A network path optimization method and device, electronic equipment and storage medium

Abstract

This invention relates to the field of network routing optimization technology, specifically to a network path optimization method, apparatus, electronic device, and storage medium. In this invention, by synchronously monitoring bit rate and packet rate load, and using the higher of their utilization rates as the criterion for link congestion judgment, the limitations of traditional single-indicator evaluation are effectively overcome, enabling cost calculation to more comprehensively and accurately reflect real-time link congestion risk. Through a cost coefficient mapping mechanism, and combined with a baseline link cost to determine dynamic link costs, real-time and accurate input is provided to the path optimization algorithm, resulting in significant optimization of the overall network in terms of throughput and transmission latency. This is particularly suitable for high-reliability industrial scenarios sensitive to jitter.

Description

Technical Field

[0001] This invention relates to the field of network routing optimization technology, specifically to a network path optimization method, apparatus, electronic device, and storage medium. Background Technology

[0002] Network path optimization refers to the process of rationally planning data transmission paths in complex network topologies to balance network load, reduce transmission latency, and improve packet delivery rate, thereby ensuring network service quality. It is a core research direction in the field of network communication. With the rapid development of industrial 5G deterministic networks, this technology is a crucial foundation supporting industrial automation, intelligent manufacturing, remote equipment control, machine vision quality inspection, and other industrial scenarios. As the core communication carrier of the Industrial Internet, industrial 5G deterministic networks aim to achieve millisecond-level end-to-end latency, microsecond-level latency jitter, and near 100% packet delivery rate through technologies such as time synchronization, resource reservation, and deterministic scheduling. This meets the high-reliability transmission requirements of equipment collaborative control, real-time data acquisition, and remote precision operation in industrial scenarios. The path optimization capability directly determines the effectiveness of these deterministic indicators. Insufficient network path optimization capabilities not only lead to a decline in the quality of ordinary network services but also cause serious problems in industrial 5G deterministic network scenarios, such as equipment control malfunctions, production process interruptions, and increased safety accident risks, resulting in huge economic losses and adverse effects.

[0003] Currently, network path optimization is mainly implemented in two ways: one is based on traditional routing protocols, and the other is based on topology construction optimization using classic graph theory models (such as Steiner trees). Regarding traditional routing protocols, AOMDV (Ad-hoc On-demand Multipath Distance Vector), AODV (Ad-hoc On-demand Plane Distance Vector), and OLSR (Optimized Link State Routing) are mainstream solutions. These protocols can provide certain QoS (Quality of Service) performance in terms of redundancy, latency control, and packet loss rate protection, and improve network throughput through path planning. However, they lack targeted optimization for traffic aggregation, and the end-to-end path selection mechanism does not consider the merging and transmission needs of multi-source data. This makes them unsuitable for the traffic aggregation scenarios of multiple devices and multiple tasks in the Internet of Things (IoT), and also difficult to meet the high-concurrency traffic aggregation and low-latency transmission requirements of multi-device collaboration in industrial 5G deterministic networks. Regarding the Steiner model, it provides a theoretical basis for efficient aggregation topology construction by introducing Steiner points to build a minimum cost tree connecting all terminal nodes. However, traditional Steiner tree-related algorithms have high time complexity, making them difficult to adapt to large-scale networks and unable to meet the stringent requirements of industrial 5G deterministic networks for real-time performance, stability, and reliability. Summary of the Invention

[0004] This invention provides a network path optimization method, apparatus, electronic device, and storage medium to solve the problem of inaccurate warning results in existing dust accumulation warning methods.

[0005] In a first aspect, the present invention provides a network path optimization method, the method comprising: A network model is constructed based on network nodes, inter-node links, and baseline performance parameters of the links, including baseline link costs. The dynamic link cost is determined based on the network status and the baseline link cost. Based on the baseline link cost, an initial Steiner tree is constructed for the set of terminal nodes, which is a set of destination nodes and multiple source nodes in the network. Based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes; Based on the dynamic link cost, calculate the first total cost between a preset number of terminal nodes connected to the candidate node and the second total cost of the path between the candidate node and the preset number of terminal nodes, and determine whether to perform path switching based on the relationship between the first total cost and the second total cost. After path switching is performed, a new Steiner tree is generated. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path switching is repeated until it is determined that no path switching should be performed, resulting in the final Steiner tree. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

[0006] This invention first utilizes real-time network status to dynamically adjust link costs, enabling path planning to proactively avoid congested links and significantly improving network resource utilization efficiency and data transmission real-time performance. Second, it introduces a pre-trained graph neural network to filter candidate nodes, transforming the combinatorial optimization problem into an efficient heuristic search. While ensuring the quality of approximate solutions, it reduces computational complexity from exponential to polynomial levels, achieving feasible deployment in large-scale network environments. Finally, through iterative optimization based on gain evaluation, only the local path exchange with the maximum benefit is performed in each iteration, ensuring stable convergence and overall load balancing during the optimization process. In summary, this method achieves real-time, intelligent, and highly adaptive optimization of network paths while meeting the low latency and high reliability requirements of industrial scenarios.

[0007] In one alternative implementation, determining the dynamic link cost based on network status and the baseline link cost includes: The bit rate load and packet rate load of each link at the current moment are determined based on the network status. The bit rate utilization and packet rate utilization are determined based on the bit rate load, packet rate load, and the baseline performance parameters, and the larger value between the bit rate utilization and packet rate utilization is taken as the link utilization. The cost coefficient is determined based on the relationship between the link utilization and the link congestion, and the dynamic link cost of each link at the current moment is determined using the cost coefficient and the baseline link cost.

[0008] This invention overcomes the limitations of traditional single-indicator evaluation by synchronously monitoring bit rate and packet rate load and using the higher of their utilization rates as the basis for link congestion judgment. This allows cost calculation to more comprehensively and accurately reflect real-time link congestion risk. Through a cost coefficient mapping mechanism and combining it with baseline link costs to determine dynamic link costs, it provides real-time and accurate input for path optimization algorithms, resulting in significant optimization of the overall network in terms of throughput and transmission latency. This is particularly suitable for high-reliability industrial scenarios sensitive to jitter.

[0009] In one alternative implementation, the relationship between link utilization and link congestion is represented by the following congestion penalty function:

[0010] In the formula, This represents the utilization penalty threshold. Indicates link utilization. Represents the stability constant. This represents the maximum penalty value.

[0011] This invention achieves a refined quantitative assessment of network link congestion by designing a congestion penalty function with clearly defined parameters. Specifically, when link utilization exceeds a preset threshold, the function significantly increases link costs in a non-linear manner, thereby objectively and sensitively reflecting real-time congestion risks. By adjusting the threshold and penalty upper limit, administrators can flexibly adapt to the sensitivity requirements of different business scenarios, providing accurate dynamic cost input for routing optimization algorithms. Ultimately, this mechanism enables the network system to proactively guide traffic to avoid high-load links, achieving preventative load balancing before resource bottlenecks occur, effectively improving overall network throughput efficiency, reducing transmission latency and packet loss risks, and enhancing service quality assurance capabilities under high-demand business scenarios.

[0012] In one optional implementation, based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes, including: Obtain the dynamic link cost and the topology in the network model at the current moment, and extract the node feature matrix; The node feature matrix is ​​processed using multi-layer graph convolution in a pre-trained graph neural network model to obtain the scores of non-terminal nodes in the network model. Based on the scoring, a preset number of candidate nodes are selected to form a candidate node set.

[0013] In this invention, node features are generated by combining dynamic link costs with network topology, and multi-layer graph convolution calculations are performed using a pre-trained graph neural network to generate accurate scores for each non-terminal node, thereby achieving intelligent screening of candidate nodes. This enables efficient and automated identification of the most promising nodes from a large number of non-terminal nodes, significantly reducing the time and computational overhead of traditional manual or exhaustive screening. Simultaneously, the deep learning capabilities of the graph neural network ensure high accuracy and adaptability of the screening results, providing a reliable and high-quality set of basic nodes for subsequent network optimization (such as path planning and resource allocation), thus improving the intelligence and efficiency of overall network analysis and decision-making.

[0014] In one optional implementation, the candidate nodes include multiple nodes. Based on the dynamic link cost, a first total cost between a preset number of terminal nodes connected to the candidate nodes and a second total cost of the path between the candidate nodes and the preset number of terminal nodes are calculated, including: Based on dynamic link cost, determine the terminal nodes that meet the preset conditions connected to each candidate node; Determine a preset number of original paths in the initial Steiner tree for any number of terminal nodes that meet the preset conditions, and calculate the first total cost of removing the original paths based on dynamic link cost. The second total cost is determined based on the dynamic link cost to establish a star-shaped path connecting any candidate node to any preset number of terminal nodes.

[0015] This invention achieves precise quantitative evaluation of network optimization schemes by systematically comparing the first total cost of removing existing connection paths with the second total cost of establishing star connection paths. This allows for a scientific and objective judgment based on real-time network conditions (dynamic link costs) on whether introducing specific relay nodes truly yields global benefits, thus avoiding performance degradation or oscillations caused by blind adjustments. This mechanism ensures that every path optimization decision is based on a clear cost-benefit analysis, significantly improving the accuracy and reliability of network reconstruction, and ultimately guiding the network topology to converge to a truly efficient, stable, and load-balanced state, enhancing the overall network adaptability and quality of service.

[0016] In one alternative implementation, determining whether to perform a path swap based on the relationship between the first total cost and the second total cost includes: The gain of path switching is determined based on the difference between the first total cost and the second total cost; The maximum gain when any candidate node is connected to any preset number of terminal nodes; When the maximum value is greater than the gain threshold, the original path is removed, a star path is replaced, and a new Steiner tree is generated.

[0017] This invention introduces gain calculation based on cost differences to ensure that each path switching decision is based on objective performance improvement expectations, effectively avoiding ineffective or negative network adjustments. By selecting the maximum value as the switching criterion, it guarantees that each iteration executes the optimization operation with the highest benefit in the current state, enabling network resources to quickly converge towards an efficient configuration. Combined with an automatic execution mechanism based on threshold determination, it improves optimization efficiency while maintaining network stability. Ultimately, it systematically reduces the total path cost of data transmission, enhancing overall network performance and agility in adapting to load changes.

[0018] In one optional implementation, the bit rate utilization and packet rate utilization are determined using the following formulas:

[0019]

[0020] In the formula, Indicates bit rate utilization. Indicates bit rate load. This refers to the bandwidth capacity in the baseline performance parameters. Indicates packet rate utilization. Indicates packet rate load. This indicates the packet processing capability in the baseline performance parameters.

[0021] In this invention, real-time load is first directly correlated with the link's baseline performance parameters (bandwidth capacity, packet processing capability), which can accurately and unambiguously reflect the actual resource utilization ratio of the link, providing a unified and objective quantitative basis for subsequent network status analysis. Secondly, by simultaneously monitoring two utilization rates, the link bottleneck (whether it is bandwidth-limited or processing capacity-limited) can be more comprehensively identified, thereby providing key data support for refined network optimization and resource scheduling. Ultimately, this helps to improve the utilization efficiency of network resources and ensure the stability and service quality of high-requirement data transmission.

[0022] In a second aspect, the present invention provides a network path optimization device, the device comprising: The network model building module is used to build a network model based on network nodes, inter-node links, and baseline performance parameters of the links, including baseline link costs. A cost determination module is used to determine dynamic link costs based on network status and the baseline link cost. The Steiner tree construction module is used to construct an initial Steiner tree for a set of terminal nodes based on the baseline link cost, wherein the set of terminal nodes is a set of destination nodes and multiple source nodes in the network nodes. The node filtering module is used to filter candidate nodes among non-terminal nodes based on the dynamic link cost using a pre-trained graph neural network model. The path switching module is used to calculate, based on the dynamic link cost, a first total cost between a preset number of terminal nodes connected to the candidate node and a second total cost of the path between the candidate node and the preset number of terminal nodes, and to determine whether to perform path switching based on the relationship between the first total cost and the second total cost. The iterative optimization module is used to generate a new Steiner tree after path swapping. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path swapping is repeated until it is determined that no path swapping should be performed, resulting in the final Steiner tree. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

[0023] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the network path optimization method described in the first aspect or any corresponding embodiment thereof.

[0024] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the network path optimization method described in the first aspect or any corresponding embodiment thereof.

[0025] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the network path optimization method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the first type of network path optimization method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second process for a network path optimization method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the dynamic link cost calculation logic according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the Steiner tree iterative optimization process according to an embodiment of the present invention; Figure 5 This is an overview diagram of the application scenarios and computing workflow according to embodiments of the present invention; Figure 6 This is a structural block diagram of a network path optimization device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0030] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0031] According to an embodiment of the present invention, a network path optimization method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] This embodiment provides a network path optimization method. Figure 1 This is a flowchart of a network path optimization method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Construct a network model based on network nodes, inter-node links, and baseline performance parameters of the links, wherein the baseline performance parameters include baseline link cost.

[0033] Specifically, the network path optimization in this embodiment is geared towards industrial 5G deterministic networks, meaning the network covers the entire end-to-end path from industrial terminal equipment, through the 5G network, to the factory intranet. For example, this network includes nodes such as IoT devices, terminals, servers, and gateways. Based on this, network nodes can be determined according to the specific application scenario of the network. The links between nodes include the links between two adjacent nodes, so the network can be abstracted as a graph structure. ,in, For a set of nodes, For a set of links.

[0034] At the same time, in order to measure the performance of the links, for each link Define baseline performance parameters, including bandwidth capacity. Packet processing capability Baseline link cost Basic propagation delay Base packet loss rate These are parameters. It should be noted that these baseline performance parameters can be understood as the inherent physical properties of the link in a no-load (idle) state. Therefore, when the network state changes, these baseline performance parameters serve as a dynamically changing reference anchor point.

[0035] Step S102: Determine the dynamic link cost based on the network status and the baseline link cost. Since the network load may change over time, this embodiment monitors the network status in real time to obtain the network load at different times. Combined with the baseline link cost (i.e., the link cost when the network is unloaded), the dynamic link cost under different loads at different times is obtained. Specifically, when determining the dynamic link cost, a cost coefficient can be determined based on the obtained load information. Multiplying this cost coefficient by the baseline link cost yields the dynamic link cost.

[0036] Step S103: Based on the baseline link cost, construct an initial Steiner tree for the set of terminal nodes, wherein the set of terminal nodes is a set of destination nodes and multiple source nodes in the network.

[0037] In this embodiment, network path optimization refers to optimizing the data transmission path from the source node to the target node. To be suitable for large-scale network environments requiring efficient multi-source data aggregation, such as the Internet of Things (IoT), data center networks, and the Industrial Internet, multiple source nodes are typically included, meaning that multi-source data from multiple source nodes is transmitted to the same target node. Thus, a set of multiple source nodes constitutes a source node set. and the target node for transmission A set of terminal nodes constitutes a terminal node. Represented as .

[0038] Steiner tree is a classic problem in combinatorial optimization and graph theory, aiming to find the minimum-weighted tree connecting a specified set of vertices (called terminal nodes) in a given graph, allowing the use of non-terminal nodes as intermediate nodes (called Steiner points) to reduce the total weight. This problem is NP-hard. To solve it, various approximation algorithms have been developed, such as the Minimum Spanning Tree (MST) approximation (also known as the 2-approximation algorithm), the 1.55-approximation algorithm, and linear programming rounding. The 2-approximation algorithm specifically involves calculating the metric closure (edge ​​weights equal to the shortest path distance) of the terminal set S in the complete graph; calculating the minimum spanning tree on the metric closure; and replacing the edges in the MST with the shortest paths in the original graph.

[0039] However, Steiner tree algorithms have high time complexity, making them difficult to adapt to large-scale networks. Furthermore, they do not consider dynamic network load changes. In industrial scenarios, device startup / shutdown and task switching often cause sudden increases and decreases in link load, making tree topologies built using traditional algorithms prone to failure due to load fluctuations. This fails to meet the stringent requirements of real-time performance, stability, and reliability in industrial 5G deterministic networks. Therefore, this embodiment introduces link costs when constructing the Steiner tree to adapt to large-scale networks and load fluctuations. Specifically, during construction, a baseline link cost is first used... The 2-approximation algorithm is used to define the set of terminal nodes. Construct the initial Steiner tree Then, the initial Steiner tree is optimized using dynamic link cost to obtain the optimized path.

[0040] Specifically, when constructing the initial Steiner tree, the cost of any two terminal nodes in the set of terminal nodes is first calculated based on the baseline link cost. The path with the lowest cumulative cost (i.e. Passing through all links The path with the smallest sum is also called the shortest path. ), and thereby construct a metric closure graph of the terminal node set. The metric closure graph can be understood as a simplified abstraction of the original network, where its nodes only contain terminal nodes, and a virtual transition connects any two terminal nodes. The weight of each edge represents the cost of the shortest path. Then, Kruskal's algorithm is used to calculate the minimum spanning tree of the metric closure graph. A spanning tree is a connected subgraph containing all nodes in the graph and without cycles, while a minimum spanning tree is... The goal is to minimize the total weight connecting all terminal nodes. Finally, the edges in the minimum spanning tree are mapped back to the paths in the original network topology graph to obtain the initial Steiner tree. The paths from u to v in the minimum spanning tree are then replaced with the paths from u to v in the original network topology graph.

[0041] Step S104: Based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes. Specifically, after obtaining the dynamic link cost at each time step, it and the node topology are input into the pre-trained graph neural network model. Then, the model is used to filter non-terminal nodes to obtain candidate nodes that meet the conditions. The selected candidate nodes may include multiple nodes.

[0042] Step S105: Based on the dynamic link cost, calculate the first total cost between a preset number of terminal nodes connected to the candidate node and the second total cost of the path between the candidate node and the preset number of terminal nodes, and determine whether to perform path switching based on the relationship between the first total cost and the second total cost.

[0043] Specifically, after determining the candidate nodes, the terminal nodes connected to each candidate node are filtered based on dynamic link cost. Then, a predetermined number of terminal nodes from the filtered terminal nodes are grouped together, and the first total cost of the connection path (original path) in the Steiner tree for each group of terminal nodes is calculated. At the same time, the second total cost of the path (new path) when connecting the candidate node to each terminal node in each group of terminal nodes is calculated. By comparing the first total cost and the second total cost, it is determined whether to replace the original path with the new path.

[0044] Step S106: After path switching is performed, a new Steiner tree is generated. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path switching is repeated until it is determined that no path switching will be performed, and the final Steiner tree is obtained. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

[0045] Specifically, when it is determined that path switching is required, a new Steiner tree is generated. Then, based on the new Steiner tree, steps S102 and S105 are repeated until it is determined that path switching is not required, and the final Steiner tree is obtained. Non-terminal nodes in the tree are removed to obtain the optimized data aggregation path ST, which is then sent to the network device for execution.

[0046] This embodiment provides a network path optimization method, which includes the following steps: Step S201: Construct a network model based on network nodes, inter-node links, and baseline performance parameters of the links. These baseline performance parameters include baseline link costs. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0047] Step S202: Determine the dynamic link cost based on the network status and the baseline link cost.

[0048] Specifically, step S202 includes: Step S2021: Determine the bit rate load and packet rate load of each link at the current moment based on the network status; wherein, when monitoring the network status, the total amount of data (number of bits) and the total number of data packets (number of packets) transmitted on each link per unit time are used to obtain the data for each link. At any moment bit rate load and packet rate load .

[0049] Step S2022: Determine the bit rate utilization and packet rate utilization based on the bit rate load, packet rate load, and the baseline performance parameters, and take the larger value of the bit rate utilization and packet rate utilization as the link utilization; specifically, the bit rate utilization and packet rate utilization are determined using the following formulas:

[0050]

[0051] In the formula, Indicates bit rate utilization. Indicates bit rate load. This refers to the bandwidth capacity in the baseline performance parameters. Indicates packet rate utilization. Indicates packet rate load. This indicates the packet processing capability in the baseline performance parameters.

[0052] After determining the bit rate utilization and packet rate utilization, the maximum value between them is taken as the link utilization (also known as the critical utilization factor), i.e., the critical utilization factor. Therefore, it can be ensured that penalties are triggered when either bit rate utilization or packet rate utilization becomes a bottleneck.

[0053] Step S2023: Determine the cost coefficient based on the relationship between link utilization and link congestion, and use the cost coefficient and the baseline link cost to determine the dynamic link cost of each link at the current moment. Specifically, the relationship between link utilization and link congestion is represented by the following congestion penalty function:

[0054] In the formula, This represents the utilization penalty threshold. Indicates link utilization. Represents the stability constant. This represents the maximum penalty value. The utilization penalty threshold and the maximum penalty value can be determined based on actual circumstances; for example, the utilization penalty threshold could be 0.7, and the maximum penalty value 10.

[0055] The cost of dynamic links is expressed by the following formula:

[0056] In the formula, Weighting factors representing benchmark costs ， This represents the scaling factor for congestion penalty, and its specific value can be determined based on the actual situation.

[0057] Based on the congestion penalty function and dynamic link cost calculation formula mentioned above, it can be seen that when the utilization rate is lower than the utilization penalty threshold, =0, at which point the dynamic link cost is 0. This means that the dynamic link cost is a fixed value close to the baseline link cost, thus providing a buffer for network load. When the utilization rate exceeds the utilization penalty threshold, the congestion penalty function is equivalent to... As utilization increases, Gradually increase until the maximum penalty value is reached. Therefore, according to the formula above, in As the dynamic link cost increases, the level of network congestion avoidance can be adjusted based on this dynamic link cost.

[0058] Step S203: Based on the baseline link cost, construct an initial Steiner tree for the set of terminal nodes, where the set of terminal nodes is a collection of destination nodes and multiple source nodes in the network; for details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0059] Step S204: Based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes.

[0060] Specifically, step S204 includes: Step S2041: Obtain the dynamic link cost at the current moment and the topology in the network model, and extract the node feature matrix; specifically, the topology includes the connection relationship between each network node in the network model, wherein the dynamic link cost can be used as an attribute of the connection relationship, and thus the connection relationship and the dynamic link cost together form the node feature matrix X.

[0061] Step S2042: The node feature matrix is ​​processed using multi-layer graph convolution in the pre-trained graph neural network model to obtain the scores of non-terminal nodes in the network model; specifically, the pre-trained graph neural network model includes multi-layer graph convolutional layers, wherein the node representation update formula of the graph convolutional layer is:

[0062] In the formula, Indicates the node at the 1st Layer representation (specifically represented by node feature matrix). Indicates the first The learnable parameters of the layer Represents a node The set of neighboring nodes, MLP represents a multilayer perceptron.

[0063] Based on the processing of multiple graph convolutional layers, the final output layer of the model outputs the scores of non-terminal nodes. That is, the score... It is expressed by the following formula:

[0064] In the formula, Represents the weight parameters of the output layer. This represents the bias parameters of the output layer.

[0065] Step S2043: Based on the scores, a preset number of candidate nodes are selected to form a candidate node set. Specifically, after determining the scores of all non-terminal nodes, the scores are sorted from high to low, and then the nodes with the highest scores are selected. (Right now A set of candidate nodes is formed by selecting 10 non-terminal nodes as candidate nodes. .in, The number can be determined based on the actual situation, for example, .

[0066] Step S205: Based on the dynamic link cost, calculate the first total cost between a preset number of terminal nodes connected to the candidate node and the second total cost of the path between the candidate node and the preset number of terminal nodes, and determine whether to perform path switching based on the relationship between the first total cost and the second total cost.

[0067] Specifically, step S205 includes: Step S2051: Based on dynamic link cost, determine the terminal nodes that meet the preset conditions connected to each candidate node; specifically, for each candidate center node... The path cost of each connected terminal node is determined based on dynamic link cost. Then, the path costs of each terminal node are sorted, and the top-ranked terminal nodes are selected as terminal nodes that meet preset conditions. For example, in this embodiment, m=5 terminal nodes are selected for each candidate node. .

[0068] Step S2052: Determine multiple original paths in the initial Steiner tree for any preset number of terminal nodes that meet preset conditions, and calculate the first total cost of removing the original paths based on dynamic link cost; specifically, for each candidate node selected by the terminal node From these, a predetermined number of terminal nodes are randomly selected as a group. For example, if 3 terminal nodes are selected from 5 terminal nodes of each candidate node, 10 combinations of terminal nodes are obtained. Each combination is a triplet. And each triplet .

[0069] For each triple of each candidate node, find the original path connecting the three terminal nodes in the current Steiner tree, and calculate the first total cost of removing the original path based on the dynamic link cost. This first total cost is calculated using the following formula:

[0070] Therefore, multiple first total costs can be calculated for each candidate node. For example, if there are 5 candidate nodes, and 3 terminal nodes are selected from the 5 terminal nodes of each candidate node, then 50 first total costs can be obtained in the end.

[0071] Step S2053: Determine the second total cost of a star-shaped path connecting any candidate node to any preset number of terminal nodes based on the dynamic link cost. Specifically, as analyzed above, each candidate node has multiple triples. For each triple of each candidate node, a direct connection is constructed from the candidate node as the center. The star topology is calculated, and the second total cost of the star topology consisting of three directly connected links is calculated based on the sum of dynamic link costs. Thus, when there are 5 candidate nodes, 50 second total costs can also be calculated.

[0072] Step S2054: Determine the gain of path switching based on the difference between the first total cost and the second total cost; specifically, when multiple first total costs are calculated... Second total cost Next, the difference between the first total cost and the corresponding second total cost of the same candidate node and its corresponding triple is calculated to obtain the path swap gain. From this, 50 gains can also be determined. This gain is specifically calculated using the following formula:

[0073] Step S2055: Select the maximum gain when any candidate node is connected to any preset number of terminal nodes; specifically, after obtaining multiple gains, determine the maximum value among the multiple gains, i.e., the maximum gain. .

[0074] Step S2056: When the maximum value is greater than the gain threshold, the original path is removed, a new star-shaped path is generated, and a new Steiner tree is created. Specifically, the maximum gain calculated for the selected candidate node set is compared with the gain threshold. If the maximum gain is greater than the gain threshold, a path swap is performed, i.e., the original path is cut off, and a new star-shaped connection is enabled. The current Steiner tree is then updated. .

[0075] Step S206: After path switching is performed, a new Steiner tree is generated. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path switching is repeated until it is determined that no path switching will be performed, and the final Steiner tree is obtained. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

[0076] Specifically, after updating the current Steiner tree to obtain a new Steiner tree, the network status is monitored to obtain dynamic link costs. Then, the candidate node set is re-selected based on the dynamic link costs, and the corresponding maximum gain is calculated. A comparison between the maximum gain and a gain threshold determines whether to switch paths. If a path switching is determined and a new Steiner tree is generated, the above process continues until it is determined that no further path switching is necessary. The Steiner tree at this point is taken as the final Steiner tree, and non-terminal nodes are removed from the final Steiner tree to obtain the optimized path.

[0077] As one or more specific application embodiments of the present invention, such as Figure 2 As shown, this network path optimization method is implemented using the following process: Step (1): Construct a network system model and abstract the network into a graph structure. ,in For a set of nodes, This is a set of links, where a link refers to a connection between two adjacent nodes. Each link... Define baseline performance parameters, including bandwidth capacity. Packet processing capability Baseline link cost Basic propagation delay Base packet loss rate .

[0078] Step (2): Define the communication task Determine its source node set , destination node and terminal node set . Step (3): Monitor network status in real time and calculate the status of each link. At any moment bit rate load and packet rate load Calculate link utilization:

[0079]

[0080] Determine the critical utilization factor:

[0081] Step (4): As Figure 3 As shown, dynamic link cost is calculated based on real-time load:

[0082] in, It is a weighting factor for the benchmark cost. It is the scaling factor for the congestion penalty. For the congestion penalty function:

[0083] The formula parameters are set as follows , , .

[0084] Step (5): As Figure 4 As shown, the baseline cost based on the link. The 2-approximation algorithm is used to define the set of terminal nodes. Construct the initial Steiner tree Computing between terminal nodes based on shortest path Construct the metric closure graph of the terminal node set. Use Kruskal's algorithm to compute the minimum spanning tree of the metric closure graph. Finally, the edges in the minimum spanning tree are mapped back to the paths in the original network topology graph to obtain the initial Steiner tree.

[0085] Step (6): As Figure 4 As shown, intelligent selection of candidate Steiner points based on GNN utilizes a pre-trained graph neural network model. Filtering high-quality candidate nodes: In each iteration of the algorithm, the system first monitors the network load in real time and calculates the dynamic link cost of all links using the formula in step (4). Then, these dynamic cost data are combined with the node connections, i.e., the topology, to assemble a node feature matrix. .

[0086] Calculate the centrality score of non-terminal nodes using a GNN model:

[0087] in Represents the weight parameters of the output layer. This represents the bias parameters of the output layer; the one with the highest score is selected. Each node is a set of candidate center nodes. ,in .

[0088] Step (7): As Figure 4 As shown, the Steiner tree is iteratively optimized. Entering the main optimization loop, for each candidate center node... : turn up The nearest in the current Steiner tree terminal nodes "Nearest" refers to the path with the lowest cost.

[0089] Triple group This refers to any combination of three terminal nodes from the previously selected m terminal nodes. For each triplet... .

[0090] In the current tree, find the connection between these three terminal nodes. Given the original paths, calculate the total cost of removing these paths:

[0091] Introducing candidate points As a new transit hub, building from Direct connection respectively Given a star-shaped path, calculate its total cost:

[0092] Calculate the gain of path switching:

[0093] The maximum gain is set to ,if If the path is not found, a path swap is performed, cutting off the original path and enabling a new star connection. The current Steiner tree is then updated. .

[0094] Record maximum gain and the corresponding optimal swapping scheme .

[0095] Step (8): As Figure 4 As shown, convergence is determined and the final output is given. The steps are repeated. To the steps until no path swap with significant gain can be found in consecutive iterations. The final Steiner tree is pruned by removing all leaf nodes of non-terminal nodes to obtain an optimized data aggregation path. And then distribute it to the network devices for execution.

[0096] In summary, this invention constructs a dynamic load-aware network model, defines multi-source data aggregation tasks, monitors link load status in real time, and quantifies link congestion using a dynamic cost function. Figure 5 The diagram illustrates a physical network scenario where this method is applied. IoT devices and industrial terminals communicate with the server via a gateway, visually demonstrating a typical industrial application model where multi-source data (from multiple devices) converges to a single aggregation point (the server). In this physical scenario, the core algorithm engine DLSAG (Dynamic Load-aware Steiner tree Approximation and Guidance algorithm) is used for path optimization: First, the system abstracts the received communication tasks into a graph model and receives communication task requests. Then, an initial Steiner tree is generated based on the baseline link cost. Subsequently, the core iterative optimization loop is entered. In this loop, the algorithm monitors the network load in real time and calculates the dynamic link cost accordingly. Simultaneously, a pre-trained graph neural network (GNN) is used to intelligently select potential candidate Steiner points. Next, the gain brought by introducing candidate points for path swapping is evaluated. If the convergence condition is not met (i.e., a significant gain exists), the optimal path swap is executed to update the current tree, and a new round of optimization is initiated. Finally, once the tree structure is stable, the algorithm outputs an optimized Steiner tree and distributes it to the network deployment, thereby achieving a load-balanced data transmission path in the physical network.

[0097] This solution combines dynamic load awareness, Steiner tree theory, and neural network heuristic search, significantly improving the efficiency and stability of multi-source data aggregation in large-scale networks. It effectively reduces maximum link utilization and increases packet delivery rate, providing an efficient solution for traffic optimization in large-scale networks such as the Internet of Things and data centers, and has significant practical application value.

[0098] This embodiment also provides a network path optimization device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0099] This embodiment provides a network path optimization device, such as... Figure 6 As shown, it includes: The network model construction module 61 is used to construct a network model based on network nodes, inter-node links, and baseline performance parameters of the links, wherein the baseline performance parameters include baseline link cost. Cost determination module 62 is used to determine dynamic link cost based on network status and the baseline link cost; Steiner tree construction module 63 is used to construct an initial Steiner tree for a set of terminal nodes based on the baseline link cost, wherein the set of terminal nodes is a set of destination nodes and multiple source nodes in the network nodes. The node filtering module 64 is used to filter candidate nodes among non-terminal nodes based on the dynamic link cost using a pre-trained graph neural network model. The path switching module 65 is used to calculate, based on the dynamic link cost, a first total cost between a preset number of terminal nodes connected to the candidate node and a second total cost of the path between the candidate node and the preset number of terminal nodes, and to determine whether to perform path switching based on the relationship between the first total cost and the second total cost. The iterative optimization module 66 is used to generate a new Steiner tree after path switching. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path switching is repeated until it is determined that no path switching is to be performed, and the final Steiner tree is obtained. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

[0100] The network path optimization apparatus provided in this embodiment of the invention can execute the network path optimization method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0101] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0102] The following is a detailed reference. Figure 7 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 11, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 12 or a program loaded from memory 18 into random access memory (RAM) 13. The RAM 13 also stores various programs and data required for the operation of the electronic device. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0103] Typically, the following devices can be connected to I / O interface 15: input devices 16 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 17 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 18 including, for example, magnetic tapes, hard disks, etc.; and communication devices 19. Communication device 19 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0104] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 19, or installed from a memory 18, or installed from a ROM 12. When the computer program is executed by the processor 11, it performs the functions defined in the network path optimization method of the embodiments of the present invention.

[0105] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0106] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the network path optimization method shown in the above embodiments is implemented.

[0107] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0108] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A network path optimization method, characterized in that, The method includes: A network model is constructed based on network nodes, inter-node links, and baseline performance parameters of the links, including baseline link costs. The dynamic link cost is determined based on the network status and the baseline link cost. Based on the baseline link cost, an initial Steiner tree is constructed for the set of terminal nodes, which is a set of destination nodes and multiple source nodes in the network. Based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes; Based on the dynamic link cost, calculate the first total cost between a preset number of terminal nodes connected to the candidate node and the second total cost of the path between the candidate node and the preset number of terminal nodes, and determine whether to perform path switching based on the relationship between the first total cost and the second total cost. After path switching is performed, a new Steiner tree is generated. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path switching is repeated until it is determined that no path switching should be performed, resulting in the final Steiner tree. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

2. The method according to claim 1, characterized in that, Determining dynamic link costs based on network status and the baseline link cost includes: The bit rate load and packet rate load of each link at the current moment are determined based on the network status. The bit rate utilization and packet rate utilization are determined based on the bit rate load, packet rate load, and the baseline performance parameters, and the larger value between the bit rate utilization and packet rate utilization is taken as the link utilization. The cost coefficient is determined based on the relationship between the link utilization and the link congestion, and the dynamic link cost of each link at the current moment is determined using the cost coefficient and the baseline link cost.

3. The method according to claim 2, characterized in that, The relationship between link utilization and link congestion is represented by the following congestion penalty function: In the formula, This represents the utilization penalty threshold. Indicates link utilization. Represents the stability constant. This represents the maximum penalty value.

4. The method according to claim 1, characterized in that, Based on the dynamic link cost, a pre-trained graph neural network model is used to filter candidate nodes among non-terminal nodes, including: Obtain the dynamic link cost and the topology in the network model at the current moment, and extract the node feature matrix; The node feature matrix is ​​processed using multi-layer graph convolution in a pre-trained graph neural network model to obtain the scores of non-terminal nodes in the network model. Based on the scoring, a preset number of candidate nodes are selected to form a candidate node set.

5. The method according to claim 1, characterized in that, The candidate nodes include multiple nodes. Based on the dynamic link cost, a first total cost between a preset number of terminal nodes connected to the candidate nodes and a second total cost of the path between the candidate nodes and the preset number of terminal nodes are calculated, including: Based on dynamic link cost, determine the terminal nodes that meet the preset conditions connected to each candidate node; Determine a preset number of original paths in the initial Steiner tree for any number of terminal nodes that meet the preset conditions, and calculate the first total cost of removing the original paths based on dynamic link cost. The second total cost is determined based on the dynamic link cost to establish a star-shaped path connecting any candidate node to any preset number of terminal nodes.

6. The method according to claim 5, characterized in that, Determining whether to perform a path swap based on the relationship between the first total cost and the second total cost includes: The gain of path switching is determined based on the difference between the first total cost and the second total cost; The maximum gain when any candidate node is connected to any preset number of terminal nodes; When the maximum value is greater than the gain threshold, the original path is removed, a star path is replaced, and a new Steiner tree is generated.

7. The method according to claim 2, characterized in that, Bit rate utilization and packet rate utilization are determined using the following formulas: In the formula, Indicates bit rate utilization. Indicates bit rate load. This refers to the bandwidth capacity in the baseline performance parameters. Indicates packet rate utilization. Indicates packet rate load. This indicates the packet processing capability in the baseline performance parameters.

8. A network path optimization device, characterized in that, The device includes: The network model building module is used to build a network model based on network nodes, inter-node links, and baseline performance parameters of the links, including baseline link costs. A cost determination module is used to determine dynamic link costs based on network status and the baseline link cost. The Steiner tree construction module is used to construct an initial Steiner tree for a set of terminal nodes based on the baseline link cost, wherein the set of terminal nodes is a set of destination nodes and multiple source nodes in the network nodes. The node filtering module is used to filter candidate nodes among non-terminal nodes based on the dynamic link cost using a pre-trained graph neural network model. The path switching module is used to calculate, based on the dynamic link cost, a first total cost between a preset number of terminal nodes connected to the candidate node and a second total cost of the path between the candidate node and the preset number of terminal nodes, and to determine whether to perform path switching based on the relationship between the first total cost and the second total cost. The iterative optimization module is used to generate a new Steiner tree after path swapping. Based on the new Steiner tree, the process of determining dynamic link cost, filtering nodes, and path swapping is repeated until it is determined that no path swapping should be performed, resulting in the final Steiner tree. Non-terminal nodes in the final Steiner tree are removed to obtain the optimized path.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the network path optimization method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the network path optimization method according to any one of claims 1 to 7.

Images