Intelligent flow management method and device based on switch test system

Abstract

The invention provides an intelligent flow management method and device based on a switch test system, and the method comprises the steps: carrying out the detection of a switch through the switch test system, and obtaining a first data package and a second data package; performing time-space joint coding on the first data packet to obtain a synchronous topological graph; inputting the synchronous topological graph into a preset space-time diagram convolutional network to obtain a dynamic path strategy library; performing packaging and simulation verification on the dynamic path strategy library to obtain a formalized verification strategy packet; determining a consistency strategy instruction according to the second data packet, the synchronous topological graph and the formalized verification strategy packet; and performing cross-domain strategy synchronization processing on the form consistency strategy instruction to generate a management strategy instruction for the switch.

Description

Technical Field

[0001] This application relates to the field of communication technologies, and in particular, to an intelligent traffic management method and device based on a switch test system. Background Art

[0002] Currently, in the time-sensitive network (TSN) of smart factories, multi-modal services such as industrial control flows and video surveillance flows coexist. The existing technologies have the following defects in dealing with path anomalies: The traditional centralized control architecture relies on global topology updates and needs to recalculate the whole network policy when a link fails, resulting in a decision-making delay exceeding the gPTP synchronization period threshold (typical value > 50 ms), which cannot meet the requirements of high-precision motion control. Summary of the Invention

[0003] This application provides an intelligent traffic management method and device based on a switch test system for managing the time-sensitive network of switches.

[0004] In a first aspect, an embodiment of this application provides an intelligent traffic management method based on a switch test system. The method includes: Probing a switch through a switch test system to obtain a first data packet and a second data packet; Performing time-space joint encoding on the first data packet to obtain a synchronization topology graph; Inputting the synchronization topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library; Encapsulating and simulating and verifying the dynamic path policy library to obtain a formal verification policy packet; Determining a consistency policy instruction according to the second data packet, the synchronization topology graph, and the formal verification policy packet.

[0005] In a second aspect, an embodiment of this application provides an intelligent traffic management device based on a switch test system. The device includes: A data probing module, configured to probe a switch through a switch test system to obtain a first data packet and a second data packet; A topology generation module, configured to perform time-space joint encoding on the first data packet to obtain a synchronization topology graph; A policy generation module, configured to input the synchronization topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library; A policy verification module, configured to encapsulate and simulate and verify the dynamic path policy library to obtain a formal verification policy packet; An instruction generation module, configured to determine a consistency policy instruction according to the second data packet, the synchronization topology graph, and the formal verification policy packet; A policy specification module for performing cross - domain policy synchronization processing on the form - consistency policy instruction to generate a management policy instruction for the switch.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, which includes a memory and a processor; The memory is used to store a computer program; The processor is configured to execute the computer program and, when executing the computer program, implement the intelligent traffic management method based on a switch test system as described in any one of the embodiments of the present application.

[0007] In a fourth aspect, an embodiment of the present application provides a computer - readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor is caused to implement the intelligent traffic management method based on a switch test system as described in any one of the embodiments of the present application.

[0008] An embodiment of the present application provides an intelligent traffic management method based on a switch test system. The method includes: detecting a switch through a switch test system to obtain a first data packet and a second data packet; performing time - space joint encoding on the first data packet to obtain a synchronization topology graph; inputting the synchronization topology graph into a pre - set spatio - temporal graph convolutional network to obtain a dynamic path policy library; encapsulating and simulating and verifying the dynamic path policy library to obtain a formal verification policy packet; determining a consistency policy instruction according to the second data packet, the synchronization topology graph, and the formal verification policy packet; performing cross - domain policy synchronization processing on the form - consistency policy instruction to generate a management policy instruction for the switch. In the above method, the original data is encoded through a time - space joint encoding technology to obtain a synchronization topology graph; the synchronization topology graph is processed by means of a spatio - temporal graph convolutional network to effectively capture the spatio - temporal characteristics of the network topology, intelligently learn traffic patterns, increase time synchronization accuracy and reduce fault recovery time. The correctness and reliability of the policy are ensured through a formal verification mechanism, reducing implementation risks. Combining cross - domain policy synchronization processing technology ensures the accuracy of conflict resolution for policy changes, while optimizing the network resource utilization efficiency and effectively reducing the labor cost of network management. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1Schematic flowchart of an intelligent traffic management method based on a switch test system provided by an embodiment of the present application; Figure 2 Schematic block diagram of an intelligent traffic management device based on a switch test system provided by an embodiment of the present application. Detailed implementation manners

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0012] The flowchart shown in the accompanying drawings is only an example illustration, and does not necessarily include all contents and operations / steps, nor does it necessarily execute in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may be changed according to the actual situation.

[0013] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0014] It should be further understood that the term "and / or" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0015] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an intelligent traffic management method based on a switch test system provided by an embodiment of the present application. As Figure 1 shown, the specific steps of the intelligent traffic management method based on the switch test system include: S101 - S106.

[0016] S101. Detect the switch through the switch test system to obtain a first data packet and a second data packet.

[0017] Exemplarily, the switch test system sends probe messages to the switch nodes in the network through an active detection mechanism to collect key data information. Among them, the first data packet mainly collects the static topology information of the switch, including the blockchain hash value of the switch node ID (used to uniquely identify each switch node), the gPTP synchronization deviation value (used to evaluate the clock synchronization accuracy), and the port link status bitmap (reflecting the physical connection status); the second data packet focuses on collecting the real-time traffic data in the network, including performance indicators such as the bandwidth occupancy, delay jitter, and packet loss rate of the data stream, and at the same time records feature information such as the source and destination addresses, priorities, and QoS requirements of the traffic. These data will provide basic data support for subsequent intelligent traffic management.

[0018] S102. Perform time-space joint encoding on the first data packet to obtain a synchronization topology graph.

[0019] Exemplarily, the unique identifier of the switch node is extracted through the blockchain hash algorithm, and cryptographic algorithms such as SHA-256 are used to ensure the uniqueness and non-forgeability of each node identifier, and a trusted node identity mapping relationship is established. Based on the in-depth analysis of the gPTP synchronization deviation value, by calculating the time deviation, propagation delay, and path delay between the master and slave clocks, the clock drift rate, clock deviation correction coefficient, and time synchronization quality index of each node are obtained, and an accurate time synchronization compensation mapping table is constructed; then, the physical layer connection relationship is extracted using the port link status bitmap, including information such as port rate, duplex mode, and link quality, and a reversible binary feature code is generated through an improved topology fingerprint compression algorithm, significantly reducing the data storage overhead; subsequently, a time topology layer is constructed based on the time synchronization compensation mapping table to generate a gPTP master-slave clock deviation matrix reflecting the clock synchronization relationship between nodes; the time dimension (clock synchronization relationship), space dimension (physical connection relationship), and resource dimension (CBS bandwidth reservation status of the TSN queue) are multi-dimensionally fused, and technologies such as tensor decomposition and feature mapping are used to ensure the effective integration of information in different dimensions and generate a complete synchronization topology graph, providing an omni-directional network view for subsequent intelligent path planning.

[0020] S103. Input the synchronization topology graph into a pre-set spatio-temporal graph convolutional network to obtain a dynamic path policy library.

[0021] Exemplarily, perform multi-modal feature decomposition on the topology graph. Use tensor decomposition technology to decompose network features into feature tensors in three dimensions: space, time, and resources, reducing the computational complexity while retaining key information. The network architecture adopts a modular design. The spatial attention sub-module extracts the topological connection relationships and link state features between nodes through a graph convolutional network, capturing the dependencies in the spatial dimension; the temporal attention sub-module adopts an innovative dilated convolutional structure to extract long-term temporal dependencies by expanding the receptive field, effectively capturing clock synchronization patterns and delay jitter features. The resource attention sub-module uses the self-attention mechanism to analyze bandwidth allocation and queue scheduling features, optimizing resource utilization efficiency; the feature fusion sub-module uses an improved gated linear unit for feature selection and weighted fusion to achieve adaptive integration of multi-dimensional features; after multi-level feature extraction and fusion, generate an initial path weight matrix through a fully connected layer and softmax normalization. Introduce the quantum annealing algorithm to optimize the path, and ensure the feasibility of the path through gated list conflict detection, output a dynamic path policy library containing multiple alternative path schemes, providing intelligent decision-making support for network traffic scheduling.

[0022] S104. Package and simulate and verify the dynamic path policy library to obtain a formal verification policy package.

[0023] Exemplarily, define the syntax structure and semantic rules of the policy through a formal method to construct a complete policy syntax tree, including elements such as the trigger conditions, execution actions, and constraint conditions of the policy, generating a standardized policy semantic model. Based on the policy semantic model, establish a strict policy verification rule set using temporal logic and state machine theory, defining multi-dimensional verification rules such as temporal constraints, resource constraints, and security constraints for policy execution; input the policy into a pre-set high-fidelity verification environment, which can simulate various network scenarios and load conditions, including normal working scenarios, fault recovery scenarios, high-load scenarios, etc., and perform large-scale policy execution simulations through the Monte Carlo method, collecting detailed execution trace data; conduct in-depth temporal consistency analysis on the execution traces, and use statistical methods to calculate performance indicators in multiple dimensions such as the time deviation rate, spatial consistency index, and resource utilization efficiency of policy execution; at the same time, evaluate the impact of policy execution on network performance, including key indicators such as end-to-end delay, throughput, and packet loss rate. Compare the execution quality data with the pre-set verification thresholds in multiple dimensions, generate a verification report containing detailed verification results and optimization suggestions, and perform secure binary packaging on the policies that pass the verification to ensure the integrity and confidentiality of the policies during transmission and storage, forming a reliable formal verification policy package.

[0024] S105. Determine the consistency policy instructions according to the second data packet, the synchronization topology graph, and the formal verification policy package.

[0025] Exemplarily, multi-dimensional feature extraction is performed on the real-time traffic data in the second data packet, including key features such as the bandwidth requirement, delay sensitivity, priority level, and service type of the traffic. At the same time, the time distribution feature and space distribution feature of the traffic are analyzed to construct a complete traffic feature vector. Based on the synchronous topology graph, in-depth topology matching degree analysis is carried out. By calculating indicators such as link bandwidth utilization, node processing capacity, and path delay, the adaptability of each available path to the current traffic demand is evaluated, and a detailed topology matching degree evaluation report is generated. The formal verification policy package and the topology matching result are associated and mapped in multiple dimensions to establish a constraint relationship model for policy execution. This model considers multiple dimensions such as time window constraint, resource capacity constraint, and service quality constraint. The comprehensive benefit of policy execution is calculated through an improved multi-objective optimization algorithm, including multiple optimization objectives such as bandwidth utilization, load balancing degree, delay performance, and energy efficiency. The Pareto optimal principle is used for weighing and selection. Based on the optimization result, the policy execution path is dynamically adjusted to generate an execution plan including primary and backup paths, and a corresponding failover mechanism is designed. The execution plan is strictly compared with the preset consistency rules to ensure the consistency and reliability of policy execution. Through a high-strength digital signature algorithm, the immutability of policy instructions is ensured, and a secure and reliable consistency policy instruction is generated.

[0026] S106. Perform cross-domain policy synchronization processing on the formal consistency policy instruction to generate a management policy instruction for the switch.

[0027] Exemplarily, cross-domain policy synchronization processing is the core mechanism for realizing large-scale network collaborative management. First, systematic policy domain division is performed on the consistency policy instruction. Based on multiple dimensions such as network topology structure, management authority, and business characteristics, the boundary of the policy domain is defined, and a clear inter-domain relationship model is established. Then, complex cross-domain policy mapping is carried out. By establishing inter-domain policy conversion rules and coordination mechanisms, the consistency of policy semantics between different domains is ensured, and at the same time, the management autonomy of each domain is maintained. Comprehensive conflict detection is performed on the mapping result. By calculating indicators such as the overlap degree of the time window of policy execution, the degree of resource competition, and the impact on service quality, potential policy conflicts are identified. Based on the detection result, a dynamic policy priority evaluation system is established. Considering multiple factors such as business importance, resource urgency, and execution timeliness, the comprehensive priority score of the policy is obtained through weighted calculation. The policy execution sequence is intelligently scheduled according to the priority score, and a heuristic algorithm is used to optimize the execution order to maximize resource utilization efficiency. At the same time, an inter-domain feedback mechanism is established to monitor the policy execution effect in real time and dynamically adjust the execution plan. The optimized execution plan and the original consistency policy instruction are deeply integrated. Through policy recombination and parameter tuning, accurate management instructions for specific switches are generated to ensure the effective execution of policies in a cross-domain environment and realize the global optimization and collaborative management of network resources.

[0028] The embodiment of the present application provides an intelligent traffic management method based on a switch test system. The method includes: detecting a switch through the switch test system to obtain a first data packet and a second data packet; performing time-space joint encoding on the first data packet to obtain a synchronization topology graph; inputting the synchronization topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library; encapsulating and simulating and verifying the dynamic path policy library to obtain a formal verification policy packet; determining a consistency policy instruction according to the second data packet, the synchronization topology graph, and the formal verification policy packet; performing cross-domain policy synchronization processing on the formal consistency policy instruction to generate a management policy instruction for the switch. In the above method, the original data is encoded by the time-space joint encoding technology to obtain a synchronization topology graph; the synchronization topology graph is processed by means of the spatio-temporal graph convolutional network to effectively capture the spatio-temporal characteristics of the network topology, intelligently learn the traffic pattern, increase the time synchronization accuracy and reduce the fault recovery time. The formal verification mechanism ensures the correctness and reliability of the policy, reduces the implementation risk, combines the cross-domain policy synchronization processing technology, ensures the accuracy of conflict resolution for policy changes, optimizes the network resource utilization efficiency at the same time, and effectively reduces the labor cost of network management.

[0029] To more clearly introduce the technical solution of the present application, the technical solution of the present application will also be introduced through specific embodiments below. It should be noted that the specific embodiment is used to expand the description of the technical solution of the present application, rather than limiting the present application.

[0030] In some embodiments, performing time-space joint encoding on the first data packet to obtain a synchronization topology graph includes: S1021-S1026.

[0031] S1021. Extract the blockchain hash value from the first data packet to obtain a blockchain hash value set, and the blockchain hash value set includes: the unique identifier of each switch node.

[0032] Exemplarily, when extracting the blockchain hash value of the first data packet, cryptographic algorithms such as SHA-256 are used to perform a hash operation on the identification information of the switch node to generate a hash value with a fixed length. These hash values serve as the unique identifiers of the nodes, possessing irreversibility and collision resistance, ensuring that each switch node has a unique identity in the network. The establishment of the blockchain hash value set provides a trusted node identity basis for subsequent network topology construction. During the extraction process, the node information in the data packet is preprocessed, including data format standardization, redundant information filtering, and integrity verification, to ensure the accuracy and reliability of the generated hash values. Meanwhile, by establishing a mapping relationship database between the hash values and the physical nodes, it facilitates subsequent node identity authentication and topology relationship tracking. In addition, to improve the security of the system, digital signatures are also applied to the hash values to prevent node identities from being tampered with or forged.

[0033] S1022. Analyze the gPTP synchronization deviation value of the first data packet according to the blockchain hash value set to obtain a time synchronization compensation mapping table, where the time synchronization compensation mapping table includes the clock drift rate, clock deviation correction coefficient, and time synchronization quality index of each node.

[0034] Exemplarily, by calculating the time deviation, propagation delay, and path delay between the master and slave clocks, the clock drift rate of each node is obtained, reflecting the stability of the clock frequency; calculate the clock deviation correction coefficient to compensate for the clock synchronization error; evaluate the time synchronization quality index, including synchronization accuracy, stability, and reliability. These parameters together constitute the time synchronization compensation mapping table, providing an accurate compensation mechanism for the time synchronization of the network. During the analysis process, considering the dynamic changes in the network environment, such as factors like link load fluctuations and temperature changes that affect clock synchronization, and dynamically adjusting the compensation parameters through an adaptive algorithm. At the same time, by establishing a historical data analysis model, through the statistical analysis of long-term synchronization data, the prediction accuracy and compensation effect of the clock drift rate are optimized. In this way, a multi-level clock synchronization quality evaluation mechanism is achieved, quantitatively evaluating the synchronization performance of nodes from multiple dimensions such as time accuracy, frequency stability, and synchronization reliability.

[0035] S1023. Extract the port link status bitmap of the first data packet based on the time synchronization compensation mapping table to obtain the connection relationship diagram of the physical topology layer.

[0036] Exemplarily, the node information in the time synchronization compensation mapping table is utilized to extract the port link status bitmap for the first data packet. The bitmap records the connection status of each switch port, including information such as port rate, duplex mode, and link quality. By analyzing the bitmap data, a connection relationship graph of the physical topology layer is constructed, clearly showing the physical connection relationships between nodes in the network, providing basic data support for subsequent topology analysis. During the bitmap extraction process, a multi-dimensional link status monitoring mechanism is adopted, including real-time monitoring of performance indicators such as link bandwidth utilization rate, bit error rate, and packet loss rate, and comprehensive evaluation of the link quality through intelligent algorithms. At the same time, by establishing a tracking mechanism for link status changes, the historical change trends of the link status are recorded, providing a basis for network fault diagnosis and performance optimization. In this way, an automated link discovery and verification mechanism is realized, which can timely detect and update link changes in the network, ensuring the accuracy and real-time nature of the physical topology map.

[0037] S1024. Perform topology fingerprint compression processing on the first data packet according to the connection relationship graph to obtain a reversible binary feature code.

[0038] Exemplarily, an improved topology fingerprint algorithm is adopted to compress the complex network topology structure into a binary feature code of a fixed length. This feature code is reversible, can accurately restore the original topology information, and at the same time significantly reduces the data storage and transmission overhead, providing an efficient data representation for subsequent topology analysis. During the compression process, a multi-level topology feature extraction method is adopted, including the extraction and encoding of topology features such as node degree distribution, path features, and clustering coefficient. Through an optimized feature selection algorithm, the key information in the topology structure is retained to achieve efficient data compression. An adaptive compression ratio adjustment mechanism is realized, which dynamically adjusts the compression parameters according to the network scale and complexity, and balances between compression efficiency and information retention. By establishing a version control mechanism for the feature code, incremental updates of topology changes are supported, reducing data processing and storage overhead.

[0039] S1025. Based on the reversible binary feature code, construct the time topology layer for the first data packet to obtain the gPTP master-slave clock deviation matrix between each node.

[0040] Exemplarily, by analyzing the clock synchronization relationship between nodes, the gPTP master-slave clock deviation between each pair of nodes is calculated to generate a clock deviation matrix. This matrix reflects the clock synchronization status among nodes in the network and provides a time synchronization basis for the precise scheduling of time-sensitive networks. During the construction process, a hierarchical clock synchronization management architecture is adopted, and through the establishment of the master-slave clock hierarchy relationship, clock synchronization transmission within the network scope is achieved. In this way, a monitoring and early warning mechanism for clock synchronization performance is established. By real-time monitoring the change trend of clock deviation, synchronization anomalies can be detected and processed in a timely manner to ensure the time synchronization quality of the network. To cope with network topology changes, the system also implements a dynamic adjustment mechanism for clock synchronization relationships, which can quickly reconstruct the clock synchronization hierarchy.

[0041] S1026. Construct a resource topology layer for the first data packet according to the time topology layer and the physical topology layer to obtain the CBS bandwidth reservation status of the TSN queue, and form a synchronization topology graph.

[0042] Exemplarily, by analyzing the Credit-Based Shaper (CBS) mechanism of the TSN queue, the bandwidth reservation status of each node is calculated, including information such as available bandwidth, reserved bandwidth, and priority queues. The topological information in three dimensions is fused to form a complete synchronization topology graph, providing a multi-dimensional topological view for network traffic management. During the construction process, through the dynamic management mechanism of resource reservation and real-time monitoring of traffic demands and network status, the bandwidth reservation strategy is dynamically adjusted. At the same time, the system also establishes a multi-level queue management model, and by precisely controlling the scheduling strategy of each priority queue, the quality of service for time-sensitive services is ensured.

[0043] In some embodiments, the synchronization topology graph is input into a pre-set spatio-temporal graph convolutional network to obtain a dynamic path policy library, including: S1031 - S1034.

[0044] S1031. Perform multi-modal feature decomposition on the synchronization topology graph to generate a spatio-temporal feature tensor.

[0045] Exemplarily, when performing multi-modal feature decomposition on the synchronization topology graph, the tensor decomposition technique is used to decompose the network features into three dimensions: space, time, and resources. The spatial dimension features include node degree distribution, link connectivity, and topological structure features, and the spatial feature matrix is extracted through the spectral decomposition algorithm; the temporal dimension features include clock synchronization status, delay distribution, and jitter characteristics, and the temporal feature matrix is extracted through Fourier transform; the resource dimension features include bandwidth utilization, queue status, and load distribution, and the resource feature matrix is extracted through principal component analysis. During the feature decomposition process, the feature dimension is set to 128, and the Tucker decomposition method is used to combine the feature matrices of the three dimensions into a unified spatio-temporal feature tensor, and the core dimension of the tensor is set to (64, 64, 64). The generated spatio-temporal feature tensor retains more than 95% of the information of the original topology graph, and the data dimension is effectively compressed.

[0046] S1032. Based on the spatio-temporal feature tensor, input the pre-set spatio-temporal graph convolutional network, and generate a path weight matrix by decoupling the feature learning of the space, time, and resource dimensions.

[0047] Exemplarily, input the spatio-temporal feature tensor into the pre-set spatio-temporal graph convolutional network for feature learning. The network structure includes 3 spatial graph convolutional layers, the convolutional kernel sizes are 16, 32, and 64 respectively, and the stride is 1; 2 temporal convolutional layers, using the causal convolutional structure, the convolutional kernel size is 3×3, and the dilation rate is 2; 2 resource attention layers, the number of heads is set to 8, and the hidden layer dimension is 256. The spatial dimension features extract the topological dependencies between nodes through graph convolutional operations, and the activation function uses ReLU; the temporal dimension features capture the temporal correlations through causal convolutions, and the attention mechanism weights are calculated through the softmax function; the resource dimension features learn the resource allocation patterns through the multi-head attention mechanism. The features of the three dimensions are fused through the cross-dimensional attention module to generate a path weight matrix with a dimension of 256×256, and each element in the matrix represents the feasibility weight of the corresponding path.

[0048] S1033. Perform quantum annealing optimization on the path weight matrix to generate a set of candidate paths.

[0049] Exemplarily, quantum annealing optimization is performed on the path weight matrix, and the optimization process uses qubit encoding to represent the path state. The annealing time of the quantum annealing system is set to 1000 μs, the quantum tunneling strength is 0.5, and the system temperature linearly decreases from 2 K to 0.1 K. During the annealing process, each path corresponds to a qubit, and path optimization is achieved by adjusting the coupling strength between qubits. The optimization objective function includes multiple evaluation metrics such as path length, bandwidth utilization, and delay performance, and the weights of each metric are dynamically adjusted in an adaptive manner. In each annealing cycle, the system randomly selects 1000 initial states for parallel optimization and quickly converges to the local optimal solution through the quantum tunneling effect. After annealing optimization, path combinations with weight values greater than 0.8 are selected from the 256×256 path weight matrix to form a candidate path set.

[0050] S1034. Perform gated list conflict detection on the candidate path set to generate a dynamic path policy library.

[0051] Exemplarily, gated list conflict detection is performed on the candidate path set, and a hierarchical detection strategy is adopted in the detection process. At the physical level, the bandwidth competition and link sharing situation between paths are detected, and the conflict threshold is set to 80% bandwidth utilization; at the time level, the delay superposition effect of paths is detected, and the end-to-end delay jitter is calculated, and the conflict threshold is set to 100 μs; at the resource level, queue scheduling conflicts are detected, and the queue blocking probability is calculated, and the conflict threshold is set to 0.1. The detection algorithm adopts a parallel processing method to simultaneously process the conflict detection of multiple path combinations. For the detected conflict paths, path adjustment is performed through a heuristic algorithm, and the adjustment parameters include path hop count, link utilization, and queue priority. The adjusted paths are re-detected for conflicts until all paths meet the conflict-free condition, and a dynamic path policy library is generated.

[0052] In some embodiments, based on the spatio-temporal feature tensor, a pre-set spatio-temporal graph convolutional network generates a path weight matrix by decoupling the feature learning of the spatial, temporal, and resource dimensions, including: S321 - S326.

[0053] S321. Input the spatio-temporal feature tensor into the pre-set spatio-temporal graph convolutional network, and the pre-set spatio-temporal graph convolutional network includes: a spatial attention sub-module, a temporal attention sub-module, a resource attention sub-module, and a feature fusion sub-module.

[0054] Exemplarily, the pre - configured spatio - temporal graph convolutional network as a whole adopts a multi - branch parallel structure, including four key functional modules: a spatial attention sub - module, a temporal attention sub - module, a resource attention sub - module, and a feature fusion sub - module. The spatial attention sub - module adopts a graph convolutional network structure and includes 3 graph convolutional layers; the temporal attention sub - module adopts a causal convolution structure and includes 2 dilated convolutional layers; the resource attention sub - module adopts a multi - head self - attention structure and includes 8 attention heads; the feature fusion sub - module adopts a gated linear unit structure and includes 2 fully - connected layers. Information transfer between each sub - module is achieved through a residual connection method to avoid the problem of gradient disappearance. The network is trained using the Adam optimizer, with the learning rate set to 0.001, the batch size to 64, and the number of training epochs to 200 rounds.

[0055] S322. Perform graph convolution operations on the spatio - temporal feature tensor in the spatial dimension through the spatial attention sub - module to extract the topological connection relationship and link state features between nodes, and generate a spatial feature vector.

[0056] Exemplarily, the spatial attention sub - module performs spatial - dimension feature extraction on the input spatio - temporal feature tensor. The convolution kernel sizes of the graph convolutional layers are set to 16, 32, and 64 respectively, with a stride of 1 for each layer, and each layer is followed by a batch normalization layer and a ReLU activation function. During the graph convolution operation, the node features are aggregated and updated through a message - passing mechanism, and the aggregation function uses an average pooling operation. For each node, calculate its topological connection strength with neighbor nodes, and the connection strength threshold is set to 0.6. The extracted features include node degree distribution features, link bandwidth features, and topological structure features. Calculate the spatial correlation weights between nodes through an attention mechanism, and the dimension of the weight matrix is 64×64. After three - layer graph convolution operations, a spatial feature vector with a dimension of 256 is generated, and this vector encodes the spatial topology information of the network.

[0057] S323. Perform causal convolution operations on the spatio - temporal feature tensor in the temporal dimension through the temporal attention sub - module to extract the clock synchronization relationship and delay jitter features between nodes, and generate a temporal feature vector, where the temporal attention sub - module adopts a dilated convolution structure to expand the receptive field.

[0058] Exemplarily, the temporal attention sub-module uses a causal convolution structure to process the temporal dimension features. The dilation rates of the dilated convolution layers are set to 2 and 4, and the kernel size is 3×3. Long-term dependencies are captured by expanding the receptive field. The number of output channels for each convolution layer is 128, and Layer Normalization is used for feature normalization. Causal convolution ensures that only historical information is used during prediction, avoiding information leakage. The temporal features extracted by the module include clock offset features, delay distribution features, and jitter features. The clock synchronization relationship is obtained by calculating the clock offset sequence between node pairs, and the delay jitter features are obtained by analyzing the trend of end-to-end transmission delays. The temporal attention weights are calculated through the softmax function, and the dimension of the weight matrix is 64×64. After two layers of dilated convolution operations, a temporal feature vector with a dimension of 256 is generated.

[0059] S324. Perform self-attention operations on the spatio-temporal feature tensor in the resource dimension through the resource attention sub-module to extract the bandwidth allocation and queue scheduling features between nodes, and generate a resource feature vector.

[0060] Exemplarily, the resource attention sub-module uses a multi-head self-attention mechanism to process the resource dimension features. The number of attention heads is set to 8, and the hidden dimension of each head is 32. Query matrix, key matrix, and value matrix are generated through linear mapping. During the self-attention operation process, the resource correlation scores between nodes are calculated, and the dimension of the score matrix is 64×64. The correlation scores are calculated through the scaled dot-product attention mechanism, and the scaling factor is set to the square root of 8. The resource features extracted by the module include bandwidth utilization features, queue length features, and load balancing features. The bandwidth allocation features are obtained by analyzing the link capacity and actual traffic, and the queue scheduling features are obtained by analyzing the buffer status and scheduling policies. After multi-head attention operations and feature concatenation, a resource feature vector with a dimension of 256 is generated.

[0061] S325. Perform adaptive weighted fusion on the spatial feature vector, temporal feature vector, and resource feature vector through the feature fusion sub-module to generate a multi-dimensional feature representation, where the feature fusion sub-module uses a gated linear unit for feature selection.

[0062] Exemplarily, the feature fusion sub-module uses a gated linear unit structure to fuse the feature vectors in three dimensions. The gated linear unit contains two parallel branches: a linear transformation branch and a gating signal branch. The linear transformation branch transforms the features through a 256-dimensional fully connected layer; the gating signal branch generates a gating value between 0 and 1 through a sigmoid activation function to control the importance of the features. During the feature fusion process, learnable weight parameters are set to weight the three feature vectors, and the initial weight values are all set to 1 / 3. The weight parameters are optimized through the backpropagation algorithm, and L2 regularization is used to prevent overfitting. The fused features are added to the original features through a residual connection and normalized through a Layer Normalization layer to generate a multi-dimensional feature representation with a dimension of 256.

[0063] S326. Perform a fully connected layer transformation and softmax normalization on the multi-dimensional feature representation to generate a path weight matrix.

[0064] Exemplarily, perform dimension conversion and normalization on the 256-dimensional multi-dimensional feature representation. The output dimension of the fully connected layer is set to 64×64, corresponding to the size of the path weight matrix. The fully connected layer initializes the weight parameters using the Xavier initialization method, and the bias term is initialized to 0. After the feature transformation, it is normalized through the softmax function to calculate the probability weights of each possible path. During the normalization process, the temperature parameter is set to 0.1 to adjust the smoothness of the probability distribution. In the generated path weight matrix, each element value represents the probability of the corresponding path being selected, and the value range is between 0 and 1. Paths with weight values greater than 0.8 are considered high-feasibility paths, and paths with weight values less than 0.2 are filtered out. The path weight matrix serves as an important basis for subsequent path planning.

[0065] In some embodiments, the dynamic path policy library is encapsulated and simulated and verified to obtain the formal verification policies including: S1041 - S1045.

[0066] S1041. Construct a policy syntax tree for the dynamic path policy library to generate a policy semantic model and obtain policy syntax structure data.

[0067] S1042. Formally describe the policy constraint conditions based on the policy syntax structure data, establish a policy verification rule set, and obtain policy constraint specification data.

[0068] S1043. Input the policy constraint specification data into a pre-set simulation verification environment to perform multi-scenario policy execution simulation and obtain policy execution trajectory data.

[0069] S1044. Perform a temporal consistency analysis on the policy execution trajectory data, calculate the time deviation rate and spatial consistency index of the policy execution, and obtain policy execution quality data.

[0070] S1045. Compare the policy execution quality data with a preset verification threshold to generate a policy verification report, and perform binary encapsulation on the policies that pass the verification to obtain a formal verification policy package.

[0071] The formal verification process of the exemplary dynamic path policy library involves multiple key links, and ensures the correctness and reliability of the policy through a systematic verification process. In the policy syntax tree construction stage, perform syntax analysis and semantic parsing on the policy content in the dynamic path policy library, and convert the unstructured policy description into standard syntax structure data, which lays the foundation for subsequent formal verification. The policy syntax structure data contains the core logic components, execution conditions, and related constraint parameters of the policy. Based on the generated policy syntax structure data, the verification system further extracts the constraint condition elements in the policy and uses formal methods to strictly describe these constraint conditions. By establishing a complete set of policy verification rules, convert the informal constraint conditions into computable and verifiable policy constraint specification data. These specification data cover verification elements in multiple dimensions such as time constraints, space constraints, and resource constraints. After obtaining the policy constraint specification data, the verification system imports it into a pre-configured simulation verification environment. This environment simulates various actual application scenarios, and generates detailed policy execution trajectory data by simulating the execution process of the policy in different scenarios. These trajectory data record the state changes, decision-making processes, and resource scheduling situations during the policy execution process. The verification system conducts in-depth temporal consistency analysis on the policy execution trajectory data, focusing on the time characteristics during the policy execution process. By calculating the time deviation rate of the policy execution, evaluate the execution accuracy of the policy in the time dimension; by analyzing the spatial consistency index, evaluate the execution accuracy of the policy in the spatial dimension. These analysis results are summarized to form policy execution quality data, which comprehensively reflects the performance characteristics of the policy execution. The verification system systematically compares the policy execution quality data with the preset verification threshold. The verification threshold includes multiple evaluation dimensions such as time accuracy requirements and space accuracy requirements. For the policies that meet the verification requirements, the system generates a detailed policy verification report, recording the key data and analysis results during the verification process. By performing binary encapsulation on the policies that pass the verification, form a standardized formal verification policy package, which is convenient for deployment and use in actual applications. This complete formal verification process ensures the reliability and correctness of the dynamic path policy. Through the organic combination of links such as syntax analysis, constraint extraction, simulation verification, and performance evaluation, a comprehensive verification of the policy is achieved. The generation of the formal verification policy package provides a reliable guarantee for the actual application of the policy, enabling the policy to operate stably and efficiently in the actual environment.

[0072] In some embodiments, determining a consistency policy instruction according to a second data packet, a synchronization topology graph, and a formal verification policy packet includes: S1051-S1056.

[0073] S1051. Extract traffic characteristics from the second data packet to generate a traffic characteristic vector.

[0074] Exemplarily, use deep packet inspection technology to perform multi-dimensional feature extraction on the second data packet, including data packet size distribution characteristics, data packet arrival time interval characteristics, flow duration characteristics, etc. A sliding window mechanism is used during the feature extraction process. The window size is set to 1000 data packets, and the sliding step is 200 data packets. Statistical analysis is performed on the data within each window. By calculating statistics such as the average size, standard deviation, kurtosis, and skewness of the data packets, combined with the periodic and bursty characteristics of the traffic, a 128-dimensional traffic characteristic vector is constructed. This characteristic vector contains the distribution characteristics of the data packets in the time and space dimensions, providing basic data support for subsequent topology matching degree analysis.

[0075] S1052. Perform topology matching degree analysis on the traffic characteristic vector based on the synchronization topology graph, calculate the topology fitness of each path, and obtain topology matching degree data.

[0076] Exemplarily, map the traffic characteristic vector to the nodes and edges of the synchronization topology graph, and use a graph neural network model for topology matching degree analysis. The model uses three layers of graph convolutional layers, with the output dimensions of each layer being 64, 32, and 16 respectively, and the ReLU activation function is used. For each possible transmission path, calculate indicators such as its node connectivity score, link bandwidth utilization score, and delay jitter score. By setting the weight coefficients α = 0.4, β = 0.3, and γ = 0.3, the scores are weighted and combined to obtain the comprehensive topology fitness of the path. The topology matching degree data is stored in matrix form, and the matrix elements represent the fitness values of the corresponding paths, providing a decision basis for policy-topology association mapping.

[0077] S1053. Perform policy-topology association mapping on the formal verification policy packet and the topology matching degree data, establish policy execution constraint relationships, and obtain policy constraint mapping data.

[0078] Exemplarily, a bidirectional attention mechanism is adopted to perform an associative mapping on the policy rules and topology matching degree data in the formal verification policy package. The hidden dimension of the attention layer is set to 256, and the number of heads is set to 8. By calculating the correlation scores between the policy rules and topology features, for each policy rule, elements such as its resource requirement constraints, time window constraints, and security level constraints are extracted and matched with the path features in the topology matching degree data. When establishing the policy execution constraint relationship, hard constraints such as path bandwidth capacity, node processing capacity, and link delay are considered, as well as soft constraints such as load balancing and energy consumption optimization, to generate a policy constraint mapping data matrix.

[0079] S1054. Perform multi-objective optimization processing on the policy constraint mapping data. By calculating the cost-benefit ratio and resource utilization rate of policy execution, policy optimization data is obtained.

[0080] Exemplarily, a multi-objective optimization model is constructed. The objective function includes three dimensions: minimizing the policy execution cost, maximizing the resource utilization rate, and optimizing the quality of service. An improved NSGA-III algorithm is used for solving. The population size is set to 200, and the number of evolutionary generations is 500. The calculation of the cost-benefit ratio considers factors such as computing resource consumption, bandwidth occupancy, and energy consumption. The resource utilization rate is evaluated using a weighted average method, and the weights are determined by the analytic hierarchy process. During the optimization process, an adaptive crossover operator and a mutation operator are introduced. The crossover probability is 0.8, and the mutation probability is 0.1. High-quality solutions are selected through non-dominated sorting and crowding degree calculation to form a policy optimization data set.

[0081] S1055. Dynamically adjust the policy execution path based on the policy optimization data to generate policy execution plan data.

[0082] Exemplarily, a reinforcement learning method is adopted to dynamically adjust the policy execution path, and a decision-making model based on Deep Q-Network is constructed. The state space includes information such as the current path load, node status, and link status. The action space includes operations such as path switching, load adjustment, and resource reallocation. The reward function design considers multiple evaluation indicators such as execution efficiency, resource utilization rate, and quality of service. The model uses a double-layer LSTM network to extract temporal features. The hidden layer dimension is 512, the learning rate is set to 0.001, and the size of the experience replay pool is 10,000. The network parameters are updated by the policy gradient method to generate policy execution plan data, including a detailed path adjustment plan and a resource allocation plan.

[0083] S1056. Compare the policy execution plan data with the preset consistency rules to generate a consistency policy instruction, and digitally sign the consistency policy instruction to obtain the consistency policy instruction.

[0084] Exemplarily, a formal verification method is used to check the consistency of the policy execution plan data to verify whether it meets a preset set of consistency rules. The consistency rules include multiple dimensions such as causal consistency, sequential consistency, and eventual consistency, and are formally described by temporal logic formulas. The comparison process uses model checking techniques to construct a state transition graph and verify key properties. The generated consistency policy instructions are signed using the Elliptic Curve Digital Signature Algorithm (ECDSA), the NIST P-256 curve is selected, the key length is 256 bits, and the message digest is calculated through SHA-256 to ensure the integrity and non-repudiation of the instructions.

[0085] In some embodiments, cross-domain policy synchronization processing is performed on the formal consistency policy instructions to generate management policy instructions for switches, including: S1061 - S1066.

[0086] S1061. Divide the consistency policy instructions into policy domains to obtain policy domain division data.

[0087] Exemplarily, a hierarchical clustering algorithm is used to divide the consistency policy instructions into domains. The clustering distance threshold is set to 0.85, and a feature vector is constructed based on the semantic similarity, execution object relevance, and resource dependence degree of the policy instructions. By calculating the cosine similarity matrix between policy instructions, the policy instructions with a similarity higher than the threshold are divided into the same policy domain. Each policy domain contains a set of policy instructions with similar semantics, high execution object association, and close resource dependence relationships, and the coupling degree between policy domains is kept low. The policy domain division data includes the boundary definition of each policy domain, the set of policy instructions within the domain, and the description of the inter-domain relationship.

[0088] S1062. Based on the policy domain division data, perform cross-domain policy mapping on the policy execution plan data, establish an association relationship between policy domains, and obtain cross-domain policy mapping data.

[0089] Exemplarily, a policy domain dependency graph is constructed, where nodes represent each policy domain and edges represent the dependency relationships between domains. For each policy domain, extract the execution preconditions, execution effects, and resource requirements of its policy instructions to generate a policy feature vector. By calculating the correlation matrix of the feature vectors of different policy domains, identify domain pairs with execution dependencies, resource sharing, or result impacts. For the identified relevant domain pairs, establish a bidirectional mapping relationship, and the mapping strength is determined by the correlation coefficient. During the mapping process, consider the temporal constraints, resource limitations, and execution environment of the policy instructions to ensure the feasibility of cross-domain policy mapping. The cross-domain policy mapping data records the mapping relationship, mapping strength, and constraint conditions between policy domains.

[0090] S1063. Perform policy conflict detection on the cross - domain policy mapping data and the policy constraint mapping data. By calculating the time - window overlap degree and resource competition degree of policy execution, obtain the policy conflict data.

[0091] Exemplarily, adopt the time - window analysis method. Divide the policy execution time into time segments of 0.1 seconds, and calculate the set of concurrent policy instructions within each time segment. For concurrent policy instructions, construct a resource - demand matrix, where the matrix elements represent the occupancy of specific resources by the policy instructions. By calculating the eigenvalues of the resource - demand matrix, evaluate the degree of resource competition. Set the time - window overlap degree threshold to 0.6 and the resource competition degree threshold to 0.8. The policy - instruction pairs exceeding the thresholds are marked as conflict pairs. For each conflict pair, record the conflict type, conflict degree, and influence scope to form the policy conflict data.

[0092] S1064. Perform policy - priority sorting on the policy conflict data, generate a policy - execution priority list, and obtain the policy - priority data.

[0093] Exemplarily, construct a priority - scoring model based on the importance, urgency, and resource consumption of the policy instructions. The importance weight is 0.4, which is determined by the business criticality affected by the policy instruction; the urgency weight is 0.35, considering the time constraints of policy execution; the resource - consumption weight is 0.25, reflecting the occupancy degree of system resources by policy execution. Calculate the comprehensive score of each policy instruction through weighted summation. The policy instructions with higher scores obtain higher execution priorities. The priority data includes policy - instruction identifiers, priority scores, and execution - order information.

[0094] S1065. Dynamically adjust the policy - execution sequence based on the policy - priority data, generate a cross - domain policy - execution plan, and obtain the cross - domain policy - execution data.

[0095] Exemplarily, adopt a heuristic algorithm to optimize the policy - execution sequence, considering policy priorities, resource constraints, and execution - dependency relationships. Set the number of algorithm iterations to 200 times. In each iteration, evaluate the feasibility of the adjusted execution plan by locally adjusting the policy - execution order. The feasibility evaluation includes checking whether resource usage exceeds the limit, whether execution dependencies are met, and whether time constraints are violated, etc. Select the execution sequence with the highest score as the optimal solution, and generate a detailed execution plan, including the execution time points of each policy instruction, resource - allocation schemes, and expected execution effects.

[0096] S1066. Perform policy - fusion processing on the cross - domain policy - execution data and the consistency policy instructions to generate management policy instructions for the switch.

[0097] Exemplarily, the cross-domain policy execution data is normalized, and the policy instructions are converted into a configuration command format recognizable by the switch. The normalization process includes instruction syntax conversion, parameter mapping, and execution condition adaptation. Instruction recombination is performed through the policy template library to ensure that the generated management policy instructions comply with the execution specifications of the switch. During the fusion process, the semantic consistency of the policy instructions is maintained, and necessary execution control information is added, such as mechanisms for timeout handling, failure recovery, and logging. The generated management policy instructions are executable, rollbackable, and traceable.

[0098] Please refer to Figure 2 , Figure 2 FIG. is a schematic block diagram of an intelligent traffic management device based on a switch test system provided by an embodiment of the present application. The intelligent traffic management device 200 based on the switch test system is used to execute the foregoing intelligent traffic management method based on the switch test system. Among them, the intelligent traffic management device 200 based on the switch test system can be configured in a server.

[0099] Among them, the server can be an independent server, a server cluster, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0100] As Figure 2 shown, the intelligent traffic management device 200 based on the switch test system includes: a data detection module 201, a topology generation module 202, a policy generation module 203, a policy verification module 204, an instruction generation module 205, and a policy specification module 206.

[0101] The data detection module 201 is used to detect the switch through the switch test system to obtain a first data packet and a second data packet.

[0102] The topology generation module 202 is used to perform time-space joint coding on the first data packet to obtain a synchronous topology graph.

[0103] The policy generation module 203 is used to input the synchronous topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library.

[0104] The policy verification module 204 is used to encapsulate and simulate the verification of the dynamic path policy library to obtain a formal verification policy package.

[0105] The instruction generation module 205 is used to determine a consistency policy instruction according to the second data packet, the synchronous topology graph, and the formal verification policy package.

[0106] A policy specification module 206 is configured to perform cross-domain policy synchronization processing on the form consistency policy instruction to generate a management policy instruction for the switch.

[0107] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is configured to execute the computer program and implement the intelligent traffic management method based on the switch test system according to any one of the embodiments of the present application when executing the computer program.

[0108] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor is caused to implement the intelligent traffic management method based on the switch test system according to any one of the embodiments of the present application.

[0109] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An intelligent traffic management method based on a switch test system, characterized in that, The method includes: Probing the switch through a switch test system to obtain a first data packet and a second data packet; Performing time-space joint encoding on the first data packet to obtain a synchronization topology graph; Inputting the synchronization topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library; Encapsulating and simulating and verifying the dynamic path policy library to obtain a formal verification policy packet; Determining a consistency policy instruction according to the second data packet, the synchronization topology graph and the formal verification policy packet; Performing cross-domain policy synchronization processing on the formal consistency policy instruction to generate a management policy instruction for the switch.

2. The intelligent traffic management method based on the switch test system according to claim 1, characterized in that The performing time-space joint encoding on the first data packet to obtain a synchronization topology graph includes: Extracting blockchain hash values from the first data packet to obtain a blockchain hash value set, where the blockchain hash value set includes: unique identifiers of each switch node; Parsing the gPTP synchronization deviation value of the first data packet according to the blockchain hash value set to obtain a time synchronization compensation mapping table, where the time synchronization compensation mapping table includes the clock drift rate, clock deviation correction coefficient and time synchronization quality index of each node; Extracting a port link state bitmap from the first data packet based on the time synchronization compensation mapping table to obtain a connection relationship graph of the physical topology layer; Performing topology fingerprint compression processing on the first data packet according to the connection relationship graph to obtain a reversible binary feature code; Constructing a time topology layer for the first data packet based on the reversible binary feature code to obtain a gPTP master-slave clock deviation matrix between nodes; Constructing a resource topology layer for the first data packet according to the time topology layer and the physical topology layer to obtain the CBS bandwidth reservation state of the TSN queue, forming a synchronization topology graph.

3. The intelligent traffic management method based on a switch test system according to claim 1, characterized in that The inputting the synchronization topology graph into a preset spatio-temporal graph convolutional network to obtain a dynamic path policy library includes: Performing multi-modal feature decomposition on the synchronization topology graph to generate a spatio-temporal feature tensor; Based on the spatio-temporal feature tensor, inputting it into a preset spatio-temporal graph convolutional network, and generating a path weight matrix through decoupled feature learning in the spatial, temporal, and resource dimensions; Performing quantum annealing optimization on the path weight matrix to generate a candidate path set; Performing gated list conflict detection on the candidate path set to generate a dynamic path policy library.

4. The intelligent traffic management method based on a switch test system according to claim 3, wherein The based on the spatio-temporal feature tensor, inputting it into a preset spatio-temporal graph convolutional network, and generating a path weight matrix through decoupled feature learning in the spatial, temporal, and resource dimensions includes: Inputting the spatio-temporal feature tensor into the preset spatio-temporal graph convolutional network, where the preset spatio-temporal graph convolutional network includes: a spatial attention sub-module, a temporal attention sub-module, a resource attention sub-module, and a feature fusion sub-module; Performing graph convolutional operation in the spatial dimension on the spatio-temporal feature tensor through the spatial attention sub-module, extracting the topological connection relationship and link state features between nodes, and generating a spatial feature vector; Perform causal convolution operation on the spatio-temporal feature tensor in the time dimension through the time attention sub-module to extract the clock synchronization relationship and delay jitter characteristics between nodes, and generate a time feature vector, where the time attention sub-module adopts a dilated convolution structure to expand the receptive field; Perform self-attention operation on the spatio-temporal feature tensor in the resource dimension through the resource attention sub-module to extract the bandwidth allocation and queue scheduling characteristics between nodes, and generate a resource feature vector; Perform adaptive weighted fusion on the spatial feature vector, the time feature vector, and the resource feature vector through the feature fusion sub-module to generate a multi-dimensional feature representation, where the feature fusion sub-module uses a gated linear unit for feature selection; Perform a fully connected layer transformation and softmax normalization on the multi-dimensional feature representation to generate a path weight matrix.

5. The intelligent traffic management method based on the switch test system according to claim 1, characterized in that, The encapsulation and simulation verification of the dynamic path policy library to obtain the formal verification policy includes: Construct a policy syntax tree for the dynamic path policy library to generate a policy semantic model and obtain policy syntax structure data; Formalize the policy constraint conditions based on the policy syntax structure data to establish a policy verification rule set and obtain policy constraint specification data; Input the policy constraint specification data into a pre-set simulation verification environment to perform multi-scenario policy execution simulation and obtain policy execution trajectory data; Perform temporal consistency analysis on the policy execution trajectory data, calculate the time deviation rate and spatial consistency index of policy execution, and obtain policy execution quality data; Compare the policy execution quality data with a pre-set verification threshold to generate a policy verification report, and perform binary encapsulation on the policies that pass the verification to obtain a formal verification policy package.

6. The intelligent traffic management method based on a switch test system according to claim 1, characterized in that The determination of the consistency policy instruction according to the second data packet, the synchronization topology graph, and the formal verification policy package includes: Extract traffic characteristics from the second data packet to generate a traffic feature vector; Perform topology matching degree analysis on the traffic feature vector based on the synchronization topology graph, calculate the topology fitness of each path, and obtain topology matching degree data; Perform policy-topology association mapping on the formal verification policy package and the topology matching degree data to establish a policy execution constraint relationship and obtain policy constraint mapping data; Perform multi-objective optimization processing on the policy constraint mapping data, and obtain policy optimization data by calculating the cost-benefit ratio and resource utilization rate of policy execution; Dynamically adjust the policy execution path based on the policy optimization data to generate policy execution plan data; Compare the policy execution plan data with a pre-set consistency rule to generate a consistency policy instruction, and digitally sign the consistency policy instruction to obtain a consistency policy instruction.

7. The intelligent traffic management method based on a switch test system according to claim 5, characterized in that The cross-domain policy synchronization processing of the formal consistency policy instruction to generate a management policy instruction for the switch includes: Perform policy domain division on the consistency policy instruction to obtain policy domain division data; Perform cross - domain policy mapping on the policy execution plan data based on the divided policy domains, establish the association relationship between policy domains, and obtain cross - domain policy mapping data; Perform policy conflict detection on the cross - domain policy mapping data and the policy constraint mapping data. By calculating the overlap degree of the time window and the resource competition degree of policy execution, obtain policy conflict data; Sort the policy priorities of the policy conflict data, generate a policy execution priority list, and obtain policy priority data; Dynamically adjust the policy execution sequence based on the policy priority data, generate a cross - domain policy execution plan, and obtain cross - domain policy execution data; Perform policy fusion processing on the cross - domain policy execution data and the consistency policy instructions to generate management policy instructions for the switch.

8. An intelligent traffic management device based on a switch test system, characterized in that, The intelligent traffic management device based on the switch test system includes: A data detection module for detecting the switch through the switch test system to obtain the first data packet and the second data packet; A topology generation module for performing time - space joint encoding on the first data packet to obtain a synchronous topology map; A policy generation module for inputting the synchronous topology map into a pre - set spatio - temporal graph convolutional network to obtain a dynamic path policy library; A policy verification module for encapsulating and simulating and verifying the dynamic path policy library to obtain a formal verification policy package; An instruction generation module for determining consistency policy instructions according to the second data packet, the synchronous topology map, and the formal verification policy package; A policy specification module for performing cross - domain policy synchronization processing on the formal consistency policy instructions to generate management policy instructions for the switch.

Images