Key flow-based SRv6 explicit path intelligent scheduling method and system

Abstract

The invention relates to a key flow-based SRv6 explicit path intelligent scheduling method and system, and belongs to the technical field of communication information. The method comprises the following steps: carrying out criticality scoring through a semi-supervised-migration mixed learning model, and outputting a key stream; constructing a full-dimensional network twinborn model, pre-judging a congestion risk trend, and verifying a link path survival rate; generating a low-delay source node list through a key flow path enhancement algorithm based on the key flow and link path survival rate; analyzing the low-delay source node list to create a micro-isolation queue; when the digital twin detects that the actual state deviates from the predicted value, calling a pre-verified backup path identification chain to complete path switching in real time; and performing closed-loop optimization based on the key operation state data by using a double-channel learning model. The technical problems that in the prior art, key flow recognition precision is low, SRv6 path scheduling and network state adaptability is poor, fault switching lags behind, and core model optimization is insufficient are solved.

Description

TECHNICAL FIELD

[0001] The application relates to a key flow-based SRv6 explicit path intelligent scheduling method and system, and belongs to the field of communication information technology. BACKGROUND

[0002] With the development of 5G, industrial internet and other scenarios, the key flow in the network has strict requirements on delay and reliability. In the prior art, key flow identification depends on manual rules or a single feature, and the precision is insufficient; SRv6 explicit path scheduling is often statically configured, and it is difficult to dynamically adapt to network load changes and topology fluctuations. When the network twin model is constructed, the data fusion degree is low, and the accuracy of congestion risk prediction and path survival rate verification is limited. When a fault occurs, path switching depends on manual intervention or a simple triggering mechanism, which easily leads to interruption of key flow transmission. At the same time, there is a lack of closed-loop optimization mechanism for key flow identification model, twin model and path algorithm, and the overall scheduling adaptability and intelligence cannot meet the needs of complex network scenarios. The present application aims to solve the technical problems of low key flow identification precision, poor adaptability of SRv6 path scheduling to network state, lagging fault switching and insufficient optimization of core models in the prior art. SUMMARY

[0003] The technical problem solved by the present application is that the present application provides a key flow-based SRv6 explicit path intelligent scheduling method and system to solve the technical problems of low key flow identification precision, poor adaptability of SRv6 path scheduling to network state, lagging fault switching and insufficient optimization of core models in the prior art.

[0004] The technical solution of the present application is: a key flow-based SRv6 explicit path intelligent scheduling method, the method comprising:

[0005] Collecting multi-dimensional dynamic characteristics of traffic, performing key degree scoring through a semi-supervised-migration hybrid learning model, and outputting key flows carrying delay tolerance threshold values;

[0006] Collecting network fine-grained state data, constructing a full-dimensional network twin model, predicting congestion risk trends, and verifying link path survival rates;

[0007] Based on the key flow and the link path survival rate, a low-delay source node list is generated through a key flow path enhancement algorithm;

[0008] Analyzing the low-delay source node list to create a micro-isolation queue, reserving resources for key flows, and guiding non-key flows to low-load links;

[0009] When the digital twin detects that the actual state deviates from the predicted value, a backup path identification chain verified in advance is called to complete path switching in real time;

[0010] Based on the key operating state data, a semi-supervised-transfer hybrid learning model, a full-dimensional network twin model, and a key flow path enhancement algorithm are optimized by using a dual-channel learning model.

[0011] Further, the multi-dimensional dynamic characteristics of the collected traffic are scored by the semi-supervised-transfer hybrid learning model, and the key flow carrying the delay tolerance threshold is output, including:

[0012] The multi-dimensional dynamic characteristics of the collected traffic are input into the semi-supervised-transfer hybrid learning model for preliminary key degree scoring to identify potential key flows.

[0013] Based on the scoring results and the delay tolerance threshold, the key flow carrying the threshold is output, and the key flow is filtered by dynamic optimization of the semi-supervised-transfer hybrid learning model.

[0014] Further, network fine-grained state data is collected to build a full-dimensional network twin model to predict congestion risk trends and verify link path survival rates, including:

[0015] The network fine-grained state data is collected to build a full-dimensional network twin model, and a high-precision digital map is created through multi-dimensional data fusion.

[0016] The full-dimensional network twin model is used to predict congestion risk trends and verify link path survival rates, and the reliability index is output through real-time simulation and prediction analysis to guide path selection.

[0017] Further, based on the key flow and the link path survival rate, a low-latency source node list is generated by the key flow path enhancement algorithm, including:

[0018] The delay tolerance threshold of the key flow and the link path survival rate are fused to build a weighted directed graph model, and a topological mining algorithm is used to filter effective source nodes.

[0019] The key flow path enhancement algorithm is triggered by the source node, and backtracking pruning is performed in the weighted directed graph model to generate a low-latency source node list that meets the key flow service level agreement.

[0020] Further, the low-latency source node list is parsed to create a micro-isolation queue, reserve resources for key flows, and guide non-key flows to low-load links, including:

[0021] The low-latency source node list is parsed to construct a micro-isolation queue container at the entry node, and a priority forwarding identifier is injected into the segment routing header to reserve bandwidth and time slot resources.

[0022] Based on real-time link load data, a non-key flow guidance strategy is generated, and non-key flows are redirected to low-load paths by combining equivalent multipath entropy calculation.

[0023] Further, when the digital twin detects that the actual state deviates from the predicted value, the pre-verified backup path identification chain is called to complete the path switching in real time, including:

[0024] By continuously comparing the actual network state with the predicted value through the digital twin, when the key indicators break through the dynamic threshold, the path switching instruction based on the network service chain is activated;

[0025] The pre-stored backup path identification chain is called to perform the segment routing identification stack switching operation to complete the traffic rerouting.

[0026] Further, the multi-dimensional dynamic characteristics of the collected traffic are input into the semi-supervised-migration hybrid learning model to perform preliminary key degree scoring and identify potential key flows; specifically including:

[0027] The multi-dimensional dynamic characteristics of the collected traffic, including bandwidth fluctuation, transmission delay change, protocol type, application identifier, and interaction frequency, are standardized and preprocessed to form a feature matrix; the feature matrix is input into the semi-supervised-migration hybrid learning model, wherein the semi-supervised module uses a small number of labeled samples and a large number of unlabeled traffic data to construct a feature space mapping, and uses a clustering algorithm to mine the internal correlation pattern of the traffic; the migration learning module introduces key flow identification knowledge in similar network scenarios, and adjusts the model parameters through a domain adaptation mechanism to adapt to the current network environment; the semi-supervised-migration hybrid learning model processes the input features in a hierarchical manner, first capturing the key feature vector through the feature extraction layer, and then outputting the preliminary key degree score of each traffic through the scoring layer: the key degree score K is calculated as follows:

[0028]

[0029] wherein, and defines the time window boundary for traffic feature collection, is the start time of collection, is the end time of collection, used to limit the observation time length of traffic dynamic characteristics; represents the instantaneous change rate of bandwidth fluctuation, where B represents the bandwidth, i.e. the amount of data that can be transmitted per unit time, and t represents the current t time; is the transmission delay change rate; is the protocol key coefficient; represents the application priority; is the interaction frequency; is the feature space mapping distance; is the feature space standard deviation; in which, is the matrix trace operation, is the labeled sample confidence matrix, is the transpose of the domain adaptation coefficient matrix M, It is the bandwidth fluctuation Laplace operator; Truncate the order of the series expansion; It represents the application priority standard deviation; n represents the nth order of the series expansion stage;

[0030] Based on the initial criticality score distribution, an initial threshold is set, and traffic with scores exceeding the threshold is filtered out and marked as potential critical traffic.

[0031] Furthermore, a full-dimensional network twin model is used to predict congestion risk trends and verify link path survival rates. Reliability indicators are output through real-time simulation and predictive analysis to guide path selection; specifically including:

[0032] The system invokes a full-dimensional network twin model, inputs real-time collected fine-grained network status data and traffic change trends, and starts a dynamic simulation engine. Time series analysis algorithms are used to extrapolate trends in link bandwidth and node load indicators, predicting the congestion risk level and propagation path of each link in the future. Simultaneously, based on graph theory-based reachability analysis, the system simulates link connectivity under various fault scenarios in the full-dimensional network twin model, calculates the survival probability of different paths within a preset time window, and verifies the link path survival rate.

[0033] Among them, link path survival rate The formula is:

[0034]

[0035] Where T represents the total duration of the time window used to analyze the link path status; m represents the number of links that make up the path; Let represent the instantaneous reliability of the i-th link at time t, reflecting the stability of the link over time; n1 is the number of time steps after discretization. It is the interval between adjacent time steps; Let f be the failure rate of the path at the k-1th time step; It is the matrix trace operation, which sums the elements on the main diagonal of a matrix. This is a topology connectivity matrix, used to describe the connectivity relationships between links and nodes in the network topology; It is a node processing capability matrix; It is its inverse matrix; This is the Hadamard product, which integrates topological connectivity information with node processing capability information through this operation; It is the fault recovery rate at the path endpoint L; It is the fault recovery rate at the starting point of the path; This reflects changes in the path fault recovery capability; It is an approximate calculation of the integral of the standard normal distribution, used to quantify the probabilistic characteristics of risks such as path congestion; x is the standardized risk statistic, derived from the path congestion probability. Mean congestion probability Standard deviation The calculation shows that u is the integral variable of the standard normal distribution.

[0036] The simulation results are integrated with the predicted data, and statistical analysis is used to generate reliability indicators such as path availability time, fault recovery probability, and congestion probability. These indicators are then quantified into evaluation parameters that can be directly used for path decision-making.

[0037] Furthermore, by integrating the latency tolerance threshold of critical flows with the link path survival rate, a weighted directed graph model is constructed, and a topology mining algorithm is used to filter effective source nodes; specifically including:

[0038] The system receives critical flows and their criticality scores, integrates link data and link path survival rates, and constructs a weighted directed graph model after correlation calibration and standardization. In the weighted directed graph model, nodes correspond to network source nodes and forwarding nodes, edges represent physical links, and edge weights are generated by a weighted fusion algorithm of critical flow criticality scores and link survival rates. Simultaneously, potential source nodes are inferred based on network topology features. Combining the communication characteristics of critical flows and the reachability constraints of network topology, a topology mining algorithm is used to identify candidate source nodes that match the critical flows. By verifying the link path survival rate and criticality scores of the nodes, effective source nodes with the ability to carry critical flows are selected, and their latency sensitivity level and initial path weight values ​​are labeled.

[0039] The source node selection formula is as follows:

[0040]

[0041] Where K is the criticality score of the critical flow, used to measure the criticality of the traffic; R represents the node resource reserve, reflecting the remaining resources of the node to handle traffic; and U is the node resource utilization rate. Ψ(t) is the resource sensitivity coefficient; S is the link path survival rate; T represents the length of the time observation window for source node screening; Ψ(t) is the node time-varying availability function, which is the availability of the source node at time t; D is the link delay jitter, which reflects the fluctuation of link transmission delay. Node association matrix With path stability matrix The largest eigenvalue of the matrix after transpose and multiplication is a comprehensive eigenvalue that reflects node association and path stability; For time delay jitter variance; It is the path length. Used to quantify the negative impact of factors such as link latency jitter on source node selection; Historical failure rate of the node; It is the vector of topological coupling coefficients; It is the topological coupling index; is the Fourier transform of the distance attenuation factor; j is the imaginary unit; f is the frequency variable; r is the distance variable.

[0042] The present invention also provides an SRv6 explicit path intelligent scheduling system based on critical flow, the system comprising: a module for executing the SRv6 explicit path intelligent scheduling method based on critical flow.

[0043] The beneficial effects of this invention are as follows: This invention improves the accuracy of critical flow identification through semi-supervised-transfer hybrid learning, ensuring accurate screening of critical flows; the full-dimensional network twin model enhances congestion prediction and path reliability verification capabilities, providing accurate basis for scheduling; the critical flow path enhancement algorithm, combined with weighted graph backtracking pruning, generates a sequence of efficient source nodes; the dynamic path switching and closed-loop optimization mechanism improves fault response speed and model adaptability, ultimately achieving low-latency, high-reliability transmission of critical flows and optimizing network resource utilization. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 A flowchart of the SRv6 explicit path intelligent scheduling method based on critical flow provided in Embodiment 1 of this application;

[0046] Figure 2 This is a schematic diagram of the SRv6 explicit path intelligent scheduling system based on critical flow, provided in Embodiment 2 of this application. Detailed Implementation

[0047] 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, not all, of the embodiments of the present invention. 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.

[0048] Example 1: As Figure 1 As shown, Embodiment 1 of this application provides an SRv6 explicit path intelligent scheduling method based on critical flows, including:

[0049] S1: Collect multi-dimensional dynamic features of traffic, perform keyness scoring through a semi-supervised-transfer hybrid learning model, and output key flows carrying latency tolerance thresholds;

[0050] The process involves collecting multi-dimensional dynamic features of traffic, performing keyness scoring using a semi-supervised-transfer hybrid learning model, and outputting key flows carrying latency tolerance thresholds. This includes the following sub-steps:

[0051] S11: Collect multi-dimensional dynamic features of traffic flow and input them into a semi-supervised-transfer hybrid learning model to perform preliminary criticality scoring and identify potential critical flows;

[0052] Multi-dimensional dynamic features of traffic are collected, including bandwidth fluctuations, transmission latency changes, protocol types, application identifiers, and interaction frequencies. These features are then standardized and preprocessed to form a feature matrix. This feature matrix is ​​input into a semi-supervised-transfer hybrid learning model. The semi-supervised module constructs a feature space mapping using a small number of labeled samples and a large amount of unlabeled traffic data, and uses clustering algorithms to uncover inherent correlation patterns within the traffic. The transfer learning module introduces key flow identification knowledge from similar network scenarios and adjusts model parameters through a domain adaptation mechanism to suit the current network environment. The semi-supervised-transfer hybrid learning model performs hierarchical processing on the input features. First, a feature extraction layer captures key feature vectors, and then a scoring layer outputs a preliminary key score for each traffic flow. The key score K is calculated using the following formula:

[0053]

[0054] in, and Define the time window boundaries for traffic feature collection. It is the start time of data collection. It is the end time of data collection, used to limit the observation duration of the dynamic characteristics of the flow. It reflects the instantaneous rate of change of bandwidth fluctuation, where B represents bandwidth, which is the amount of data that can be transmitted per unit time, and t represents the current time t; The rate of change of transmission delay; It is a key coefficient of the protocol; Represents application priority; Interaction frequency; It is the feature space mapping distance; The standard deviation of the feature space; middle, It is the matrix trace operation. To label the confidence matrix of the samples, It is the transpose of the domain fitness coefficient matrix M. It is the bandwidth fluctuation Laplace operator; Truncate the order of the series expansion; It represents the application priority standard deviation; n represents the nth order of the series expansion stage;

[0055] Based on the initial criticality score distribution, an initial threshold is set, and traffic with scores exceeding the threshold is filtered out and marked as potential critical traffic.

[0056] S12: Based on the scoring results and the latency tolerance threshold, output the key flow carrying the threshold, and accurately filter the key flow through dynamic optimization of the hybrid learning model.

[0057] The system receives potential critical flows and their initial criticality scores, while also introducing a latency tolerance threshold for each flow. The scores are correlated and matched with the latency tolerance thresholds to construct a comprehensive evaluation dataset containing flow features, criticality scores, and threshold parameters. A dynamic optimization module of a semi-supervised-transfer hybrid learning model is used to incrementally train the model with newly collected traffic data in real time, adjusting feature weights and scoring algorithm parameters to improve the adaptability of critical flow identification. Based on the optimized model, a second scoring calculation is performed on potential critical flows, and a double verification is performed using the latency tolerance thresholds to eliminate flows with scores that fluctuate excessively or fail to meet latency constraints. Finally, a precise set of critical flows with clearly defined latency tolerance thresholds is output.

[0058] S2: Collect fine-grained network status data, construct a full-dimensional network twin model, predict congestion risk trends, and verify link path survival rate;

[0059] The process of collecting fine-grained network status data, constructing a full-dimensional network twin model, predicting congestion risk trends, and verifying link path survival rates includes the following sub-steps:

[0060] S21: Collect fine-grained network state data, construct a full-dimensional network twin model, and create a high-precision digital mapping through multi-dimensional data fusion;

[0061] Fine-grained network status data is collected through distributed monitoring nodes, encompassing port traffic, CPU utilization, memory usage, link bandwidth utilization, packet loss rate, latency, and topology connections for each device. This data is then standardized to form a multi-source heterogeneous dataset. Based on this dataset, a full-dimensional network twin model is constructed. A hierarchical architecture of physical, logical, and service layers is first established. The physical layer maps network devices and link entities, the logical layer abstracts routing protocols and forwarding rules, and the service layer associates the application traffic characteristics it carries. Spatiotemporal correlation is performed on device status, link performance, and topology data to eliminate data redundancy and conflicts. A dynamic mapping mechanism binds real-time collected data to corresponding entities in the twin model, achieving a high-precision mapping from physical network status to the digital space.

[0062] S22: Utilize a full-dimensional network twin model to predict congestion risk trends and verify link path survival rates. Output reliability indicators through real-time simulation and predictive analysis to guide path selection.

[0063] The system invokes a full-dimensional network twin model, inputs real-time collected fine-grained network status data and traffic change trends, and starts a dynamic simulation engine. Time series analysis algorithms are used to extrapolate trends in link bandwidth and node load indicators, predicting the congestion risk level and propagation path of each link in the future. Simultaneously, based on graph theory-based reachability analysis, the system simulates link connectivity under various fault scenarios in the full-dimensional network twin model, calculates the survival probability of different paths within a preset time window, and verifies the link path survival rate.

[0064] Among them, link path survival rate The formula is:

[0065]

[0066] Where T represents the total duration of the time window used to analyze the link path status; m represents the number of links that make up the path; Let represent the instantaneous reliability of the i-th link at time t, reflecting the stability of the link over time; n1 is the number of time steps after discretization. It is the interval between adjacent time steps; Let f be the failure rate of the path at the k-1th time step; It is the matrix trace operation, which sums the elements on the main diagonal of a matrix. This is a topology connectivity matrix, used to describe the connectivity relationships between links and nodes in the network topology; It is a node processing capability matrix; It is its inverse matrix; This is the Hadamard product, which integrates topological connectivity information with node processing capability information through this operation; It is the fault recovery rate at the path endpoint L; It is the fault recovery rate at the starting point of the path; This reflects changes in the path fault recovery capability; It is an approximate calculation of the integral of the standard normal distribution, used to quantify the probabilistic characteristics of risks such as path congestion; x is the standardized risk statistic, derived from the path congestion probability. Mean congestion probability Standard deviation The calculation shows that u is the integral variable of the standard normal distribution.

[0067] The simulation results are integrated with the predicted data, and statistical analysis is used to generate reliability indicators such as path availability time, fault recovery probability, and congestion probability. These indicators are then quantified into evaluation parameters that can be directly used for path decision-making.

[0068] S3: Based on the critical flow and link path survival rate, generate a low-latency source node list through the critical flow path enhancement algorithm;

[0069] The process of generating a low-latency source node list based on the critical flow and link path survival rate includes the following sub-steps:

[0070] S31: Integrate the latency tolerance threshold of critical flows with the link path survival rate to construct a weighted directed graph model, and use a topology mining algorithm to filter effective source nodes;

[0071] The system receives critical flows and their criticality scores, integrates link data and link path survival rates, and constructs a weighted directed graph model after correlation calibration and standardization. In the weighted directed graph model, nodes correspond to network source nodes and forwarding nodes, edges represent physical links, and edge weights are generated by a weighted fusion algorithm of critical flow criticality scores and link survival rates. Simultaneously, potential source nodes are inferred based on network topology features. Combining the communication characteristics of critical flows and the reachability constraints of network topology, a topology mining algorithm is used to identify candidate source nodes that match the critical flows. By verifying the link path survival rate and criticality scores of the nodes, effective source nodes with the ability to carry critical flows are selected, and their latency sensitivity level and initial path weight values ​​are labeled.

[0072] The source node selection formula is as follows:

[0073]

[0074] Where K is the criticality score of the critical flow, used to measure the criticality of the traffic; R represents the node resource reserve, reflecting the remaining resources of the node to handle traffic; and U is the node resource utilization rate. Ψ(t) is the resource sensitivity coefficient; S is the link path survival rate; T represents the length of the time observation window for source node screening; Ψ(t) is the node time-varying availability function, which is the availability of the source node at time t; D is the link delay jitter, which reflects the fluctuation of link transmission delay. Node association matrix With path stability matrix The largest eigenvalue of the matrix after transpose and multiplication is a comprehensive eigenvalue that reflects node association and path stability; For time delay jitter variance; It is the path length. Used to quantify the negative impact of factors such as link latency jitter on source node selection; Historical failure rate of the node; It is the vector of topological coupling coefficients; It is the topological coupling index; is the Fourier transform of the distance attenuation factor; j is the imaginary unit; f is the frequency variable; r is the distance variable.

[0075] S32: Trigger the critical flow path enhancement algorithm through the source node, perform backtracking pruning inference in the weighted directed graph model, and generate a low-latency source node sequence list that satisfies the critical flow service level protocol.

[0076] Based on a weighted directed graph model, a critical flow path enhancement algorithm is triggered. Using the service level agreement (SLA) constraints of the critical flow as pruning conditions, a backtracking search is performed in the graph. By dynamically evaluating the cumulative weight of paths and constraint satisfaction, substandard path branches are pruned and removed, retaining effective paths with low latency and high survival rates. Finally, the starting point information of eligible paths is integrated to generate a priority-sorted list of low-latency source node sequences.

[0077] S4: Parse the list of low-latency source nodes to create a micro-segmentation queue, reserve resources for critical flows, and guide non-critical flows to low-load links;

[0078] The process of parsing the list of low-latency source nodes to create a micro-segmentation queue, reserving resources for critical flows, and simultaneously directing non-critical flows to low-load links includes the following sub-steps:

[0079] S41: Parse the list of low-latency source nodes, construct a micro-isolation queue container at the ingress node, and inject a priority forwarding identifier through the segment route header to reserve bandwidth and time slot resources;

[0080] The system receives low-latency source node sequences, parses their structure, and extracts the identification information, priority order, and corresponding critical flow service level protocol parameters for each source node. Based on the parsing results, a micro-segmentation queue container is deployed at the network ingress node. Independent queues are created based on the priority and latency sensitivity level of the critical flows, and resource occupancy thresholds and scheduling weights are configured for each queue. Simultaneously, according to the segment routing protocol specification, an SRv6 header containing a priority forwarding identifier is generated. This identifier is injected into the header of critical flow packets through a protocol conversion interface, establishing a mapping relationship between critical flows and reserved resources. Through interaction with the resource management module of the ingress node, bandwidth resource reservation and time slot scheduling table configuration are completed based on the bandwidth requirements and time slot allocation strategies of the critical flows, ensuring that critical flows receive priority processing rights during forwarding.

[0081] S42: Generate non-critical flow redirection strategies based on real-time link load data, and redirect non-critical traffic to low-load paths by combining the entropy value calculation of equivalent multipath.

[0082] The system collects load data for each link in real time, including bandwidth utilization, packet queuing length, and forwarding latency. This data is then normalized using a data preprocessing module to construct a link load state matrix. A non-critical flow redirection strategy framework is generated, clearly defining the forwarding constraints and path selection range for non-critical flows. Entropy values ​​are calculated for the load balancing of each candidate path; higher entropy values ​​indicate a more balanced load distribution. Based on the calculation results, the system selects low-load paths with acceptable entropy values ​​and redirects non-critical flow traffic to these paths. The system monitors the load changes of the redirected links in real time. If load fluctuations exceed a threshold, a dynamic policy adjustment mechanism is triggered to ensure that non-critical flows are always transmitted on low-load links, avoiding the crowding out of resources reserved for critical flows.

[0083] S5: When the digital twin detects that the actual state deviates from the predicted value, it calls the pre-verified backup path identifier chain to complete the path switching in real time;

[0084] Specifically, when the digital twin detects that the actual state deviates from the predicted value, it invokes the pre-verified backup path identifier chain to complete the path switch in real time, including the following sub-steps:

[0085] S51: By continuously comparing the network status with the predicted value through a digital twin, when the key indicator exceeds the dynamic threshold, a path switching command based on the network service chain is activated.

[0086] Through a full-dimensional network twin, key operational status data of the physical network is collected in real time and continuously compared with the predicted values ​​generated by the twin model. A dynamic threshold adjustment mechanism is set up to update the deviation tolerance thresholds of various indicators in real time based on the service level protocol parameters of key flows and the current network load. When the comparison results show that a certain key indicator exceeds the corresponding dynamic threshold and the deviation lasts for more than a preset window, it is determined that the network status deviates from the predicted trend and may affect the transmission quality of key flows. Then, the path switching instruction generation module based on the network service chain is activated, and a suitable switching strategy is initially selected based on the identification information of the key flows and the current path characteristics.

[0087] S52: Retrieve the pre-stored backup path identifier chain and perform segment route identifier stack switching operation to complete traffic rerouting.

[0088] In response to a path switching command, the system retrieves a backup path identifier chain matching the current critical flow from a pre-defined backup path library. This identifier chain contains a pre-verified SRv6 segment routing identifier sequence and corresponding node forwarding rules. The identifier chain is validated for legality. After confirming that its link liveness and latency parameters meet the service level agreement requirements for the critical flow, the segment routing identifier stack switching mechanism is initiated. Through interaction with the forwarding plane of nodes along the route, the identifier stack information of the original path is replaced sequentially, and a path switching announcement is sent to relevant nodes to update forwarding table entries. During the switching process, the transmission status of the critical flow is monitored in real time, and a smooth transition algorithm is used to avoid traffic interruptions or jitter. Once the traffic carried by the new path is stable and the indicators meet the standards, the forwarding process of the original path is terminated, completing the seamless rerouting of the critical flow.

[0089] S6: Utilize a dual-channel learning model to optimize a semi-supervised-transfer hybrid learning model, a full-dimensional network twin model, and a key flow path enhancement algorithm based on closed-loop optimization of key operational status data.

[0090] The system collects key operational status data, including at least the accuracy deviation of critical flow identification, the prediction error of the full-dimensional network twin model, and the matching degree between the output of the path enhancement algorithm and the service level agreement. After standardization, this data is input into a dual-channel learning model. One channel uses a graph neural network to extract the correlation features between each model and algorithm and the network dynamics, analyzing the feature extraction deviation of the semi-supervised-transfer model, the topology mapping error of the twin model, and the constraint adaptation defects of the path enhancement algorithm. The other channel uses linear programming to model the constraint relationships of each optimization objective and quantifies the range of parameters that need to be adjusted. Through dual-channel fusion analysis, feature weight correction values ​​for the semi-supervised-transfer model, dynamic mapping coefficient adjustment amounts for the twin model, and pruning threshold optimization parameters for the path enhancement algorithm are generated. These parameters are then back-injected into the corresponding models and algorithms to initiate iterative optimization. By continuously comparing the deviation between the optimized output and the expected target, the optimization strategy of the dual-channel learning model is dynamically adjusted to form a closed-loop mechanism.

[0091] Example 2, as Figure 2 As shown, Embodiment 2 of this application provides an SRv6 explicit path intelligent scheduling system based on critical flow, including:

[0092] Key Flow Intelligent Identification Module 21: Used to collect multi-dimensional dynamic features of traffic, perform keyness scoring through a semi-supervised-transfer hybrid learning model, and output key flows carrying latency tolerance thresholds;

[0093] Congestion prediction and path verification module 22: used to collect fine-grained network status data, build a full-dimensional network twin model, predict congestion risk trends, and verify link path survival rate;

[0094] Low-latency source node generation module 23: used to generate a list of low-latency source nodes based on critical flow and link path survival rate through a critical flow path enhancement algorithm;

[0095] Critical Flow Traffic Scheduling Module 24: Used to parse the list of low-latency source nodes, create micro-segmentation queues, reserve resources for critical flows, and guide non-critical flows to low-load links;

[0096] Fault detection and path switching module 25: When the digital twin detects that the actual state deviates from the predicted value, it calls the pre-verified backup path identifier chain to complete the path switching in real time;

[0097] Iterative optimization module 26: Used to optimize the semi-supervised-transfer hybrid learning model, the full-dimensional network twin model, and the key flow path enhancement algorithm based on the closed loop of key operating state data using a dual-channel learning model.

[0098] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0099] The memory is used to store one or more program instructions;

[0100] The processor is used to run one or more program instructions to execute the critical flow-based SRv6 explicit path intelligent scheduling method.

[0101] Corresponding to the above embodiments, the present invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor as a critical flow-based SRv6 explicit path intelligent scheduling method.

[0102] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions that, when executed on a computer, cause the computer to perform the aforementioned critical flow-based SRv6 explicit path intelligent scheduling method.

[0103] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0104] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0105] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0106] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0107] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM).

[0108] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0109] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0110] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

[0111] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for SRv6 explicit path intelligent scheduling based on key flow, characterized in that: The method comprises: Collecting multi-dimensional dynamic characteristics of traffic, performing key degree scoring through a semi-supervised-transfer hybrid learning model, and outputting key flows carrying a time delay tolerance threshold; Collecting network fine-grained state data, constructing a full-dimensional network twin model, predicting congestion risk trends, and verifying link path survival rates; Based on the key flow and the link path survival rate, a low time delay source node list is generated through a key flow path enhancement algorithm; Analyzing the low time delay source node list to create a micro-isolation queue, reserving resources for key flows, and guiding non-key flows to low-load links; When the digital twin detects that the actual state deviates from the predicted value, the pre-verified backup path identification chain is called to complete path switching in real time; Using a dual-channel learning model, based on key operating state data, the semi-supervised-transfer hybrid learning model, the full-dimensional network twin model, and the key flow path enhancement algorithm are closed-loop optimized. 2.The key flow based SRv6 explicit path intelligent scheduling method according to claim 1, characterized in that, Collecting multi-dimensional dynamic characteristics of traffic, performing key degree scoring through a semi-supervised-transfer hybrid learning model, and outputting key flows carrying a time delay tolerance threshold, comprising: Collecting multi-dimensional dynamic characteristics of traffic and inputting them into a semi-supervised-transfer hybrid learning model to perform preliminary key degree scoring and identify potential key flows; Based on the scoring results and the time delay tolerance threshold, key flows carrying the threshold are outputted, and key flow screening is performed through dynamic optimization of the semi-supervised-transfer hybrid learning model. 3.The key flow based SRv6 explicit path intelligent scheduling method according to claim 1, characterized in that, Collecting network fine-grained state data, constructing a full-dimensional network twin model, predicting congestion risk trends, and verifying link path survival rates, comprising: Collecting network fine-grained state data, constructing a full-dimensional network twin model, and creating a high-precision digital mapping through multi-dimensional data fusion; Using the full-dimensional network twin model to predict congestion risk trends and verify link path survival rates, outputting reliability indicators through real-time simulation and prediction analysis to guide path selection. 4.The key flow based SRv6 explicit path intelligent scheduling method according to claim 1, characterized in that, Based on the key flow and the link path survival rate, a low time delay source node list is generated through a key flow path enhancement algorithm, comprising: Fusing the time delay tolerance threshold of the key flow and the link path survival rate, constructing a weighted directed graph model, and using a topology mining algorithm to screen effective source nodes; Triggering the key flow path enhancement algorithm through the source nodes, performing backtracking pruning and deduction in the weighted directed graph model, and generating a low time delay source node list that meets the key flow service level agreement.

5. The key flow based SRv6 explicit path intelligent scheduling method according to claim 1, characterized in that, Analyzing the low time delay source node list to create a micro-isolation queue, reserving resources for key flows, and guiding non-key flows to low-load links, comprising: Analyzing the low time delay source node list, constructing a micro-isolation queue container at the entry node, and injecting a priority forwarding identifier into the segment routing header to reserve bandwidth and time slot resources; Generating a non-key flow guidance strategy based on real-time link load data, and redirecting non-key flows to low-load paths by combining equivalent multipath entropy calculation.

6. The key flow based SRv6 explicit path intelligent scheduling method according to claim 1, characterized in that, When the digital twin detects that the actual state deviates from the predicted value, the pre-verified backup path identification chain is called to complete path switching in real time, comprising: Through continuous comparison of the actual network state and the predicted value by the digital twin, when the key indicators break through the dynamic threshold, the path switching instruction based on the network service chain is activated; Calling the pre-stored backup path identification chain, performing segment routing identifier stack switching operation to complete flow rerouting. 7.The key flow based SRv6 explicit path intelligent scheduling method according to claim 2, characterized in that, The multi-dimensional dynamic characteristic input of the flow is collected to input a semi-supervised-transfer hybrid learning model to perform preliminary key degree scoring and identify potential key flows; specifically including: The multi-dimensional dynamic characteristics of the flow are collected, including bandwidth fluctuation, transmission delay change, protocol type, application identification and interaction frequency of the flow, and a feature matrix is formed after standardization preprocessing; the feature matrix is input into a semi-supervised-transfer hybrid learning model, wherein the semi-supervised module uses a small amount of labeled samples and a large amount of unlabeled flow data to construct a feature space mapping, and the internal correlation mode of the flow is mined through a clustering algorithm; the transfer learning module introduces the key flow identification knowledge in the similar network scene, and adjusts the model parameters through the domain adaptation mechanism to adapt to the current network environment; the semi-supervised-transfer hybrid learning model performs hierarchical processing on the input features, first captures the key feature vector through the feature extraction layer, and then outputs the preliminary key degree score of each flow through the scoring layer: the key degree score K calculation formula is: ； wherein, and defines the time window boundary of the traffic feature collection, is the collection start time, is the collection end time, used to limit the observation length of the traffic dynamic feature; embodies the instantaneous change rate of bandwidth fluctuation, where B represents the bandwidth, indicating the amount of data that can be transmitted per unit time, and t represents the current t time; is the transmission delay change rate; is the protocol key coefficient; represents the application priority; is the interaction frequency; is the feature space mapping distance; is the feature space standard deviation; wherein, is the matrix trace operation, is the labeled sample confidence matrix, is the transpose of the domain adaptation coefficient matrix M, is the bandwidth fluctuation Laplace operator; is the series expansion truncation order; is the application priority standard deviation; n represents the nth order of the series expansion stage. According to the preliminary key degree score distribution, a preliminary threshold is set, and the flow with a score exceeding the threshold is marked as a potential key flow. 8.The key flow based SRv6 explicit path intelligent scheduling method according to claim 3, characterized in that, The full-dimensional network twin model is used to predict the congestion risk trend and verify the link path survival rate, and the reliability indicators are output through real-time simulation and prediction analysis to guide path selection; specifically including: The full-dimensional network twin model is called, the real-time collected network fine-grained state data and flow change trend are input, and the dynamic simulation engine is started; the trend of link bandwidth and node load indicators is deduced through time series analysis algorithm, the congestion risk level and spread path of each link in the future period are predicted; at the same time, based on the reachability analysis method of graph theory, the link connectivity in various fault scenarios is simulated in the full-dimensional network twin model, and the survival probability of different paths in the preset time window is calculated to verify the link path survival rate; Wherein, the link path survival rate The formula is: ； wherein T represents the total length of the time window for analyzing the link path state; m represents the number of links constituting the path; is the instantaneous reliability of the ith link at time t, reflecting the stability degree of the link over time; n1 is the number of time steps after discretization; is the interval of adjacent time steps; is the failure rate of the path at the corresponding time of the k1th time step; is the matrix trace operation, summing the main diagonal elements of the matrix; is the topological connectivity matrix, used to describe the connectivity relationship of links and nodes in the network topology; is the node processing capacity matrix; is the inverse matrix thereof; is the Hadamard product, which integrates the topological connectivity information and the node processing capacity information through the operation; is the failure recovery rate at the end L of the path; is the failure recovery rate at the start of the path; reflects the change in the path failure recovery capability; is the approximate calculation of the integral of the standard normal distribution, used to quantify the probability characteristics of the path congestion risk and the like; , x is the normalized risk statistic, calculated by the path congestion probability , the mean , and the standard deviation of the congestion probability; u is the integral variable of the standard normal distribution; The simulation results and prediction data are fused, and statistical analysis is used to generate reliability indicators such as path available duration, fault recovery probability and congestion occurrence probability, and these indicators are quantified as evaluation parameters that can be directly used for path decision. 9.The key flow based SRv6 explicit path intelligent scheduling method according to claim 4, characterized in that, The time delay tolerance threshold of the key flow and the link path survival rate are fused to construct a weighted directed graph model, and a topological mining algorithm is used to select effective source nodes; specifically including: The key flow and its key degree score are received, the link data and the link path survival rate are fused, and after correlation calibration and standardization processing, a weighted directed graph model is constructed; in the weighted directed graph model, the nodes correspond to network source nodes and forwarding nodes, the edges represent physical links, and the edge weights are generated through the weighted fusion algorithm of the key degree score of the key flow and the link survival rate; at the same time, the potential source nodes are deduced based on the network topology characteristics; combined with the communication characteristics of the key flow and the reachability constraint of the network topology, a topological mining algorithm is used to identify the candidate source nodes matched with the key flow, the link path survival rate and the key degree score of the nodes are verified, the effective source nodes with the ability to carry the key flow are selected, and their time delay sensitivity levels and path weight initial values are marked; The source node selection formula is: ； Where K is the criticality score of the critical flow, used to measure the criticality of the traffic; R represents the node resource reserve, reflecting the remaining resources of the node to handle traffic; and U is the node resource utilization rate. Ψ(t) is the resource sensitivity coefficient; S is the link path survival rate; T represents the length of the time observation window for source node screening; Ψ(t) is the node time-varying availability function, which is the availability of the source node at time t; D is the link delay jitter, which reflects the fluctuation of link transmission delay. Node association matrix With path stability matrix The largest eigenvalue of the matrix after transpose and multiplication is a comprehensive eigenvalue that reflects node association and path stability; For time delay jitter variance; It is the path length. Used to quantify the negative impact of factors such as link latency jitter on source node selection; Historical failure rate of the node; It is the vector of topological coupling coefficients; It is the topological coupling index; is the Fourier transform of the distance attenuation factor; j is the imaginary unit; f is the frequency variable; r is the distance variable. 10.A system for key flow based SRv6 explicit path intelligent scheduling, characterized in that, The system includes a module for executing the key flow-based SRv6 explicit path intelligent scheduling method according to any one of claims 1 to 9.

Images