Intelligent traffic scheduling optimization method and device, equipment and storage medium

Abstract

The invention relates to the field of communication, and discloses an intelligent traffic scheduling optimization method, device and equipment and a storage medium, and the method is used for 5G network intelligent traffic scheduling optimization. The method comprises the following steps: receiving a service request, analyzing the service request to obtain a 5G standard service type, an application scene and a 5G network slice identifier, then obtaining an SLA parameter corresponding to the service request and a candidate link required by service transmission, and converting the SLA parameter into an SLA preference vector; acquiring global data of the candidate link, and predicting by using the network state prediction model to obtain a predicted network state of the candidate link; the SLA preference vector is used as optimization target guidance, the predicted network state is used as a constraint condition, and a global optimization routing strategy is calculated through a multi-target optimization algorithm; and generating a control message packet based on the global optimization routing strategy, issuing the control message packet to the network forwarding equipment, and adjusting parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data after execution.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to an intelligent traffic scheduling optimization method, apparatus, device, and storage medium. Background Technology

[0002] With the large-scale commercialization and diversified development of 5G networks, the IP bearer network, as the key link between the radio access network and the core network, faces unprecedented challenges. The three major application scenarios of 5G—enhanced mobile broadband (eMBB), ultra-reliable low-latency communication (uRLLC), and massive machine-type communication (mMTC)—place differentiated and stringent requirements on the bearer network: eMBB requires a peak bandwidth of 10Gbps, uRLLC requires end-to-end latency of <1ms and ultra-high reliability of 99.999%, and mMTC needs to support massive connections. Traditional IP bearer network traffic scheduling methods are no longer adequate, and the core issues are as follows: First, static configuration is incompatible with dynamic services. Traditional IP networks rely on statically configured IGP protocols (such as OSPF and IS-IS), which are based on the shortest path first algorithm and only consider the single factor of link cost, failing to perceive real-time service demands and network status changes. This "one-size-fits-all" strategy is difficult to meet the differentiated QoS requirements of 5G services. For example, when uRLLC and ordinary internet services are transmitted together, the low latency requirements of the former cannot be guaranteed.

[0003] Secondly, there is a disconnect between service priorities and network resources. Although the DiffServ model supports DSCP-based service classification and priority queue scheduling, it is limited to single-hop queue management and is not organically integrated with end-to-end routing. This may result in situations where the shortest path selected for high-priority services is already overloaded or at risk of congestion, thus reducing the quality of service.

[0004] Third, there is a lack of accurate traffic prediction and proactive optimization capabilities. 5G service traffic has strong bursts and spatiotemporal correlations (such as large-scale events and tidal effects). Existing methods cannot accurately predict traffic patterns and are unable to proactively adjust paths before congestion occurs, and can only passively adapt to network changes.

[0005] Fourth, the network slicing resource guarantee mechanism is imperfect. 5G network slicing requires the bearer network to provide isolated and guaranteed resources. Traditional IP networks find it difficult to achieve strict hard slice isolation, and dynamically allocating and adjusting slice bandwidth, latency, and other resources in cross-domain and cross-vendor environments faces enormous challenges.

[0006] In existing technologies, Chinese patent CN201910123456.7 proposes a traffic scheduling method based on SDN, which calculates routes by collecting global state data through a centralized controller, but still uses simple weight adjustment based on real-time utilization. US patent US2020156789A1 discloses a method for predicting traffic and adjusting routes using machine learning, but its prediction model is simple and does not fully consider the deep coupling between service intent and network state. These methods still have problems such as coarse optimization granularity, poor adaptability, and insufficient guarantee capability when dealing with the extreme and diverse needs of 5G bearer networks. Summary of the Invention

[0007] This invention provides an intelligent traffic scheduling optimization method, apparatus, device, and storage medium for intelligent traffic scheduling optimization in 5G networks.

[0008] The first aspect of this invention provides an intelligent traffic scheduling optimization method, comprising: receiving a service request; parsing the service request to obtain a 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request; obtaining SLA parameters and candidate links required for service transmission corresponding to the service request based on the 5G standard service type, the application scenario, and the 5G network slice identifier, and converting the SLA parameters into an SLA preference vector; obtaining global data of the candidate links and inputting the global data into a pre-trained network state prediction model to predict the predicted network state of the candidate links; calculating a global optimized routing strategy using the SLA preference vector as the optimization objective and the predicted network state as the constraint through a multi-objective optimization algorithm; verifying the security and effectiveness of the global optimized routing strategy; generating a control message packet based on the global optimized routing strategy after the global optimized routing strategy has been verified; sending the control message packet to the network forwarding device using a smooth migration mechanism, collecting actual performance data related to the execution effect of the control message packet, and adjusting the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data.

[0009] Preferably, the step of receiving a service request and parsing the service request to obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request includes: receiving the service request from a service orchestrator or network management system and parsing the service request to obtain the service intent and 5G network slice identifier; mapping the service intent to the 5G standard service type corresponding to the service request based on a predefined service feature library and rule engine, wherein the 5G standard service type includes enhanced mobile broadband, ultra-reliable low-latency communication, and massive machine-type communication; and determining the application scenario corresponding to the service request based on the context information in the service intent, wherein the application scenario includes virtual reality / augmented reality, industrial control, or live video streaming.

[0010] Preferably, the step of obtaining the SLA parameters corresponding to the service request and the candidate links required for service transmission based on the 5G standard service type, the application scenario, and the 5G network slice identifier, and converting the SLA parameters into an SLA preference vector, includes: matching and extracting key SLA parameters from a preset SLA template library based on the 5G standard service type and the application scenario, wherein the key SLA parameters include maximum allowable latency, maximum allowable latency jitter, minimum required bandwidth, target bandwidth, reliability level, and service lifetime; determining the set of physical or logical links preferred for the service request as candidate links based on the 5G network slice identifier and a preset network topology and slice resource mapping strategy; and using the analytic hierarchy process (AHP) to assign optimization weights to each dimension of the SLA parameters based on the 5G standard service type and the application scenario, and converting them into an SLA preference vector.

[0011] Preferably, the step of acquiring the global data of the candidate links and inputting the global data into a pre-trained network state prediction model to predict the predicted network state of the candidate links includes: acquiring real-time state data of the candidate links and their associated network devices, wherein the real-time state data includes link utilization, latency, packet loss rate, error frame rate, node resource utilization, and queue depth; preprocessing the acquired real-time state data and constructing a spatiotemporal data sequence based on the preprocessed real-time state data; constructing and training a network state prediction model based on a spatiotemporal graph convolutional network and a Transformer architecture, and inputting the spatiotemporal data sequence into the network state prediction model; and using the network state prediction model to continuously output the predicted network state of each candidate link within the next 5 to 30 minutes, wherein the predicted network state includes predicted load, predicted latency, and predicted packet loss rate.

[0012] Preferably, the step of calculating the global optimized routing strategy using the SLA preference vector as the optimization objective and the predicted network state as the constraint through a multi-objective optimization algorithm includes: constructing a multi-objective optimization problem based on a multi-commodity flow model, with the objectives of minimizing the weighted deviation of the SLA preference vector, minimizing the maximum link utilization, and minimizing route changes; and using the predicted network state as the capacity and performance constraint for each link within the future time window. The service flows with hard latency requirements in the SLA preference vector are identified. Based on the capacity and performance constraints, K candidate feasible paths that meet the latency requirements of each such service flow are calculated using the constrained shortest path algorithm. Then, for the remaining network resources after satisfying the hard constraints and deducting the resources occupied by the hard-constrained flow, a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem for the remaining flexible service flows, generating a Pareto optimal solution set. Based on a preset decision strategy, each solution in the Pareto optimal solution set is comprehensively scored, and the solution with the highest comprehensive score is selected as the global optimization routing strategy. The global optimization routing strategy defines the primary path, backup path, and corresponding bandwidth allocation ratio for all service flows or service aggregation flows to be scheduled in the network.

[0013] Preferably, the step of verifying the security and effectiveness of the global optimized routing strategy, and generating a control message packet based on the global optimized routing strategy after the verification is passed, includes: using graph theory algorithms to detect whether the global optimized routing strategy has routing loops, conflicts with existing security policies, or causes node overload; if the global optimized routing strategy does not have routing loops, conflicts with existing security policies, or causes node overload, then simulating the execution of the global optimized routing strategy in a preset digital twin network environment to estimate the impact of the global optimized routing strategy on key network performance indicators; if the improvement of key performance indicators reaches the expected threshold, then determining that the global optimized routing strategy has passed verification; converting the verified global optimized routing strategy into control instructions, and encapsulating the control instructions into control message packets.

[0014] Preferably, the step of using a smooth migration mechanism to send the control message packet to the network forwarding device, collecting actual performance data related to the execution effect of the control message packet, and adjusting the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data includes: using a multi-stage activation mechanism to send the control message packet to the network forwarding device and collecting actual performance data related to the execution effect of the control message packet; calculating the quantized deviation between the measured values ​​of each service quality parameter in the actual performance data and the corresponding expected service quality target values ​​in the SLA preference vector, and evaluating the SLA compliance of each scheduled service flow or service aggregation flow in the current network based on the quantized deviation; if the SLA compliance is lower than a predetermined threshold or the prediction error is consistently high, then using the actual performance data related to the execution effect of the control message packet to fine-tune the parameters of the network state prediction model and the multi-objective optimization algorithm online.

[0015] A second aspect of the present invention provides an intelligent traffic scheduling optimization device, comprising: a parsing module, configured to receive a service request, parse the service request, and obtain a 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request; a conversion module, configured to obtain SLA parameters and candidate links required for service transmission corresponding to the service request based on the 5G standard service type, the application scenario, and the 5G network slice identifier, and convert the SLA parameters into an SLA preference vector; and a prediction module, configured to obtain full-domain data of the candidate links, input the full-domain data into a pre-trained network state prediction model, and predict the prediction of the candidate links. The network state prediction model includes an optimization module, which uses the SLA preference vector as the optimization objective and the predicted network state as the constraint to calculate a global optimized routing strategy using a multi-objective optimization algorithm. A verification module verifies the security and effectiveness of the global optimized routing strategy; after successful verification, it generates a control message packet based on the global optimized routing strategy. An adjustment module uses a smooth migration mechanism to send the control message packet to the network forwarding device, collects actual performance data related to the execution effect of the control message packet, and adjusts the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data.

[0016] A third aspect of the present invention provides an intelligent traffic scheduling optimization device, comprising: a memory and at least one processor, wherein the memory stores computer-readable instructions, and the memory and the at least one processor are interconnected via a line; the at least one processor invokes the computer-readable instructions in the memory to cause the intelligent traffic scheduling optimization device to perform the various steps of the intelligent traffic scheduling optimization method described above.

[0017] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-readable instructions that, when executed on a computer, cause the computer to perform the steps of the intelligent traffic scheduling optimization method described above.

[0018] The technical solution provided by this invention extracts 5G standard service types, application scenarios, and slice identifiers from service requests, enabling precise anchoring of service requirements and scheduling resources and avoiding resource mismatch. Simultaneously, a pre-trained network state prediction model is introduced to predict the network state for a preset time period based on the full-domain data of candidate links, upgrading routing strategies from traditional static adjustments to proactive planning and mitigating network congestion risks in advance. Furthermore, a multi-objective optimization algorithm is designed based on SLA preference vectors, balancing multiple objectives such as service SLA compliance, balanced network resource utilization, and routing stability, avoiding the one-sidedness of single-objective optimization. In addition, dual verification for security and effectiveness is added. Graph theory algorithms and digital twin simulations are used to proactively identify risks such as routing loops and node overload. A smooth migration mechanism is adopted to distribute strategies in stages, and the model and algorithm parameters are fine-tuned using actual performance data, ensuring strategy feasibility and enabling continuous iteration of scheduling capabilities. Attached Figure Description

[0019] Figure 1 A flowchart of the intelligent traffic scheduling optimization method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the intelligent traffic scheduling and optimization device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the intelligent traffic scheduling and optimization device provided in an embodiment of the present invention. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The intelligent traffic scheduling optimization method in this embodiment of the invention includes: S101. Receive a service request, parse the service request, and obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request. S102. Obtain the SLA parameters corresponding to the service request and the candidate links required for service transmission based on the 5G standard service type, application scenario and 5G network slice identifier, and convert the SLA parameters into an SLA preference vector. S103. Obtain the global data of the candidate links and input the global data into the pre-trained network state prediction model to predict the predicted network state of the candidate links. S104. Using the SLA preference vector as the optimization objective and the predicted network state as the constraint, a global optimal routing strategy is calculated through a multi-objective optimization algorithm. S105. Verify the security and effectiveness of the global optimization routing policy. Once the global optimization routing policy is verified, generate a control message packet based on the global optimization routing policy. S106. A smooth migration mechanism is used to send control message packets to network forwarding devices, and actual performance data related to the execution effect of the control message packets is collected. Based on the actual performance data, the parameters of the network state prediction model and the multi-objective optimization algorithm are adjusted.

[0022] It is understood that the executing entity of this invention can be an intelligent traffic scheduling and optimization device, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0023] In this embodiment, by parsing service requests to extract 5G standard service types, application scenarios, and slice identifiers, the precise anchoring of service requirements and scheduling resources is achieved, avoiding resource mismatch. Simultaneously, a pre-trained network state prediction model is introduced to predict the network state for a preset time period based on the full-domain data of candidate links, upgrading the routing strategy from traditional static adjustment to proactive planning and avoiding network congestion risks in advance. Furthermore, a multi-objective optimization algorithm is designed based on SLA preference vectors to balance multiple objectives such as service SLA compliance, balanced network resource utilization, and routing stability, avoiding the one-sidedness of single-objective optimization. In addition, dual verification for security and effectiveness is added. Graph theory algorithms and digital twin simulations are used to proactively identify risks such as routing loops and node overload. A smooth migration mechanism is adopted to distribute strategies in stages, and the model and algorithm parameters are fine-tuned using actual performance data, ensuring the feasibility of the strategy and enabling continuous iteration of scheduling capabilities.

[0024] In this embodiment, step S101 involves receiving a service request, parsing the service request to obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request, including: receiving the service request from a service orchestrator or network management system, and parsing the service request to obtain the service intent and 5G network slice identifier; mapping the service intent to the 5G standard service type corresponding to the service request based on a predefined service feature library and rule engine, wherein the 5G standard service type includes enhanced mobile broadband, ultra-reliable low-latency communication, and massive machine-type communication; determining the application scenario corresponding to the service request based on the context information in the service intent, wherein the application scenario includes virtual reality / augmented reality, industrial control, or live video streaming. By constructing a mapping mechanism of a predefined service feature library and rule engine, the abstract service intent is transformed into a standard 5G standard service type. At the same time, through standardized feature matching of the feature library and automated judgment of the rule engine, the subjectivity and error of manual classification are completely avoided. Moreover, by deeply mining the context information in the service intent to define the application scenario, the parsing results not only cover the attributes of 5G standard service types but also better meet the needs of actual operating scenarios.

[0025] In this embodiment, service requests are first received from the service orchestrator or network management system through a pre-defined standardized interface (such as RESTful API or NETCONF). These service requests can be service activation requests or service change requests. Then, the received request data undergoes format validation and legality verification, eliminating abnormal requests with incorrect formats or without valid identifiers. For legitimate requests, message parsing is performed. The 5G network slice identifier is extracted by parsing the identifier field in the message header, and the service intent is extracted by parsing the semantic information in the message payload. This clarifies the core transmission objectives that the service needs to achieve (such as high-speed data transmission and low-latency control). The extracted information is then structured and stored to provide foundational data for subsequent mapping and definition work. In this embodiment, a service feature library is pre-built. The library stores core feature parameters according to three types of 5G standard services: eMBB, uRLLC, and mMTC. These parameters include transmission rate thresholds, latency limits, connection density requirements, and reliability levels.

[0026] In this embodiment, a rule engine is pre-built with built-in feature matching rules and priority determination logic. The extracted business intent is broken down into specific transmission requirement features, which are then compared with standard features in the business feature library by the rule engine. If the feature matching degree of a certain type of standard business reaches a preset threshold (e.g., above 85%), then that type is used as the mapping result, realizing the automated and accurate conversion of business intent to standard type. In this embodiment, the contextual information in the business intent is analyzed in depth, focusing on extracting key information such as terminal type, geographical location of business initiation, type of transmitted content, and service object. Based on this information, a scenario judgment rule base is constructed. For example, if the terminal type is a VR / AR device and the transmitted content is high-definition video, it is defined as a virtual reality / augmented reality scenario; if the terminal is an industrial control sensor and the transmitted content is control commands, it is defined as an industrial control scenario; if the terminal is a live streaming device and the transmitted content is a real-time video stream, it is defined as a live video streaming scenario.

[0027] In this embodiment, step S102 involves obtaining the SLA parameters corresponding to the service request and the candidate links required for service transmission based on the 5G standard service type, application scenario, and 5G network slice identifier, and converting the SLA parameters into an SLA preference vector. This includes: matching and extracting key SLA parameters from a preset SLA template library based on the 5G standard service type and application scenario. The key SLA parameters include maximum allowable latency, maximum allowable latency jitter, minimum required bandwidth, target bandwidth, reliability level, and service lifetime. Based on the 5G network slice identifier and combined with a preset network topology and slice resource mapping strategy, a set of physical or logical links preferred for the service request is determined as candidate links. Using the analytic hierarchy process (AHP), optimization weights are assigned to each dimension of the SLA parameters based on the 5G standard service type and application scenario, and then converted into an SLA preference vector.

[0028] In this embodiment, a pre-built SLA template library is constructed, and SLA templates are stored according to the combination dimension of 5G standard service types and application scenarios. Each template contains the core SLA parameters for service transmission in that scenario, specifically covering key indicators such as maximum allowable latency, maximum allowable latency jitter, minimum required bandwidth, target bandwidth, reliability level, and service lifetime. Based on the 5G standard service type and application scenario, a precise search and matching is performed in the template library to locate the appropriate SLA template. The parameters in the template are fine-tuned according to the specific transmission requirements of the service (such as the amount of data transmitted and the service object level), and finally, a set of key SLA parameters that fit the current service is extracted.

[0029] In this embodiment, a preset network topology map and slice resource mapping strategy are first retrieved. The slice resource mapping strategy specifies the physical resource pool, logical link range, and QoS guarantee standard corresponding to each 5G network slice. Based on the 5G network slice identifier, all physical and logical links belonging to the resource pool range of that slice are selected from the network topology map. The availability of the selected links is verified, and links that are in a faulty state, have a load rate exceeding a preset threshold (e.g., 80%), or do not meet the slice QoS requirements are removed using real-time status monitoring data. The verified links are integrated to form a link set, which serves as candidate links for current service transmission. In this embodiment, a weight allocation model is constructed using the analytic hierarchy process (AHP), with 5G standard service types and application scenarios as core influencing factors, and each SLA parameter dimension as an evaluation indicator. First, a judgment matrix is ​​constructed, and the relative importance of each SLA parameter in the current scenario is determined through pairwise comparisons. The eigenvectors of the judgment matrix are calculated to obtain the initial weights of each parameter. A consistency check is performed on the initial weights; if the check passes, they are determined as the final optimized weights; otherwise, the judgment matrix is ​​adjusted and recalculated. The standardized values ​​of each SLA parameter are multiplied by their corresponding weights, and combined in a preset order to form an SLA preference vector, thus completing the quantitative modeling of the scheduling objective. In this embodiment, key parameters are matched from a pre-set SLA template library based on the dual dimensions of 5G standard service types and application scenarios. The template library pre-sets targeted parameter standards such as latency and bandwidth for different scenarios to ensure that SLA requirements are highly adapted to actual service needs and avoid parameter selection chaos or being out of touch with reality. Moreover, candidate links are screened by combining 5G network slice identifiers, and services are bound to dedicated link resources through slice resource mapping strategies to ensure that link resources meet the QoS guarantee requirements of slices, while eliminating faulty and high-load links to improve link availability. In addition, the hierarchical analysis method is used to assign optimized weights to SLA parameters, fully considering the priority differences of each parameter under different service scenarios (such as latency weight being higher than bandwidth in industrial control scenarios), transforming the scattered multi-dimensional SLA requirements into a unified quantitative preference vector.

[0030] In this embodiment, step S103 involves acquiring the global data of candidate links and inputting the global data into a pre-trained network state prediction model to predict the predicted network state of the candidate links. This includes: acquiring real-time state data of the candidate links and their associated network devices, where the real-time state data includes link utilization, latency, packet loss rate, error frame rate, node resource utilization, and queue depth; preprocessing the acquired real-time state data and constructing a spatiotemporal data sequence based on the preprocessed real-time state data; constructing and training a network state prediction model based on a spatiotemporal graph convolutional network and a Transformer architecture, and inputting the spatiotemporal data sequence into the network state prediction model; and using the network state prediction model to continuously output the predicted network state of each candidate link within the next 5 to 30 minutes, where the predicted network state includes predicted load, predicted latency, and predicted packet loss rate.

[0031] In this embodiment, a comprehensive data acquisition system covering candidate links and associated network devices (routers, switches, base stations, etc.) is constructed. The acquisition metrics are clearly defined, including link-level utilization, latency, packet loss rate, and error frame rate, as well as node-level CPU utilization, memory usage, and queue depth. Lightweight acquisition protocols (such as SNMP and sFlow) are used to collect data in real-time at preset acquisition intervals (1-5 seconds) to avoid interfering with normal network operation. Then, the collected data is initially filtered through edge nodes to remove outliers, duplicates, and invalid data. The filtered valid data is then transmitted to the data processing center for centralized storage. In this embodiment, the collected valid data undergoes standardized preprocessing. The Min-Max normalization method is used to transform indicators with different dimensions (such as latency in milliseconds or utilization rate as a percentage) to the [0,1] interval, avoiding the impact of dimensional differences on model prediction performance. Then, missing values ​​are filled into the preprocessed data. Linear interpolation is used to fill in short-term missing data, and moving average is used to fill in long-term missing data, ensuring data continuity. Finally, a spatiotemporal data sequence is constructed based on the data timestamps and network topology. The time dimension is sorted according to the collection period to form a one-dimensional time series, and the spatial dimension associates candidate links and associated node data under the same topology to form a two-dimensional spatial matrix, ultimately constructing a three-dimensional spatiotemporal data sequence encompassing time, space, and indicators. In this embodiment, a network state prediction model is constructed based on a Spatiotemporal Graph Convolutional Network (STGCN) and a Transformer architecture. The STGCN module is used to capture the spatial correlation features of network states (such as the transmission relationship of load changes between adjacent links), while the Transformer module captures long-term temporal dependency features (such as the variation pattern of link load during peak periods) through a self-attention mechanism. Historical network state data is selected as the training set, and training samples are constructed according to the pattern of outputting the state at the next time step corresponding to the input spatiotemporal data sequence. During training, mean squared error (MSE) is used as the loss function, and the model parameters are iteratively optimized through the Adam optimizer until the prediction accuracy of the model on the validation set reaches a preset threshold (such as prediction error below 10%). The model training is then completed, and the optimal model parameters are saved.

[0032] In this embodiment, the prediction time window of the network state prediction model is set to the next 5 to 30 minutes. A rolling prediction mechanism is adopted. The network state prediction model first predicts the network state for the first future time step (consistent with the data acquisition period). This prediction result is used as one of the input features, and combined with historical data, it continues to predict the state for the next time step, iterating until the state prediction for the entire prediction time window is completed. The model outputs the predicted load, predicted latency, and predicted packet loss rate for each candidate link at each future time step. Finally, the prediction results are denormalized to restore them to numerical values ​​with actual physical meaning, forming a structured predicted network state report, providing constraints for subsequent optimization strategy calculations. In this embodiment, a comprehensive data acquisition system is constructed to fully cover multi-dimensional real-time data of candidate links and associated network devices, including transmission indicators such as link utilization and latency, and device indicators such as node resource utilization and queue depth, ensuring the integrity and comprehensiveness of the prediction data foundation. Moreover, through data preprocessing and spatiotemporal data sequence construction, the temporal continuity and spatial correlation of the data are fully explored to improve the effectiveness of the data. A prediction model using a fusion architecture of spatiotemporal graph convolutional network and Transformer is adopted. STGCN can accurately capture the spatial correlation characteristics of network state, while Transformer can effectively explore long-term temporal dependencies. The two work together to overcome the prediction limitations of traditional single models and improve prediction accuracy. The prediction time window of 5 to 30 minutes in the future is clearly defined, and core indicators such as predicted load, latency, and packet loss rate are accurately output, providing a reasonable forward-looking planning basis for scheduling strategies. In this embodiment, step S104, which uses the SLA preference vector as the optimization objective and the predicted network state as the constraint, calculates a global optimized routing strategy using a multi-objective optimization algorithm. This includes: constructing a multi-objective optimization problem based on a multi-commodity flow model, aiming to minimize the weighted deviation of the SLA preference vector, minimize the maximum link utilization, and minimize route changes; using the predicted network state as the capacity and performance constraint for each link within a future time window; identifying service flows with hard constraints on latency requirements in the SLA preference vector; and, based on the capacity and performance constraints, calculating a global optimized routing strategy using a multi-objective optimization algorithm. The shortest path algorithm calculates K candidate feasible paths that meet the latency requirements for each such service flow. Then, using the remaining network resources after satisfying hard constraints and deducting the resources occupied by the current flow, a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem for the remaining flexible service flows, generating a Pareto optimal solution set. Based on a preset decision strategy, each solution in the Pareto optimal solution set is comprehensively scored, and the solution with the highest comprehensive score is selected as the global optimization routing strategy. The global optimization routing strategy defines the primary path, backup path, and corresponding bandwidth allocation ratio for all service flows or service aggregation flows to be scheduled in the network.

[0033] In this embodiment, an optimization framework is constructed based on a multi-commodity flow model. This model can accurately describe the transmission and allocation patterns of multiple service flows in the network, treating each service flow as a commodity and link capacity as transportation capacity. Three optimization objectives are clearly defined and mathematically quantified: first, minimizing the weighted deviation of the SLA preference vector, i.e., minimizing the sum of the weighted deviations between the actual and target values ​​of each SLA parameter; second, minimizing the maximum link utilization, i.e., minimizing the highest utilization value among all links to achieve resource balance; and third, minimizing route changes, i.e., minimizing the path difference between the current and historical strategies to improve stability. Variable constraints are set, including service flow bandwidth demand constraints, link capacity constraints, and path continuity constraints, forming a complete multi-objective optimization mathematical model. In this embodiment, the predicted network state of candidate links is transformed into constraints for the optimization model. Predicted latency and predicted packet loss rate are used as link performance constraints, setting the transmission latency of service flows on the link to not exceed the upper limit of predicted latency and the packet loss rate to not exceed the predicted packet loss rate threshold. The remaining bandwidth corresponding to the predicted load is used as a link capacity constraint, meaning the total bandwidth occupied by all service flows on a certain link does not exceed the predicted remaining bandwidth of that link. Simultaneously, link connectivity constraints are set in conjunction with the network topology to ensure the feasibility of service flow transmission paths. These constraints are integrated into the optimization model to limit the scope of the optimization solution. In this embodiment, service flows with hard latency requirements in the SLA preference vector are first identified (such as uRLLC-type industrial control services). These services are extremely sensitive to latency and must be prioritized. Based on the transformed capacity and performance constraints, the Dijkstra algorithm with latency constraints is used to calculate feasible paths for each hard-constrained service flow. To improve the reliability of service transmission, K (K≥2) candidate feasible paths that meet the latency requirements are calculated for each service flow, and the bandwidth usage, latency, and other key parameters of each path are recorded. This path information and resource usage are entered into a resource reservation table to reserve resource space for subsequent elastic service flow solutions. In this embodiment, within the remaining network resources after deducting the resources occupied by hard-constrained service flows, the focus is on other elastic service flows (such as eMBB-type live video streaming services and mMTC-type IoT services). A non-dominated sorting genetic algorithm (NSGA-III) is used to solve the multi-objective optimization problem. This algorithm selects excellent individuals through non-dominated sorting and congestion calculation, and introduces adaptive crossover and mutation operators to improve convergence speed and solution accuracy. During the algorithm iteration process, new solutions are continuously generated and constraint verification is performed, ultimately obtaining a Pareto optimal solution set that satisfies all constraints. This solution set contains multiple optimization schemes with different objective trade-offs. In this embodiment, the non-dominated sorting genetic algorithm is an improved non-dominated sorting genetic algorithm (NSGA-III). The decision weights for each objective are set based on factors such as service priority, network resource utilization target, and operation and maintenance cost. Then, each solution in the Pareto optimal solution set is comprehensively scored based on a preset decision strategy. The scoring formula is: Comprehensive Score = SLA Compliance Score × 0.4 + Resource Balancing Score × 0.3 + Stability Score × 0.3. Finally, the solution with the highest comprehensive score is selected as the final optimization scheme as the global optimization routing strategy. The global optimization routing strategy clarifies the primary path, backup path (for failover), and corresponding bandwidth allocation ratio for all scheduled service flows or service aggregation flows in the network, forming a structured routing strategy document.

[0034] In this embodiment, a multi-objective optimization problem is constructed based on a multi-commodity flow model. It clearly defines three objectives: minimizing the weighted deviation of the SLA preference vector, minimizing the maximum link utilization, and minimizing route changes. This approach balances service SLA compliance, balanced network resource utilization, and route stability, avoiding the unintended consequences of optimizing a single objective. Furthermore, a hierarchical solution strategy is adopted. First, feasible paths are calculated for latency-constrained service flows (such as industrial control), prioritizing the rigid requirements of high-priority services. Then, Pareto optimal solutions are found for flexible service flows, improving solution efficiency and policy compliance. In addition, a pre-defined decision strategy selects the final solution from the Pareto optimal solution set, clearly defining the primary and backup paths and bandwidth allocation ratios for each service flow, ensuring the practicality and executability of the strategy. In this embodiment, step S105 involves verifying the security and effectiveness of the global optimized routing strategy. After the global optimized routing strategy passes verification, a control message packet is generated based on the global optimized routing strategy. This includes: using graph theory algorithms to detect whether the global optimized routing strategy has routing loops, conflicts with existing security policies, or causes node overload; if the global optimized routing strategy does not have routing loops, conflicts with existing security policies, or causes node overload, then the global optimized routing strategy is simulated and executed in a preset digital twin network environment to estimate the impact of the global optimized routing strategy on key network performance indicators; if the improvement in key performance indicators reaches the expected threshold, then the global optimized routing strategy is determined to have passed verification; the verified global optimized routing strategy is converted into control instructions, and the control instructions are encapsulated into a control message packet.

[0035] In this embodiment, graph theory algorithms are used to model and analyze the global optimization routing strategy. The network topology is abstracted as an undirected graph structure of nodes (network devices) and edges (links), and the path planning in the routing strategy is mapped to a set of paths in the graph. A depth-first search (DFS) algorithm is used to traverse the path set to determine if routing loops exist (i.e., data packets are repeatedly transmitted in the network and cannot reach their destination). Existing security policies in the network (such as access control lists (ACLs) and firewall rules) are retrieved to compare the paths in the routing strategy with the service flow information to check for policy conflicts (such as prohibiting a certain type of service flow from passing through a specific link). Based on the bandwidth allocation parameters and node resource capacity in the strategy, the resource utilization rate of each node is calculated to determine if there is a risk of node overload (resource utilization rate exceeding a preset threshold of 90%). If any problem is detected, the strategy is adjusted in the optimization stage; if no problem is detected, the process proceeds to the next verification step.

[0036] In this embodiment, a digital twin network environment that is completely consistent with the actual network topology, resource configuration, and service distribution is pre-built. This environment maps the actual network's operating status through a real-time data synchronization module, accurately simulating the service transmission process and changes in network resource usage. A global optimization routing strategy is imported into the digital twin environment, and a service traffic model consistent with reality is set up to simulate the entire execution process of the strategy. Key network performance indicators during the simulation process are collected through monitoring modules within the environment, including the actual latency, bandwidth utilization, packet loss rate of each service flow, and the resource usage of network nodes. The differences between the simulated performance indicators and the preset expected targets are compared and analyzed. If the improvement in key performance indicators reaches the expected threshold (e.g., a reduction in service latency of more than 20% or an increase in network resource utilization balance of more than 30%), the routing strategy is deemed to have passed verification; otherwise, the process returns to the optimization stage to adjust the strategy.

[0037] In this embodiment, the verified global optimization routing policy undergoes instruction conversion. Based on the type of network forwarding device (e.g., SDN switch, traditional router), the corresponding southbound protocol is selected (e.g., OpenFlow for SDN switches, NETCONF / YANG for traditional routers). The path planning, bandwidth allocation, priority settings, and other requirements in the policy are converted into control instructions conforming to the protocol specifications. Syntax and integrity checks are performed on the control instructions to ensure correct instruction format, complete parameters, and correct parsing and execution by the target device. The verified control instructions are encapsulated into control message packets according to a preset format. These packets contain key information such as instruction type, execution time, target device identifier, parameter configuration, and checksum. The checksum is generated using the MD5 algorithm to ensure the integrity and security of the message packet during transmission and prevent tampering.

[0038] In this embodiment, graph theory algorithms are used to conduct systematic security detection. Through loop detection, security policy conflict verification, and node overload calculation, potential risks such as routing loops, conflicts with existing ACL rules, and node resource overflow can be accurately identified, eliminating insecure policies at their source. Moreover, simulation verification is performed in a digital twin network environment, which is completely consistent with the actual network topology and resource configuration. This environment can realistically reproduce the policy execution effect, assess the impact on key network performance indicators in advance, and ensure that the policy can achieve the expected optimization goals. In addition, the verified policies are converted into standardized control commands that conform to specific southbound protocols, adapting to different types of network forwarding devices and avoiding command incompatibility issues.

[0039] In this embodiment, step S106 involves using a smooth migration mechanism to send control message packets to the network forwarding device and collecting actual performance data related to the execution effect of the control message packets. Based on this actual performance data, the parameters of the network state prediction model and the multi-objective optimization algorithm are adjusted. This includes: using a multi-stage activation mechanism to send the control message packets to the network forwarding device and collecting actual performance data related to the execution effect of the control message packets; calculating the quantized deviation between the measured values ​​of each service quality parameter in the actual performance data and the corresponding expected service quality target values ​​in the SLA preference vector; evaluating the SLA compliance of each scheduled service flow or service aggregation flow in the current network based on the quantized deviation; and if the SLA compliance is lower than a predetermined threshold or the prediction error remains high, then the parameters of the network state prediction model and the multi-objective optimization algorithm are fine-tuned online using the actual performance data related to the execution effect of the control message packets.

[0040] In this embodiment, a multi-stage smooth migration mechanism of preheating, switching, and stabilization is employed to distribute control message packets. During the preheating phase (1-5 minutes), control message packets are distributed to the target network forwarding device, but the new routing policy is not activated; only the device configuration reception status and command parsing correctness are verified. During the switching phase, service flows are gradually switched to the new routing policy according to a preset ratio. First, 10% of service flows are switched to the new path; after 5 minutes of monitoring without anomalies, 50% of service flows are switched, and finally the remaining 40% are switched. Service transmission status and network performance are monitored in real time throughout the process. During the stabilization phase, after confirming that all service flows have switched successfully and transmission is stable, the new routing policy configuration is solidified, and the policy is distributed. Throughout the process, if any service transmission anomalies are detected, the switching is immediately paused and the original policy is rolled back.

[0041] In this embodiment, a multi-channel performance data collection system is constructed. Through the monitoring functions built into network devices, third-party network monitoring tools, and feedback from service terminals, actual performance data related to the execution effect of message packets is collected. This includes service quality parameters such as actual latency, latency jitter, bandwidth usage, packet loss rate, and reliability for each service flow. The quantitative deviation between the measured value of each service quality parameter and the corresponding expected target value in the SLA preference vector is calculated using the relative deviation formula: Deviation = |Actual value - Target value| / Target value × 100%. Based on the optimized weights of each parameter, a weighted summation method is used to calculate the overall SLA compliance score: SLA compliance score = Σ(1 - Deviation) × Parameter weight. If the score is higher than a preset threshold (e.g., 85 points), the SLA is deemed compliant; otherwise, it is deemed non-compliant, and a parameter adjustment process needs to be initiated.

[0042] In this embodiment, for the network state prediction model, training samples are supplemented using collected actual performance data, and incremental learning is employed to fine-tune model parameters, including the convolutional kernel weights of the STGCN module and the attention weights of the Transformer module, to improve the model's prediction accuracy of actual network states. For the multi-objective optimization algorithm, algorithm parameters are adjusted based on actual execution results. For example, the crossover probability (from 0.8 to 0.75) and mutation probability (from 0.1 to 0.12) of the improved NSGA-Ⅲ algorithm are adjusted to optimize the algorithm's convergence speed and solution accuracy. Simultaneously, the objective weights in the decision strategy are adjusted to make the optimization objectives more aligned with actual business needs. After parameter adjustments, small-scale verification is conducted in a digital twin environment. Once the adjustment effects are confirmed, the parameters are fixed, completing one iteration of optimization.

[0043] In this embodiment, a multi-stage smooth migration mechanism is adopted to deploy the strategy. Through a warm-up, switching, and stable gradual execution approach, service transmission jitter or interruption caused by sudden policy changes is avoided, ensuring the continuous operation of critical services. Moreover, by constructing a multi-channel, multi-dimensional actual performance data collection system, the actual effect after policy execution can be accurately captured. By quantifying the SLA compliance score, the actual performance of the strategy can be objectively evaluated. In addition, when the SLA compliance is not up to standard or the prediction error is too high, the prediction model and optimization algorithm parameters are fine-tuned online using actual data to achieve dynamic self-adaptation of scheduling capabilities.

[0044] The intelligent traffic scheduling optimization method in the embodiments of the present invention has been described above. The apparatus in the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 2 The implementation methods of the intelligent traffic scheduling optimization device in this invention include: The parsing module 201 is used to receive service requests, parse the service requests, and obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request. The conversion module 202 is used to obtain the SLA parameters corresponding to the service request and the candidate links required for service transmission according to the 5G standard service type, the application scenario and the 5G network slice identifier, and convert the SLA parameters into an SLA preference vector; The prediction module 203 is used to acquire the global data of the candidate link and input the global data into the pre-trained network state prediction model to predict the predicted network state of the candidate link. Optimization module 204 is used to calculate a global optimized routing strategy by taking the SLA preference vector as the optimization objective and the predicted network state as the constraint. Verification module 205 is used to verify the security and effectiveness of the global optimized routing policy. When the global optimized routing policy is verified, a control message packet is generated based on the global optimized routing policy. The adjustment module 206 is used to send the control message packet to the network forwarding device using a smooth migration mechanism, collect actual performance data related to the execution effect of the control message packet, and adjust the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data.

[0045] In this embodiment, by parsing service requests to extract 5G standard service types, application scenarios, and slice identifiers, the precise anchoring of service requirements and scheduling resources is achieved, avoiding resource mismatch. Simultaneously, a pre-trained network state prediction model is introduced to predict the network state for a preset time period based on the full-domain data of candidate links, upgrading the routing strategy from traditional static adjustment to proactive planning and avoiding network congestion risks in advance. Furthermore, a multi-objective optimization algorithm is designed based on SLA preference vectors to balance multiple objectives such as service SLA compliance, balanced network resource utilization, and routing stability, avoiding the one-sidedness of single-objective optimization. In addition, dual verification for security and effectiveness is added. Graph theory algorithms and digital twin simulations are used to proactively identify risks such as routing loops and node overload. A smooth migration mechanism is adopted to distribute strategies in stages, and the model and algorithm parameters are fine-tuned using actual performance data, ensuring the feasibility of the strategy and enabling continuous iteration of scheduling capabilities.

[0046] Figure 2 The structure of the intelligent traffic scheduling optimization device shown does not constitute a limitation on the intelligent traffic scheduling optimization device, and can implement the steps of the intelligent traffic scheduling optimization method provided in the above-described method embodiments.

[0047] above Figure 2 The intelligent traffic scheduling and optimization device in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The intelligent traffic scheduling and optimization device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0048] Figure 3 This is a schematic diagram of the structure of an intelligent traffic scheduling optimization device provided in an embodiment of the present invention. The device 300 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown), each module including a series of instruction operations on the device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media on the device 300.

[0049] Device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.

[0050] This invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the intelligent traffic scheduling optimization method.

[0051] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0052] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent traffic scheduling optimization, characterized in that, The intelligent traffic scheduling optimization method includes: Receive a service request, parse the service request, and obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request; Based on the 5G standard service type, the application scenario, and the 5G network slice identifier, obtain the SLA parameters corresponding to the service request and the candidate links required for service transmission, and convert the SLA parameters into an SLA preference vector; The global data of the candidate link is obtained, and the global data is input into the pre-trained network state prediction model to predict the predicted network state of the candidate link. Using the SLA preference vector as the optimization objective and the predicted network state as the constraint, a global optimized routing strategy is calculated through a multi-objective optimization algorithm. The global optimization routing policy is verified for security and effectiveness. Once the global optimization routing policy is verified, a control message packet is generated based on the global optimization routing policy. A smooth migration mechanism is used to send the control message packet to the network forwarding device, and actual performance data related to the execution effect of the control message packet is collected. Based on the actual performance data, the parameters of the network state prediction model and the multi-objective optimization algorithm are adjusted.

2. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The process of receiving a service request involves parsing the request to obtain the corresponding 5G standard service type, application scenario, and 5G network slice identifier, including: Receive service requests from the service orchestrator or network management system, and parse the service requests to obtain the service intent and 5G network slice identifier; Based on a predefined service feature library and rule engine, the service intent is mapped to the 5G standard service type corresponding to the service request. The 5G standard service type includes enhanced mobile broadband, ultra-reliable low-latency communication and massive machine-type communication. Based on the contextual information in the business intent, the application scenario corresponding to the business request is determined, and the application scenario includes virtual reality / augmented reality, industrial control, or live video streaming.

3. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The step of obtaining the SLA parameters and candidate links required for service transmission corresponding to the service request based on the 5G standard service type, the application scenario, and the 5G network slice identifier, and converting the SLA parameters into an SLA preference vector, includes: Based on the 5G standard service type and the application scenario, key SLA parameters are matched and extracted from the preset SLA template library. The key SLA parameters include maximum allowable latency, maximum allowable latency jitter, minimum required bandwidth, target bandwidth, reliability level, and service lifetime. Based on the 5G network slice identifier, and combined with the preset network topology and slice resource mapping strategy, a set of physical or logical links that the service request is preferably used is determined as candidate links. Using the analytic hierarchy process (AHP), optimization weights are assigned to each dimension of the SLA parameters based on the 5G standard service type and the application scenario, and then converted into an SLA preference vector.

4. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The step of obtaining the global data of the candidate link and inputting the global data into a pre-trained network state prediction model to predict the predicted network state of the candidate link includes: Obtain real-time status data of the candidate links and their associated network devices. The real-time status data includes link utilization, latency, packet loss rate, error frame rate, node resource utilization, and queue depth. The acquired real-time status data is preprocessed, and a spatiotemporal data sequence is constructed based on the preprocessed real-time status data; A network state prediction model is constructed and trained based on a spatiotemporal graph convolutional network and a Transformer architecture, and the spatiotemporal data sequence is input into the network state prediction model. Using the network state prediction model, the predicted network state of each candidate link is output in a rolling manner for the next 5 to 30 minutes. The predicted network state includes predicted load, predicted latency, and predicted packet loss rate.

5. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The process of calculating a global optimized routing strategy using the SLA preference vector as the optimization objective and the predicted network state as the constraint through a multi-objective optimization algorithm includes: Based on the multi-commodity flow model, a multi-objective optimization problem is constructed with the objectives of minimizing the weighted deviation of the SLA preference vector, minimizing the maximum link utilization, and minimizing route changes. The predicted network state is used as a constraint on the capacity and performance of each link within the future time window. Identify the service flows with hard latency requirements in the SLA preference vector. Based on the capacity and performance constraints, calculate K candidate feasible paths that meet the latency requirements for each such service flow using the constrained shortest path algorithm. On the remaining network resources after satisfying the hard constraints and deducting the resources occupied by the hard constraints, solve the multi-objective optimization problem using a non-dominated sorting genetic algorithm for the remaining flexible service flows to generate a Pareto optimal solution set. Based on a preset decision-making strategy, each solution in the Pareto optimal solution set is comprehensively scored, and the solution with the highest comprehensive score is selected as the global optimization routing strategy. The global optimization routing strategy defines the primary path, backup path and corresponding bandwidth allocation ratio for all service flows or service aggregation flows to be scheduled in the network.

6. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The global optimized routing policy is verified for security and effectiveness. Once the global optimized routing policy is verified, a control message packet is generated based on the global optimized routing policy, including: Using graph theory algorithms, detect whether the global optimization routing strategy has routing loops, conflicts with existing security policies, or causes node overload; If the global optimized routing strategy does not have routing loops, conflicts with existing security strategies, or causes node overload, the global optimized routing strategy is simulated and executed in a preset digital twin network environment to estimate the impact of the global optimized routing strategy on key network performance indicators. If the improvement of key performance indicators reaches the expected threshold, the global optimized routing strategy is deemed to have passed the verification. The verified global optimization routing policy is converted into control commands, and the control commands are encapsulated into control message packets.

7. The intelligent traffic scheduling optimization method according to claim 1, characterized in that, The process of using a smooth migration mechanism to send the control message packet to the network forwarding device, collecting actual performance data related to the execution effect of the control message packet, and adjusting the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data includes: A multi-stage activation mechanism is used to send the control message packet to the network forwarding device, and actual performance data related to the execution effect of the control message packet is collected. Calculate the quantified deviation between the measured values ​​of each service quality parameter in the actual performance data and the corresponding expected service quality target values ​​in the SLA preference vector, and evaluate the SLA compliance of each service flow or service aggregation flow scheduled in the current network based on the quantified deviation. If the SLA compliance is lower than a predetermined threshold or the prediction error remains high, the parameters of the network state prediction model and the multi-objective optimization algorithm are fine-tuned online using actual performance data related to the execution effect of the control message packet.

8. An intelligent traffic scheduling and optimization device, characterized in that, include: The parsing module is used to receive service requests, parse the service requests, and obtain the 5G standard service type, application scenario, and 5G network slice identifier corresponding to the service request. The conversion module is used to obtain the SLA parameters corresponding to the service request and the candidate links required for service transmission based on the 5G standard service type, the application scenario and the 5G network slice identifier, and convert the SLA parameters into an SLA preference vector. The prediction module is used to acquire the global data of the candidate link and input the global data into the pre-trained network state prediction model to predict the predicted network state of the candidate link. The optimization module is used to calculate a global optimized routing strategy by taking the SLA preference vector as the optimization objective and the predicted network state as the constraint. The verification module is used to verify the security and effectiveness of the global optimized routing policy. When the global optimized routing policy is verified, a control message packet is generated based on the global optimized routing policy. The adjustment module is used to send the control message packet to the network forwarding device using a smooth migration mechanism, collect actual performance data related to the execution effect of the control message packet, and adjust the parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data.

9. An intelligent traffic scheduling and optimization device, characterized in that, It includes a memory and at least one processor, wherein the memory stores computer-readable instructions; The at least one processor invokes the computer-readable instructions in the memory to perform the steps of the intelligent traffic scheduling optimization method as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer-readable instructions thereon, characterized in that, When the computer-readable instructions are executed by a processor, they implement the steps of the intelligent traffic scheduling optimization method as described in any one of claims 1-7.

Images