Intelligent optimization and resource dynamic reallocation method and system based on graph neural network

Abstract

The present invention discloses a method and system for intelligent optimization and dynamic resource reallocation based on a graph neural network. The method includes the steps of data acquisition and modeling, network topology feature extraction, congestion detection and prediction, data flow reallocation and intelligent optimization, and real-time monitoring and feedback mechanism. The present invention uses a graph neural network model to achieve real-time modeling and intelligent optimization of network topology and data flow, accurately identifying and locating potential congested areas. Combined with a traffic reallocation strategy based on the maximum entropy principle, the system can dynamically adjust the allocation of data flows and resources, alleviating congestion and improving overall transmission efficiency. By introducing improved rules for cascading failures, the system effectively prevents secondary congestion caused by the failure of congested nodes, thereby enhancing network stability.

Description

Technical Field

[0001] The present invention relates to the field of computer networks and communication technologies, and in particular discloses a method and system for intelligent optimization and dynamic resource reallocation based on graph neural networks. Background Art

[0002] In modern communications networks, with the rapid growth of data traffic and the increasing complexity of network topologies, network congestion is becoming increasingly serious. Traditional congestion management and traffic distribution methods, such as fixed-rule flow control and static resource allocation, are often unable to effectively adapt to the rapid changes in network conditions and struggle to meet the needs of efficient, real-time data transmission.

[0003] In recent years, graph neural networks (GNNs) have been increasingly applied to network data flow management due to their advantages in processing graph-structured data. GNNs can capture the dynamic characteristics of network nodes and links, and through training and prediction, they can model congestion in complex networks. However, existing methods still have shortcomings in real-time identification of congested nodes, dynamic optimization of traffic distribution, and resource reallocation. This is especially true when dynamic changes in network structure or cascading failures occur, making it difficult to respond promptly, resulting in degraded network performance.

[0004] Furthermore, traffic allocation methods that consider cascading failure rules and the principle of maximum entropy in GNNs help ensure network resource balance and system stability, but they have not yet been fully utilized in the intelligent scheduling of network data flows. Therefore, this paper proposes an intelligent optimization method in graph neural networks that considers data congestion and maximum entropy redistribution. Through a self-learning module and feedback mechanism, dynamic traffic redistribution and congestion management are achieved to improve the overall transmission efficiency and resource utilization of the network. Summary of the Invention

[0005] The present invention provides a method and system for intelligent optimization and dynamic resource reallocation based on graph neural networks, which aims to solve one of the technical problems of network data flow congestion control and intelligent optimization in the above-mentioned prior art to a certain extent.

[0006] One aspect of the present invention relates to a method for intelligent optimization and dynamic resource reallocation based on a graph neural network, comprising the following steps:

[0007] Data Collection and Modeling: Collect historical data flow features of nodes and links in the network and build a graph neural network model for data flow congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic features are attached as attributes to the graph structure. Historical data flow features include the number of transmitted packets, number of bytes, transmission rate, and protocol type information.

[0008] Network topology feature extraction: Analyze historical data flow characteristics and extract feature vectors for each node and link through a multi-layer graph convolutional neural network structure to capture dynamic changes in network structure and traffic characteristics.

[0009] Congestion detection and prediction: Graph neural network training models are used to identify nodes and links in the network where congestion may occur. Based on cascading failure detection rules and network traffic characteristics, potential congested areas are predicted. The prediction accuracy of the graph neural network training model is dynamically adjusted by calculating the bandwidth utilization of nodes and links, as well as traffic variance indicators.

[0010] Data flow redistribution and intelligent optimization: When congestion risk is detected or actual congestion occurs, data flows are dynamically scheduled based on intelligent redistribution strategies. Traffic is redistributed using the maximum entropy principle to achieve a more balanced traffic load. Specifically, traffic is shifted along congested links by calculating optimized traffic distribution probabilities.

[0011] Real-time monitoring and feedback mechanism: The optimized network is monitored in real time. If performance indicators exceed set thresholds, the feedback mechanism triggers dynamic adjustment of data flows and automatically redistributes traffic to ensure efficient allocation of network resources and path optimization. Performance indicators include average transmission delay, bandwidth utilization, and traffic balance.

[0012] Furthermore, in the data collection and modeling steps, network data flow information is collected, a graph structure is generated, and the nodes that transmit data in the network are represented as Node collection, node With node The transmission relationship between two nodes is expressed as The edge set, the specific data flow size during the transmission process is expressed as Traffic collection, where Indicates the first A node is a specific device or system unit for data transmission. Representation node With node The connection edge between them is used to describe the transmission path or relationship between two nodes. If there is transmission, it is 1, otherwise it is 0. Represents a slave node To Node The amount of data traffic transmitted, Represents a slave node To Node The amount of data traffic transmitted, Meaning or express or ; Represents the set of nodes that have established valid connections in the network, which is a subset of the node set V, that is, ,Network data flow information includes network topology, ,node and edge attributes, and traffic conditions.

[0013] Furthermore, in the step of network topology feature extraction, the graph neural network model is used to characterize the network topology structure and traffic characteristics and generate node embeddings. and transfer relation embedding , the update rules are as follows:

[0014]

[0015] in, Representation node In the Layer embedding, Representation node In the Layer embedding, The trainable weight matrix of the kth layer is represented by the data flow matrix between nodes as described in point 2. Representation node With node The edge feature vector between the two nodes is as described in point 2. Representation node In the Layer embedding, represents the neighbor set of a node, is the information aggregation function, is the activation function.

[0016] Furthermore, in the congestion detection and prediction step, the node and edge embeddings generated by the graph neural network are used to predict potential congestion areas and their occurrence probability. , calculate the congestion threshold of each edge , where the congestion prediction formula is:

[0017]

[0018] in, and are model parameters, For nodes and nodes The embedding features of the edges.

[0019] When congestion is detected on a link, the cascading failure improvement rules are used to determine the impact of the congestion on surrounding links to prevent chain failures caused by single-point congestion. The specific improvement rules are as follows:

[0020]

[0021] in, Represents an edge The cascade risk coefficient, is the weight coefficient, For the Transmission nodes, Does not contain nodes The rest of the node set, , For nodes To Node The amount of data traffic transmitted, To represent a node To Node The congestion threshold of the transmission edge, For nodes To Node The amount of data traffic transmitted, The cascade risk coefficient is used to determine whether dynamic redistribution of traffic between transmission node pairs is required; if If the set threshold is exceeded, it is considered that there is a risk of chain congestion.

[0022] Furthermore, in the steps of data flow redistribution and intelligent optimization, in order to balance the redistribution of data flow in the network, the maximum entropy principle is first used to determine the probability of optimizing the flow, ensuring that the flow is evenly distributed on each path, improving the overall network efficiency, and defining the flow distribution probability. Satisfy the entropy maximization principle:

[0023]

[0024] Optimize the results under the premise of meeting the following constraints; redistribute traffic ratio constraints: ; Maximum capacity constraint for allocation: ,in, Represents the entropy value of the network system, Represents the edge set in the network, including all node pairs There are transmission relationships between edges, For the edge capacity, is the current flow, which is solved using the Lagrange multiplier method. Can be used for traffic redistribution;

[0025] Under the guidance of cascading failure judgment and maximum entropy principle, data flow distribution is dynamically adjusted based on the congestion area predicted by GNN to optimize network resources. The adjusted traffic distribution rules are as follows:

[0026]

[0027] in, is a node With node After the data traffic is redistributed between is a node With node The original data traffic between is a node With node The original data traffic between is the flow balance coefficient, represents the cascading failure impact weight, Assign probabilities to data streams based on maximum entropy optimization.

[0028] Furthermore, in the step of the real-time monitoring and feedback mechanism, the average transmission delay is defined as real-time monitoring of the average delay of data packet transmission within the network, calculated by the following formula:

[0029]

[0030] in, represents the average transmission delay of the network, For nodes To Node The transmission delay between Indicates the number of links in the network.

[0031] Furthermore, in the step of the real-time monitoring and feedback mechanism, bandwidth utilization is defined as the bandwidth occupancy of each link, calculated using the following formula:

[0032]

[0033] in, For nodes To Node Bandwidth utilization between is the data flow between the current node pair, The maximum bandwidth capacity of the link; when the bandwidth utilization exceeds the set threshold, the system will trigger traffic redistribution.

[0034] Furthermore, in the real-time monitoring and feedback mechanism, traffic balance is defined as the traffic variance between different links, calculated using the following formula:

[0035]

[0036] in, Indicates the traffic balance of the network. For nodes To Node Bandwidth utilization between represents the number of links in the network, It is the average bandwidth utilization of all links. If the traffic imbalance exceeds the set threshold, the system will start the traffic optimization module. When the risk of cascading failure is detected or the key performance indicators exceed the set threshold, the real-time adjustment mechanism will be triggered.

[0037] Another aspect of the present invention relates to a graph neural network-based intelligent optimization and dynamic resource reallocation system, which is used to implement the graph neural network-based intelligent optimization and dynamic resource reallocation method. The graph neural network-based intelligent optimization and dynamic resource reallocation system includes:

[0038] The data collection and modeling module is used to collect historical data flow characteristics of nodes and links in the network and build a graph neural network model for data flow congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic characteristics are attached as attributes to the graph structure. Historical data flow characteristics include the number of transmitted packets, number of bytes, transmission rate, and protocol type information.

[0039] The network topology feature extraction module is used to analyze the characteristics of historical data flows. It extracts the feature vectors of each node and link through a multi-layer graph convolutional neural network structure to capture the dynamic changes in network structure and traffic characteristics.

[0040] The congestion detection and prediction module uses a graph neural network training model to identify nodes and links that may be congested in the network. Based on cascading failure detection rules and network traffic characteristics, it predicts potential congested areas. By calculating the bandwidth utilization of nodes and links, as well as traffic variance indicators, it dynamically adjusts the prediction accuracy of the graph neural network training model.

[0041] The data flow redistribution and intelligent optimization module is used to dynamically schedule data flows based on intelligent redistribution strategies when congestion risks are detected or actual congestion occurs. It uses the maximum entropy principle to redistribute traffic to achieve a more balanced traffic load. Specifically, it calculates the optimized traffic distribution probability and then shifts traffic to congested links.

[0042] The real-time monitoring and feedback mechanism module is used to monitor the optimized network in real time. If performance indicators exceed the set threshold, the feedback mechanism is used to trigger dynamic adjustment of data flow and automatically redistribute traffic to ensure efficient allocation of network resources and path optimization. Performance indicators include average transmission delay, bandwidth utilization and traffic balance.

[0043] The beneficial effects achieved by the present invention are:

[0044] This invention provides a method and system for intelligent optimization and dynamic resource reallocation based on a graph neural network. Using a graph neural network (GNN) model, this system implements real-time modeling and intelligent optimization of network topology and data flows, accurately identifying and locating potential congested areas. Combined with a traffic reallocation strategy based on the maximum entropy principle, the system dynamically adjusts the allocation of data flows and resources, alleviating congestion and improving overall transmission efficiency. By introducing improved rules for cascading failures, it effectively prevents secondary congestion caused by the failure of congested nodes, enhancing network stability. Furthermore, a self-learning module continuously optimizes the GNN model using historical data, enabling adaptive adjustments and improving the model's congestion management capabilities. Furthermore, a real-time feedback mechanism enables immediate response and dynamic adjustment to changes in network status, ensuring efficient data flow transmission and rational resource allocation. By integrating real-time monitoring, intelligent optimization, and dynamic scheduling capabilities, this system provides an intelligent, automated, and efficient solution for data flow congestion management in complex network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of an embodiment of a method for intelligent optimization and dynamic resource reallocation based on a graph neural network according to the present invention;

[0046] Figure 2 This is a schematic diagram of a specific embodiment of the present invention's graph neural network prediction of network node congestion;

[0047] Figure 3 A schematic diagram of a specific embodiment of chain congestion prediction considering cascading failure rules in a graph neural network of the present invention;

[0048] Figure 4 Schematic diagram of a specific embodiment of intelligent redistribution of congestion considering the maximum entropy principle in the graph neural network of the present invention. DETAILED DESCRIPTION

[0049] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0050] like Figure 1 and Figure 2 As shown, the first embodiment of the present invention proposes an intelligent optimization and dynamic resource reallocation method based on a graph neural network, comprising the following steps:

[0051] Step S100, data collection and modeling: Collect historical data flow features of nodes and links in the network, build a graph neural network model for data flow congestion detection, use nodes to represent servers or routers in the network, use edges to represent links, and attach traffic features as attributes to the graph structure. Historical data flow features include the number of transmitted data packets, number of bytes, transmission rate, and protocol type information.

[0052] Regarding data collection and modeling, imagine an enterprise network or data center. The network administrator first needs to collect historical data for each node (such as servers and routers) and link (such as network interfaces and communication links). Specifically, this data includes: the number of packets transmitted by each node and the size (in bytes); the transmission rate and protocol type (such as TCP or UDP) of each link; and the time periods and frequency of peak network traffic. This data can be automatically collected by a network monitoring system (such as SNMP or NetFlow), updated in real time, and stored in a database. Based on this data, a network topology model is constructed, defining nodes and links as nodes and edges in a graph to facilitate subsequent feature extraction and analysis.

[0053] Step S200, network topology feature extraction: Analyze the historical data flow characteristics, extract the feature vectors of each node and link through a multi-layer graph convolutional neural network structure, and capture the dynamic changes of network structure and traffic characteristics.

[0054] After modeling is complete, network topology feature extraction uses a multi-layer graph neural network to extract feature vectors for each node and link. For example, the hierarchical structure of a graph neural network (GNN) can capture information such as the degree centrality of each node, the characteristics of neighboring nodes, and the strength of connections between nodes. The extracted feature vector contains information such as the node or link's traffic status, congestion, and historical transmission rate changes, describing the real-time state of the network. For example, suppose server A typically has high network traffic, while servers B and C have lower loads. A GNN can extract the high-load characteristics of node A and identify the potential congestion risk for that node in the future.

[0055] Step S300, congestion detection and prediction: Use the graph neural network training model to identify nodes and links where congestion may occur in the network, predict potential congested areas based on cascading failure detection rules and network traffic characteristics, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization of nodes and links, as well as the traffic variance index.

[0056] Congestion prediction and detection uses a trained graph neural network model to predict and detect potential congested nodes in the network. The model considers cascading failure rules. For example, by inputting node feature data and historical traffic data, the graph neural network identifies that server A may experience data congestion during peak hours. During the prediction process, the model detects that the transmission rate of the link connecting server A to other nodes has increased significantly and is approaching the maximum bandwidth of the link, which may cause congestion in adjacent nodes and lead to cascading congestion. Once these potential congestion areas are detected, the system will generate an early warning and intelligently implement subsequent traffic distribution optimization plans.

[0057] Step S400, data flow reallocation and intelligent optimization: When congestion risk is detected or actual congestion occurs, data flows are dynamically scheduled based on an intelligent reallocation strategy; traffic is redistributed using the maximum entropy principle to make the traffic load more balanced. Specifically, traffic is transferred to congested links by calculating the optimized traffic distribution probability.

[0058] By calculating the optimized traffic distribution probability, traffic on congested links is transferred, network transmission bottlenecks are reduced, and the overall network transmission efficiency is improved.

[0059] Data traffic allocation and intelligent optimization: when it is detected that server A is at risk of congestion and may lead to potential chain congestion, the system automatically starts the intelligent optimization strategy and combines the maximum entropy principle to allocate traffic in a probabilistic manner. Redistribute traffic to adjacent links. For example, the system can redistribute some traffic from server A to servers B and C, which have lower loads, to reduce pressure on server A. Based on graph neural network analysis, the system optimizes traffic distribution across the entire network, selecting load-balanced paths or links to minimize traffic concentration on a single link. For example, if congestion risk is detected on the link from server A to B, the system can select a backup path from A to C and then to B to reduce congestion on the primary link.

[0060] Step S500, real-time monitoring and feedback mechanism: The optimized network is monitored in real time. If performance indicators exceed the set threshold, the feedback mechanism is used to trigger dynamic adjustment of data flow and automatically redistribute traffic to ensure efficient allocation of network resources and path optimization. Performance indicators include average transmission delay, bandwidth utilization and traffic balance.

[0061] Real-time monitoring and feedback control: After network optimization is completed, the system monitors the network status in real time and implements feedback control. Specifically: the system continuously monitors the status of each node and link, and automatically adjusts the traffic distribution plan once a new traffic peak or congestion is detected. If the load of node A reaches a high level again, the system can dynamically reduce the traffic input to node A through a feedback mechanism, or pre-allocate it to other links before the traffic increases to prevent congestion. For example, when server A experiences a new traffic peak in the afternoon, the system will automatically allocate traffic to B or C again based on previous feedback experience to ensure the smooth operation of the network.

[0062] like Figure 2 The figure shows a specific implementation diagram of the network node congestion prediction process based on the graph neural network of the present invention. The figure is divided into three main parts: input layer, hidden layer and output layer. The specific implementation is as follows:

[0063] 1) Input layer: Assume the initial state of the graph, which contains three nodes i As well as the data traffic on their links, these initial data include the connection relationship between nodes and the weight of the connection, representing the current traffic distribution of the network.

[0064] 2) Hidden layer: During the training process of the graph neural network, the model captures the dynamic relationship between nodes through multi-layer feature propagation. The hidden layer consists of three layers, each of which updates the features of each node and aggregates the information of neighboring nodes through multiple iterations: Aggregate from and Information, while the node and It also receives status updates from other nodes. and The same is true; through multiple propagations of hidden layers, the graph neural network gradually captures the traffic status and congestion risk between each node and its neighboring nodes.

[0065] 3) Output layer: After feature extraction and training, the model outputs the prediction results. Nodes are displayed here. The state of has changed, indicating that the graph neural network has detected the node The results of the output layer will be used for further traffic redistribution or early warning, helping administrators to manage congestion in advance.

[0066] like Figure 3 As shown in the figure, it is a schematic diagram of a specific embodiment of chain congestion prediction considering cascading failure rules in the graph neural network of the present invention. This figure is divided into four sub-graphs, which respectively describe the normal situation of node traffic and the potential chain congestion prediction process under traffic congestion state: (a) Normal node traffic: This figure shows the network node The traffic is in normal state, and the node and There is a certain amount of outflow and , there is no congestion at the beginning, and the traffic is balanced in the network; (b) Node Traffic congestion, when the node When congestion occurs due to excessive traffic, in this case, in order to reduce the congestion pressure, the node Inflow According to the coefficient Attenuation is performed to alleviate its load. This mechanism helps the system When congestion occurs, prevent the congestion from getting worse; (c) detect the node Traffic will be congested due to the node The sum of received traffic (i.e. ) exceeds its maximum allowable load (i.e. ), so the node Congestion occurs; in this case, the system can detect the abnormal traffic through the graph neural network model and predict the node Potential congestion risk occurs; (d) Node is detected Traffic is also congested, similar to (c), due to the congested traffic received, the node The traffic demand exceeds its processing capacity, resulting in the risk of node congestion. At this time, the graph neural network can identify the congestion pattern, predict and mark the node As shown in the four subgraphs above, the graph neural network, incorporating the cascading failure rule in its hidden layer, can dynamically monitor and predict congestion at different nodes, providing support for chain congestion management.

[0067] like Figure 4 As shown, it is a schematic diagram of a specific embodiment of congestion intelligent redistribution considering the maximum entropy principle in the graph neural network of the present invention, which is as follows:

[0068] When the system detects a node and When chain congestion occurs, the intelligent redistribution mechanism starts to work, and by adjusting the traffic load of each node, the congestion in the network is alleviated. According to the maximum entropy principle, the model redistributes the traffic of the congested nodes and guides the traffic to the surrounding non-congested nodes, thereby reducing the load pressure of the central node. Indicates possible traffic distribution paths. The traffic of each congested node is redistributed according to the maximum entropy principle. The redistribution satisfies the formula: ,node for ,node for ,here, etc. represent the traffic weights between connected nodes, Representation node The congestion threshold, Representation node The maximum entropy principle distributes traffic appropriately, ensuring balanced load across nodes and reducing the risk of congestion across the entire network. This intelligent redistribution of congestion distributes excess traffic to non-congested nodes, achieving efficient network traffic management and preventing further congestion.

[0069] Furthermore, the intelligent optimization and dynamic resource reallocation method based on graph neural network provided in this embodiment, in step S100, collects network data flow information, generates a graph structure, and represents the nodes performing data transmission in the network as Node collection, node With node The transmission relationship between two nodes is expressed as The edge set, the specific data flow size during the transmission process is expressed as Traffic collection, where Indicates the first A node is a specific device or system unit for data transmission. Representation node With node The connection edge between them is used to describe the transmission path or relationship between two nodes. If there is transmission, it is 1, otherwise it is 0. Represents a slave node To Node The amount of data traffic transmitted, Represents a slave node To Node The amount of data traffic transmitted, Meaning or express or ; Represents the set of nodes that have established valid connections in the network, which is a subset of the node set V, that is, ,Network data flow information includes network topology, ,node and edge attributes, and traffic conditions.

[0070] Preferably, the intelligent optimization and dynamic resource reallocation method based on graph neural network provided in this embodiment, in step S200, uses the graph neural network model to characterize and learn the network topology and traffic characteristics, and generates node embeddings and transfer relation embedding , the update rules are as follows:

[0071] (1)

[0072] In formula (1), Representation node In the Layer embedding, Representation node In the Layer embedding, The trainable weight matrix of the kth layer is represented by the data flow matrix between nodes as described in point 2. Representation node With node The edge feature vector between the two nodes is as described in point 2. Representation node In the Layer embedding, represents the neighbor set of a node, is the information aggregation function, is the activation function.

[0073] Furthermore, the intelligent optimization and dynamic resource reallocation method based on graph neural network provided in this embodiment predicts potential congestion areas and their occurrence probability based on the node and edge embedding generated by the graph neural network in step S300. , calculate the congestion threshold of each edge , where the congestion prediction formula is:

[0074] (2)

[0075] In formula (2), and are model parameters, For nodes and nodes The embedding features of the edges.

[0076] When congestion is detected on a link, the cascading failure improvement rules are used to determine the impact of the congestion on surrounding links to prevent chain failures caused by single-point congestion. The specific improvement rules are as follows:

[0077] (3)

[0078] In formula (3), Represents an edge The cascade risk coefficient, is the weight coefficient, For the Transmission nodes, Does not contain nodes The rest of the node set, , For nodes To Node The amount of data traffic transmitted, To represent a node To Node The congestion threshold of the transmission edge, For nodes To Node The amount of data traffic transmitted, The cascade risk coefficient is used to determine whether dynamic redistribution of traffic between transmission node pairs is required; if If the set threshold is exceeded, it is considered that there is a risk of chain congestion.

[0079] Preferably, in the intelligent optimization and dynamic resource reallocation method based on graph neural network provided in this embodiment, in step S400, in order to balance the redistribution of data flow in the network, the maximum entropy principle is first used to determine the probability of optimizing the flow, ensuring that the flow is evenly distributed on each path, improving the overall network efficiency, and defining the flow distribution probability. Satisfy the entropy maximization principle:

[0080] (4)

[0081] In formula (4), the optimization result is obtained under the premise of satisfying the following constraints: redistribution flow ratio constraint: ; Maximum capacity constraint for allocation: , Represents the entropy value of the network system, Represents the edge set in the network, including all node pairs There are transmission relationships between edges, For the edge capacity, is the current flow, which is solved using the Lagrange multiplier method. Can be used for traffic redistribution;

[0082] Under the guidance of cascading failure judgment and maximum entropy principle, data flow distribution is dynamically adjusted based on the congestion area predicted by GNN to optimize network resources. The adjusted traffic distribution rules are as follows:

[0083] (5)

[0084] In formula (5), is a node With node After the data traffic is redistributed between is a node With node The original data traffic between is a node With node The original data traffic between is the flow balance coefficient, represents the cascading failure impact weight, Assign probabilities to data streams based on maximum entropy optimization.

[0085] Furthermore, in the graph neural network-based intelligent optimization and dynamic resource reallocation method provided in this embodiment, in step S500, the average transmission delay is defined as real-time monitoring of the average delay of data packet transmission within the network, calculated using the following formula:

[0086] (6)

[0087] In formula (6), represents the average transmission delay of the network, For nodes To Node The transmission delay between Indicates the number of links in the network.

[0088] Bandwidth utilization is defined as the bandwidth usage of each link and is calculated using the following formula:

[0089] (7)

[0090] In formula (7), For nodes To Node Bandwidth utilization between is the data flow between the current node pair, The maximum bandwidth capacity of the link; when the bandwidth utilization exceeds the set threshold, the system will trigger traffic redistribution.

[0091] Traffic balance is defined as the traffic variance between different links, calculated using the following formula:

[0092] (8)

[0093] In formula (8), Indicates the traffic balance of the network. For nodes To Node Bandwidth utilization between represents the number of links in the network, It is the average bandwidth utilization of all links. If the traffic imbalance exceeds the set threshold, the system will start the traffic optimization module. When the risk of cascading failure is detected or the key performance indicators exceed the set threshold, the real-time adjustment mechanism will be triggered.

[0094] The present invention relates to an intelligent optimization and dynamic resource reallocation system based on a graph neural network, which is used to implement an intelligent optimization and dynamic resource reallocation method based on a graph neural network. The intelligent optimization and dynamic resource reallocation system based on a graph neural network includes a data acquisition and modeling module, a network topology feature extraction module, a congestion detection and prediction module, a data flow reallocation and intelligent optimization module, and a real-time monitoring and feedback mechanism module. Among them, the data acquisition and modeling module is used to collect historical data flow features of nodes and links in the network, and construct a graph neural network model for data flow congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic features are attached to the graph structure as attributes. Historical data flow features include the number of transmitted data packets, the number of bytes, the transmission rate, and protocol type information; the network topology feature extraction module is used to analyze historical data flow features, extract feature vectors of each node and link through a multi-layer graph convolutional neural network structure, and capture dynamic changes in network structure and traffic features. ; Congestion detection and prediction module, used to use the graph neural network training model to identify nodes and links that may be congested in the network, predict potential congested areas based on cascading failure detection rules and network traffic characteristics, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization of nodes and links, as well as the traffic variance index; Data flow redistribution and intelligent optimization module, used to dynamically schedule data flows based on intelligent redistribution strategies when congestion risks are detected or actual congestion occurs; Use the maximum entropy principle to redistribute traffic to make the traffic load more balanced, specifically by calculating the optimized traffic distribution probability to transfer traffic on congested links; Real-time monitoring and feedback mechanism module, used to monitor the optimized network in real time. If the performance indicators exceed the set threshold, the feedback mechanism is used to trigger dynamic adjustment of the data flow and automatically redistribute traffic to ensure efficient allocation of network resources and path optimization. Performance indicators include average transmission delay, bandwidth utilization and traffic balance.

[0095] 1) Data Collection Module: This module is responsible for collecting real-time data from the network, including network traffic data, node status data, and connection relationship data. This data will be used for subsequent modeling and analysis.

[0096] 2) GNN Modeling Module: This module uses graph neural networks (GNNs) to model the collected data. GNNs can capture the complex relationships between network nodes and data flow patterns, providing a foundation for subsequent congestion detection and prediction.

[0097] 3) Congestion Detection and Prediction Module: This module detects and predicts network congestion based on a GNN model that considers cascading failure rules. When potential congestion is detected, the system can react in advance to prevent further congestion.

[0098] 4) Redistribution and Intelligent Optimization Module: This module is responsible for handling detected congestion issues. By using a GNN model that takes into account the principle of maximum entropy, the system redistributes traffic to non-congested nodes, optimizing the use of network resources and alleviating congestion.

[0099] 5) Real-time Monitoring and Feedback Module: This module monitors network status in real time and continuously adjusts and optimizes traffic distribution strategies based on feedback, ensuring stable network operation in a dynamic environment. The entire system, through the collaborative work of five modules—data collection, modeling, detection and prediction, reallocation and optimization, and real-time monitoring and feedback—achieves intelligent optimization and dynamic resource reallocation to address network congestion. This not only improves overall network performance and stability but also enhances the system's ability to respond to sudden congestion events.

[0100] This embodiment provides a method and system for intelligent optimization and dynamic resource reallocation based on a graph neural network. Compared to existing technologies, this method utilizes a graph neural network (GNN) model to achieve real-time modeling and intelligent optimization of network topology and data flows, accurately identifying and locating potential congested areas. Combined with a traffic reallocation strategy based on the maximum entropy principle, the system dynamically adjusts the allocation of data flows and resources, alleviating congestion and improving overall transmission efficiency. By introducing improved rules for cascading failures, it effectively prevents secondary congestion caused by congested node failures, enhancing network stability. Furthermore, a self-learning module utilizes historical data to continuously optimize the GNN model, enabling adaptive adjustments and improving the model's congestion management capabilities. Furthermore, a real-time feedback mechanism enables immediate response and dynamic adjustment to changes in network status, ensuring efficient data flow transmission and rational resource allocation. By integrating real-time monitoring, intelligent optimization, and dynamic scheduling capabilities, this system provides an intelligent, automated, and efficient solution for data flow congestion management in complex network environments.

[0101] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. A method for intelligent optimization and dynamic resource reallocation based on graph neural network, characterized in that: The following steps are involved: Data Collection and Modeling: Collect historical data flow features of nodes and links in the network and build a graph neural network model for data flow congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic features are attached as attributes to the graph structure. The historical data flow features include the number of transmitted packets, number of bytes, transmission rate, and protocol type information. Network topology feature extraction: Analyze the historical data flow characteristics and extract the feature vectors of each node and link through a multi-layer graph convolutional neural network structure to capture the dynamic changes in network structure and traffic characteristics; Congestion detection and prediction: Utilizes a graph neural network training model to identify nodes and links that may be congested in the network. Based on cascading failure detection rules and network traffic characteristics, it predicts potential congested areas. By calculating the bandwidth utilization of nodes and links, as well as traffic variance indicators, it dynamically adjusts the prediction accuracy of the graph neural network training model. Data flow redistribution and intelligent optimization: When congestion risk is detected or actual congestion occurs, data flows are dynamically scheduled based on intelligent redistribution strategies; It uses the maximum entropy principle to redistribute traffic, making the traffic load more balanced. Specifically, it calculates the optimized traffic distribution probability and transfers traffic to congested links. Real-time monitoring and feedback mechanism: The optimized network is monitored in real time. If performance indicators exceed set thresholds, the feedback mechanism triggers dynamic adjustment of data flows and automatic traffic redistribution to ensure efficient allocation of network resources and path optimization. Performance indicators include average transmission delay, bandwidth utilization, and traffic balance. In the congestion detection and prediction step, the potential congestion area and its probability of occurrence are predicted based on the node and edge embedding generated by the graph neural network. , calculate the congestion threshold of each edge , where the congestion prediction formula is: in, and are model parameters, For nodes and nodes Embedding features of edges; When congestion is detected on a link, the cascading failure improvement rules are used to determine the impact of the congestion on surrounding links to prevent chain failures caused by single-point congestion. The specific improvement rules are as follows: in, Represents an edge The cascade risk coefficient, is the weight coefficient, For the Transmission nodes, Does not contain nodes The rest of the node set, , For nodes To Node The amount of data traffic transmitted, To represent a node To Node The congestion threshold of the transmission edge, For nodes To Node The amount of data traffic transmitted, The cascade risk coefficient is used to determine whether dynamic redistribution of traffic between transmission node pairs is required; if If the set threshold is exceeded, it is considered that there is a risk of chain congestion.

2. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the data collection and modeling steps, network data flow information is collected, a graph structure is generated, and the nodes performing data transmission in the network are represented as Node collection, node With node The transmission relationship between two nodes is expressed as The edge set, the specific data flow size during the transmission process is expressed as Traffic collection, where Indicates the first A node is a specific device or system unit that performs data transmission; Representation node With node The connection edge between them is used to describe the transmission path or relationship between two nodes. If there is transmission, it is 1, otherwise it is 0; Represents a slave node To Node The amount of data traffic transmitted, Represents a slave node To Node The amount of data traffic transmitted, Meaning or express or ; Represents the set of nodes that have established valid connections in the network, which is a subset of the node set V, that is, The network data flow information includes network topology, node and edge attributes, and traffic conditions.

3. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the step of extracting network topology features, the graph neural network model is used to characterize the network topology structure and traffic characteristics and generate node embeddings. and transfer relation embedding , the update rules are as follows: in, Representation node In the Layer embedding, Representation node In the Layer embedding, Table k-th layer trainable weight matrix; Representation node With node The edge eigenvectors between Representation node In the Layer embedding, represents the neighbor set of a node, is the information aggregation function, is the activation function.

4. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the steps of data flow redistribution and intelligent optimization, in order to balance the redistribution of data flow in the network, the maximum entropy principle is first used to determine the probability of optimizing the flow, ensuring that the flow is evenly distributed on each path, improving the overall network efficiency, and defining the flow distribution probability. Satisfy the entropy maximization principle: Optimize the results under the premise of meeting the following constraints; redistribute traffic ratio constraints: ; Maximum capacity constraint for allocation: ,in, Represents the entropy value of the network system, Represents the edge set in the network, including all node pairs There are transmission relationships between edges, For the edge capacity, is the current flow, which is solved using the Lagrange multiplier method. Can be used for traffic redistribution; Guided by cascading failure judgment and the principle of maximum entropy, data flow distribution is dynamically adjusted based on the congested areas predicted by GNN to optimize network resources. The adjusted traffic distribution rules are as follows: in, is a node With node After the data traffic is redistributed between is a node With node The original data traffic between is a node With node The original data traffic between is the flow balance coefficient, represents the cascading failure impact weight, Assign probabilities to data streams based on maximum entropy optimization.

5. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the step of the real-time monitoring and feedback mechanism, the average transmission delay is defined as real-time monitoring of the average delay of data packet transmission in the network, calculated by the following formula: in, represents the average transmission delay of the network, For nodes To Node The transmission delay between Indicates the number of links in the network.

6. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the step of the real-time monitoring and feedback mechanism, the bandwidth utilization is defined as the bandwidth occupancy of each link, calculated by the following formula: in, For nodes To Node Bandwidth utilization between is the data flow between the current node pair, The maximum bandwidth capacity of the link; when the bandwidth utilization exceeds the set threshold, the system will trigger traffic redistribution.

7. The method for intelligent optimization and dynamic resource reallocation based on graph neural network according to claim 1, characterized in that: In the step of the real-time monitoring and feedback mechanism, the traffic balance is defined as the traffic variance between different links, calculated by the following formula: in, Indicates the traffic balance of the network. For nodes To Node Bandwidth utilization between represents the number of links in the network, It is the average bandwidth utilization of all links. If the traffic imbalance exceeds the set threshold, the system will start the traffic optimization module. When the risk of cascading failure is detected or the key performance indicators exceed the set threshold, the real-time adjustment mechanism will be triggered.

8. A system for intelligent optimization and dynamic resource reallocation based on a graph neural network, for implementing the method for intelligent optimization and dynamic resource reallocation based on a graph neural network according to any one of claims 1 to 7, characterized in that: The intelligent optimization and dynamic resource reallocation system based on graph neural network includes: The data collection and modeling module is used to collect historical data flow features of nodes and links in the network and build a graph neural network model for data flow congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic features are attached as attributes to the graph structure. The historical data flow features include the number of transmitted packets, number of bytes, transmission rate, and protocol type information. A network topology feature extraction module is used to analyze the historical data flow characteristics and extract the feature vectors of each node and link through a multi-layer graph convolutional neural network structure to capture the dynamic changes of network structure and traffic characteristics; A congestion detection and prediction module, which uses a graph neural network training model to identify nodes and links in the network where congestion may occur. Based on cascading failure detection rules and network traffic characteristics, it predicts potential congested areas and dynamically adjusts the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization of nodes and links and traffic variance indicators. The data flow redistribution and intelligent optimization module is used to dynamically schedule data flows based on intelligent redistribution strategies when congestion risks are detected or actual congestion occurs. It uses the maximum entropy principle to redistribute traffic to achieve a more balanced traffic load. Specifically, it calculates the optimized traffic distribution probability and then shifts traffic to congested links. The real-time monitoring and feedback mechanism module is used to monitor the optimized network in real time. If performance indicators exceed the set threshold, the feedback mechanism is used to trigger dynamic adjustment of data flow and automatically redistribute traffic to ensure efficient allocation of network resources and path optimization. The performance indicators include average transmission delay, bandwidth utilization and traffic balance.

Images