A method and system for allocating edge computing resources

Abstract

The present invention relates to the field of resource allocation technology, specifically to an edge computing resource allocation method and system, comprising the following steps: collecting network node information, identifying key nodes through embedded learning and analyzing communication patterns. Analyzing and encoding node connectivity and resource status using graph convolutional networks and self-organizing mapping algorithms, and iteratively optimizing network structures. Monitor and update network status, adjust resource allocation strategies, and generate dynamic resource allocation models. In the present invention, by performing data processing tasks near the data source, the data transmission distance and delay are significantly reduced, the processing speed is accelerated, and thus the application performance is directly improved. The simulated mapping and dynamic adjustment of resource status and demand make resource allocation more adaptable to actual needs and reduce resource waste. The adaptive adjustment of resource allocation is performed based on real-time network data to ensure the real-time and accuracy of resource allocation. This ability to respond and predict instantly optimizes the overall network operation.

Description

Technical Field

[0001] The present invention relates to the technical field of resource allocation, and in particular to an edge computing resource allocation method and system. Background Art

[0002] The field of resource allocation technology focuses on how to efficiently allocate and manage limited computing resources to meet the performance requirements of various applications and services in the system or network. This field covers resource optimization strategies from traditional data center resource management to modern cloud computing and edge computing. In resource allocation, key issues include how to dynamically allocate processing power, storage space, and network bandwidth based on application priority, resource availability, and other environmental parameters. Effective resource allocation strategies can improve the overall efficiency and performance of the system, reduce energy consumption, and improve user satisfaction.

[0003] Among them, edge computing resource allocation methods refer to a series of technologies and strategies for implementing resource management and optimization in edge computing environments. Edge computing, as a distributed computing framework, aims to migrate data processing tasks from centralized data centers to the edge of the network, that is, close to data sources and users. The main purpose of this resource allocation method is to reduce the data transmission distance, reduce latency, and increase processing speed, thereby optimizing application performance and user experience. By processing data locally, edge computing also helps save bandwidth and enhance data security and privacy protection.

[0004] In the existing technology, resource management is mostly concentrated in data centers, and processing tasks are highly centralized, which often leads to data transmission delays and slow processing speeds. Inflexible resource allocation cannot quickly adapt to sudden changes in demand, which often leads to waste or shortage of resources. In addition, centralized data processing increases the risk of data leakage and affects data security and privacy protection. These problems are particularly evident when processing real-time data that requires a quick response, which seriously limits the system's response speed and user experience. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose an edge computing resource allocation method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an edge computing resource allocation method, comprising the following steps:

[0007] S1: Collect node information in the edge network, convert it into the format required for embedding learning, identify key nodes, analyze communication patterns, adjust node feature weights, iteratively optimize and calculate the primary embedding vector as the node feature embedding vector;

[0008] S2: using the node feature embedding vector, applying a graph convolutional network to deepen the processing vector, analyzing the direct and indirect connectivity between nodes, extracting and encoding the connection characteristics of the network, and generating a network structure code;

[0009] S3: Based on the network structure coding, a self-organizing mapping algorithm is used to simulate the node resource status and demand changes, collect node demand and status data, analyze resource distribution characteristics, adjust node parameters to match resource distribution, and construct a two-dimensional resource distribution map;

[0010] S4: monitor the network status, update the node feature embedding vector and the network structure encoding, reallocate network resources based on the strength and stability of network connections, correct the weights and hierarchical parameters of the graph convolutional network, update the mapping rules of the self-organizing map, synchronize the node status and the prediction model, and form a strategy update feedback;

[0011] S5: Adjust the two-dimensional resource distribution map according to the strategy update feedback, evaluate the deviation between the current resource distribution map and the actual demand, quantitatively analyze the changes in resource usage patterns, adjust the resource allocation strategy and priority according to the analysis results, use the current network data to update the resource status of multiple nodes in the two-dimensional map, and generate a dynamic resource allocation model.

[0012] The node feature embedding vector specifically includes feature weights, iterative optimization results, and node characteristics. The network structure coding includes direct connection features, indirect connection features, and connectivity coding. The two-dimensional resource distribution map includes node resource status, resource demand changes, and resource distribution characteristics. The strategy update feedback specifically refers to network connection strength, network connection stability, and prediction model synchronization. The dynamic resource allocation model includes resource allocation strategy, resource allocation priority, and resource usage pattern.

[0013] As a further solution of the present invention, the step of acquiring the node feature embedding vector is specifically as follows:

[0014] S111: Collect information about each node from the edge network, including location, connection status, and communication frequency, combine the node activity records, transform the information into a structured format required for embedded learning, and generate structured node information;

[0015] S112: Identify key nodes for the structured node information, sort the nodes according to their communication frequency and the number of connected nodes, use an enhanced sorting algorithm to optimize the selection of key nodes, select nodes with high communication mode centrality, and generate a key node list;

[0016] S113: Adjust the feature weight of each node from the key node list by using a weight adjustment function, using the formula:

[0017]

[0018] Strengthen the influence of the central node of the network and generate the adjusted feature weight list w i ;

[0019] Among them, x i is the number of connected nodes of node i;

[0020] S114: Based on the adjusted feature weight list, a node feature embedding algorithm is applied, using the formula:

[0021]

[0022] Calculate and generate node feature embedding vector;

[0023] Among them, v i represents the embedding vector of node i, N(i) represents the set of neighbor nodes of node i, α ij represents the learning rate, which regulates the influence of neighbor nodes during the learning process, w ij represents the feature weights of node i and node j, emphasizing the interaction between key nodes, x j Represents the feature vector of node j.

[0024] As a further solution of the present invention, the steps of obtaining the network structure code are specifically as follows:

[0025] S211: Load each node feature embedding vector into the graph convolutional network, update the node feature representation through multiple layers of the graph convolutional layer, deepen the integration of node features in each iteration, and generate deep node features;

[0026] S212: Based on the deep node features, calculate the direct and indirect connectivity between nodes, strengthen the feature interaction between nodes through the product operation of the adjacency matrix and the feature matrix, and generate weighted node connectivity features;

[0027] S213: encoding the weighted node connectivity features, performing weighted summation on the features using a linear transformation combined with a weight matrix without an activation function, extracting a high-order network connection pattern from the connectivity features, and generating a network structure feature before encoding;

[0028] S214: Utilizing the network structure features before encoding, extracting the sets of directly adjacent nodes of multiple nodes based on the adjacent relationships between nodes, and calculating the relationship weights of each node to the adjacent nodes, using the formula:

[0029]

[0030] Generate network structure code;

[0031] Among them, NC irepresents the network structure encoding of node i, β ij represents the connection weight between nodes i and j, highlighting the connection strength between differentiated nodes, H j is the characteristic representation of node j in the network.

[0032] As a further solution of the present invention, the steps of obtaining the two-dimensional resource distribution map are specifically as follows:

[0033] S311: Load the network structure code into the self-organizing map algorithm, adjust parameters to simulate resource status and demand changes between nodes, refine the resource status model through repeated iterations, and generate node resource simulation results;

[0034] S312: Based on the node resource simulation result, collect node demand and status data in the entire network, use real-time data stream technology to record resource usage and demand changes of each node, and generate a node status data set for the entire network;

[0035] S313: Analyze the network-wide node status data set, use principal component analysis to identify resource usage patterns and trends, extract key resource distribution features, and generate resource distribution analysis results;

[0036] S314: According to the resource distribution analysis result, adjust the parameters of the nodes in the self-organizing map algorithm, using the formula:

[0037]

[0038] Match resource distribution and draw a two-dimensional resource distribution map;

[0039] Among them, R i represents the position of node i in the resource distribution graph, λ ij Adjust and optimize the efficiency of resource allocation, S j Represents the resource status information of node j obtained through data analysis.

[0040] As a further solution of the present invention, the step of obtaining the strategy update feedback is specifically as follows:

[0041] S411: Start the network status monitoring module to collect and analyze the strength and stability of network connections in real time, measure network load and traffic patterns through data sensing technology, and generate real-time network status results;

[0042] S412: Based on the real-time network status result, the node feature embedding vector and the network structure code are updated, and the consistency and update frequency of all node features and network codes are ensured by using data synchronization technology, so as to generate updated node feature embedding vector and network structure code;

[0043] S413: Analyze the updated node feature embedding vector and network structure coding, reallocate network resources, correct the weights and hierarchical parameters of the graph convolutional network, and use the optimization formula:

[0044]

[0045] Generate optimized graph convolutional network parameters;

[0046] Among them, B i represents the weight vector of node i in the graph convolutional network, θ ij is the adjusted resource allocation coefficient, reflecting the new strategy of network resource allocation, X j is the updated node feature vector;

[0047] S414: Based on the optimized graph convolutional network parameters, synchronize the node status and prediction model, update the mapping rules of the self-organizing map, integrate all update data and parameter adjustment results, and form a strategy update feedback.

[0048] As a further solution of the present invention, the steps of acquiring the dynamic resource allocation model are specifically as follows:

[0049] S511: extracting data from the policy update feedback, making key adjustments to the two-dimensional resource distribution map to meet the strength and stability of the current network connection, ensuring that the distribution map reflects the current network status, and generating an adjusted resource distribution map;

[0050] S512: Evaluate the deviation between the adjusted resource distribution map and the actual demand, use variance analysis to determine the size and impact range of the deviation, verify the key changes in the resource usage pattern, and generate a resource distribution deviation result;

[0051] S513: Based on the resource distribution deviation result, redefine the resource allocation strategy and priority, adjust the weights and hierarchical parameters in the graph convolutional network, and use the formula:

[0052] ΔW ij =α·(D j -D i )

[0053] Adjust to minimize the resource allocation error of the overall network and generate an optimized resource allocation strategy;

[0054] Where, ΔW ij represents the weight adjustment between nodes i and j in the resource allocation strategy, α is the adjustment factor, which controls the adjustment speed and amplitude, and D i and D j Indicates the resource requirements of the corresponding node to ensure the timeliness and effectiveness of the resource allocation strategy;

[0055] S514: Based on the optimized resource allocation strategy, using current network data, updating the resource status of multiple nodes in the two-dimensional graph to form a dynamic resource allocation model.

[0056] An edge computing resource allocation system, the edge computing resource allocation system is used to execute the above-mentioned edge computing resource allocation method, the system comprising:

[0057] The feature embedding module collects node information in the edge network, converts it into the target format, identifies key nodes, analyzes the communication mode between nodes, adjusts the node feature weights, performs iterative optimization, and generates node feature vectors;

[0058] The structural coding module uses the node feature vector to analyze the direct and indirect connectivity between nodes, extracts and encodes network connection characteristics, and generates network structure coding;

[0059] The resource mapping module simulates the node resource status and demand changes based on the network structure coding, collects node demand and status data, analyzes resource distribution characteristics, and constructs a resource distribution map;

[0060] The resource allocation monitoring module monitors the network status, updates the node feature vector and the network structure code, reallocates resources according to the strength and stability of the network connection, and generates a dynamic resource allocation model.

[0061] Compared with the prior art, the advantages and positive effects of the present invention are:

[0062] In the present invention, by performing data processing tasks near the data source, the data transmission distance and delay are significantly reduced, the processing speed is accelerated, and thus the application performance is directly improved. The simulated mapping and dynamic adjustment of resource status and demand make resource allocation more adaptable to actual needs and reduce resource waste. The adaptive adjustment of resource allocation is carried out according to the real-time data of the network, which ensures the real-time and accuracy of resource allocation. This ability of instant response and prediction optimizes the overall network operation and improves efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0064] Figure 2 A flowchart of the steps for obtaining a node feature embedding vector of the present invention;

[0065] Figure 3 A flowchart of the steps for obtaining the network structure code of the present invention;

[0066] Figure 4 A flowchart of the steps for obtaining a two-dimensional resource distribution map of the present invention;

[0067] Figure 5 A flow chart of the steps for obtaining feedback for the strategy update of the present invention;

[0068] Figure 6 This is a flow chart of the acquisition steps of the dynamic resource allocation model of the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0070] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0071] Embodiment 1

[0072] See also Figure 1 The present invention provides a technical solution: an edge computing resource allocation method, comprising the following steps:

[0073] S1: Collect node information in the edge network, convert it into the format required for embedding learning, identify key nodes, analyze communication patterns, adjust node feature weights, iteratively optimize and calculate the primary embedding vector as the node feature embedding vector;

[0074] S2: Use node feature embedding vectors, apply graph convolutional networks to deepen the processing vectors, analyze direct and indirect connectivity between nodes, extract and encode the connection characteristics of the network, and generate network structure encoding;

[0075] S3: Based on network structure coding, the self-organizing mapping algorithm is used to simulate the node resource status and demand changes, collect node demand and status data, analyze resource distribution characteristics, adjust node parameters to match resource distribution, and construct a two-dimensional resource distribution map;

[0076] S4: Monitor network status, update node feature embedding vectors and network structure encoding, reallocate network resources based on the strength and stability of network connections, correct the weights and hierarchical parameters of graph convolutional networks, update the mapping rules of self-organizing maps, synchronize node status and prediction models, and form strategy update feedback;

[0077] S5: Adjust the two-dimensional resource distribution map according to the policy update feedback, evaluate the deviation between the current resource distribution map and the actual demand, quantitatively analyze the changes in resource usage patterns, adjust the resource allocation strategy and priority according to the analysis results, use the current network data to update the resource status of multiple nodes in the two-dimensional map, and generate a dynamic resource allocation model.

[0078] The node feature embedding vector specifically includes feature weights, iterative optimization results, and node characteristics. The network structure encoding includes direct connection features, indirect connection features, and connectivity encoding. The two-dimensional resource distribution map includes node resource status, resource demand changes, and resource distribution characteristics. The strategy update feedback specifically refers to network connection strength, network connection stability, and prediction model synchronization. The dynamic resource allocation model includes resource allocation strategy, resource allocation priority, and resource usage pattern.

[0079] See also Figure 2 , the specific steps for obtaining the node feature embedding vector are:

[0080] S111: Collect information about each node from the edge network, including location, connection status, and communication frequency, combine the node activity records, transform the information into a structured format required for embedded learning, and generate structured node information;

[0081] The location, connection status and communication frequency of the nodes are collected from the edge network, combined with the node activity records, and converted into a structured format required for embedded learning. The efficiency of subsequent processing is improved by structuring the data. Structured data can be directly applied to a variety of data analysis models to extract the interaction patterns and communication frequencies between nodes, thereby generating structured node information.

[0082] S112: Identify key nodes for structured node information, sort nodes according to their communication frequency and the number of connected nodes, use an enhanced sorting algorithm to optimize the selection of key nodes, select nodes with high communication mode centrality, and generate a key node list;

[0083] The identification of key nodes is based on the communication frequency of the nodes and the number of connected nodes. The enhanced sorting algorithm is applied to optimize the selection process. This optimization allows the algorithm to maintain high efficiency and accuracy when processing batch network data. The enhanced sorting algorithm dynamically evaluates the criticality of nodes based on the actual communication data of the nodes in the network and determines which nodes play a central role in the network. In this way, a list of key nodes is generated.

[0084] S113: Adjust the feature weight of each node from the key node list through the weight adjustment function, using the formula:

[0085]

[0086] Strengthen the influence of the central node of the network and generate the adjusted feature weight list w i ;

[0087] Among them, x i is the number of connected nodes of node i;

[0088] formula:

[0089]

[0090] The benefit of the formula is that it converts the number of connections of a node into a weight between 0 and 1 through the sigmoid function, reflecting the centrality and activity of the node in the network. This conversion process emphasizes the criticality of highly connected nodes.

[0091] Detailed explanation of the formula and the process of formula calculation and derivation:

[0092] Set the number of connections x for node i i =5, the weight adjustment formula is:

[0093]

[0094] By calculation:

[0095]

[0096] The results show that the feature weight of node i is close to 1, which shows its high connectivity and centrality in the network.

[0097] S114: Based on the adjusted feature weight list, a node feature embedding algorithm is applied, using the formula:

[0098]

[0099] Calculate and generate node feature embedding vector;

[0100] Among them, v i represents the embedding vector of node i, N(i) represents the set of neighbor nodes of node i, α ij represents the learning rate, which regulates the influence of neighbor nodes during the learning process, w ij represents the feature weights of node i and node j, emphasizing the interaction between key nodes, x j Represents the feature vector of node j.

[0101] formula:

[0102]

[0103] The benefit of the formula is that it effectively captures the network embedding characteristics of the node by combining the weighted features of the neighboring nodes. This formula optimizes the calculation of the node embedding vector by dynamically adjusting the learning rate and weights, improving the model's expressiveness and prediction accuracy.

[0104] Detailed explanation of the formula and the process of formula calculation and derivation:

[0105] Refer to the three neighbors of node i and w calculated in paragraph 3 i =0.993, let α ij =0.1, and the feature vector x of each neighbor node j For [1, 2], [2, 1], [1, 1] respectively, calculate:

[0106] v i =0.1×0.993×[1,2]+0.1×0.993×[2,1]+0.1×0.993×[1,1]

[0107] v i =[0.0993, 0.1986]+[0.1986, 0.0993]+[0.0993, 0.0993]

[0108] v i =[0.3972, 0.3972]

[0109] The results show that the embedding vector of node i is [0.3972, 0.3972], showing its comprehensive network position weighted by connectivity and features of neighboring nodes.

[0110] See also Figure 3 , the steps for obtaining the network structure encoding are as follows:

[0111] S211: Load each node feature embedding vector into the graph convolutional network, update the node feature representation through multiple layers of the graph convolutional layer, deepen the integration of node features in each iteration, and generate deep node features;

[0112] In the initialization stage of the graph convolutional network, the embedding vector of each node is set as the feature input. The role of the graph convolutional layer is to update the feature representation of each node through information transmission between neighboring nodes. This process simulates the flow of information in real-world networks. Through multiple iterations, these features are gradually integrated into more complex graphical expressions that can capture deeper graph structure characteristics. The generated deep node features are a comprehensive reflection of the role and function of each node in the graph.

[0113] S212: Based on deep node features, calculate the direct and indirect connectivity between nodes, strengthen the feature interaction between nodes through the product operation of adjacency matrix and feature matrix, and generate weighted node connectivity features;

[0114] In the iterative analysis of node features, the calculation of direct and indirect connectivity is based on the adjacency matrix and feature matrix of the graph. The connectivity information between nodes can be obtained through the product operation of these two matrices. This process strengthens the feature interaction between nodes. This is because the graph convolutional network refers to the direct connection between nodes and the cumulative influence of indirect connections between other nodes when propagating features. The enhanced connectivity features help to reveal the information flow and structural dynamics in the network.

[0115] S213: Encode the weighted node connectivity features, use linear transformation combined with weight matrix to perform weighted summation on the features without activation function, extract high-order network connection patterns from the connectivity features, and generate network structure features before encoding;

[0116] The encoding of weighted node connectivity features is performed through a combination of linear transformation and weight matrix. This processing step does not use an activation function and directly extracts and encodes the high-order connection patterns of the network from the connectivity features. This method can reduce computational complexity while retaining key structural information. The network structure features before encoding provide basic data for subsequent steps. These data have been converted into a format that is more suitable for revealing network complexity, providing preparation for network coding.

[0117] S214: Using the network structure features before encoding, extract the set of directly adjacent nodes of multiple nodes based on the adjacent relationship between nodes, and calculate the relationship weight of each node to the adjacent node using the formula:

[0118]

[0119] Generate network structure code;

[0120] Among them, NC i represents the network structure encoding of node i, β ij represents the connection weight between nodes i and j, highlighting the connection strength between differentiated nodes, H j is the characteristic representation of node j in the network.

[0121] formula:

[0122]

[0123] The benefit of the formula is that by referring to the strength of the relationship between each node and its neighbor nodes and weighting the characteristic vectors of the neighbor nodes, the structural position of the nodes in the network can be more accurately drawn, thereby assisting in parsing and predicting network behavior.

[0124] Detailed explanation of the formula and the process of formula calculation and derivation:

[0125] Assume that node i has three neighboring nodes, with connection weights of 0.2, 0.5, and 0.3, and corresponding node feature vector values ​​of 10, 20, and 30. Then the network structure encoding of node i is calculated as:

[0126] NC i =0.2×10+0.5×20+0.3×30

[0127] NC i =2+10+9

[0128] NC i =21

[0129] The results show that the network structure code of node i is 21, which reflects the weighted sum of the features of its neighboring nodes and is a quantitative representation of the node structure and function in the network.

[0130] See also Figure 4 , the specific steps for obtaining the two-dimensional resource distribution map are:

[0131] S311: Load the network structure code into the self-organizing map algorithm, adjust the parameters to simulate the resource status and demand changes between nodes, refine the resource status model through repeated iterations, and generate node resource simulation results;

[0132] The node resource simulation process involves using network structure coding in the self-organizing map algorithm to simulate the resource status and demand changes of multiple nodes in the network. It is necessary to adjust the parameters in the algorithm based on the real-time interaction data of each node to ensure that the simulation results can reflect the actual situation. After multiple iterations, the system can generate a model that reflects the resource status of each node. This model will directly affect the efficiency and fairness of resource allocation in the network. Through this method, the results of various resource allocation strategies can be foreseen and the strategies can be optimized before actually adjusting the resource allocation.

[0133] S312: Based on the node resource simulation results, collect node demand and status data in the entire network, use real-time data stream technology to record the resource usage and demand changes of each node, and generate a node status data set for the entire network;

[0134] The collection of node status data for the entire network is carried out through real-time data streaming technology, which can continuously monitor the resource usage and demand changes of each node in the network. The application of real-time data streaming technology ensures the real-time update and integrity of the data, allowing network managers to understand the network status in a timely manner and adjust network resources accordingly to meet the real-time needs of the nodes, thereby improving the operating efficiency and resource utilization of the entire network. The implementation of this method relies on the data capture and processing system to ensure that all data can be accurately recorded and analyzed.

[0135] S313: Analyze the node status data set of the entire network, use principal component analysis to identify resource usage patterns and trends, extract key resource distribution features, and generate resource distribution analysis results;

[0136] The analysis of resource distribution characteristics involves the use of statistical methods such as principal component analysis and clustering to process node status data. The data includes the resource utilization rate and demand change information of the node. Through data analysis methods, the patterns and trends of resource usage are identified. The analysis results will assist network managers in analyzing the current status of resource distribution and potential improvement directions. The key to this process is to accurately extract and process data, and select matching analysis models to reveal hidden patterns in the data, so as to formulate more effective resource allocation and optimization strategies.

[0137] S314: According to the resource distribution analysis result, adjust the parameters of the nodes in the self-organizing map algorithm, using the formula:

[0138]

[0139] Match resource distribution and draw a two-dimensional resource distribution map;

[0140] Among them, R i represents the position of node i in the resource distribution graph, λ ij Adjust and optimize the efficiency of resource allocation, S j Represents the resource status information of node j obtained through data analysis.

[0141] formula:

[0142]

[0143] The benefit of the formula is that by adjusting the resource allocation coefficient λ between nodes ij , can flexibly respond to dynamic changes in the network and optimize the allocation of resources among nodes.

[0144] Detailed explanation of the formula and the process of formula calculation and derivation:

[0145] Set the resource status of node i and its neighboring node j to S j =20 units, resource allocation coefficient λij =0.5, then the resource distribution of node i is calculated as:

[0146] R i =0.5×20=10 units

[0147] The results show that by adjusting the coefficients and the resource status of adjacent nodes, node i can adjust its resource distribution according to the actual needs of the network, thereby optimizing its resource utilization while ensuring network efficiency.

[0148] See also Figure 5 , the steps for obtaining strategy update feedback are as follows:

[0149] S411: Start the network status monitoring module to collect and analyze the strength and stability of network connections in real time, measure network load and traffic patterns through data sensing technology, and generate real-time network status results;

[0150] Start the network monitoring module to capture and analyze network load data and traffic patterns in real time, and measure through data sensing technology. The purpose is to obtain the real-time load situation and traffic distribution of the network in order to respond to any changes in the network status. This process ensures that the collected data can reflect the most realistic network status, which is crucial for subsequent network resource allocation and optimization.

[0151] S412: Based on the real-time network status results, the node feature embedding vector and the network structure code are updated, and the data synchronization technology is used to ensure the consistency and update frequency of all node features and network codes, and generate updated node feature embedding vectors and network structure codes;

[0152] Based on the collected network status data, the system will automatically update the node feature embedding vector and network structure coding to ensure that all node data is synchronized with the current network status. This step is performed through data synchronization technology, which not only updates the node data, but also adjusts the overall coding of the network, thereby ensuring that the feature vector and network coding of each node in the network can reflect the current network environment.

[0153] S413: Analyze the updated node feature embedding vector and network structure encoding, reallocate network resources, correct the weights and hierarchical parameters of the graph convolutional network, and use the optimization formula:

[0154]

[0155] Generate optimized graph convolutional network parameters;

[0156] Among them, B i represents the weight vector of node i in the graph convolutional network, θ ijis the adjusted resource allocation coefficient, reflecting the new strategy of network resource allocation, X j is the updated node feature vector;

[0157] formula:

[0158]

[0159] The benefit of the formula is that by adjusting θ ij , that is, the resource allocation coefficient between nodes, can flexibly reallocate network resources to match changes in network load, thereby optimizing the performance and response speed of the entire network.

[0160] Detailed explanation of the formula and the process of formula calculation and derivation:

[0161] Set the resource allocation coefficient θ between node i and node j ij According to the load rate adjustment of node j, if the load rate of node j is 0.8, then

[0162] θ ij =1.25×0.8=1.0

[0163] If X j , the feature vector of node j is [1, 2, 3], then for node i, the calculation is:

[0164] B i =1.0·[1,2,3]=[1,2,3]

[0165] The results show that: the weight vector W of node i i It is directly driven by the resource status of node j, which shows that when resources are reallocated, node i can use the resource status of node j to optimize its own parameter settings and ensure the reasonable allocation and utilization of network resources. The result of the derivation step is the optimized position and configuration of node i in the network.

[0166] S414: Based on the optimized graph convolutional network parameters, synchronize the node status and prediction model, update the mapping rules of the self-organizing map, integrate all updated data and parameter adjustment results, and form strategy update feedback.

[0167] The system will synchronize the status and prediction model of all nodes by updating the rules of self-organizing mapping. This synchronization not only includes the update of status data, but also involves the optimization and adjustment of mapping rules to ensure that the behavior of each node is consistent with the overall network strategy. In this process, the system comprehensively refers to the current status of multiple nodes and the output of the prediction model to form a set of dynamically adjusted network resource management strategies, aiming to achieve the optimal configuration of network resources through continuous feedback and adjustment.

[0168] See also Figure 6,The specific steps for obtaining the dynamic resource allocation model are:

[0169] S511: extract data from the policy update feedback, make key adjustments to the two-dimensional resource distribution map to match the strength and stability of the current network connection, ensure that the distribution map reflects the current network status, and generate an adjusted resource distribution map;

[0170] The adjusted two-dimensional resource distribution map refers to the strength and stability of the network connection. By monitoring the network status in real time, the resource allocation reflects the current network demand, thereby adjusting the strategy to match the cyclically changing conditions. This process is performed through data analysis technology to ensure that each adjustment is based on the current network operation data. Data analysis technology includes real-time flow analysis and trend prediction models to evaluate changes in network status. These technologies update the resource distribution map in real time to reflect the current network demand and priority, thereby ensuring the optimal configuration of network resources.

[0171] S512: Evaluate the deviation between the adjusted resource distribution map and the actual demand, use variance analysis to determine the size and impact of the deviation, verify the key changes in resource usage patterns, and generate resource distribution deviation results;

[0172] The deviation between the resource distribution map and the actual demand is assessed through statistical methods, including variance analysis and regression analysis, which assist in quantitatively analyzing how resource usage patterns change over time. This analysis involves the processing of batch real-time data through an advanced data analysis platform, which is equipped with automated data cleaning and processing capabilities to identify deficiencies in resource allocation and estimated overloaded or underutilized resources.

[0173] S513: Based on the resource distribution deviation results, redefine the resource allocation strategy and priority, adjust the weights and layer parameters in the graph convolutional network, and use the formula:

[0174] ΔW ij =α·(D j -D i )

[0175] Adjust to minimize the resource allocation error of the overall network and generate an optimized resource allocation strategy;

[0176] Where, ΔW ij represents the weight adjustment between nodes i and j in the resource allocation strategy, α is the adjustment factor, which controls the adjustment speed and amplitude, and D i and D j Indicates the resource requirements of the corresponding node to ensure the timeliness and effectiveness of the resource allocation strategy;

[0177] formula:

[0178] ΔWij =α·(D j -D i )

[0179] The benefit of the formula is that by adjusting the node weights in the graph convolutional network, the formula can directly reflect the dynamic adjustment of demand differences between network nodes, thereby improving the flexibility and efficiency of network resource allocation.

[0180] Detailed explanation of the formula and the process of formula calculation and derivation:

[0181] Set the demand of node i to D i = 150 units, the demand at node j is D j = 200 units, learning rate α = 0.1. According to the formula,

[0182] ΔW ij =0.1·(200-150)=0.1·50=5

[0183] This means that the weight from node i to node j needs to be increased by 5 units in order to better balance the resource allocation between the two.

[0184] The results show that the weight adjustment ΔW obtained by calculation ij =5, which means that in order to meet the needs of node j relative to node i, the resource flow is optimized by increasing the weight, which helps to alleviate the resource pressure of node i and enhance the resource support of node j.

[0185] S514: Based on the optimized resource allocation strategy, using the current network data, the resource status of multiple nodes in the two-dimensional graph is updated to form a dynamic resource allocation model.

[0186] The generated dynamic resource allocation model is based on data analysis and machine learning algorithms, which can learn from real-time data and predict changes in resource demand and optimize resource allocation. The implementation of the model includes multi-level decision support systems that integrate the functions of resource status monitoring, demand forecasting and resource optimization allocation. Through this integrated approach, the model not only responds to current resource allocation needs, but also predicts future estimated changes, providing a mechanism for continuous updating and self-optimization for resource management.

[0187] An edge computing resource allocation system, the edge computing resource allocation system is used to execute the above-mentioned edge computing resource allocation method, the system includes:

[0188] The feature embedding module collects node information in the edge network, converts it into the target format, identifies key nodes, analyzes the communication mode between nodes, adjusts the node feature weights, performs iterative optimization, and generates node feature vectors;

[0189] The structural coding module uses node feature vectors to analyze the direct and indirect connectivity between nodes, extracts and encodes network connection characteristics, and generates network structure coding;

[0190] The resource mapping module is based on network structure coding, simulates node resource status and demand changes, collects node demand and status data, analyzes resource distribution characteristics, and constructs a resource distribution map;

[0191] The resource allocation monitoring module monitors the network status, updates the node feature vector and network structure coding, reallocates resources according to the strength and stability of network connections, and generates a dynamic resource allocation model.

[0192] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for allocating edge computing resources, characterized in that: The following steps are involved: Collect node information in the edge network, convert it into the format required for embedding learning, identify key nodes, analyze communication patterns, adjust node feature weights, iteratively optimize and calculate primary embedding vectors as node feature embedding vectors; Using the node feature embedding vector, applying a graph convolutional network to deepen the processing vector, analyzing the direct and indirect connectivity between nodes, extracting and encoding the connection characteristics of the network, and generating a network structure code; Based on the network structure coding, a self-organizing mapping algorithm is used to simulate the node resource status and demand changes, collect node demand and status data, analyze resource distribution characteristics, adjust node parameters to match resource distribution, and construct a two-dimensional resource distribution map; Monitor the network status, update the node feature embedding vector and the network structure encoding, reallocate network resources based on the strength and stability of network connections, correct the weights and hierarchical parameters of the graph convolutional network, update the mapping rules of the self-organizing map, synchronize the node status and the prediction model, and form strategy update feedback; The two-dimensional resource distribution map is adjusted according to the strategy update feedback, the deviation between the current resource distribution map and the actual demand is evaluated, the changes in resource usage patterns are quantitatively analyzed, the resource allocation strategy and priority are adjusted according to the analysis results, the resource status of multiple nodes in the two-dimensional map is updated using the current network data, and a dynamic resource allocation model is generated.

2. The edge computing resource allocation method according to claim 1, characterized in that: The node feature embedding vector specifically includes feature weights, iterative optimization results, and node characteristics. The network structure coding includes direct connection features, indirect connection features, and connectivity coding. The two-dimensional resource distribution map includes node resource status, resource demand changes, and resource distribution characteristics. The strategy update feedback specifically refers to network connection strength, network connection stability, and prediction model synchronization. The dynamic resource allocation model includes resource allocation strategy, resource allocation priority, and resource usage pattern.

3. The edge computing resource allocation method according to claim 2, characterized in that: The steps for obtaining the node feature embedding vector are specifically as follows: Collect information about each node from the edge network, including location, connection status, and communication frequency, combine it with node activity records, transform the information into a structured format required for embedded learning, and generate structured node information; Identify key nodes for the structured node information, sort the nodes according to their communication frequency and the number of connected nodes, use an enhanced sorting algorithm to optimize the selection of key nodes, select nodes with high communication mode centrality, and generate a key node list; Adjust the feature weight of each node from the key node list through the weight adjustment function, using the formula: Strengthen the influence of the central node of the network and generate the adjusted feature weight list w i ; Among them, x i is the number of connected nodes of node i; Based on the adjusted feature weight list, a node feature embedding algorithm is applied using the formula: Calculate and generate node feature embedding vector; Among them, v i represents the embedding vector of node i, N(i) represents the set of neighbor nodes of node i, α ij represents the learning rate, which regulates the influence of neighbor nodes during the learning process, w ij represents the feature weights of node i and node j, emphasizing the interaction between key nodes, x j Represents the feature vector of node j.

4. The edge computing resource allocation method according to claim 3, characterized in that: The steps of obtaining the network structure code are specifically as follows: Loading each of the node feature embedding vectors into the graph convolutional network, updating the node feature representation through multiple layers of the graph convolutional layer, deepening the integration of node features in each iteration, and generating deep node features; Based on the deep node features, the direct and indirect connectivity between nodes is calculated, and the feature interaction between nodes is strengthened through the product operation of the adjacency matrix and the feature matrix to generate weighted node connectivity features; Encoding the weighted node connectivity features, using linear transformation combined with a weight matrix to perform weighted summation on the features without an activation function, extracting high-order network connection patterns from the connectivity features, and generating network structure features before encoding; Using the network structure features before encoding, the sets of directly adjacent nodes of multiple nodes are extracted based on the adjacent relationships between nodes, and the relationship weights of each node to the adjacent nodes are calculated using the formula: Generate network structure code; Among them, NC i represents the network structure encoding of node i, β ij represents the connection weight between nodes i and j, highlighting the connection strength between differentiated nodes, H j is the characteristic representation of node j in the network.

5. The edge computing resource allocation method according to claim 4, characterized in that: The steps for obtaining the two-dimensional resource distribution map are specifically as follows: Loading the network structure code into the self-organizing map algorithm, adjusting parameters to simulate resource status and demand changes between nodes, refining the resource status model through repeated iterations, and generating node resource simulation results; Based on the node resource simulation results, node demand and status data in the entire network are collected, and the resource utilization rate and demand changes of each node are recorded using real-time data stream technology to generate a node status data set for the entire network; Analyze the network-wide node status data set, use principal component analysis to identify resource usage patterns and trends, extract key resource distribution features, and generate resource distribution analysis results; According to the resource distribution analysis results, the parameters of the nodes in the self-organizing map algorithm are adjusted using the formula: Match resource distribution and draw a two-dimensional resource distribution map; Among them, R i represents the position of node i in the resource distribution graph, λ ij Adjust and optimize the efficiency of resource allocation, S j Represents the resource status information of node j obtained through data analysis.

6. The edge computing resource allocation method according to claim 5, characterized in that: The steps for obtaining the strategy update feedback are specifically as follows: Start the network status monitoring module to collect and analyze the strength and stability of network connections in real time, measure network load and traffic patterns through data sensing technology, and generate real-time network status results; Based on the real-time network status result, the node feature embedding vector and the network structure code are updated, and the consistency and update frequency of all node features and network codes are ensured by using data synchronization technology to generate updated node feature embedding vectors and network structure codes; The updated node feature embedding vector and network structure encoding are analyzed, network resources are reallocated, weights and hierarchical parameters of the graph convolutional network are corrected, and the optimization formula is adopted: Generate optimized graph convolutional network parameters; Among them, B i represents the weight vector of node i in the graph convolutional network, θ ij is the adjusted resource allocation coefficient, reflecting the new strategy of network resource allocation, X j is the updated node feature vector; Based on the optimized graph convolutional network parameters, the node status and prediction model are synchronized, the mapping rules of the self-organizing map are updated, and all update data and parameter adjustment results are integrated to form a strategy update feedback.

7. The edge computing resource allocation method according to claim 6, characterized in that: The steps for obtaining the dynamic resource allocation model are specifically as follows: Extracting data from the policy update feedback, making key adjustments to the two-dimensional resource distribution map to match the strength and stability of the current network connection, ensuring that the distribution map reflects the current network status, and generating an adjusted resource distribution map; Evaluate the deviation between the adjusted resource distribution map and the actual demand, determine the size and impact of the deviation using variance analysis, verify key changes in resource usage patterns, and generate resource distribution deviation results; According to the resource distribution deviation results, the resource allocation strategy and priority are redefined, and the weights and layer parameters in the graph convolutional network are adjusted using the formula: ΔW ij =α·(D j -D i ) Adjust to minimize the resource allocation error of the overall network and generate an optimized resource allocation strategy; Where, ΔW ij represents the weight adjustment between nodes i and j in the resource allocation strategy, α is the adjustment factor, which controls the adjustment speed and amplitude, and D i and D j Indicates the resource requirements of the corresponding node to ensure the timeliness and effectiveness of the resource allocation strategy; Based on the optimized resource allocation strategy, the current network data is used to update the resource status of multiple nodes in the two-dimensional graph to form a dynamic resource allocation model.

8. An edge computing resource allocation system, characterized in that: According to the edge computing resource allocation method according to any one of claims 1 to 7, the system comprises: The feature embedding module collects node information in the edge network, converts it into the target format, identifies key nodes, analyzes the communication mode between nodes, adjusts the node feature weights, performs iterative optimization, and generates node feature vectors; The structural coding module uses the node feature vector to analyze the direct and indirect connectivity between nodes, extracts and encodes network connection characteristics, and generates network structure coding; The resource mapping module simulates the node resource status and demand changes based on the network structure coding, collects node demand and status data, analyzes resource distribution characteristics, and constructs a resource distribution map; The resource allocation monitoring module monitors the network status, updates the node feature vector and the network structure code, reallocates resources according to the strength and stability of the network connection, and generates a dynamic resource allocation model.

Images