Heterogeneous unmanned node data access permission control method based on graph neural network

Abstract

The present application relates to the technical field of data security access control, and particularly relates to a heterogeneous unmanned node data access permission control method based on a graph neural network. The method comprises: performing node capability evaluation based on the residual power and available computing power of each networking node in a target area to obtain an initial capability value of each networking node; performing environment interference evaluation based on environment state observation information of each networking node to obtain an environment interference value of each networking node; performing weight analysis based on communication information between different networking nodes to determine a plurality of neighbor weight values; taking the networking nodes as graph nodes, taking the initial capability values and the environment interference values of the networking nodes as features of the graph nodes, taking the plurality of neighbor weight values as adjacency matrices corresponding to the graph nodes, and performing model inference according to a pre-trained graph neural network to obtain node permission allocation information. The present application can realize dynamic configuration of permissions of a plurality of networking nodes participating in networking.

Description

Technical Field

[0001] This invention relates to the technical field of data security access control, and specifically to a method for controlling data access permissions for heterogeneous unmanned nodes based on graph neural networks. Background Technology

[0002] With the rapid development of the Internet of Things (IoT), edge computing, and urban digital infrastructure, a large number of fixed-location unmanned nodes (such as meteorological monitoring stations, air quality sensors, water quality monitoring nodes, traffic flow monitoring equipment, smart street light nodes, 5G communication base station anchors, and industrial IoT gateways) are widely deployed in urban areas to undertake tasks such as environmental sensing, data collection, and information aggregation. These nodes typically possess certain sensing, computing, and communication capabilities. In actual operation, these nodes usually handle their respective dedicated tasks independently, such as meteorological stations continuously reporting temperature and humidity data, traffic nodes monitoring vehicle flow and periodically transmitting data, and air quality nodes collecting pollutant concentrations. Frequent data interaction between nodes is not common.

[0003] When a sudden collaborative computing need arises in the region (e.g., rapid simulation of urban flooding, analysis of pollution diffusion paths, joint perception of emergency events, regional data fusion processing, etc.), it is necessary to temporarily select several nodes from multiple fixed nodes distributed in the urban area to form a collaborative computing network. This allows each node to share its idle computing resources while completing its own basic tasks, and jointly complete complex computing or data fusion work that a single node cannot independently handle.

[0004] Existing access control methods are mainly designed for scenarios with long-term networking and fixed roles, and are difficult to adapt to the emerging application mode of multiple unmanned nodes forming temporary networks on demand for collaborative computing. Summary of the Invention

[0005] The purpose of this invention is to provide a method for controlling access permissions to heterogeneous unmanned nodes based on graph neural networks, which can solve the problem of permission allocation when multiple unmanned nodes are temporarily networked on demand for collaborative computing.

[0006] One embodiment of the present invention provides a method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks, the method comprising:

[0007] In the target area, among multiple network nodes, the node capabilities are evaluated based on the remaining power and available computing power of each network node to obtain the initial capability value of each network node.

[0008] In the multiple network nodes, environmental interference assessment is performed based on the environmental status observation information of each network node to obtain the environmental interference value of each network node;

[0009] Among the multiple network nodes, weight analysis is performed based on the communication information between different network nodes to determine multiple nearest neighbor weight values. The nearest neighbor weight values ​​are used to indicate the influence weight of a corresponding network node on another network node with which it has a communication relationship.

[0010] Using the network nodes as graph nodes, the initial capability value and environmental interference value of the network nodes as features of the graph nodes, and the multiple nearest neighbor weight values ​​as the adjacency matrix corresponding to the graph nodes, the node permission allocation information is obtained by model inference based on the pre-trained graph neural network. The node permission allocation information is used to indicate the permission size of each of the network nodes.

[0011] Optionally, the remaining power is positively correlated with the corresponding initial capability value, and the available computing power is positively correlated with the corresponding initial capability value.

[0012] Optionally, the step of evaluating the node capabilities of each network node in the target area based on its remaining power and available computing power to obtain the initial capability value of each network node includes:

[0013] Calculate the ratio of the remaining power of the target network node to its reference power to obtain the power index of the target network node, wherein the target network node is one of the multiple network nodes, and the reference power is the power of the corresponding network node when it is fully charged;

[0014] The ratio of the available computing power of the target network node to its maximum computing power is calculated to obtain the computing power index of the target network node;

[0015] Calculate the average of the power index and computing power index of the target network node to obtain the capability index of the target network node;

[0016] The initial capability value of the target network node is obtained by numerically transforming the capability index of the target network node based on the hyperbolic tangent function.

[0017] Optionally, the communication information includes the node type, transmission bandwidth, and communication latency of the two corresponding communication nodes;

[0018] Among the multiple network nodes, the step of performing weight analysis based on the communication information between different network nodes to determine multiple nearest neighbor weight values ​​includes:

[0019] The communication link quality is evaluated based on the transmission bandwidth and communication latency of the first network node and the second network node to obtain the link quality factor. The first network node and the second network node are any two network nodes that have a communication relationship among the plurality of network nodes.

[0020] A consistency assessment is performed between the node types of the first group of network nodes and the node types of the second group of network nodes to obtain a type matching factor;

[0021] The link quality factor and the type matching factor are weighted and calculated to obtain the nearest neighbor weight value that is common to the first network node and the second network node.

[0022] Optionally, the calculated weight of the link quality factor is greater than the calculated weight of the type matching factor.

[0023] Optionally, the step of evaluating the communication link quality based on the transmission bandwidth and communication latency of the first network node and the second network node to obtain the link quality factor includes:

[0024] The transmission bandwidth and communication delay of the first network node and the second network node are normalized respectively to obtain the bandwidth normalization value and the delay normalization value.

[0025] The time delay normalization value is numerically transformed to obtain a time delay excellence value, wherein the sum of the time delay excellence value and the time delay normalization value is 1;

[0026] The harmonic mean of the normalized bandwidth value and the good latency value is calculated to obtain the link quality factor.

[0027] Optionally, the step of performing a consistency assessment based on the node types of the first group of network nodes and the node types of the second group of network nodes to obtain a type matching factor includes:

[0028] If the node type of the first group of network nodes is the same as the node type of the second group of network nodes, the preset first matching value is determined as the type matching factor.

[0029] When the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is a type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset second matching value is determined as the type matching factor.

[0030] If the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is no type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset third matching value is determined as the type matching factor.

[0031] Wherein, the first matching value is greater than the second matching value, and the second matching value is greater than the third matching value.

[0032] Optionally, the environmental state observation information includes the measured signal-to-noise ratio and measured packet loss rate of the corresponding network nodes during communication;

[0033] The steps for assessing environmental interference based on the environmental status observation information of each network node among the multiple network nodes, and obtaining the environmental interference value of each network node, include:

[0034] Based on the maximum and minimum rated signal-to-noise ratio of the target network node, the measured signal-to-noise ratio of the target network node is numerically standardized to obtain the first interference factor of the target network node, wherein the target network node is one of the multiple network nodes.

[0035] By analyzing the similarity between the measured packet loss rate of the target network node and its set maximum packet loss rate, the second interference factor of the target network node is obtained.

[0036] The ratio of the second interference factor to the first interference factor of the target network node is calculated to obtain the environmental interference value of the target network node.

[0037] Optionally, the steps for obtaining multiple network nodes in the target area include:

[0038] Multiple candidate nodes in the target area are randomly combined to obtain multiple node combinations, wherein the total number of candidate nodes included in each node combination is equal to the preset number.

[0039] In the multiple node combinations, the degree of concentration of the spatial distribution of all candidate nodes included in each node combination is analyzed to obtain the node aggregation factor of each node combination;

[0040] In the multiple node combinations, the spatial matching degree between each node combination and the target region is analyzed to obtain the regional matching factor of each node combination;

[0041] The node aggregation factor and region matching factor of each node combination are fused to obtain the combination score of each node combination;

[0042] Among the multiple node combinations, the node combination with the highest combination score is determined as the target node combination, and all candidate nodes included in the target node combination are determined as the network nodes.

[0043] Optionally, the step of analyzing the spatial matching degree between each node combination and the target region in the plurality of node combinations to obtain the regional matching factor of each node combination includes:

[0044] In the plurality of node combinations, a region matching factor for each node combination is determined based on the distance between the combination center of each node combination and the region center of the target region. The region matching factor is negatively correlated with the corresponding distance, and the combination center is the average value of the node positions of all candidate nodes included in the corresponding node combination. Alternatively, in the plurality of node combinations, the ratio of the combination area of ​​each node combination to the region area of ​​the target region is calculated to obtain the region matching factor for each node combination. The combination area is the convex hull area of ​​the projection points of all candidate nodes included in the corresponding node combination on the standard horizontal plane.

[0045] The present invention has the following beneficial effects:

[0046] Node capability is assessed based on the remaining power and available computing power of each network node to determine the computing resource contribution each node can support during network formation, ensuring that permission allocation matches the actual carrying capacity of the nodes. Environmental interference assessment is performed based on environmental status observation information of each network node to determine the interference intensity of the environment in which each network node is located. The severity of the communication environment is used as a penalty factor for permission allocation, preventing network nodes in highly interfered environments from being assigned excessive permissions. Then, weight analysis is performed based on communication information between different network nodes to construct adjacency matrices for different network nodes as graph nodes. Finally, a pre-trained graph neural network is used to process the above three factors, comprehensively evaluating the contribution of each network node after network formation from aspects such as node's own capabilities, external interference, and interaction between nodes. The permission score output by the graph neural network is directly mapped to the task weight and importance level authorized to the corresponding network node in collaborative computing tasks, realizing dynamic permission configuration for multiple network nodes participating in the network formation, improving the security, resource utilization efficiency, and task execution effect of the distributed collaborative system temporarily composed of multiple network nodes. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating a heterogeneous unmanned node data access permission control method based on graph neural networks provided in an embodiment of the present invention. Detailed Implementation

[0048] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the heterogeneous unmanned node data access permission control method based on graph neural networks proposed by the present invention. 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.

[0049] 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 invention pertains.

[0050] The specific scheme of the heterogeneous unmanned node data access permission control method based on graph neural network provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0051] In one embodiment, the present invention provides a method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks, such as... Figure 1 As shown, the method includes:

[0052] Step S1: In the target area, among multiple network nodes, evaluate the node capabilities based on the remaining power and available computing power of each network node to obtain the initial capability value of each network node.

[0053] In this invention, the target area can be understood as the geographic spatial range where there is a need for temporary collaborative computing. This area has a clear spatial boundary, which can be defined by administrative divisions, geographic coordinate ranges, communication network coverage, or task requirement boundaries.

[0054] For example, the target area can be an urban functional area, an industrial park, a transportation hub, an ecological monitoring area, etc. When the target area is an urban functional area, its corresponding temporary collaborative computing needs can include: rapid simulation of sudden urban flooding, real-time analysis of large-scale population density, and joint perception of regional environmental emergency events. When the target area is an industrial park, its corresponding temporary collaborative computing needs can include: rapid simulation of the leakage and diffusion path of toxic and harmful gases, overall energy consumption optimization and scheduling of the park, and security situation assessment based on multi-source data fusion. When the target area is a transportation hub, its corresponding temporary collaborative computing needs can include: emergency evacuation simulation of large passenger flows in airport terminals / waiting halls, multi-source security data fusion analysis, and joint fault diagnosis of facilities and equipment. When the target area is an ecological monitoring area, its corresponding temporary collaborative computing needs can include: multi-source ecological environment data fusion modeling, pollutant diffusion trend estimation, and joint analysis of the activity patterns of rare species.

[0055] A network node can be understood as a candidate node that meets the set networking conditions among multiple candidate nodes in the target area.

[0056] Candidate nodes can be understood as unmanned independent device units deployed in the target area, with fixed locations and possessing sensing, computing, and communication capabilities, capable of participating in temporary collaborative computing networks on demand (requiring that their remaining power is greater than the minimum power threshold and their available computing power is greater than the minimum computing power threshold, so as to ensure that each candidate node can support the processing of the most basic computing tasks).

[0057] For example, the candidate nodes mentioned above may include environmental monitoring nodes, smart street light nodes, traffic flow monitoring nodes, meteorological monitoring nodes, gateway nodes, and noise monitoring nodes.

[0058] In one implementation, all candidate nodes within the target area can be identified as the network nodes to fully utilize node resources within the target area and improve the processing efficiency of temporary collaborative computing needs.

[0059] In another implementation, some candidate nodes can be selected from all candidate nodes in the target area to be used as network nodes, thereby reducing the network size, communication overhead, and computational complexity of permission management.

[0060] In this embodiment, the steps for obtaining multiple network nodes in the target area include:

[0061] Multiple candidate nodes in the target area are randomly combined to obtain multiple node combinations, wherein the total number of candidate nodes included in each node combination is equal to the preset number.

[0062] In the multiple node combinations, the degree of concentration of the spatial distribution of all candidate nodes included in each node combination is analyzed to obtain the node aggregation factor of each node combination;

[0063] In the multiple node combinations, the spatial matching degree between each node combination and the target region is analyzed to obtain the regional matching factor of each node combination;

[0064] The node aggregation factor and region matching factor of each node combination are fused to obtain the combination score of each node combination;

[0065] Among the multiple node combinations, the node combination with the highest combination score is determined as the target node combination, and all candidate nodes included in the target node combination are determined as the network nodes.

[0066] The aforementioned multiple node combinations are used to indicate all possible combinations that can be obtained by randomly combining multiple candidate nodes in the target area according to a preset number. The preset number can be understood as the rounded-up value of the ratio of the total computing power requirement corresponding to the temporary collaborative computing needs to the aforementioned minimum computing power threshold.

[0067] In the above settings, a preset number is used to achieve dynamic adaptation between the network scale and computing needs, so as to avoid insufficient computing resources due to too few nodes, and to avoid the problems of increased communication overhead and increased complexity of permission management due to too many nodes.

[0068] The higher the concentration of the spatial distribution of all candidate nodes in each node combination, the shorter the communication links between multiple nodes in the temporary network system and the higher the collaborative interaction efficiency when networking based on the corresponding node combination. The original data corresponding to the temporary collaborative computing needs depends at least partially on the candidate nodes for collection. Therefore, the higher the spatial matching degree between the node combination and the target area, the better the spatial coverage effect of the temporary network system on the target area, and the better the corresponding data collection effect. Thus, the data reliability of the processing results output is stronger.

[0069] Combining the above two aspects, multiple node combinations can be scored from both the internal collaborative effect and the external coverage effect, which can take into account both computational efficiency and computational quality, and ensure that the selected node combination is fully adapted to the computational needs of the corresponding target area.

[0070] In this invention, the node aggregation factor and the region matching factor of each node combination are specifically weighted and calculated to obtain the combination score of each node combination. The calculation weights of the node aggregation factor and the region matching factor can be dynamically set according to the actual scenario. For example, in scenarios with high timeliness requirements, the calculation weight of the node aggregation factor can be set to be greater than the calculation weight of the region matching factor, while in scenarios with high accuracy requirements, the calculation weight of the region matching factor can be set to be greater than the calculation weight of the node aggregation factor.

[0071] Specifically, the steps for obtaining the node aggregation factor of a node combination include:

[0072] Calculate the distance between each candidate node in the node combination and the combination center corresponding to the node combination to obtain the multiple center deviation distances of the node combination;

[0073] Calculate the variance of the center deviation distance of the node combination, and then calculate the ratio of this variance to the reference area value to obtain the spatial discrete value of the node combination;

[0074] Then, the spatial discrete values ​​of the node combination are reverse normalized to obtain the node aggregation factor of the node combination.

[0075] The reference area value can be understood as one-tenth or one-twentieth of the area of ​​the target region. The reverse normalization process is based on a negative exponential function with the natural constant e as the base.

[0076] The calculation of spatial discrete values ​​can be combined with subsequent reverse normalization to complete numerical normalization. In this process, the introduction of reference area values ​​can not only eliminate the dimension of the calculated variance, allowing the calculated node aggregation factor to participate in subsequent calculations as a dimensionless index value, but also amplify the differences in spatial distribution between different node combinations.

[0077] The step of analyzing the spatial matching degree between each node combination and the target region to obtain the regional matching factor for each node combination includes:

[0078] In the plurality of node combinations, a region matching factor for each node combination is determined based on the distance between the combination center of each node combination and the region center of the target region. The region matching factor is negatively correlated with the corresponding distance, and the combination center is the average value of the node positions of all candidate nodes included in the corresponding node combination. Alternatively, in the plurality of node combinations, the ratio of the combination area of ​​each node combination to the region area of ​​the target region is calculated to obtain the region matching factor for each node combination. The combination area is the convex hull area of ​​the projection points of all candidate nodes included in the corresponding node combination on the standard horizontal plane.

[0079] Based on the above settings, the calculation of the regional matching factor is made more flexible, thereby adapting to the ever-changing business needs under actual working conditions and enhancing the versatility of this solution in practical applications.

[0080] Specifically, when the target area includes several key locations, the average value of the key locations can be calculated first and defined as the area center. Then, the area matching factor of each node combination can be determined based on the distance between the combination center of each node combination and the area center of the target area. When the target area does not include key locations, the area matching factor of each node combination can be obtained by calculating the ratio of the combination area of ​​each node combination to the area of ​​the target area.

[0081] Among them, key locations can be understood as specific locations or coordinate points within the target area that have special importance, representativeness, or functional significance. They play a core role in data collection, event monitoring, or decision support during the execution of temporary collaborative computing tasks.

[0082] For example, in a rapid simulation of urban flooding, several low-lying points / culverts in the target area that have historically experienced severe flooding can be designated as key locations; however, in a joint environmental quality assessment of an industrial park, since a uniform environmental quality assessment of the entire park is required, key locations may not be designated in the target area.

[0083] In this invention, the distance between the combination center of each node combination and the regional center of the target region is calculated. Then, the ratio of this distance to the reference distance is calculated. Finally, the result of subtracting the ratio from the value 1 is determined as the regional matching factor for each node combination. The reference distance is the distance from the regional center to the regional boundary of the target region. It should be noted that if there are multiple distances from the regional center to the regional boundary of the target region, the smallest distance is determined as the reference distance.

[0084] The remaining power is positively correlated with the corresponding initial capability value, and the available computing power is positively correlated with the corresponding initial capability value.

[0085] Specifically, the steps for evaluating the capabilities of multiple network nodes in the target area based on the remaining power and available computing power of each network node to obtain the initial capability value of each network node include:

[0086] Calculate the ratio of the remaining power of the target network node to its reference power to obtain the power index of the target network node, wherein the target network node is one of the multiple network nodes, and the reference power is the power of the corresponding network node when it is fully charged;

[0087] The ratio of the available computing power of the target network node to its maximum computing power is calculated to obtain the computing power index of the target network node;

[0088] Calculate the average of the power index and computing power index of the target network node to obtain the capability index of the target network node;

[0089] The initial capability value of the target network node is obtained by numerically transforming the capability index of the target network node based on the hyperbolic tangent function.

[0090] In this invention, the remaining power is represented by the remaining available time of the corresponding network node, so as to balance the differences in energy storage capacity and power consumption of different network nodes and ensure the comparability of power index among different network nodes.

[0091] The maximum computing power can be understood as the computing power of the corresponding network node under full load.

[0092] The reason for using the hyperbolic tangent function is to amplify the subtle differences in capability indices among different network nodes, so that the initial capability value can more sensitively distinguish the capability differences among different network nodes.

[0093] Step S2: In the multiple network nodes, environmental interference assessment is performed based on the environmental status observation information of each network node to obtain the environmental interference value of each network node.

[0094] Specifically, the environmental state observation information includes the measured signal-to-noise ratio and measured packet loss rate of the corresponding network nodes during communication;

[0095] The steps for assessing environmental interference based on the environmental status observation information of each network node among the multiple network nodes, and obtaining the environmental interference value of each network node, include:

[0096] Based on the maximum and minimum rated signal-to-noise ratio of the target network node, the measured signal-to-noise ratio of the target network node is numerically standardized to obtain the first interference factor of the target network node, wherein the target network node is one of the multiple network nodes.

[0097] By analyzing the similarity between the measured packet loss rate of the target network node and its set maximum packet loss rate, the second interference factor of the target network node is obtained.

[0098] The ratio of the second interference factor to the first interference factor of the target network node is calculated to obtain the environmental interference value of the target network node.

[0099] The measured signal-to-noise ratio can be understood as the ratio of the useful signal power received by the corresponding network node during its own dedicated task within a set statistical time window (usually the first 30 minutes of the current time) to the noise power (such as environmental noise, thermal noise, interference signals, etc.) received at the same time. The measured packet loss rate should be understood as the ratio of the number of data packets lost by the network node when sending or receiving data packets during its own dedicated task within a set statistical time window to the total number of data packets expected to be received or sent.

[0100] By calculating the maximum and minimum values ​​of the rated signal-to-noise ratio, the measured signal-to-noise ratio is numerically standardized. This maps the measured signal-to-noise ratio values ​​of different network nodes under different communication protocols to the [0,1] interval, which can eliminate the adverse effects caused by the difference in the range of signal-to-noise ratio values ​​of different communication modules and make the environmental interference values ​​between different types of network nodes comparable.

[0101] Analyzing the similarity between the measured packet loss rate and the maximum tolerable packet loss rate set by the system is also to map the measured packet loss rate to the [0,1] interval, so as to support the calculation of the mean of the first interference factor and the second interference factor.

[0102] This involves using the measured signal-to-noise ratio to evaluate the potential communication capabilities of the corresponding network nodes at the physical layer in a real-world environment, and using the measured packet loss rate to evaluate the actual communication capabilities of the corresponding network nodes at the data link layer in a real-world environment. Finally, the environmental interference jointly exhibited by the physical layer and the data link layer is accurately quantified through ratio calculation.

[0103] The specific calculation process for the first interference factor is as follows:

[0104] The difference between the measured signal-to-noise ratio and the minimum value of the rated signal-to-noise ratio is calculated to obtain the first difference; and the difference between the maximum value of the rated signal-to-noise ratio and the minimum value of the rated signal-to-noise ratio is calculated to obtain the second difference.

[0105] Calculate the ratio of the first difference to the second difference to obtain the first interference factor.

[0106] The above-mentioned maximum rated signal-to-noise ratio should be understood as: the typical optimal signal-to-noise ratio that a network node can achieve with a base station or peer communication node in a standard reference environment (i.e., an open area with no obstructions, no significant interference, and line-of-sight signal propagation).

[0107] The minimum rated signal-to-noise ratio should be understood as: the minimum signal-to-noise ratio threshold at which the communication links of network nodes can maintain basic effective data communication.

[0108] Furthermore, when the measured signal-to-noise ratio is less than or equal to the minimum value of the rated signal-to-noise ratio, the calculation of the first interference factor and the second interference factor is skipped, and the environmental interference value is forcibly set to the preset default maximum value (e.g., 100).

[0109] In addition, when the measured packet loss rate of the target network node is greater than its set maximum packet loss rate, its second interference factor is forcibly set to 1. When the measured packet loss rate of the target network node is less than or equal to its set maximum packet loss rate, its second interference factor is the ratio of the measured packet loss rate of the target network node to its set maximum packet loss rate.

[0110] Step S3: Among the multiple network nodes, perform weight analysis based on the communication information between different network nodes to determine multiple nearest neighbor weight values.

[0111] The nearest neighbor weight value is used to indicate the influence weight of a corresponding network node on another network node with which it has a communication relationship.

[0112] The communication information includes the node types, transmission bandwidth, and communication latency of the two corresponding communication nodes.

[0113] Specifically, the step of determining multiple nearest neighbor weight values ​​based on weight analysis of communication information between different network nodes in the aforementioned multiple network nodes includes:

[0114] The communication link quality is evaluated based on the transmission bandwidth and communication latency of the first network node and the second network node to obtain the link quality factor. The first network node and the second network node are any two network nodes that have a communication relationship among the plurality of network nodes.

[0115] A consistency assessment is performed between the node types of the first group of network nodes and the node types of the second group of network nodes to obtain a type matching factor;

[0116] The link quality factor and the type matching factor are weighted and calculated to obtain the nearest neighbor weight value that is common to the first network node and the second network node.

[0117] After networking, nodes exchange data and control information through communication links. When a node emits abnormal status information due to environmental degradation (such as signal attenuation or multipath interference) or being controlled by an attacker, this anomaly will spread to its neighboring nodes through the communication links, realizing the spread of unauthorized access attacks. By introducing corresponding nearest neighbor weight values ​​for different networked nodes with communication relationships, the degree to which one node is affected by another node can be evaluated accordingly. This simulates the situation in the permission allocation process where the status information of any node can be propagated to its neighbors along the communication links, and the permission score of each node depends not only on its own characteristics but also on the status of its neighboring nodes.

[0118] In practical applications, each network node spends most of its time performing its own dedicated task, meaning that communication information between different network nodes may be missing. Based on this, the present invention sets up a system where, after determining multiple network nodes, each network node sends test text to other network nodes through broadcasting or other means to obtain communication information between different network nodes. After the communication information between different network nodes is collected, dynamic permission configuration is then performed for multiple network nodes.

[0119] In a temporary distributed network composed of multiple network nodes, the effective propagation of information influence depends primarily on the availability of information. In other words, if the communication link between two nodes is of extremely poor quality (insufficient bandwidth or excessive latency), even if the node types are completely identical, status information and control information cannot be transmitted between them in a timely and complete manner, and the propagation of influence cannot be realized at the physical level.

[0120] In other words, the link quality factor is the primary determinant of the nearest neighbor weight value, and the type matching factor is only a modification of it.

[0121] Therefore, in this invention, the calculated weight of the link quality factor is greater than the calculated weight of the type matching factor.

[0122] In one example, the weight of the link quality factor can be set to 0.65, and the weight of the type matching factor can be set to 0.35.

[0123] Specifically, the steps for evaluating the communication link quality based on the transmission bandwidth and communication latency of the first and second network nodes to obtain the link quality factor include:

[0124] The transmission bandwidth and communication delay of the first network node and the second network node are normalized respectively to obtain the bandwidth normalization value and the delay normalization value.

[0125] The time delay normalization value is numerically transformed to obtain a time delay excellence value, wherein the sum of the time delay excellence value and the time delay normalization value is 1;

[0126] The harmonic mean of the normalized bandwidth value and the good latency value is calculated to obtain the link quality factor.

[0127] The reason for using harmonic averaging to process the bandwidth normalization value and latency excellence value in the above settings is to amplify the impact of the weaker of the two on the link quality factor, so as to make the determined link quality factor more accurate and reliable, thereby ensuring the reliability of the subsequent calculated permission score.

[0128] The steps for evaluating the consistency between the node types of the first group of network nodes and the node types of the second group of network nodes to obtain the type matching factor include:

[0129] If the node type of the first group of network nodes is the same as the node type of the second group of network nodes, the preset first matching value is determined as the type matching factor.

[0130] When the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is a type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset second matching value is determined as the type matching factor.

[0131] If the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is no type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset third matching value is determined as the type matching factor.

[0132] Wherein, the first matching value is greater than the second matching value, and the second matching value is greater than the third matching value.

[0133] For example, the first matching value, the second matching value and the third matching value mentioned above can be set to 0.9, 0.5 and 0.2 respectively based on experience.

[0134] In this invention, the node types of the network nodes are classified according to their business functions, including but not limited to: sensing nodes responsible for collecting physical world data and converting analog / digital signals into structured data; computing nodes responsible for processing, analyzing, fusing and reasoning the sensing data; relay nodes responsible for data forwarding and communication link maintenance; storage nodes responsible for persistent data storage and historical data management; and management nodes responsible for network control and task scheduling.

[0135] In this invention, type association refers to the data dependency, collaborative working relationship, or control and management relationship between two network nodes in terms of business functions. Its essence is that the two network nodes have direct or indirect interaction needs in collaborative computing tasks—that is, the state change or abnormal behavior of one node may affect the task execution of the other node. Therefore, the influence between the two should be appropriately propagated. The absence of type association means that the functions of the two are independent of each other, the reference value of the state change can be ignored, and the propagation of influence should be suppressed.

[0136] For example, perception nodes are associated with compute nodes, compute nodes are associated with storage nodes, and management nodes are associated with compute nodes.

[0137] The relationships between different node types can be preset. That is, after determining the node types of two network nodes, the presence of a type relationship between the two network nodes can be determined by querying the data table that stores the relationships between node types. If the table is empty, it means that there is no type relationship between the two network nodes.

[0138] Step S4: Using the network nodes as graph nodes, the initial capability value and environmental interference value of the network nodes as features of the graph nodes, and the multiple nearest neighbor weight values ​​as the adjacency matrix corresponding to the graph nodes, perform model reasoning based on the pre-trained graph neural network to obtain node permission allocation information.

[0139] The node permission allocation information is used to indicate the permission level of each of the network nodes.

[0140] In practical applications, the above-mentioned graph neural network can be constructed based on any one of graph convolutional networks, graph attention networks, graph gating networks, and graph sampling aggregation networks.

[0141] In this invention, the graph neural network receives the initial capability value and environmental interference value of each network node as the node's initial feature vector, and fuses these two different-dimensional features in the latent space through its internal multi-layer nonlinear transformation, automatically learning the intrinsic relationship between the two and their joint influence on permission allocation.

[0142] The adjacency matrix, composed of nearest neighbor weights, defines the information propagation path between nodes. Through a message passing mechanism, each node receives feature information from its neighboring nodes at each network layer update and aggregates it according to the weights in the adjacency matrix. After multiple layers of propagation, the final feature representation of each node not only includes the node's own initial capability value and environmental interference value, but also the state information of its multi-hop neighboring nodes. This setting enables the permission allocation decision to fully perceive the network position of the node in the local topology and the influence of the state of the surrounding nodes, thus improving the reliability of permission evaluation.

[0143] The node permission allocation information mentioned above includes multiple permission scores, which correspond one-to-one with multiple network nodes. The permission score ranges from 0 to 1. The higher the permission score, the higher the permissions that the corresponding network node can be assigned in the future.

[0144] In one example, five discrete permission levels can be set, from highest to lowest: high permission, medium-high permission, medium permission, low permission, and restricted permission.

[0145] In this example, nodes with high privileges have a privilege score greater than or equal to 0.8. Nodes at this level are authorized to undertake core computing tasks in collaborative computing, including data fusion, model inference, and critical decision calculations, and can access all system data resources, including raw perception data, intermediate calculation results, and final decision data.

[0146] Nodes with medium to high privileges have a privilege score greater than or equal to 0.6 and less than 0.8. Nodes at this level are authorized to undertake the main computing tasks in collaborative computing tasks, including parallel subtask computing and data preprocessing, and can access most data resources, but they do not have access to final decision data or highly sensitive data involving system security.

[0147] Nodes with medium permissions have a permission score greater than or equal to 0.4 and less than 0.6. Nodes at this level are authorized to undertake auxiliary computing tasks in collaborative computing tasks, including data format conversion, simple statistical calculations, etc., and can access some data resources, limited to the subset of data necessary for their task.

[0148] Nodes with low privileges have a privilege score greater than or equal to 0.2 and less than 0.4. Nodes at this level are authorized to undertake edge computing tasks in collaborative computing tasks, including data collection and reporting, simple filtering, etc., and can access limited data resources, limited to the data they collect themselves and necessary control commands.

[0149] Nodes with restricted permissions have a permission score of less than 0.2. Nodes at this level do not participate in collaborative computing tasks, only perform basic local functions, have no right to access data from any other node, and only receive control commands issued by the system.

[0150] In summary, this invention assesses node capabilities based on the remaining power and available computing power of each network node to determine the computational resource contribution each node can support during network formation, ensuring that permission allocation matches the actual carrying capacity of the nodes. Environmental interference assessment based on environmental status observation information of each network node determines the interference intensity of the environment in which each network node is located, using the severity of the communication environment as a penalty factor for permission allocation to prevent network nodes in highly interfered environments from being assigned excessive permissions. Then, weight analysis is performed based on communication information between different network nodes to construct adjacency matrices for different network nodes as graph nodes. Finally, a pre-trained graph neural network is used to process the above three factors, comprehensively evaluating the contribution of each network node after network formation from aspects such as node capabilities, external interference, and inter-node interactions. The permission score output by the graph neural network is directly mapped to the task weight and importance level authorized to the corresponding network node in collaborative computing tasks, realizing dynamic permission configuration for multiple network nodes participating in the network formation, and improving the security, resource utilization efficiency, and task execution effect of a distributed collaborative system temporarily composed of multiple network nodes.

[0151] It should be noted that the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some implementations, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks, characterized in that, The method includes: In multiple network nodes in the target area, the node capabilities are evaluated based on the remaining power and available computing power of each network node to obtain the initial capability value of each network node. In the multiple network nodes, environmental interference assessment is performed based on the environmental status observation information of each network node to obtain the environmental interference value of each network node; Among the multiple network nodes, weight analysis is performed based on the communication information between different network nodes to determine multiple nearest neighbor weight values. The nearest neighbor weight values ​​are used to indicate the influence weight of a corresponding network node on another network node with which it has a communication relationship. Using the network nodes as graph nodes, the initial capability value and environmental interference value of the network nodes as features of the graph nodes, and the multiple nearest neighbor weight values ​​as the adjacency matrix corresponding to the graph nodes, the node permission allocation information is obtained by model inference based on the pre-trained graph neural network. The node permission allocation information is used to indicate the permission size of each of the network nodes.

2. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 1, characterized in that, The remaining power is positively correlated with the corresponding initial capability value, and the available computing power is also positively correlated with the corresponding initial capability value.

3. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 2, characterized in that, The steps for evaluating the capabilities of multiple network nodes in the target area, based on the remaining power and available computing power of each network node, to obtain the initial capability value of each network node, include: Calculate the ratio of the remaining power of the target network node to its reference power to obtain the power index of the target network node, wherein the target network node is one of the multiple network nodes, and the reference power is the power of the corresponding network node when it is fully charged; Calculate the ratio of the available computing power of the target network node to its maximum computing power to obtain the computing power index of the target network node; Calculate the average of the power index and computing power index of the target network node to obtain the capability index of the target network node; The initial capability value of the target network node is obtained by numerically transforming the capability index of the target network node based on the hyperbolic tangent function.

4. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 1, characterized in that, The communication information includes the node types, transmission bandwidth, and communication latency of the two corresponding communication nodes; Among the multiple network nodes, the step of performing weight analysis based on the communication information between different network nodes to determine multiple nearest neighbor weight values ​​includes: The communication link quality is evaluated based on the transmission bandwidth and communication latency of the first network node and the second network node to obtain the link quality factor. The first network node and the second network node are any two network nodes that have a communication relationship among the plurality of network nodes. A consistency assessment is performed between the node types of the first group of network nodes and the node types of the second group of network nodes to obtain a type matching factor; The link quality factor and the type matching factor are weighted and calculated to obtain the nearest neighbor weight value that is common to the first network node and the second network node.

5. The heterogeneous unmanned node data access control method based on graph neural networks according to claim 4, characterized in that, The calculated weight of the link quality factor is greater than the calculated weight of the type matching factor.

6. The heterogeneous unmanned node data access control method based on graph neural networks according to claim 4, characterized in that, The steps for evaluating the communication link quality and obtaining the link quality factor based on the transmission bandwidth and communication latency of the first and second network nodes include: The transmission bandwidth and communication delay of the first network node and the second network node are normalized respectively to obtain the bandwidth normalization value and the delay normalization value. The time delay normalization value is numerically transformed to obtain a time delay excellence value, wherein the sum of the time delay excellence value and the time delay normalization value is 1; The harmonic mean of the normalized bandwidth value and the good latency value is calculated to obtain the link quality factor.

7. The heterogeneous unmanned node data access control method based on graph neural networks according to claim 4, characterized in that, The steps for obtaining a type matching factor by evaluating the consistency between the node types of the first group of network nodes and the node types of the second group of network nodes include: If the node type of the first group of network nodes is the same as the node type of the second group of network nodes, the preset first matching value is determined as the type matching factor. When the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is a type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset second matching value is determined as the type matching factor. If the node types of the first group of network nodes and the node types of the second group of network nodes are different, and there is no type association between the node types of the first group of network nodes and the node types of the second group of network nodes, the preset third matching value is determined as the type matching factor. Wherein, the first matching value is greater than the second matching value, and the second matching value is greater than the third matching value.

8. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 1, characterized in that, The environmental status observation information includes the measured signal-to-noise ratio and measured packet loss rate of the corresponding network nodes during communication; The steps for assessing environmental interference based on the environmental status observation information of each network node among the multiple network nodes, and obtaining the environmental interference value of each network node, include: Based on the maximum and minimum rated signal-to-noise ratio of the target network node, the measured signal-to-noise ratio of the target network node is numerically standardized to obtain the first interference factor of the target network node, wherein the target network node is one of the multiple network nodes. By analyzing the similarity between the measured packet loss rate of the target network node and its set maximum packet loss rate, the second interference factor of the target network node is obtained. The ratio of the second interference factor to the first interference factor of the target network node is calculated to obtain the environmental interference value of the target network node.

9. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 1, characterized in that, The steps for obtaining multiple network nodes in the target area include: Multiple candidate nodes in the target area are randomly combined to obtain multiple node combinations, wherein the total number of candidate nodes included in each node combination is equal to the preset number. In the multiple node combinations, the degree of concentration of the spatial distribution of all candidate nodes included in each node combination is analyzed to obtain the node aggregation factor of each node combination; In the multiple node combinations, the spatial matching degree between each node combination and the target region is analyzed to obtain the regional matching factor of each node combination; The node aggregation factor and region matching factor of each node combination are fused to obtain the combination score of each node combination; Among the multiple node combinations, the node combination with the highest combination score is determined as the target node combination, and all candidate nodes included in the target node combination are determined as the network nodes.

10. The method for controlling access permissions to heterogeneous unmanned node data based on graph neural networks according to claim 9, characterized in that, The step of analyzing the spatial matching degree between each node combination and the target region in the plurality of node combinations to obtain the regional matching factor of each node combination includes: In the plurality of node combinations, a region matching factor for each node combination is determined based on the distance between the combination center of each node combination and the region center of the target region. The region matching factor is negatively correlated with the corresponding distance, and the combination center is the average value of the node positions of all candidate nodes included in the corresponding node combination. Alternatively, in the plurality of node combinations, the ratio of the combination area of ​​each node combination to the region area of ​​the target region is calculated to obtain the region matching factor for each node combination. The combination area is the convex hull area of ​​the projection points of all candidate nodes included in the corresponding node combination on the standard horizontal plane.

Images