the three nodes with the highest reconfiguration coefficients from node i, and reconfigure them according to the coefficient ratio; Formula used: Configure the network reconfiguration ratio, where... This represents the number of branches that were removed from node k. Let be the initial number of branches for node k, and select 20%, 30%, and 40% as typical reconfiguration ratios to represent the proportion of branches that are removed from the node. After each directed hierarchical network perturbation, determine whether all remaining nodes satisfy the condition. If a fixed reconfiguration ratio is used, the branch structure of the node that has been removed will be reconfigured according to the determined reconfiguration pattern. This process is repeated until the network is stable, and then the next perturbation is performed until all directed hierarchical networks have been removed. Record the variation curves of the key parameters, and calculate the integral to obtain the network resilience characterized by the key parameters.

Images