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

Abstract

The invention discloses an intelligent optimization and resource dynamic redistribution method and system based on a graph neural network. The method comprises the steps of data acquisition and modeling, network topology feature extraction, congestion detection and prediction, data flow redistribution and intelligent optimization, and real-time monitoring and feedback mechanism. According to the invention, real-time modeling and intelligent optimization of a network topology structure and a data flow are realized through the graph neural network model, so that a potential congestion area can be accurately identified and positioned; in combination with a flow redistribution strategy of a maximum entropy principle, the system can dynamically adjust the distribution of data streams and resources, relieve congestion and improve the overall transmission efficiency; by introducing an improvement rule of cascade failure, secondary congestion caused by failure of the congested nodes is effectively prevented, and the stability of the network is enhanced.

Description

Technical Field

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

[0002] In modern communication networks, with the rapid growth of data traffic and the increasing complexity of network topologies, the problem of network congestion has become increasingly serious. Traditional congestion management and traffic allocation methods, such as traffic control based on fixed rules and static resource allocation, usually cannot effectively adapt to the rapid changes in network states and are difficult to meet the requirements of efficient and real-time data transmission.

[0003] In recent years, due to its advantages in processing graph-structured data, the Graph Neural Network (GNN) has gradually been applied to the field of network data stream management. GNN can capture the dynamic characteristics of network nodes and links and model the congestion situation in complex networks through training and prediction. However, existing methods still have deficiencies in real-time identification of congested nodes, dynamic optimization of traffic allocation, and resource reallocation, especially in the case of dynamic changes in network structure or cascading failures, it is often difficult to respond in a timely manner, resulting in a decline in network performance.

[0004] In addition, the traffic allocation method considering cascading failure rules and the principle of maximum entropy in GNN helps to ensure the balance of network resources and the stability of the system, but it has not been fully utilized in the intelligent scheduling of network data streams. For this reason, the present invention proposes an intelligent optimization method considering data chain congestion and maximum entropy reallocation in a graph neural network, and realizes dynamic traffic reallocation and congestion management through a self-learning module and a feedback mechanism to improve the overall transmission efficiency and resource utilization rate of the network. Summary of the Invention

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

[0006] One aspect of the present invention relates to an intelligent optimization and resource dynamic reallocation method based on a graph neural network, including the following steps: Data collection and modeling: Collect the historical data stream characteristics of nodes and links in the network, and construct a graph neural network model for data stream congestion detection. Use nodes to represent servers or routers in the network, edges to represent links, and traffic characteristics as attributes attached to the graph structure. The historical data stream characteristics include the number of transmitted data packets, bytes, transmission rate, and protocol type information; Network topology feature extraction: Analyze the features of historical data streams, 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 features; Congestion detection and prediction: Use a graph neural network training model to identify the nodes and links in the network that may experience congestion. Based on the cascade failure detection rules and network traffic features, predict potential congestion areas, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization rate of nodes and links, as well as the traffic variance index; Data stream reallocation and intelligent optimization: When congestion risk or actual congestion is detected, dynamically schedule the data stream based on the intelligent reallocation strategy; Use the principle of maximum entropy for traffic reallocation to make the traffic load more balanced. Specifically, calculate the optimized traffic allocation probability to transfer the traffic of congested links; Real-time monitoring and feedback mechanism: Monitor the optimized network in real time. If the performance indicators exceed the set thresholds, use the feedback mechanism to trigger the dynamic adjustment of the data stream and automatically perform traffic reallocation to ensure the efficient allocation of network resources and path optimization. The performance indicators include average transmission delay, bandwidth utilization rate, and traffic balance.

[0007] Furthermore, in the steps of data collection and modeling, collect network data stream information, generate a graph structure, and represent the nodes that perform data transmission in the network as a set of nodes, where the node and the node The transmission relationship between two nodes is represented as a set of edges, and the specific data stream size during the transmission process is represented as a set of traffic, where, represents the th node in the network, that is, the specific device or system unit that performs data transmission, represents the node and the node The connection edge between them is used to describe the transmission path or relationship between the two nodes. If there is a transmission, it is 1, otherwise it is 0, represents the data traffic size transmitted from the node to the node , represents the data traffic size transmitted from the node to the node , means the meaning of "or", represents or ; represents the set of nodes with established effective connections in the network, which is a subset of the set of nodes V, that is, , the network data flow information includes the network topology structure, the attributes of nodes and edges, and the traffic conditions.

[0008] Furthermore, in the step of network topology feature extraction, a graph neural network model is used to perform representation learning on the network topology structure and traffic features to generate node embeddings and transmission relationship embeddings , and the update rule is as follows: Among them, represents the embedding of node at the th layer, represents the embedding of node at the th layer, represents the trainable weight matrix of the kth layer, such as the data flow matrix between nodes described in point 2, represents the edge feature vector between node and node , such as the transmission relationship between two nodes described in point 2, represents the embedding of node at the th layer, represents the neighbor set of the node, is the information aggregation function, is the activation function.

[0009] Furthermore, in the step of congestion detection and prediction, based on the node and edge embeddings generated by the graph neural network, potential congestion regions and their occurrence probabilities are predicted , and the congestion threshold of each edge is calculated , where the congestion prediction formula is: Among them, and are model parameters, is the embedding feature of the edge between node and node .

[0010] When congestion of a certain link is detected, the impact of the congestion on the surrounding links is judged through the cascade failure improvement rule to prevent chain failures caused by single-point congestion. The specific improvement rule is: Among them, represents the cascade risk coefficient of edge , is the weight coefficient, is the th transmission node, is the set of the remaining nodes that do not contain a node ; , is the data traffic volume transmitted from node to node ; represents the congestion threshold of the transmission edge from node to node ; is the data traffic volume transmitted from node to node ; judges whether it is necessary to dynamically re - allocate the traffic between transmission node pairs through the cascading risk coefficient; if exceeds the set threshold, it is regarded as having a cascading congestion risk.

[0011] Furthermore, in the steps of data flow re - allocation and intelligent optimization, in order to balance the re - allocation of data flows in the network, first, the probability of optimizing the adjustment of the traffic is determined by using the principle of maximum entropy to ensure that the traffic is evenly distributed on each path and improve the overall network efficiency. Define the traffic allocation probability to satisfy the principle of entropy maximization: Under the premise of satisfying the following constraint conditions, optimize the result; re - allocation traffic ratio constraint: ; maximum capacity constraint for allocation: , where represents the entropy value of the network system, represents the set of edges in the network, including all the edges with a transmission relationship between node pairs ; is the capacity of edge ; is the current traffic, and the obtained by using the Lagrange multiplier method can be used for traffic re - allocation; Under the guidance of cascading failure judgment and the principle of maximum entropy, based on the congested area predicted by GNN, dynamically adjust the data flow allocation and optimize the network resources. The adjusted traffic allocation rule is: where is the data traffic volume after re - allocation between node and node ; is the original data traffic volume between node and node ; is the original data traffic volume between node and node ; is the traffic balance coefficient, represents the impact weight of cascading failure, is the data flow allocation probability based on maximum entropy optimization.

[0012] Furthermore, in the steps of the real-time monitoring and feedback mechanism, the average transmission delay is defined as the real-time monitoring of the average delay of data packet transmission in the network, and is calculated by the following formula: where, represents the average transmission delay of the network, is the node to the node the transmission delay between, represents the number of links in the network.

[0013] Furthermore, in the steps of the real-time monitoring and feedback mechanism, the bandwidth utilization rate is defined as by monitoring the bandwidth occupancy of each link, and is calculated by the following formula: where, is the node to the node the bandwidth utilization rate between, is the data traffic between the current node pairs, is the maximum bandwidth capacity of this link; when the bandwidth utilization rate is higher than the set threshold, the system will trigger traffic reallocation.

[0014] Furthermore, in the steps of the real-time monitoring and feedback mechanism, the traffic balance is defined as by calculating the traffic variance between different links, and is calculated by the following formula: where, represents the traffic balance of the network, is the node to the node the bandwidth utilization rate between, represents the number of links in the network, is the average bandwidth utilization rate of all links. If the degree of traffic imbalance exceeds the set threshold, the system starts the traffic optimization module; when detecting the risk of cascading failure or the key performance indicators exceed the set threshold, the real-time adjustment mechanism is triggered.

[0015] Another aspect of the present invention relates to an intelligent optimization and resource dynamic reallocation system based on a graph neural network, used to implement the intelligent optimization and resource dynamic reallocation method based on a graph neural network above. The intelligent optimization and resource dynamic reallocation system based on a graph neural network includes: The data collection and modeling module is used to collect the historical data stream characteristics of nodes and links in the network, and construct a graph neural network model for data stream congestion detection. Servers or routers in the network are represented by nodes, links are represented by edges, and traffic characteristics are attached as attributes to the graph structure. The historical data stream characteristics 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 the historical data stream 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 the network structure and traffic characteristics; The congestion detection and prediction module is used to train the model using the graph neural network, identify the nodes and links in the network that may experience congestion, predict potential congestion areas based on cascade 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 rate of nodes and links and the traffic variance index; The data stream reallocation and intelligent optimization module is used to dynamically schedule the data stream based on an intelligent reallocation strategy when congestion risk or actual congestion is detected; the maximum entropy principle is used for traffic reallocation to make the traffic load more balanced. Specifically, the traffic transfer of congested links is performed by calculating the optimized traffic allocation probability; The real-time monitoring and feedback mechanism module is used to monitor the optimized network in real time. If the performance indicators exceed the set thresholds, the feedback mechanism is used to trigger the dynamic adjustment of the data stream, automatically perform traffic reallocation, and ensure the efficient allocation of network resources and path optimization. The performance indicators include average transmission delay, bandwidth utilization rate, and traffic balance.

[0016] The beneficial effects achieved by the present invention are as follows: The present invention provides an intelligent optimization and resource dynamic reallocation method and system based on a graph neural network. Through the graph neural network (GNN) model, real-time modeling and intelligent optimization of the network topology structure and data stream are realized to accurately identify and locate potential congestion areas. Combining the traffic reallocation strategy of the maximum entropy principle, the system can dynamically adjust the allocation of data streams and resources, relieve congestion and improve the overall transmission efficiency. By introducing an improved rule for cascade failure, secondary congestion caused by the failure of congested nodes is effectively prevented, and the stability of the network is enhanced. In addition, the self-learning module continuously optimizes the GNN model using historical data to adaptively adjust and improve the congestion management ability of the model. At the same time, the real-time feedback mechanism realizes instant response and dynamic adjustment to network state changes, ensuring the efficient transmission of data streams and the reasonable allocation of resources. Integrating real-time monitoring, intelligent optimization, and dynamic scheduling functions, it provides an intelligent and automated efficient solution for data stream congestion management in complex network environments. Brief Description of the Drawings

[0017] Figure 1 Schematic diagram of a process of an embodiment of the intelligent optimization and resource dynamic reallocation method based on a graph neural network according to the present invention; Figure 2 Schematic diagram of a specific embodiment of predicting node congestion in a graph neural network of the present invention; Figure 3 Schematic diagram of a specific embodiment of cascade failure rule - considered chain congestion prediction in a graph neural network of the present invention; Figure 4 Schematic diagram of a specific embodiment of congestion intelligent reallocation considering the principle of maximum entropy in a graph neural network of the present invention. Specific embodiments

[0018] To better understand the above - mentioned technical solutions, the following will describe the above - mentioned technical solutions in detail in conjunction with the accompanying drawings of the specification and specific embodiments.

[0019] As Figure 1 and Figure 2 shown, a first embodiment of the present invention proposes an intelligent optimization and resource dynamic reallocation method based on a graph neural network, including the following steps: Step S100, data collection and modeling: Collect historical data stream features of nodes and links in the network, and construct a graph neural network model for data stream congestion detection. Represent nodes as servers or routers in the network, represent edges as links, and attach traffic features as attributes to the graph structure. The historical data stream features include the number of transmitted data packets, the number of bytes, the transmission rate, and protocol type information.

[0020] For data collection and modeling, assume that in an enterprise internal network or data center, a network administrator first needs to collect historical data of each node (such as servers, routers) and link (such as network interfaces and communication links) in the network. Specifically, the data includes: the number of data packets and data size (number of bytes) transmitted by each node; the transmission rate and protocol type (such as TCP, UDP) of each link; the peak traffic time period and occurrence frequency in the network. These data can be automatically collected through a network monitoring system (such as SNMP or NetFlow), and the data will be updated in real - time and stored in a database. Based on these data, a topological model of the network is established, and nodes and links are defined as nodes and edges of a graph to facilitate subsequent feature extraction and analysis.

[0021] Step S200, network topology feature extraction: Analyze the historical data stream features, and extract feature vectors of each node and link through a multi - layer graph convolutional neural network structure to capture the dynamic changes of the network structure and traffic features.

[0022] Network topology feature extraction: After the modeling is completed, a multi-layer graph neural network is used to extract the feature vectors of each node and link. For example, through the hierarchical structure of the graph neural network (GNN), information such as the degree centrality of each node, the features of neighboring nodes, and the connection strength between nodes can be captured. The extracted feature vectors contain information such as the traffic status, congestion situation, and historical transmission rate changes of the nodes or links, which are used to describe the real-time state of the network. Suppose a certain server A usually has a high transmission volume in the network, while the loads of other servers B and C are low. Through GNN, the high-load feature of node A can be extracted, and the possible congestion risk of this node in the future can be identified.

[0023] Step S300: Congestion detection and prediction: Use the graph neural network training model to identify the nodes and links in the network that may experience congestion. Based on the cascade failure detection rules and network traffic characteristics, predict potential congestion areas, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization rate of nodes and links and the traffic variance index.

[0024] Congestion prediction and detection: Use the trained graph neural network model to predict and detect potential congestion nodes in the network. The model considers the cascade failure rules. For example, through the input of node feature data and historical traffic data, the graph neural network can identify that server A may experience data congestion during peak hours. During the prediction process, the model detects that in the connection links between server A and other nodes, the transmission rate has increased significantly and is close to the maximum bandwidth of the link, which may cause congestion problems for adjacent nodes and lead to cascading congestion. Once these possible congestion areas are detected, the system will generate a warning and intelligently optimize the subsequent traffic allocation plan.

[0025] Step S400: Data flow reallocation and intelligent optimization: When congestion risk or actual congestion is detected, dynamically schedule the data flow based on the intelligent reallocation strategy; use the principle of maximum entropy to reallocate the traffic to make the traffic load more balanced. Specifically, calculate the optimized traffic allocation probability to transfer the traffic of the congested link.

[0026] By calculating the optimized traffic allocation probability, transfer the traffic of the congested link to reduce the network transmission bottleneck and improve the overall network transmission efficiency.

[0027] Data traffic allocation and intelligent optimization: When it is detected that server A has a congestion risk leading to potential cascading congestion, the system automatically activates the intelligent optimization strategy and combines the principle of maximum entropy to distribute the traffic with a probability Reallocate to adjacent links. For example: The system reallocates part of the traffic from server A to servers B and C with lower loads, reducing the pressure on server A. Based on the analysis results of the graph neural network, the system optimizes the traffic distribution of the entire network, selects load-balanced paths or links, and tries to avoid traffic concentration on a certain link. Suppose a congestion risk is detected on the link from server A to B, the system can select an alternative path from A to C and then to B to relieve the congestion on the main link.

[0028] Step S500, Real-time monitoring and feedback mechanism: Real-time monitor the optimized network. If the performance metrics exceed the set thresholds, use the feedback mechanism to trigger the dynamic adjustment of the data stream and automatically perform traffic redistribution to ensure the efficient allocation of network resources and path optimization. The performance metrics include average transmission delay, bandwidth utilization, and traffic balance.

[0029] Real-time monitoring and feedback control. After the network optimization is completed, the system performs real-time monitoring on the network status and implements feedback control. Specifically: The system continuously monitors the status of each node and link. Once a new traffic peak or congestion phenomenon is found, it will automatically adjust the traffic distribution plan. If the load of node A reaches a high level again, the system can dynamically reduce the traffic input to node A through the feedback mechanism, or pre-allocate it to other links before the traffic increases to prevent congestion. For example, when a new traffic peak appears on server A in the afternoon, the system will automatically redistribute the traffic to B or C again according to the previous feedback experience to ensure the smooth operation of the network.

[0030] As Figure 2 shown, it is a schematic diagram of the specific implementation of the network node congestion prediction process based on the graph neural network of the present invention. This figure is divided into three main parts: the input layer, the hidden layer, and the output layer. The specific implementation is as follows: 1) Input layer: Assume the initial state of the graph, which contains three nodes i and the data traffic on their links. These initial data include the connection relationships between nodes and the weights of the connections, representing the current traffic distribution of the network.

[0031] 2) Hidden layer: During the training process of the graph neural network, the model captures the dynamic relationships between nodes through multi-layer feature propagation. The hidden layer consists of three layers, and each layer updates the features of each node, aggregating the information of neighbor nodes through multiple iterations: Node aggregates the information from and , and at the same time, nodes and will also obtain status updates from other nodes. Nodes and The same is true; through multiple propagations of the hidden layer, the graph neural network gradually captures the traffic status and congestion risk between each node and its neighbor nodes.

[0032] 3) Output layer: After the model goes through feature extraction and training, it outputs the prediction results. Here, it shows that the status of the node has changed, indicating that the graph neural network has detected that the node may face congestion risk. The results of the output layer will be used for further traffic redistribution or early warning to help the administrator conduct congestion management in advance.

[0033] As Figure 3 shown, it is a schematic diagram of a specific embodiment of the cascading congestion prediction considering the cascading failure rule in the graph neural network of the present invention. This figure is divided into four subgraphs, which respectively describe the potential cascading congestion prediction processes in the normal situation of node traffic and the traffic congestion state: (a) Normal node traffic: This figure represents the normal traffic state of the network node , and the nodes and have certain outflow traffic and . Initially, there is no congestion phenomenon, and the traffic remains balanced in the network; (b) Detecting congestion of node . When the node becomes congested due to excessive traffic, in this case, to relieve the congestion pressure, the inflow traffic of the node is attenuated according to the coefficient to relieve its load. This mechanism helps the system avoid the further deterioration of the congestion situation when the node is congested; (c) Detecting that the traffic of node will be congested. Since the sum of the traffic received by the node (i.e., ) exceeds its allowed maximum carrying capacity (i.e., ), the node becomes congested; in this case, the system can detect this abnormal traffic through the graph neural network model and predict the potential congestion risk of the node ; (d) Detecting that the traffic of node is also congested. Similar to (c), due to the received congested traffic, the traffic demand of the node exceeds its processing capacity, resulting in a congestion risk for the node. At this time, the graph neural network can identify this congestion pattern, predict and mark the node is a high-risk node, which is the cascading failure rule. Through the processes illustrated by the above four sub-diagrams, the graph neural network combines the cascading failure rule in the hidden layer, enabling dynamic monitoring and prediction of the congestion situation of different nodes, and providing support for chain congestion management.

[0034] As Figure 4 shown, it is a schematic diagram of a specific embodiment of congestion intelligent reallocation considering the principle of maximum entropy in the graph neural network of the present invention, specifically as follows: When the system detects that nodes and have a chain congestion, the intelligent reallocation mechanism starts to work. By adjusting the traffic load of each node, the congestion phenomenon in the network is alleviated. According to the principle of maximum entropy, the model redirects the traffic of congested nodes to non-congested nodes around, reducing the load pressure on the central node. The nodes in the figure represent possible traffic allocation paths. The traffic of each congested node is reallocated according to the principle of maximum entropy. The reallocation for node satisfies the formula: , for node it is , for node it is , where, etc. represent the traffic weights between connected nodes, represents the congestion threshold of node , represents the total capacity of node . The principle of maximum entropy ensures load balance of each node through reasonable traffic allocation, thereby reducing the congestion risk of the entire network. This congestion intelligent reallocation method enables overloaded traffic to be dispersed to non-congested nodes, achieving efficient traffic management of the network and avoiding further spread of congestion.

[0035] Furthermore, in the intelligent optimization and resource dynamic reallocation method based on the graph neural network provided in this embodiment, in step S100, network data stream information is collected to generate a graph structure. The nodes for data transmission in the network are represented as the node set, and the transmission relationship between two nodes and node is represented as the edge set, and the specific data stream size during the transmission process is represented as the traffic set. Among them, represents the th node in the network, that is, the specific device or system unit for data transmission, represents the connection between node and node The connecting edges between are used to describe the transmission path or relationship between two nodes. If there is transmission, it is 1; otherwise, it is 0. Indicates from node To node The size of the data flow transmitted, Indicates from node To node The size of the data flow transmitted, Indicates the meaning of OR, Indicates OR ; Indicates the set of nodes with effective connections established in the network, which is a subset of the node set V, that is , The network data flow information includes the network topology structure, the attributes of nodes and edges, and the traffic situation.

[0036] Preferably, in the intelligent optimization and resource dynamic reallocation method based on graph neural network provided in this embodiment, in step S200, the graph neural network model is used to perform representation learning on the network topology structure and traffic characteristics to generate node embeddings And transmission relationship embeddings , The update rule is as follows: (1) In formula (1), Indicates node At the Layer embedding, Indicates node At the Layer embedding, Represents the trainable weight matrix of the kth layer, such as the data flow matrix between nodes described in point 2, Indicates node And node The edge feature vector between, such as the transmission relationship between two nodes described in point 2, Indicates node At the Layer embedding, Indicates the neighbor set of the node, Is an information aggregation function, Is an activation function.

[0037] Furthermore, in the intelligent optimization and resource dynamic reallocation method based on graph neural network provided in this embodiment, in step S300, based on the node and edge embeddings generated by the graph neural network, predict the potential congestion areas and their occurrence probabilities , Calculate the congestion threshold of each edge , Where the congestion prediction formula is: (2) In formula (2), and are model parameters, is the node and node the embedding feature of the edge between them.

[0038] When a certain link congestion is detected, the impact of the congestion on the surrounding links is judged through the cascading failure improvement rule to prevent the cascading failure caused by single-point congestion. The specific improvement rule is: (3) In formula (3), represents the cascading risk coefficient of edge , is the weight coefficient, is the th transmission node, is the set of the remaining nodes excluding node , , is the node transmitting data traffic volume to node , represents the congestion threshold of the transmission edge from node to node , is the node transmitting data traffic volume to node , judges whether it is necessary to dynamically reallocate the traffic between the transmission node pairs through the cascading risk coefficient; if exceeds the set threshold, it is regarded as having the risk of cascading congestion.

[0039] Preferably, in the intelligent optimization and resource dynamic reallocation method based on the graph neural network provided in this embodiment, in step S400, in order to balance the reallocation of the data flow in the network, first, the probability of optimizing and adjusting the traffic is determined by using the maximum entropy principle to ensure that the traffic is evenly distributed on each path and improve the overall network efficiency. Define the traffic allocation probability satisfies the principle of entropy maximization: (4) In formula (4), on the premise of satisfying the following constraints, the optimization result; the constraint of the reallocated traffic ratio: ; the constraint of the maximum allocated capacity: , represents the entropy value of the network system, represents the set of edges in the network, including all the edges with a transmission relationship between node pairs , is the edge Capacity For the current traffic, it is obtained by solving using the Lagrange multiplier method Can be used for traffic reallocation; Under the guidance of cascade failure judgment and the maximum entropy principle, based on the congestion area predicted by GNN, dynamically adjust the data flow allocation and optimize network resources. The adjusted traffic allocation rule is: (5) In formula (5), Is the node And the node The data traffic after being reallocated between them, Is the node And the node The original data traffic between them, Is the node And the node The original data traffic between them, Is the traffic balance coefficient, Represents the cascade failure impact weight, Is the data flow allocation probability optimized based on the maximum entropy.

[0040] Furthermore, in the intelligent optimization and resource dynamic reallocation method based on graph neural network provided in this embodiment, in step S500, the average transmission delay is defined as the real-time monitoring of the average delay of packet transmission in the network, and is calculated by the following formula: (6) In formula (6), Represents the average transmission delay of the network, Is the node To the node The transmission delay between them, Represents the number of links in the network.

[0041] The bandwidth utilization rate is defined as monitoring the bandwidth occupancy of each link and is calculated by the following formula: (7) In formula (7), Is the node To the node The bandwidth utilization rate between them, Is the data traffic between the current node pairs, Is the maximum bandwidth capacity of this link; when the bandwidth utilization rate is higher than the set threshold, the system will trigger traffic reallocation.

[0042] The traffic balance is defined as calculating the traffic variance between different links and is calculated by the following formula: (8) In formula (8), represents the traffic balance of the network, is the node to node the bandwidth utilization rate between them, represents the number of links in the network, is the average bandwidth utilization rate of all links. If the degree of traffic imbalance exceeds the set threshold, the system starts the traffic optimization module; when the risk of cascading failure or key performance indicators exceed the set threshold is detected, the real-time adjustment mechanism is triggered.

[0043] The present invention relates to an intelligent optimization and resource dynamic reallocation system based on a graph neural network, which is used to implement an intelligent optimization and resource dynamic reallocation method based on a graph neural network. The intelligent optimization and resource dynamic reallocation system based on a graph neural network includes a data collection 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 collection and modeling module is used to collect the historical data flow characteristics 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 characteristics are attached as attributes to the graph structure. The historical data flow characteristics include the number of transmitted data packets, bytes, transmission rate, and protocol type information; the 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 the network structure and traffic characteristics; the congestion detection and prediction module is used to use the graph neural network training model to identify the nodes and links in the network that may be congested, and predict potential congestion areas based on the cascading failure detection rules and network traffic characteristics. By calculating the bandwidth utilization rate of nodes and links, as well as the traffic variance index, the prediction accuracy of the graph neural network training model is dynamically adjusted; the data flow reallocation and intelligent optimization module is used to dynamically schedule the data flow based on the intelligent reallocation strategy when congestion risk or actual congestion occurs; the maximum entropy principle is used for traffic reallocation to make the traffic load more balanced. Specifically, by calculating the optimized traffic allocation probability, the traffic of congested links is transferred; the real-time monitoring and feedback mechanism module is used to perform real-time monitoring on the optimized network. If the performance indicators exceed the set threshold, the feedback mechanism is used to trigger the dynamic adjustment of the data flow, and the traffic is automatically reallocated to ensure the efficient allocation of network resources and path optimization. The performance indicators include average transmission delay, bandwidth utilization rate, and traffic balance.

[0044] 1) Data Acquisition Module: This module is responsible for collecting real-time data from the network, including network traffic data, node status data, connection relationship data, etc. These data will be used for subsequent modeling and analysis.

[0045] 2) GNN Modeling Module: This module uses the Graph Neural Network (GNN) to model the collected data. Through GNN, the complex relationships between network nodes and data flow patterns can be captured, providing a basis for subsequent congestion detection and prediction.

[0046] 3) Congestion Detection and Prediction Module: This module detects and predicts congestion situations in the network based on the GNN model considering the cascade failure rule. When potential congestion is detected, the system can react in advance to prevent the congestion from expanding further.

[0047] 4) Reallocation and Intelligent Optimization Module: This module is responsible for handling the detected congestion problems. Through the GNN model considering the maximum entropy principle, the system redistributes the traffic to non-congested nodes, optimizes the use of network resources, and alleviates the congestion situation.

[0048] 5) Real-time Monitoring and Feedback Module: This module monitors the network status in real-time and continuously adjusts and optimizes the traffic allocation strategy according to the feedback to ensure the stable operation of the network in a dynamic environment. Through the collaborative work of the five modules of data acquisition, modeling, detection and prediction, reallocation and optimization, and real-time monitoring and feedback, the entire system realizes the intelligent optimization of network data flow congestion and dynamic resource reallocation. This not only improves the overall performance and stability of the network but also enhances the system's ability to handle sudden congestion situations.

[0049] This embodiment provides an intelligent optimization and dynamic resource reallocation method and system based on graph neural network. Compared with the prior art, through the Graph Neural Network (GNN) model, it realizes the real-time modeling and intelligent optimization of the network topology structure and data flow to accurately identify and locate potential congestion areas. Combining the traffic reallocation strategy based on the maximum entropy principle, the system can dynamically adjust the allocation of data flow and resources, alleviate congestion, and improve the overall transmission efficiency. By introducing an improved rule for cascade failure, it effectively prevents secondary congestion caused by the failure of congested nodes and enhances the stability of the network. In addition, the self-learning module continuously optimizes the GNN model using historical data to adaptively adjust and improve the congestion management ability of the model. At the same time, the real-time feedback mechanism realizes the immediate response and dynamic adjustment to network state changes, ensuring the efficient transmission of data flow and the reasonable allocation of resources. Integrating real-time monitoring, intelligent optimization, and dynamic scheduling functions, it provides an intelligent and automated efficient solution for data flow congestion management in complex network environments.

[0050] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent optimization and resource dynamic reallocation method based on graph neural network, characterized in that, Including the following steps: Data collection and modeling: Collect the historical data stream characteristics of nodes and links in the network, and construct a graph neural network model for data stream congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic characteristics are attached as attributes to the graph structure. The historical data stream characteristics include the number of transmitted packets, bytes, transmission rate, and protocol type information; Network topology feature extraction: Analyze the historical data stream 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; Congestion detection and prediction: Use the graph neural network training model to identify the nodes and links in the network that may experience congestion. Based on the cascade failure detection rule and network traffic characteristics, predict potential congestion areas, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization rate of nodes and links and the traffic variance index; Data stream reallocation and intelligent optimization: When congestion risk or actual congestion is detected, dynamically schedule the data stream based on the intelligent reallocation strategy; Adopt the principle of maximum entropy to reallocate the traffic, making the traffic load more balanced. Specifically, calculate the optimized traffic allocation probability and transfer the traffic of congested links; Real-time monitoring and feedback mechanism: Monitor the optimized network in real time. If the performance index exceeds the set threshold, use the feedback mechanism to trigger the dynamic adjustment of the data stream, automatically perform traffic reallocation, and ensure the efficient allocation of network resources and path optimization. The performance index includes average transmission delay, bandwidth utilization rate, and traffic balance; 2. The intelligent optimization and resource dynamic reallocation method based on graph neural network according to claim 1, characterized in that, In the steps of data collection and modeling, network data stream information is collected to generate a graph structure. Nodes that perform data transmission in the network are represented as a set of nodes, where the node and the node The transmission relationship between two nodes is represented as a set of edges, and the specific data stream size during the transmission process is represented as a set of traffic. Among them, represents the th node in the network, that is, the specific device or system unit that performs data transmission; represents the node and the node The connection edge between them is used to describe the transmission path or relationship between two nodes. If there is a transmission, it is 1, otherwise it is 0; represents the data traffic size transmitted from the node to the node ; represents the data traffic size transmitted from the node to the node ; represents the meaning of "or", represents or ; represents the set of nodes with effective connections established in the network, which is a subset of the set of nodes V, that is , and the network data stream information includes the network topology structure, the attributes of nodes and edges, and the traffic situation.

3. The intelligent optimization and resource dynamic reallocation method based on a graph neural network according to claim 1, wherein In the steps of extracting the network topology features, a graph neural network model is used to perform representation learning on the network topology structure and traffic features to generate node embeddings and transmission relationship embeddings , and the update rule is as follows: Among them, represents the embedding of node at the -th layer, represents the embedding of node at the -th layer, represents the trainable weight matrix of the k-th layer; represents the edge feature vector between node and node , represents the embedding of node at the -th layer, represents the neighbor set of the node, is the information aggregation function, is the activation function.

4. The intelligent optimization and resource dynamic reallocation method based on a graph neural network according to claim 1, wherein In the steps of congestion detection and prediction, based on the node and edge embeddings generated by the graph neural network, predict potential congestion regions and their occurrence probabilities , calculate the congestion threshold for each edge , where the congestion prediction formula is: Among them, and are model parameters, is the and embedded feature of the edge between nodes; When a certain link congestion is detected, judge the impact of the congestion on surrounding links through the improved cascade failure rule to prevent the chain failure caused by single-point congestion. The specific improvement rule is: Among them, represents the cascading risk coefficient of the edge , is the weight coefficient, is the th transmission node, is the set of the remaining nodes excluding the node , , is the data traffic volume transmitted from the node to the node , represents the congestion threshold of the transmission edge from the node to the node , is the data traffic volume transmitted from the node to the node , judges whether it is necessary to dynamically reallocate the traffic between the transmission node pairs through the cascading risk coefficient; if exceeds the set threshold, it is regarded as having a risk of chain congestion.

5. The intelligent optimization and resource dynamic reallocation method based on a graph neural network according to claim 1, wherein In the steps of the data flow reallocation and intelligent optimization, in order to balance the reallocation of the data flow in the network, first, the probability of optimizing and adjusting the traffic is determined by using the principle of maximum entropy to ensure that the traffic is evenly distributed on each path and improve the overall network efficiency, and the traffic allocation probability is defined satisfies the principle of entropy maximization: Optimize the result on the premise of satisfying the following constraints; traffic ratio redistribution constraint: ; maximum capacity allocation constraint: , where represents the entropy value of the network system, represents the set of edges in the network, including all edges with a transmission relationship between node pairs ; is the edge capacity, is the current traffic, obtained by using the Lagrange multiplier method can be used for traffic redistribution; Under the guidance of cascade failure judgment and the principle of maximum entropy, based on the congestion area predicted by GNN, dynamically adjust the data stream allocation, optimize network resources, and the adjusted traffic allocation rule is: Among them, is the data traffic redistributed between node and node is the original data traffic between node and node is the original data traffic between node and node is the traffic balance coefficient, represents the cascading failure impact weight, is the data flow allocation probability based on maximum entropy optimization.

6. The intelligent optimization and resource dynamic reallocation method based on a graph neural network according to claim 1, wherein In the steps of the real-time monitoring and feedback mechanism, the average transmission delay is defined as the real-time monitoring of the average delay of data packet transmission in the network, and is calculated by the following formula: Among them, represents the average transmission delay of the network, is the transmission delay from node to node and represents the number of links in the network.

7. The intelligent optimization and resource dynamic reallocation method based on graph neural network according to claim 1, characterized in that, In the steps of the real-time monitoring and feedback mechanism, the bandwidth utilization rate is defined as the monitoring of the bandwidth occupancy of each link, and is calculated by the following formula: Among them, is the bandwidth utilization rate from node to node is the data traffic between the current node pairs, is the maximum bandwidth capacity of the link; when the bandwidth utilization rate is higher than the set threshold, the system will trigger traffic reallocation.

8. The intelligent optimization and resource dynamic reallocation method based on a graph neural network according to claim 1, characterized in that In the steps of the real-time monitoring and feedback mechanism, the traffic balance is defined as the calculation of the traffic variance between different links, and is calculated by the following formula: Among them, represents the traffic balance of the network, is the bandwidth utilization rate from node to node , represents the number of links in the network, is the average bandwidth utilization rate of all links. If the degree of traffic imbalance exceeds the set threshold, the system starts the traffic optimization module; when the risk of cascading failure or key performance indicators exceed the set threshold is detected, the real-time adjustment mechanism is triggered.

9. An intelligent optimization and resource dynamic reallocation system based on a graph neural network, for implementing the intelligent optimization and resource dynamic reallocation method based on a graph neural network according to any one of claims 1 to 8, characterized in that, The intelligent optimization and resource dynamic reallocation system based on the graph neural network includes: A data collection and modeling module, which is used to collect the historical data stream characteristics of nodes and links in the network, construct a graph neural network model for data stream congestion detection. Nodes represent servers or routers in the network, edges represent links, and traffic characteristics are attached as attributes to the graph structure. The historical data stream characteristics include the number of transmitted packets, bytes, transmission rate, and protocol type information; A network topology feature extraction module, which is used to analyze the historical data stream features, extract the feature vectors of each node and link through a multi-layer graph convolutional neural network structure, and capture the dynamic changes of the network structure and traffic features; A congestion detection and prediction module, which is used to train a graph neural network model to identify the nodes and links that may experience congestion in the network, predict potential congestion areas based on cascade failure detection rules and network traffic features, and dynamically adjust the prediction accuracy of the graph neural network training model by calculating the bandwidth utilization rate of nodes and links and the traffic variance index; A data stream reallocation and intelligent optimization module, which is used to dynamically schedule the data stream based on an intelligent reallocation strategy when congestion risk or actual congestion is detected; adopt the principle of maximum entropy to reallocate the traffic to make the traffic load more balanced, specifically by calculating the optimized traffic allocation probability to transfer the traffic of congested links; A real-time monitoring and feedback mechanism module, which is used to monitor the optimized network in real time. When the performance indicators exceed the set thresholds, the feedback mechanism is used to trigger the dynamic adjustment of the data stream and automatically reallocate the traffic to ensure the efficient allocation of network resources and path optimization. The performance indicators include average transmission delay, bandwidth utilization rate, and traffic balance.

Images