Cross-domain QoS mapping method, system and device based on three-way decision and medium

Abstract

The invention relates to a cross-domain QoS (Quality of Service) mapping method, system and equipment based on three-way decision, and a medium. The method comprises the following steps: constructing an attribute matrix by extracting key service quality attributes such as priority and delay budget of time-sensitive network data flow; carrying out weighted clustering by adopting a comprehensive weight fusing attribute dispersion and discrimination, and generating a clustering center set and a membership relationship; dividing the 5G service quality identifier into a core domain set and a boundary domain set based on a three-way decision mechanism; a dynamic weight is calculated in combination with the load residual rate and the service quality similarity, and a layered mapping decision with core domain priority is realized; dynamic remapping is triggered by closed-loop load monitoring. According to the method, the three technical problems of static mapping rule stiffness, attribute differentiation processing deficiency and contradiction between load balancing and service quality management in the industrial internet are solved, and the network resource utilization rate and the service quality guarantee capability in a high burst traffic scene are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of industrial communication network technology, specifically to a cross-domain QoS mapping method, system, device, and medium based on three-way decision-making. Background Technology

[0002] With the development trend of industrial internet and 5G convergence networking, cross-domain Quality of Service (QoS) mapping technology has become a core component in ensuring the end-to-end performance of critical services such as real-time control in intelligent manufacturing and power grid relay protection. This technology aims to solve the problem of QoS parameter conversion between heterogeneous networks by mapping attributes such as priority and latency budget of TSN traffic to 5G QoS identifiers (5QI), thereby achieving coordinated scheduling of network resources. As high-burst traffic becomes the norm in industrial scenarios, such as surges in sensor data and synchronization of equipment commands, cross-domain QoS mapping needs to dynamically adapt to network load fluctuations while ensuring low latency and high reliability.

[0003] However, existing QoS mapping algorithms have three shortcomings: First, they lack in-depth modeling of QoS attribute differences, resulting in insufficient clustering accuracy, which affects subsequent mapping results; second, existing partitioning methods rely on fixed thresholds and cannot adaptively adjust according to network conditions, resulting in insufficient mapping flexibility; and third, the mapping strategy ignores the coupling relationship between load remaining rate and QoS similarity, which can easily lead to single 5QI overload or QoS consistency degradation. Summary of the Invention

[0004] Based on this, the purpose of this invention is to provide a cross-domain QoS mapping method based on three-branch decision-making that dynamically balances load and quality of service and adapts to traffic mutations.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a cross-domain QoS mapping method based on three-branch decision-making, comprising the following steps:

[0007] S1: Perform multi-dimensional attribute vector construction processing on the input TSN data stream, extract the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size of each data stream, combine the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and construct a matrix to generate the TSN data stream attribute matrix.

[0008] S2: Based on the TSN data stream attribute matrix, the weight value of each QoS attribute is calculated by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminative weight. The weight value of each QoS attribute is combined to perform weighted calculation of the distance between cluster centers and samples and iteratively update the membership relationship to generate a cluster center set and a membership matrix.

[0009] S3: Based on the cluster center set, comprehensive attribute difference weights, and the preset 5G QoS identifier attribute vector, a three-branch decision region is divided. The relative membership degree of each 5G QoS identifier to the cluster center is calculated, and the core domain and boundary domain division threshold is automatically generated according to the distribution of the relative membership degree. The 5G QoS identifier is divided into the core domain set and boundary domain set corresponding to each cluster.

[0010] S4: Based on the membership matrix, core domain set, boundary domain set, and real-time load status of 5G QoS identifiers, calculate the dynamic weight value of the target mapping according to the load remaining rate and QoS similarity, map the TSN data stream to the 5G QoS identifier with the largest dynamic weight, generate and output the mapping relationship table between TSN stream and 5G QoS identifier, and use the mapping relationship table to instruct network devices to perform traffic scheduling.

[0011] In one embodiment, S1 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0012] S11: Extract QoS attributes from the input TSN data stream, and parse the priority value, latency budget value, packet loss rate value, bandwidth requirement value and burst traffic size value in the data stream one by one to generate the original QoS attribute dataset.

[0013] S12: Vectorize the original QoS attribute dataset, combine the five QoS attribute values ​​of each data stream into a five-dimensional vector in a fixed-dimensional order, and generate a multi-dimensional vector set of TSN data streams;

[0014] S13: Perform matrix construction processing on the TSN data stream multidimensional vector set, arrange all data stream vectors in the input order into an N-row, 5-column numerical matrix, and generate the TSN data stream attribute matrix.

[0015] In one embodiment, S2 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0016] S21: Perform objective weight calculation on the TSN data stream attribute matrix, calculate the CRITIC-Entropy weight value based on the standard deviation, information entropy and correlation coefficient between attributes of each QoS attribute, and generate an objective weight vector.

[0017] S22: Perform discriminant weight calculation on the initial clustering results, calculate Fisher discriminant weight values ​​by analyzing the proportional relationship between inter-class variance and intra-class variance, and generate a discriminant weight vector;

[0018] S23: The objective weight vector and the discriminant weight vector are fused together. The two types of weight values ​​are normalized and then linearly superimposed to generate a comprehensive attribute difference weight vector.

[0019] S24: Perform weighted distance calculation on the comprehensive attribute difference weight vector and the current cluster center, and introduce the weight values ​​of each QoS attribute into the Euclidean distance formula for dimensional weighting to generate a weighted distance matrix;

[0020] S25: Perform membership update processing on the weighted distance matrix, redistribute the membership probabilities of samples to each cluster according to the inverse distance ratio, and generate the updated membership matrix;

[0021] S26: Update the cluster centers of the membership matrix and the TSN data stream attribute matrix, calculate the new centroid positions of each cluster by weighted average, and generate a new set of cluster centers.

[0022] In one embodiment, the formula for calculating the comprehensive attribute difference weight vector of a cross-domain QoS mapping method based on three-way decision-making provided by the present invention is as follows:

[0023]

[0024] Among them, w j Let be the comprehensive attribute difference weight vector of the traffic QoS attribute in the j-th dimension. Let be the CRITIC-Entropy objective weight vector of the traffic QoS attribute in the j-th dimension. Let be the Fisher discrimination weight of the traffic QoS attribute in the j-th dimension, and D be the total number of dimensions of the QoS attribute.

[0025] In one embodiment, S3 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0026] S31: Calculate the weighted membership degree of the cluster center set based on the preset 5G QoS identifier attribute vector, adjust the dimension contribution ratio based on the comprehensive attribute difference weight, and generate the identifier-cluster membership degree matrix;

[0027] S32: Perform relative membership transformation on the identifier-cluster membership matrix, perform maximum value normalization on the membership vector of each 5G QoS identifier, and divide the membership of each dimension by the maximum value of its row to generate the relative membership.

[0028] S33: Generate an adaptive threshold for relative membership, group and statistically analyze the difference between the maximum and second largest membership of each identifier according to resource type, calculate the average value of the difference of each group as the boundary threshold, derive the core threshold based on the boundary threshold according to a preset ratio, and generate the core domain threshold and the boundary domain threshold.

[0029] S34: Divide the relative membership degree into three regions, assign the identifiers with a relative membership degree not lower than the core domain threshold to the core domain, and assign the identifiers with a relative membership degree between the boundary domain threshold and the core domain threshold to the boundary domain, thereby generating a core domain set and a boundary domain set.

[0030] In one embodiment, the formula for calculating the relative membership degree of a cross-domain QoS mapping method based on three-branch decision-making provided by the present invention is as follows:

[0031]

[0032] Among them, g jk To represent the relative membership of the j-th 5G QoS identifier to the k-th cluster center, To show the basic membership degree of the j-th 5G QoS identifier to the k-th cluster center, Let j be the maximum value of the membership degree of the j-th 5G QoS identifier to all cluster centers. Let i be the weighted Euclidean distance from data point i to cluster center k. Let be the weighted distance from data point i to cluster center p.

[0033] In one embodiment, S4 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0034] S41: Perform traffic clustering and grouping processing on the membership matrix, assign each TSN data stream to the cluster corresponding to its maximum membership, and generate a clustered traffic group set;

[0035] S42: Perform normalized distance calculation on each data stream in the clustered traffic group set, calculate the weighted Euclidean distance of each data stream to each 5G QoS identifier in the core domain and boundary domain based on the comprehensive attribute difference weight, and linearly scale the distance to the [0,1] interval through the maximum and minimum value method to generate a normalized similarity index.

[0036] S43: Perform remaining carrying capacity assessment on the real-time load status of 5G QoS identifiers, query the difference between the current carrying capacity of each identifier and its preset load limit, calculate the ratio of the difference to the load limit, and generate a load remaining rate index.

[0037] S44: Dynamically weight and fuse the normalized similarity index and the load balancer index. By using a preset weight factor, the normalized similarity index is inverted and then linearly weighted and combined with the load balancer to balance the service quality matching degree and the load balancing requirement, and generate a dynamic weight decision value.

[0038] S45: Perform weight priority mapping on the identifiers within the core domain set, select available identifiers to map data streams in descending order of dynamic weight values. If all core domain identifiers are fully loaded, automatically switch to the boundary domain set to continue weight priority mapping, complete the mapping and allocation of all data streams, and generate a mapping relationship table between TSN streams and 5G QoS identifiers.

[0039] Secondly, the present invention provides a cross-domain QoS mapping system based on three-way decision-making, which is configured with the following modules:

[0040] The TSN data stream attribute matrix construction module is used to perform multi-dimensional attribute vector construction processing on the input TSN data stream. It extracts the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size of each data stream, combines the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and constructs a matrix to generate the TSN data stream attribute matrix.

[0041] The comprehensive weighted clustering analysis module is used to calculate the weight value of each QoS attribute based on the TSN data stream attribute matrix by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminative weight. It combines the weight values ​​of each QoS attribute to perform weighted calculation of the distance between cluster centers and samples and iteratively update the membership relationship to generate a set of cluster centers and a membership matrix.

[0042] The three-branch decision region partitioning module is used to partition the decision region based on the cluster center set, the comprehensive attribute difference weight, and the preset 5G QoS identifier attribute vector. It calculates the relative membership degree of each 5G QoS identifier to the cluster center and automatically generates the core domain and boundary domain partitioning threshold according to the distribution of relative membership degree, dividing the 5G QoS identifier into the core domain set and boundary domain set corresponding to each cluster.

[0043] The TSN-5G QoS dynamic mapping module is used to calculate the dynamic weight value of the target mapping based on the membership matrix, core domain set, boundary domain set and real-time load status of 5G QoS identifier, according to the load remaining rate and QoS similarity. It maps TSN data streams to the 5G QoS identifier with the largest dynamic weight, generates and outputs a mapping relationship table between TSN streams and 5G QoS identifiers, and uses the mapping relationship table to instruct network devices to perform traffic scheduling.

[0044] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned cross-domain QoS mapping methods based on three-way decision-making.

[0045] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned cross-domain QoS mapping methods based on three-way decision-making.

[0046] In summary, the cross-domain QoS mapping method based on three-way decision-making provided in this application achieves three core breakthroughs in high-burst traffic scenarios in the Industrial Internet by integrating weighted clustering, three-way decision-making, and dynamic weighting mechanisms: First, the comprehensive attribute difference weight calculation based on CRITIC-Entropy and Fisher discriminant weights can enhance the differentiated processing capability of key QoS attributes such as latency budget, solving the defect of traditional clustering algorithms that ignore attribute sensitivity differences, thereby improving mapping accuracy; Second, by dynamically generating core domain and boundary domain partitioning thresholds through three-way decision-making, the mapping range can be adaptively adjusted according to the relative membership distribution of 5G QoS identifiers and cluster centers, overcoming the limitation of static mapping rules being unable to respond to traffic mutations and ensuring the real-time performance of network resource scheduling; Third, the dynamic weight calculation mechanism that couples load remaining rate and QoS similarity can automatically balance service quality matching degree and load balancing requirements under the core domain priority mapping framework, thereby resolving the contradiction between single-node overload and service latency degradation during resource allocation. Ultimately, a full-link technical closed loop is formed, consisting of "attribute-weighted modeling → dynamic regional partitioning → load-aware mapping → closed-loop reconfiguration optimization". This can significantly improve network resource utilization, reduce the risk of failures such as industrial control command timeouts and video stream stuttering, and provide high reliability assurance for deterministic services in 5G+TSN converged networks.

[0047] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0048] Figure 1 A flowchart illustrating a cross-domain QoS mapping method based on three-branch decision-making provided in this application embodiment;

[0049] Figure 2 This is a schematic diagram of the process for generating a mapping table between TSN streams and 5G QoS identifiers, provided in an embodiment of this application.

[0050] Figure 3 This is a schematic diagram of the structure of a cross-domain QoS mapping system based on three-branch decision-making, which is provided as another embodiment of this application. Detailed Implementation

[0051] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0053] In one embodiment, such as Figure 1 As shown, a cross-domain QoS mapping method based on three-way decision-making is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0054] S1: Perform multi-dimensional attribute vector construction processing on the input TSN data stream, extract the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size for each data stream, combine the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and construct a matrix to generate the TSN data stream attribute matrix.

[0055] Specifically, when the system performs multi-dimensional attribute vector construction on the input TSN data streams, it identifies the QoS attributes of each data stream, including priority, latency budget, packet loss rate, bandwidth requirement, and burst traffic size. The system determines the acquisition method for each attribute based on TSN-related standards in industrial communication networks. Priority is extracted based on the standardized TSN protocol; latency budget is determined according to the communication requirements of industrial services; packet loss rate is obtained by analyzing parameters related to the error control mechanism during data stream transmission; bandwidth requirement is calculated based on the data stream's transmission rate and data volume characteristics; and burst traffic size is obtained by statistically analyzing the traffic fluctuations of the data streams within a unit of time.

[0056] Furthermore, the system combines the QoS attribute values ​​of each data stream into a multi-dimensional vector in a fixed order. This fixed order is determined based on the parameter correspondence of the 5G QoS identifier to ensure compatibility with subsequent cross-domain mapping. After generating the multi-dimensional vectors for all data streams, the system arranges these vectors sequentially according to the data stream reception order to construct a TSN data stream attribute matrix. The number of rows in this matrix corresponds to the total number of input TSN data streams, and the number of columns corresponds to the number of selected QoS attributes. Each element in the matrix represents a specific QoS attribute value for a TSN data stream.

[0057] S2: Based on the TSN data stream attribute matrix, the weight value of each QoS attribute is calculated by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminative weight. The weight value of each QoS attribute is combined to calculate the weighted distance between the cluster center and the sample and iteratively update the membership relationship to generate the cluster center set and membership matrix.

[0058] Specifically, based on the generated TSN data stream attribute matrix, the system first calculates the CRITIC-Entropy objective weights. When calculating the CRITIC weights, the system first obtains the standard deviation of each QoS attribute in the attribute matrix. The standard deviation reflects the dispersion of each attribute value. Then, it calculates the correlation coefficient between different QoS attributes. The correlation coefficient reflects the degree of association between attributes. The system obtains the CRITIC weight components of each attribute by combining the standard deviation and the correlation coefficient. When calculating the Entropy weights, the system first normalizes the value of each attribute in the attribute matrix to obtain the proportion of each data stream under each attribute. Then, it calculates the entropy value of each attribute based on the proportion. The entropy value reflects the uncertainty of the information contained in the attribute. The system obtains the Entropy weight components of each attribute by complementary calculation of the entropy values.

[0059] Furthermore, the system normalizes and fuses the CRITIC and Entropy weight components to obtain the CRITIC-Entropy objective weights. The system then calculates the Fisher discriminant weights. For a preset number of clusters, the system first divides the initial sample categories, calculates the mean of each QoS attribute in each category and the mean of each QoS attribute across all samples globally, then calculates the sum of squared deviations within each category based on the difference between the intra-category sample mean and the class mean, and calculates the sum of squared deviations between each category based on the difference between each category mean and the global mean. The system obtains the Fisher discriminant weight component for each attribute by the ratio of the sum of squared deviations between the classes to the sum of squared deviations within the classes. This component is then normalized to obtain the Fisher discriminant weight.

[0060] The system integrates the CRITIC-Entropy objective weights and Fisher discriminative weights according to a preset weight allocation coefficient to obtain the comprehensive attribute difference weights. Preferably, the system can employ a weighted fuzzy C-means clustering algorithm to calculate the weighted distance between each sample and each cluster center based on the comprehensive attribute difference weights. During the calculation of the weighted distance, the system uses the comprehensive attribute difference weights as weight factors for each attribute, incorporating them into the calculation logic of Euclidean distance. The system updates the membership degree of each sample to each cluster based on the weighted distance. The update of the membership degree follows the membership degree constraint of fuzzy clustering, i.e., the sum of the membership degrees of each sample to all clusters is 1. The system then adjusts the position of the cluster centers based on the updated membership degrees. The adjustment of the cluster centers is based on the weighted calculation of the sample value and the corresponding membership degree. The system repeats the process of weighted distance calculation, membership degree update, and cluster center adjustment until the difference between the cluster centers obtained from two adjacent iterations meets the preset iteration termination condition. At this point, the system outputs the final set of cluster centers and the membership degree matrix.

[0061] S3: Based on the cluster center set, comprehensive attribute difference weights, and the preset 5G QoS identifier attribute vector, a three-branch decision region is divided. The relative membership degree of each 5G QoS identifier to the cluster center is calculated, and the core domain and boundary domain division threshold is automatically generated according to the distribution of the relative membership degree. The 5G QoS identifier is divided into the core domain set and boundary domain set corresponding to each cluster.

[0062] Specifically, the system performs three-branch decision region division based on the cluster center set, comprehensive attribute difference weights, and pre-set 5G QoS identifier attribute vectors obtained in the previous step. The system first obtains the pre-set 5G QoS identifier attribute vectors, which contain various parameters of the 5G QoS identifier. The parameters are set according to relevant 3GPP standards to ensure a correspondence with the QoS attributes of the TSN data stream. The system calculates the membership degree of each 5G QoS identifier attribute vector to each cluster center, using the same method as the sample membership degree calculation in the previous step. That is, the 5G QoS identifier attribute vector is treated as a sample, substituted into the membership degree calculation logic of weighted fuzzy C-means clustering, and the membership degree result is obtained by combining the comprehensive attribute difference weights and the cluster center set.

[0063] The system extracts the maximum membership degree of each 5G QoS identifier attribute vector to all cluster centers. It then calculates the ratio of this membership degree to the maximum value to obtain the relative membership degree of the 5G QoS identifier attribute vector to each cluster center. For each cluster center, the system collects the relative membership degrees of all 5G QoS identifier attribute vectors to that cluster center, extracts the maximum and second-largest values ​​from these relative membership degrees, and calculates the core and boundary domain division thresholds corresponding to that cluster center using the difference between the maximum and second-largest values. The system classifies the 5G QoS identifier attribute vectors according to the division thresholds. For each cluster center, if the relative membership degree of a 5G QoS identifier attribute vector to that cluster center is greater than or equal to the difference between the maximum value and the division threshold, then the 5G QoS identifier is assigned to the core domain set corresponding to that cluster center; if the relative membership degree of a 5G QoS identifier attribute vector to that cluster center is greater than or equal to the second-largest value and less than the difference between the maximum value and the division threshold, then the 5G QoS identifier is assigned to the boundary domain set corresponding to that cluster center.

[0064] After classifying all 5G QoS identifiers, the system outputs the core domain set and boundary domain set corresponding to each cluster center, providing a classification basis for subsequent dynamic mapping. During the classification process, the system records the attribution status of each 5G QoS identifier to ensure that each identifier is assigned to only one cluster center's core domain or boundary domain, avoiding mapping conflicts caused by duplicate attribution. The recorded data will serve as the basis for subsequent mapping relationship verification.

[0065] S4: Based on the membership matrix, core domain set, boundary domain set, and real-time load status of 5G QoS identifiers, calculate the dynamic weight value of the target mapping according to the load remaining rate and QoS similarity, map the TSN data stream to the 5G QoS identifier with the largest dynamic weight, generate and output the mapping relationship table between TSN stream and 5G QoS identifier, and use the mapping relationship table to instruct network devices to perform traffic scheduling.

[0066] Specifically, the system performs dynamic weight calculation and mapping based on the membership matrix, core domain set, boundary domain set, and real-time load status of 5G QoS identifiers. The system obtains the real-time load status of each 5G QoS identifier through the network device's communication protocol or traffic acquisition module. The real-time load status includes the load limit and current load of the 5G QoS identifier. The load limit is determined by the network resource configuration, and the current load is the number of TSN data streams already carried by the 5G QoS identifier. The system calculates the load remaining rate of each 5G QoS identifier based on the difference between the load limit and the current load, combined with the load limit. For each TSN data stream, the system determines the cluster center corresponding to the data stream based on the membership matrix, and then determines the core domain set and boundary domain set corresponding to the cluster center. 5G QoS identifiers in these two sets are used as candidate 5G QoS identifiers for that data stream.

[0067] Furthermore, the system calculates the QoS similarity between each TSN data stream and each candidate 5G QoS identifier. During the calculation, the system first calculates the weighted distance between the data stream and the candidate 5G QoS identifier based on the comprehensive attribute difference weights. Then, it extracts the maximum value among the weighted distances between the data stream and all candidate 5G QoS identifiers. The normalized weighted distance is calculated by the ratio of the weighted distance to the maximum value. The result of subtracting the normalized weighted distance from 1 is taken as the QoS similarity. The system then integrates the load remaining rate and QoS similarity calculation according to a preset weight adjustment coefficient to obtain the dynamic weight of each candidate 5G QoS identifier for the corresponding TSN data stream.

[0068] During the fusion calculation process, the weight adjustment coefficient is used to balance the proportion of load remaining rate and QoS similarity in the dynamic weight. For each TSN data stream, the system selects the candidate 5G QoS identifier with the largest dynamic weight as the mapping target for that data stream, establishing a mapping relationship between the TSN data stream and the 5G QoS identifier. The system organizes all mapping relationships into a mapping relationship table, which includes fields such as TSN stream ID, TSN stream QoS attribute, mapped 5G QoS identifier, 5G QoS identifier load remaining rate, and dynamic weight. The system distributes the mapping relationship table to the network control device, which then issues flow table instructions to the 5G base station and TSN switch based on the mapping relationship table, enabling the TSN data stream to be scheduled according to the mapped 5G QoS identifier, thus achieving cross-domain QoS guarantee. After distributing the mapping relationship table, the system continuously monitors the scheduling execution status of the network devices to ensure that the transmission of TSN data streams meets the QoS requirements in the mapping relationship table. The monitoring data is obtained through the status feedback mechanism of the network devices and used to verify the effectiveness of the mapping method.

[0069] In summary, the cross-domain QoS mapping method based on three-way decision-making provided in this application achieves three core breakthroughs in high-burst traffic scenarios in the Industrial Internet by integrating weighted clustering, three-way decision-making, and dynamic weighting mechanisms: First, the comprehensive attribute difference weight calculation based on CRITIC-Entropy and Fisher discriminant weights can enhance the differentiated processing capability of key QoS attributes such as latency budget, solving the defect of traditional clustering algorithms that ignore attribute sensitivity differences, thereby improving mapping accuracy; Second, by dynamically generating core domain and boundary domain partitioning thresholds through three-way decision-making, the mapping range can be adaptively adjusted according to the relative membership distribution of 5G QoS identifiers and cluster centers, overcoming the limitation of static mapping rules being unable to respond to traffic mutations and ensuring the real-time performance of network resource scheduling; Third, the dynamic weight calculation mechanism that couples load remaining rate and QoS similarity can automatically balance service quality matching degree and load balancing requirements under the core domain priority mapping framework, thereby resolving the contradiction between single-node overload and service latency degradation during resource allocation. Ultimately, a full-link technical closed loop is formed, consisting of "attribute-weighted modeling → dynamic regional partitioning → load-aware mapping → closed-loop reconfiguration optimization". This can significantly improve network resource utilization, reduce the risk of failures such as industrial control command timeouts and video stream stuttering, and provide high reliability assurance for deterministic services in 5G+TSN converged networks.

[0070] In one embodiment, S1 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0071] S11: Extract QoS attributes from the input TSN data stream, and parse the priority value, latency budget value, packet loss rate value, bandwidth requirement value and burst traffic size value in the data stream one by one to generate the original QoS attribute dataset.

[0072] Specifically, the system receives input TSN data streams through the built-in traffic monitoring mechanism of the industrial communication network. This monitoring mechanism covers real-time control service data streams, monitoring and acquisition service data streams, and multimedia transmission service data streams, ensuring that attribute extraction covers the QoS requirements of various services. The system parses the received TSN data streams one by one, extracting five core attribute values. Priority value extraction is based on the IEEE 802.1Q standard, obtained by parsing the priority identifier field in the data stream frame structure; latency budget value extraction is based on the Industrial Service Level Agreement, determined by reading the latency clause of the corresponding service in the agreement; packet loss rate value is obtained by tracking the statistical data stream transmission process, recording the total number of data packets sent and received within a specific period, and calculating it through comparative analysis; bandwidth requirement value is determined based on the statistical results of the data stream transmission volume per unit time, covering the complete data stream transmission cycle; burst traffic magnitude value is obtained by analyzing the peak data stream transmission within a specific period in conjunction with the TSN standard frame transmission mechanism.

[0073] Because the units of measurement for each QoS attribute differ, the system normalizes the extracted raw attribute values ​​using the following formula:

[0074]

[0075] Where, x' ij Let x represent the original value of the j-th QoS attribute of the i-th data stream, min(x) j ) represents the minimum value of the j-th QoS attribute in all data streams, max(x) j () represents the maximum value of the j-th QoS attribute in all data streams, x' ij This represents the normalized QoS attribute value of the j-th data stream. The system associates and stores the original attribute values ​​of each data stream with the normalized values. The five attribute values ​​of all data streams together constitute the original QoS attribute dataset. The dataset is indexed by the data stream identifier to ensure that each data stream corresponds one-to-one with its attribute value.

[0076] S12: Vectorize the original QoS attribute dataset by combining the values ​​of the five QoS attributes of each data stream into a five-dimensional vector in a fixed-dimensional order to generate a multi-dimensional vector set of TSN data streams.

[0077] Specifically, the system performs uniform processing on the five attribute values ​​of each data stream in the original QoS attribute dataset. First, it determines a fixed order for the five dimensions of the vector: priority value, latency budget value, packet loss rate value, bandwidth requirement value, and burst traffic size value. This order remains consistent throughout all subsequent technical processes to avoid calculation errors caused by inconsistent dimension order. This order setting is based on the impact logic of QoS attributes in industrial business transmission, while also taking into account the data retrieval habits of subsequent comprehensive weight calculations and clustering operations.

[0078] The system extracts the normalized five attribute values ​​from a single data stream in the original QoS attribute dataset, arranges them in a preset fixed-dimensional order, and forms a five-dimensional vector, which is represented as follows:

[0079] X i =[x' i1 ,x' i2 ,x' i3 ,x' i4 ,x′ i5 ]

[0080] Where, x' i1 The corresponding normalized priority value, x' i2 The corresponding normalized delay budget value, x' i3 The corresponding normalized packet loss rate value, x'i4 The corresponding normalized bandwidth requirement value, x' i5 This corresponds to the normalized burst traffic magnitude. The values ​​in each dimension of the vector are directly derived from the corresponding normalized values ​​in the original QoS attribute dataset, ensuring that the vector accurately reflects the QoS attribute characteristics of the data stream. The system performs five-dimensional vector construction on all data streams using the same process, integrating all vectors to form a TSN data stream multi-dimensional vector set. The vector arrangement order in the vector set is consistent with the data stream storage order in the original QoS attribute dataset, ensuring a traceable correspondence between the vectors and the original data streams.

[0081] S13: Perform matrix construction processing on the TSN data stream multidimensional vector set, arrange all data stream vectors in the input order into an N-row, 5-column numerical matrix, and generate the TSN data stream attribute matrix.

[0082] Specifically, the system takes the TSN data stream multidimensional vector set as the processing object. First, it uses the timestamp information recorded when the data stream is received to check and determine the data stream input order corresponding to each five-dimensional vector in the vector set. This input order is the time order in which the data stream initially enters the system, ensuring that the subsequent matrix arrangement reflects the timing relationship of the data stream input.

[0083] The system arranges the five-dimensional vectors in the multidimensional vector set sequentially according to the determined data stream input order, forming a numerical matrix, which is represented as follows:

[0084] X'=[x′ ij ]N ×5

[0085] Where N represents the total number of TSN data streams, 5 represents the number of dimensions for QoS attributes, and x' ij The element in the i-th row and j-th column of the matrix corresponds to the j-th QoS attribute value after normalization of the i-th data stream. The number of rows in the matrix is ​​determined by the number of five-dimensional vectors in the vector set, and the number of columns is fixed at 5. The first column corresponds to the priority value, the second column corresponds to the latency budget value, the third column corresponds to the packet loss rate value, the fourth column corresponds to the bandwidth requirement value, and the fifth column corresponds to the burst traffic size value. The row index of the element corresponds to the data stream input order identifier, and the column index corresponds to the QoS attribute dimension identifier.

[0086] In one embodiment, S2 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0087] S21: Perform objective weight calculation on the TSN data stream attribute matrix, calculate the CRITIC-Entropy weight value based on the standard deviation, information entropy and correlation coefficient between attributes of each QoS attribute, and generate an objective weight vector.

[0088] Specifically, the system takes the TSN data stream attribute matrix as input and calculates the standard deviation of each QoS attribute using the following formula:

[0089]

[0090] Where, σ j Let x' represent the standard deviation of the j-th QoS attribute. ij μ represents the normalized value of the j-th attribute of the i-th data stream. j Let represent the mean of the j-th attribute, and N represent the total number of data streams. Further, the system calculates the correlation coefficient between attributes using the following formula:

[0091]

[0092] Where, r jk Cov(x') represents the correlation coefficient between the j-th and k-th attributes. j ,x' k () represents the covariance of two attributes. The raw CRITIC weights are calculated based on the standard deviation and correlation coefficient.

[0093]

[0094] Normalize the original values ​​to obtain the CRITIC weights Then, the entropy weights are calculated to determine the attribute probability distribution and information entropy:

[0095]

[0096] Where k = 1 / lnN is the normalization coefficient, and the entropy weight is:

[0097]

[0098] Preferably, the system uses a weighted average method to fuse the two types of weights, with a balance coefficient α, and the formula is:

[0099] w objj =α×w cj +(1-α)×w ej

[0100] For w objj After normalization, an objective weight vector W is formed. obj =[w obj1 ,w obj2 ,w obj3 ,w obj4 ,w obj5 ], where W obj This represents the objective weight of the j-th attribute.

[0101] S22: Perform discriminant weight calculation on the initial clustering results. Calculate the Fisher discriminant weight value by analyzing the ratio between-class variance and within-class variance, and generate a discriminant weight vector.

[0102] Specifically, the system first presets the number of clusters C according to the industrial business type, and then initializes the cluster center set using the K-means++ algorithm. in Let k be the initial cluster center. Calculate the mean and variance of the j-th attribute within each cluster:

[0103]

[0104] Where Ω k Let n be the sample set of the k-th cluster. k =|Ω k | represents the sample size, μ kj Let j be the mean of the j-th attribute in the k-th cluster. The variance is the within-class variance. Fisher's discrimination ratio is calculated based on the between-class and within-class differences:

[0105]

[0106] Among them, F j Let F represent the Fisher discriminant rate for the j-th attribute, with the numerator being the sum of inter-class differences and the denominator being the weighted sum of within-class differences (weights being the sample size). j Normalization:

[0107]

[0108] Form the discriminant weight vector W f =[w f1 ,w f2 ,w f3 ,w f4 ,w f5 ], where w fj This represents the Fisher discriminant weight of the j-th attribute.

[0109] S23: The objective weight vector and the discriminant weight vector are fused together. The two types of weight values ​​are normalized and then linearly superimposed to generate a comprehensive attribute difference weight vector.

[0110] Specifically, the system will use the objective weight vector w objj With the discriminant weight vector W f As input, the two types of weights are first normalized and validated to ensure... and Perform the weighted calculation before linear superposition according to the preset fusion formula:

[0111]

[0112] Among them, w j Let be the comprehensive attribute difference weight vector of the traffic QoS attribute in the j-th dimension. Let be the CRITIC-Entropy objective weight vector of the traffic QoS attribute in the j-th dimension. Let be the Fisher discrimination weight of the traffic QoS attribute in the j-th dimension, and D be the total number of dimensions of the QoS attributes. After calculation, the system integrates all w j The values ​​are used to generate a comprehensive attribute difference weight vector W = [w1, w2, w3, w4, w5]. The values ​​of each element in this vector reflect the weight ratio of the corresponding QoS attribute in the subsequent clustering operation. The higher the weight of the attribute, the greater its impact on the clustering result.

[0113] S24: Perform weighted distance calculation on the comprehensive attribute difference weight vector and the current cluster center, and introduce the weight values ​​of each QoS attribute into the Euclidean distance formula for dimensional weighting to generate a weighted distance matrix.

[0114] Specifically, the system calls the comprehensive attribute difference weight vector W and the current cluster center set \(F^{(t-1)}\) (\(t\) is the iteration round), and calculates the weighted distance between each sample \(x'_i\) (the normalized attribute vector of the i-th data stream) in the TSN data stream attribute matrix and each cluster center \(f_k^{(t-1)}\), with the formula as follows:

[0115]

[0116] in, w is the weighted distance between the i-th data stream and the k-th cluster center in the t-th iteration. j Let x' be the overall weight of the j-th attribute. ij Let j be the normalized value of the j-th attribute of the i-th data stream. This represents the value of the j-th attribute of the k-th cluster center in the (t-1)-th iteration.

[0117] The system calculates the weighted distance between all samples and cluster centers, and arranges the results according to the index relationship of "sample-cluster center" to form an N×C dimensional weighted distance matrix. The matrix row index corresponds to the data stream identifier (1 to N), the column index corresponds to the cluster center identifier (1 to C), and the matrix elements directly reflect the weighted difference between the sample and the cluster center.

[0118] S25: Perform membership update processing on the weighted distance matrix, redistribute the membership probabilities of samples to each cluster according to the inverse distance ratio, and generate the updated membership matrix.

[0119] Specifically, the system uses a weighted distance matrix D (t) Based on this, and combined with the fuzzy coefficient m (m>1), the membership relationship of the sample to each cluster is updated according to the fuzzy C-means (FCM) algorithm rules, as follows:

[0120]

[0121] in, is the membership degree of the i-th data stream in the t-th iteration to the k-th cluster, with a value range of [0,1]. is the weighted distance between the i-th data stream and the l-th cluster center in the t-th iteration; m is the fuzzy coefficient, used to control the degree of fuzziness of the membership degree.

[0122] Preferably, the system calculates the membership degree of all samples to ensure that for any i-th data stream, the following condition is met. That is, the sum of the membership degrees of a sample to all clusters is 1. The calculation results are arranged by "sample-cluster" index to generate an updated N×C dimensional membership matrix. The matrix elements reflect the probability of a sample belonging to a cluster.

[0123] S26: Update the cluster centers of the membership matrix and the TSN data stream attribute matrix, calculate the new centroid positions of each cluster by weighted average, and generate a new set of cluster centers.

[0124] Specifically, the system calls the updated membership matrix U. (t) Using the TSN data stream attribute matrix \(X'\), the new cluster centers are calculated according to the weighted average rule, as follows:

[0125]

[0126] in, Let j be the new value of the j-th attribute of the k-th cluster center in the t-th iteration. x' is the weighted value of the membership degree of the i-th data stream to the k-th cluster, with the weight being a power of m. ij Let be the normalized value of the j-th attribute in the i-th data stream. The numerator is the weighted sum of the sample attribute values, and the denominator is the sum of the weighted membership values, ensuring that the cluster center can reflect the attribute mean characteristics of the belonging sample.

[0127] The system calculates all attribute dimensions (1-5) for each cluster center and integrates them to obtain the new set of cluster centers in the t-th iteration:

[0128]

[0129] in, Let be the five-dimensional attribute vector of the k-th new cluster center. The set of new cluster centers will serve as the input for the next iteration (if it does not converge) or as the output of the final clustering result.

[0130] In one embodiment, S3 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0131] S31: Calculate the weighted membership degree of the cluster center set based on the preset 5G QoS identifier attribute vector, adjust the dimensional contribution ratio based on the comprehensive attribute difference weight, and generate the identifier-cluster membership degree matrix.

[0132] Specifically, the system uses a pre-set 5G QoS identifier attribute vector and a cluster center set F = [f1, f2, ..., f C The system takes the 5G QoS identifier attribute vector and the comprehensive attribute difference weight vector W = [w1, w2, ..., w5] as input. First, it performs normalization processing on the 5G QoS identifier attribute vector, using the same processing method as the TSN data stream attribute normalization, ensuring consistent attribute dimensions and numerical ranges. Preferably, the system calculates the weighted Euclidean distance between each 5G QoS identifier and each cluster center, using the following formula:

[0133]

[0134] in, Let y' be the weighted Euclidean distance from the j-th 5G QoS identifier to the k-th cluster center. jl f is the normalized value of the l-th attribute of the j-th 5G QoS identifier. kl w is the value of the l-th attribute of the k-th cluster center. j , where is the comprehensive attribute difference weight for the l-th attribute.

[0135] Based on the weighted Euclidean distance, the system calculates the basic membership degree of the j-th 5G QoS identifier to the k-th cluster center according to a preset formula:

[0136]

[0137] in, Let j be the basic membership degree of the j-th 5G QoS identifier to the k-th cluster center. Let be the weighted Euclidean distance from the j-th 5G QoS identifier to the p-th cluster center, C be the number of clusters, and m be the fuzzy coefficient. The system calculates the basic membership degree between all 5G QoS identifiers and cluster centers one by one, arranges them according to the "identifier-cluster center" index relationship, and generates an M×C dimensional identifier-cluster membership degree matrix. Where M represents the total number of 5G QoS identifiers.

[0138] S32: Perform relative membership transformation on the identifier-cluster membership matrix, perform maximum value normalization on the membership vector of each 5G QoS identifier, and divide the membership degree of each dimension by the maximum value of its row to generate the relative membership degree.

[0139] Specifically, the system uses the identifier-clustering membership matrix U * For input, the membership vector corresponding to each 5G QoS identifier (i.e., all elements in the j-th row of the matrix) Perform maximum value extraction to determine the maximum value of the membership degree of the j-th 5G QoS identifier to all cluster centers. Where n is the cluster center index corresponding to the maximum value (1≤n≤C).

[0140] The system assigns membership to each basic degree according to a preset formula. Perform normalization to generate relative membership degrees:

[0141]

[0142] Among them, g jk To represent the relative membership of the j-th 5G QoS identifier to the k-th cluster center, To show the basic membership degree of the j-th 5G QoS identifier to the k-th cluster center, Let g be the maximum value among all cluster center memberships of the j-th 5G QoS identifier. This formula eliminates the influence of differences in the range of basic membership values ​​between different identifiers by normalizing the maximum value, making g... jk ∈(0,1], which more intuitively reflects the relative matching degree of the identifier to the cluster center.

[0143] The system calculates the relative membership degree between all 5G QoS identifiers and cluster centers, forming a relative membership degree matrix G = [g] based on the "identifier-cluster center" index relationship. jk ] M×C Each element in the matrix directly quantifies the relative matching characteristics between the corresponding identifier and the cluster center.

[0144] S33: Generate an adaptive threshold for relative membership, group and count the difference between the maximum and second largest membership of each identifier according to resource type, calculate the average value of the difference of each group as the boundary threshold, derive the core threshold based on the boundary threshold according to a preset ratio, and generate the core domain threshold and the boundary domain threshold.

[0145] Specifically, based on the relative membership degree of each sample, a threshold ε is obtained, which is used for the logical judgment of numerical screening:

[0146]

[0147] Where α∈{1,2,3} represent DC-GBR, GBR, and Non-GBR, respectively; L a This indicates the number of 5QI values ​​for resource type a; j a Represents the set of 5QIs of resource type a; g jp and g jq g jk The maximum and second-highest values ​​of the relative membership degree of the 5QI set. Observing the formula, we can see that the threshold ε is the average difference between the maximum and second-highest relative membership degrees of the 5QI set, reflecting the average distribution level of the 5QI set. That is, the larger the ε value, the stronger the membership degree of set element I. j Distance from the center of aggregation αf p The closer they are, the greater the distance to αf q The farther away, the easier it is to judge; conversely, the smaller the ε value, the weaker I is. j Distance αf p and distance αf q They are similar, meaning it's difficult to judge.

[0148] Then, based on the relative membership degree and the threshold, the evaluation function and the splitting threshold pair are calculated. The former corresponds to the minimum distance and relative distance based on the distance-based splitting method; the latter corresponds to the minimum distance threshold and the relative distance threshold. The calculation method of the evaluation function is as follows:

[0149]

[0150] Among them, L jk =g jk >ε, taking the value 0 or 1. Threshold pair for dividing regions (α) j ,β j The calculation method is as follows:

[0151] α j =1

[0152]

[0153] S34: Divide the relative membership degree into three regions, assign the identifiers with a relative membership degree not lower than the core domain threshold to the core domain, and assign the identifiers with a relative membership degree between the boundary domain threshold and the core domain threshold to the boundary domain, thereby generating a core domain set and a boundary domain set.

[0154] Specifically, the system uses relative membership degree g jk Core domain threshold α j and boundary domain threshold β j As input, divide all 5G QoS identifiers corresponding to each cluster center into regions. For the k-th cluster center, iterate through all the relative membership degrees g corresponding to the 5G QoS identifiers. jk :

[0155] If gjk ≥α j If the j-th 5G QoS identifier is assigned to the core domain of the k-th cluster center, then the 5G QoS identifier within the core domain has the highest QoS requirement matching degree with the k-th cluster center and possesses stable QoS guarantee capabilities. In subsequent mapping processes, it will be given priority as the mapping target for the TSN data stream corresponding to this cluster center.

[0156] If (β) j <g jk <α j If the 5G QoS identifier is assigned to the boundary domain of the k-th cluster center, then the j-th 5G QoS identifier will be assigned to the boundary domain of the k-th cluster center. The 5G QoS identifiers within the boundary domain have a moderate match degree with the QoS requirements of the k-th cluster center, and will serve as a supplementary mapping target when the core domain 5G QoS identifier resources are insufficient.

[0157] If g jk ≤β j If the j-th 5G QoS identifier is assigned to the negative domain of the k-th cluster center, the 5G QoS identifier in the negative domain has a low match with the QoS requirement of the k-th cluster center and will not participate in the mapping of the TSN data stream corresponding to that cluster center.

[0158] The system performs the above division on all C cluster centers one by one, integrates the core domain 5G QoS identifier of each cluster center, and generates a core domain set CR = {CR1, CR2, ..., CR...} C}, where CR k Let be the core domain 5G QoS identifier set for the k-th cluster center; simultaneously, integrate the boundary domain 5G QoS identifiers of each cluster center to generate the boundary domain set BR = {BR1, BR2, ..., BR...} C}, where BR k This is the set of 5G QoS identifiers for the boundary domain of the k-th cluster center. The core domain set and the boundary domain set together constitute the target identifier pool for subsequent TSN data stream QoS mapping.

[0159] In one embodiment, S4 of the cross-domain QoS mapping method based on three-branch decision provided by the present invention specifically includes the following steps:

[0160] S41: Perform traffic clustering and grouping processing on the membership matrix, assign each TSN data stream to the cluster corresponding to its maximum membership, and generate a clustered traffic group set.

[0161] Specifically, the system calls the updated membership matrix. Where N is the total number of TSN data streams, and C is the number of clusters. Let represent the membership degree of the i-th data stream to the k-th cluster, with a value in the range [0,1], and satisfying . The system performs clustering and affiliation determination for each TSN data stream by traversing all elements of the corresponding row in the membership matrix. Select the cluster identifier corresponding to the highest membership value and assign the data stream to that cluster.

[0162] For example, if the membership row vector of the i-th data stream contains If the maximum value is found, the system assigns the data stream to the 3rd cluster. The system performs the above-mentioned attribution and allocation operations on all N TSN data streams one by one, grouping the data streams according to the cluster identifier to form a clustered traffic group set G = {G1, G2, ..., G...}. C}, where G k Let the subset of data stream corresponding to the k-th cluster satisfy the following condition: and (No duplicate data stream allocation). This set clarifies the cluster affiliation of each data stream, providing a grouping basis for subsequent matching of 5G QoS identifiers by cluster.

[0163] S42: Perform normalized distance calculation on each data stream in the clustered traffic group set, calculate the weighted Euclidean distance of each data stream to each 5G QoS identifier in the core domain and boundary domain based on the comprehensive attribute difference weight, and linearly scale the distance to the [0,1] interval through the maximum and minimum value method to generate a normalized similarity index.

[0164] Specifically, the system targets each group G in the clustered traffic group set G. k The data stream calls the comprehensive attribute difference weight vector W = [w1, w2, ..., w D ], and the core domain set CR k Boundary domain set BR k Calculate the weighted Euclidean distance between the data stream and the identifier.

[0165] For the data stream i (i∈G) of the k-th cluster k ) and 5G QoS identifier j (j∈CR) k ∪BR k The weighted Euclidean distance formula is:

[0166]

[0167] in, Let x' be the weighted Euclidean distance between data stream i and 5G QoS identifier j. it Let y' be the normalized value of the t-th QoS attribute of data stream i, taken from the TSN data stream attribute matrix X', y' jt w is the normalized value of the t-th QoS attribute of the 5G QoS identifier j, taken from a preset 5G QoS identifier attribute vector set. tdenoted as the comprehensive attribute difference weight for the t-th QoS attribute.

[0168] The system is for all Max-min normalization is performed to convert the distance into a normalized similarity index in the [0,1] interval. The formula is as follows:

[0169]

[0170] Among them, s ij The normalized similarity index s between data stream i and 5G QoS identifier j ij The closer to 1, the higher the QoS matching degree, max(d w ) represents the maximum weighted Euclidean distance between all data streams and their corresponding identifiers within the current cluster k, min(d w The minimum distance is 0. The system records all data streams (s) according to the "data stream-identifier" index relationship. ij Generate a normalized similarity index set S.

[0171] S43: Perform remaining capacity assessment on the real-time load status of 5G QoS identifiers, query the difference between the current traffic volume carried by each identifier and its preset load limit, calculate the ratio of the difference to the load limit, and generate a load remaining rate index.

[0172] Specifically, the system queries the load parameters of each 5G QoS identifier j in real time through the resource management interface of the 5G core network (following the 3GPP TS28.531 standard): one is the load limit L. j (The preset value corresponds to the maximum number of TSN data streams that identifier j can carry, determined by network resource configuration), and the second is the current load C. j (Real-time value, corresponding to the number of TSN data stream sessions carried by identifier j, counted by the session management function SMF).

[0173] Based on the query L j With C j The system calculates the load balancing rate of identifier j using the following formula:

[0174]

[0175] Among them, R j The load balancing rate metric (R) for 5G QoS identifier j j >0) indicates that the identifier still has the capacity to carry data, R j =0 indicates that the flag is full (this indicator reflects I). j The remaining load-bearing capacity, the larger the value, the greater the I. j The more data streams that can be added, the more the system can calculate R for each 5G QoS identifier in all core and boundary domains. jGenerate a load margin metric set R.

[0176] S44: Dynamically weight and fuse the normalized similarity index and the load balancing index. By using a preset weight factor, the normalized similarity index is inverted and then linearly weighted and combined with the load balancing index to balance the service quality matching degree and the load balancing requirement, and generate a dynamic weight decision value.

[0177] Specifically, when performing dynamic weighted QoS mapping, the system can use a greedy approximate mapping, prioritizing the 5QIs with the highest similarity (i.e., membership degree) to the clustered flows, concentrating the load on a few 5QIs. Furthermore, the membership degree is calculated between the cluster center and the 5QIs, rather than by calculating it between each element's traffic and the 5QIs. Although the greedy approximate mapping reduces computation (i.e., the number of clustered flows C < the number of data flows N), it ignores the QoS differences between data flows and different 5QIs, roughly replacing them with the QoS differences between the cluster center and the 5QIs.

[0178] When the remaining amount of 5QIA is greater than the total number of elements to be mapped in the k-th aggregate stream, 5QIA is closer to the center than 5QIB. A greedy approximate mapping is performed, mapping all elements of the k-th aggregate stream to 5QIA. Clearly, there is a significant QoS difference; some sample points closer to 5QIB are mapped to the more distant 5QIA.

[0179] When the remaining amount of 5QIA is less than the total number of elements to be mapped in the k-th aggregate stream, if the remaining amount of 5QIA is greater than 0, a greedy approximate mapping is performed. The mapping result depends on the execution order of the streams. Ideally, elements closer to 5QIA are mapped to 5QIA, and elements closer to 5QIB are mapped to 5QIB; in the worst case, elements closer to 5QIA are mapped to 5QIB, and elements closer to 5QIB are mapped to 5QIA.

[0180] After using the region partitioning method, each aggregated flow has a corresponding core domain and boundary domain, both of which contain elements of 5QI. Dynamic weighted QoS mapping is performed on the TSN traffic of the aggregated flow set elements to 5QI, following a "core domain first, then boundary domain" order. This involves iterating through the elements f of the aggregated flow set. ki , relative to the core domain or boundary domain element I j Calculate its dynamic weights:

[0181]

[0182] Where r is the weighting factor, with a value satisfying 0 ≤ r ≤ 1; N0 is the maximum load limit set for a single 5QI; u j For the current y j The number of loads; d ki,j f ki To Ij The normalized distance.

[0183] Weight The first one considers I j The load margin, the latter taking into account f ki and I j The QoS similarity is determined by a weighted sum of two factors to obtain a dynamic weight. The process in this step involves first filtering out I... where the load margin is not zero. j Then calculate the corresponding dynamic weights, and finally f ki Mapped to the I with the largest corresponding weight j Up, until the aggregate stream set is empty or the load margin of elements in both the core domain and the boundary domain is 0.

[0184] S45: Perform weight priority mapping on the identifiers in the core domain set, select available identifiers to map data streams in descending order of dynamic weight values. If all core domain identifiers are full, automatically switch to the boundary domain set to continue weight priority mapping, complete the mapping and allocation of all data streams, and generate a mapping table between TSN streams and 5G QoS identifiers.

[0185] Specifically, the system uses the core domain set CR k The 5G QoS identifier within the data is prioritized for weighted priority mapping. The clustered traffic groups C are traversed. k Each data stream f in i Filter core domains where the load margin is not 0 (i.e.) The identifier I j ∈CR k Calculate each f i To these I j Dynamic weights and according to Sort in descending order. Select the identifier with the highest weight. f i Mapped to Simultaneously updated Current load

[0186] If the core domain set CR k When the load margin of all identifiers in the set is 0, the system automatically switches to the boundary domain set BR. k , for BR k For traffic groups with a non-zero load margin, the same dynamic weight calculation and descending mapping operation are performed. This process is repeated until the traffic group C is clustered. k It is empty, or the load availability of all identifiers in the core domain and the boundary domain is 0.

[0187] After mapping all clustered traffic groups, the system integrates all mapping relationships to generate a mapping table between TSN flows and 5G QoS identifiers. This table contains information such as data flow identifier, mapped 5G QoS identifier, dynamic weight value, and load availability rate at the time of mapping. It is then sent to network devices such as TSN switches and 5G base stations via the southbound interface to guide the devices in performing traffic scheduling and ensure that TSN data flows obtain matching QoS guarantees during cross-domain transmission.

[0188] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0189] Based on the same inventive concept, this application also provides a cross-domain QoS mapping system based on three-way decision-making for implementing the cross-domain QoS mapping method based on three-way decision-making described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the cross-domain QoS mapping system based on three-way decision-making provided below can be found in the limitations of the cross-domain QoS mapping method based on three-way decision-making described above, and will not be repeated here.

[0190] Preferably, such as Figure 3 As shown, the present invention provides a cross-domain QoS mapping system 500 based on three-way decision-making, which is configured with the following modules:

[0191] The TSN data stream attribute matrix construction module 510 is used to perform multi-dimensional attribute vector construction processing on the input TSN data stream, extract the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size of each data stream, combine the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and construct a matrix to generate the TSN data stream attribute matrix.

[0192] The comprehensive weighted clustering analysis module 520 is used to calculate the weight value of each QoS attribute based on the TSN data stream attribute matrix by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminative weight. It combines the weight values ​​of each QoS attribute to perform weighted calculation of the distance between cluster centers and samples and iteratively update the membership relationship to generate a set of cluster centers and a membership matrix.

[0193] The three-branch decision region division module 530 is used to divide the decision region into three branches based on the cluster center set, the comprehensive attribute difference weight and the preset 5G QoS identifier attribute vector. It calculates the relative membership degree of each 5G QoS identifier to the cluster center and automatically generates the core domain and boundary domain division threshold according to the distribution of the relative membership degree, dividing the 5G QoS identifier into the core domain set and boundary domain set corresponding to each cluster.

[0194] The TSN-5G QoS dynamic mapping module 540 is used to calculate the dynamic weight value of the target mapping based on the real-time load status of the membership matrix, core domain set, boundary domain set and 5G QoS identifier, according to the load remaining rate and QoS similarity, and map the TSN data stream to the 5G QoS identifier with the largest dynamic weight. It generates and outputs a mapping relationship table between TSN stream and 5G QoS identifier, and the mapping relationship table is used to instruct network devices to perform traffic scheduling.

[0195] Preferably, the TSN data stream attribute matrix construction module 510 provided in this application is configured with the following units:

[0196] The QoS attribute extraction unit is used to extract QoS attributes from the input TSN data stream, and parse the priority value, latency budget value, packet loss rate value, bandwidth requirement value and burst traffic size value in the data stream one by one to generate the original QoS attribute dataset.

[0197] The data stream vector transformation unit is used to vectorize the original QoS attribute dataset, combining the five QoS attribute values ​​of each data stream into a five-dimensional vector in a fixed-dimensional order to generate a TSN data stream multi-dimensional vector set.

[0198] The attribute matrix construction unit is used to perform matrix construction processing on the multidimensional vector set of TSN data streams, arranging all data stream vectors in the input order into an N-row, 5-column numerical matrix to generate the TSN data stream attribute matrix.

[0199] Preferably, the comprehensive weighted clustering analysis module 520 provided in this application is configured with the following units:

[0200] The objective weight calculation unit is used to perform objective weight calculation on the TSN data stream attribute matrix. It calculates the CRITIC-Entropy weight value based on the standard deviation, information entropy and correlation coefficient between attributes of each QoS attribute, and generates an objective weight vector.

[0201] The Fisher discriminant weight calculation unit is used to calculate the discriminant weights of the initial clustering results. It calculates the Fisher discriminant weight values ​​by analyzing the ratio of between-class variance to within-class variance and generates a discriminant weight vector.

[0202] The comprehensive weight fusion unit is used to fuse the objective weight vector and the discrimination weight vector, and linearly superimpose the normalized weight values ​​of the two types to generate a comprehensive attribute difference weight vector.

[0203] The weighted distance calculation unit is used to perform weighted distance calculation on the comprehensive attribute difference weight vector and the current cluster center. It introduces the weight values ​​of each QoS attribute into the Euclidean distance formula for dimensional weighting and generates a weighted distance matrix.

[0204] The membership update unit is used to update the membership of the weighted distance matrix. It redistributes the membership probabilities of samples to each cluster according to the inverse distance ratio, and generates the updated membership matrix.

[0205] The cluster center update unit is used to update the cluster centers of the membership matrix and the TSN data stream attribute matrix. It calculates the new centroid positions of each cluster by weighted averaging and generates a new set of cluster centers.

[0206] Preferably, the three decision-making region division module 530 provided in this application is configured with the following units:

[0207] The identifier-cluster weighted membership calculation unit is used to calculate the weighted membership of the cluster center set based on the preset 5G QoS identifier attribute vector, and to adjust the dimensional contribution ratio based on the comprehensive attribute difference weight to generate the identifier-cluster membership matrix.

[0208] The relative membership transformation unit is used to perform relative membership transformation on the identifier-cluster membership matrix. It performs maximum value normalization processing on the membership vector of each 5G QoS identifier, and divides the membership degree of each dimension by the maximum value of its row to generate the relative membership degree.

[0209] The adaptive domain threshold generation unit is used to generate adaptive thresholds for relative membership degrees. It groups and counts the difference between the maximum and second largest membership degrees of each identifier according to resource type, calculates the average value of the difference of each group as the boundary threshold, and derives the core threshold based on the boundary threshold according to a preset ratio to generate the core domain threshold and the boundary domain threshold.

[0210] The three-branch decision region division execution unit is used to divide the relative membership degree into three regions. The identifiers with a relative membership degree not lower than the core domain threshold are assigned to the core domain, and the identifiers with a relative membership degree between the boundary domain threshold and the core domain threshold are assigned to the boundary domain, generating a core domain set and a boundary domain set.

[0211] Preferably, the TSN-5G QoS dynamic mapping module 540 provided in this application is configured with the following units:

[0212] The traffic clustering grouping unit is used to perform traffic clustering grouping processing on the membership matrix, assigning each TSN data stream to the cluster corresponding to its maximum membership degree, and generating a clustered traffic grouping set.

[0213] The normalized similarity calculation unit is used to perform normalized distance calculation on each data stream in the clustered traffic group set. Based on the comprehensive attribute difference weight, it calculates the weighted Euclidean distance of each data stream to each 5G QoS identifier in the core domain and boundary domain. The distance is linearly scaled to the [0,1] interval by the maximum and minimum value method to generate a normalized similarity index.

[0214] The load remaining capacity assessment unit is used to assess the remaining carrying capacity of the real-time load status of 5G QoS identifiers, query the difference between the current carrying traffic of each identifier and its preset load limit, calculate the ratio of the difference to the load limit, and generate the load remaining rate index.

[0215] The dynamic weight fusion calculation unit is used to dynamically fuse the normalized similarity index and the load remaining rate index. By using preset weight factors, the normalized similarity is inverted and then linearly weighted and combined with the load remaining rate to balance the service quality matching degree and the load balancing requirement, and generate dynamic weight decision values.

[0216] The QoS identifier weight priority mapping unit is used to perform weight priority mapping on identifiers within the core domain set. It selects available identifiers to map data streams in descending order of dynamic weight values. If all identifiers in the core domain are fully loaded, it automatically switches to the boundary domain set to continue performing weight priority mapping, completes the mapping and allocation of all data streams, and generates a mapping relationship table between TSN streams and 5G QoS identifiers.

[0217] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described cross-domain QoS mapping method based on three-way decision-making.

[0218] In one embodiment, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described cross-domain QoS mapping method based on three-way decision-making.

[0219] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0220] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0221] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A cross-domain QoS mapping method based on three-branch decision-making, characterized in that, Includes the following steps: S1: Perform multi-dimensional attribute vector construction processing on the input TSN data stream, extract the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size of each data stream, combine the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and construct a matrix to generate the TSN data stream attribute matrix. S2: Based on the TSN data stream attribute matrix, calculate the weight value of each QoS attribute by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminative weight, combine the weight values ​​of each QoS attribute to perform weighted calculation of the distance between cluster centers and samples and iteratively update the membership relationship to generate a cluster center set and a membership matrix. S3: Based on the cluster center set, the comprehensive attribute difference weight, and the preset 5G QoS identifier attribute vector, perform three-branch decision region division, calculate the relative membership degree of each 5G QoS identifier to the cluster center, and automatically generate core domain and boundary domain division thresholds according to the relative membership degree distribution, and divide the 5G QoS identifier into the core domain set and boundary domain set corresponding to each cluster. S4: Based on the membership matrix, the core domain set, the boundary domain set, and the real-time load status of the 5G QoS identifier, calculate the dynamic weight value of the target mapping according to the load remaining rate and QoS similarity, map the TSN data stream to the 5G QoS identifier with the largest dynamic weight, generate and output the mapping relationship table between TSN stream and 5G QoS identifier, and use the mapping relationship table to instruct network devices to perform traffic scheduling.

2. The method according to claim 1, characterized in that, S1 includes: S11: Extract QoS attributes from the input TSN data stream, and parse the priority value, latency budget value, packet loss rate value, bandwidth requirement value and burst traffic size value in the data stream one by one to generate the original QoS attribute dataset. S12: The original QoS attribute dataset is vectorized by combining the five QoS attribute values ​​of each data stream into a five-dimensional vector in a fixed-dimensional order to generate a multi-dimensional vector set of TSN data streams. S13: Perform matrix construction processing on the TSN data stream multidimensional vector set, arrange all data stream vectors in the input order into an N-row, 5-column numerical matrix, and generate the TSN data stream attribute matrix.

3. The method according to claim 1, characterized in that, S2 includes: S21: Perform objective weight calculation processing on the TSN data stream attribute matrix, calculate CRITIC-Entropy weight values ​​based on the standard deviation, information entropy and correlation coefficient between attributes of each QoS attribute, and generate an objective weight vector. S22: Perform discriminant weight calculation on the initial clustering results, calculate Fisher discriminant weight values ​​by analyzing the proportional relationship between inter-class variance and intra-class variance, and generate a discriminant weight vector; S23: The objective weight vector and the discrimination weight vector are fused together. The two types of weight values ​​are normalized and then linearly superimposed to generate a comprehensive attribute difference weight vector. S24: Perform weighted distance calculation on the comprehensive attribute difference weight vector and the current cluster center, and introduce the weight values ​​of each QoS attribute into the Euclidean distance formula for dimensional weighting to generate a weighted distance matrix; S25: Perform membership update processing on the weighted distance matrix, and redistribute the membership probabilities of samples to each cluster according to the inverse distance ratio to generate an updated membership matrix; S26: Update the cluster centers of the membership matrix and the TSN data stream attribute matrix, calculate the new centroid positions of each cluster by weighted average, and generate a new set of cluster centers.

4. The method according to claim 3, characterized in that, The formula for calculating the comprehensive attribute difference weight vector is as follows: Among them, w j Let be the comprehensive attribute difference weight vector of the traffic QoS attribute in the j-th dimension. Let be the CRITIC-Entropy objective weight vector of the traffic QoS attribute in the j-th dimension. Let be the Fisher discrimination weight of the traffic QoS attribute in the j-th dimension, and D be the total number of dimensions of the QoS attribute.

5. The method according to claim 1, characterized in that, S3 includes: S31: Calculate the weighted membership degree of the cluster center set based on the preset 5G QoS identifier attribute vector, adjust the dimension contribution ratio based on the comprehensive attribute difference weight, and generate an identifier-cluster membership degree matrix. S32: Perform relative membership transformation on the identifier-cluster membership matrix, perform maximum value normalization on the membership vector of each 5G QoS identifier, and divide the membership degree of each dimension by the maximum value of its row to generate the relative membership degree; S33: Generate an adaptive threshold for the relative membership degree, group and statistically analyze the difference between the maximum and second largest membership degree of each identifier according to resource type, calculate the average value of the difference of each group as the boundary threshold, derive the core threshold based on the boundary threshold according to a preset ratio, and generate the core domain threshold and the boundary domain threshold. S34: Divide the relative membership degree into three regions, assign the identifiers with a relative membership degree not lower than the core domain threshold to the core domain, and assign the identifiers with a relative membership degree between the boundary domain threshold and the core domain threshold to the boundary domain, thereby generating a core domain set and a boundary domain set.

6. The method according to claim 5, characterized in that, The formula for calculating the relative membership degree is: Among them, g jk To represent the relative membership of the j-th 5G QoS identifier to the k-th cluster center, To show the basic membership degree of the j-th 5G QoS identifier to the k-th cluster center, Let j be the maximum value of the membership degree of the j-th 5G QoS identifier to all cluster centers. Let i be the weighted Euclidean distance from data point i to cluster center k. Let be the weighted distance from data point i to cluster center p.

7. The method according to any one of claims 1-6, characterized in that, S4 includes: S41: Perform traffic clustering and grouping processing on the membership matrix, assign each TSN data stream to the cluster corresponding to its maximum membership degree, and generate a clustered traffic group set; S42: Perform normalized distance calculation processing on each data stream in the clustered traffic group set, calculate the weighted Euclidean distance of each data stream to each 5G QoS identifier in the core domain and boundary domain based on the comprehensive attribute difference weight, and linearly scale the distance to the [0,1] interval through the maximum and minimum value method to generate a normalized similarity index. S43: Perform remaining carrying capacity assessment on the real-time load status of 5G QoS identifiers, query the difference between the current carrying capacity of each identifier and its preset load limit, calculate the ratio of the difference to the load limit, and generate a load remaining rate index. S44: Dynamically weight fusion of the normalized similarity index and the load balancer index, and linearly weighted combination of the normalized similarity index and the load balancer index by means of a preset weight factor, to balance the service quality matching degree and the load balancing requirement, and generate dynamic weight decision value. S45: Perform weight priority mapping on the identifiers in the core domain set, select available identifiers to map data streams in descending order of dynamic weight values, and automatically switch to the boundary domain set to continue weight priority mapping if all core domain identifiers are fully loaded, complete the mapping and allocation of all data streams, and generate a mapping relationship table between TSN streams and 5G QoS identifiers.

8. A cross-domain QoS mapping system based on three-branch decision-making, characterized in that, The system includes: The TSN data stream attribute matrix construction module is used to perform multi-dimensional attribute vector construction processing on the input TSN data stream. It extracts the QoS attribute values ​​of priority, latency budget, packet loss rate, bandwidth requirement and burst traffic size of each data stream, combines the QoS attribute values ​​into a multi-dimensional vector according to the data stream order and constructs a matrix to generate the TSN data stream attribute matrix. The comprehensive weighted clustering analysis module is used to calculate the weight value of each QoS attribute based on the TSN data stream attribute matrix by integrating the comprehensive attribute difference weight of CRITIC-Entropy objective weight and Fisher discriminant weight, and to perform weighted calculation of the distance between cluster centers and samples by combining the weight values ​​of each QoS attribute and iteratively updating the membership relationship to generate a cluster center set and a membership matrix. The three-branch decision region division module is used to perform three-branch decision region division based on the cluster center set, the comprehensive attribute difference weight and the preset 5G QoS identifier attribute vector, calculate the relative membership degree of each 5G QoS identifier to the cluster center, and automatically generate the core domain and boundary domain division threshold according to the relative membership degree distribution, so as to divide the 5G QoS identifier into the core domain set and boundary domain set corresponding to each cluster. The TSN-5G QoS dynamic mapping module is used to calculate the dynamic weight value of the target mapping based on the membership matrix, the core domain set, the boundary domain set, and the real-time load status of the 5G QoS identifier, according to the load remaining rate and QoS similarity. It maps the TSN data stream to the 5G QoS identifier with the largest dynamic weight, generates and outputs a mapping relationship table between TSN streams and 5G QoS identifiers, and the mapping relationship table is used to instruct network devices to perform traffic scheduling.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images