Multi-terminal cooperative access resource allocation system under 5G network slice

Abstract

The invention relates to the technical field of network slices, in particular to a multi-terminal collaborative access resource allocation system under 5G network slices, which comprises a terminal demand parameter standardization processing module for collecting hard QoS (Quality of Service) indexes corresponding to all terminals; the service attribute parameters of each terminal are analyzed in sequence by adopting an analytic hierarchy process, the rigid QoS indexes and the service attribute parameters are processed in a standardized mode, and demand parameters of each terminal are constructed; the terminal clustering and cluster division module adopts a K-means algorithm to cluster all terminals to obtain a plurality of terminal clusters, and the cluster configuration requirement extraction module sets network slices with the number corresponding to that of the terminal clusters for each terminal cluster in the terminal clustering and cluster division module; extracting a common demand and a core QoS constraint of each terminal cluster to form a configuration demand corresponding to each terminal cluster; and the slice resource quantitative distribution module converts the configuration demand in the cluster configuration demand extraction module into the slice resource of the corresponding network slice.

Description

Technical Field

[0001] This invention relates to the field of network slicing technology, and more specifically, to a multi-terminal collaborative access resource allocation system under 5G network slicing. Background Technology

[0002] With the deep penetration and widespread application of 5G technology in diverse scenarios such as industrial control, telemedicine, and autonomous driving, the access demand for various types of terminals has surged explosively. Terminals in different scenarios are equipped with differentiated hard QoS indicators. For example, industrial control terminals explicitly require end-to-end latency ≤10ms and packet loss rate ≤10⁻ 6 The stringent requirements of remote medical terminals for stable transmission bandwidth and low latency, as well as the core requirements of autonomous driving terminals for highly reliable and low-latency data transmission, etc. To address the aforementioned issues, existing solutions rely on the mature 5G core network (5GC) and access network (RAN) standardization technology system, combined with inter-terminal collaborative communication protocols. First, the overall 5G network is logically sliced, constructing multiple isolated network slice instances with independently allocated resources. Then, based on the differences in hard QoS indicators for terminals in different scenarios, the core configuration requirements for each network slice are clearly defined (covering key parameters such as bandwidth resource quotas, latency thresholds, packet loss rate limits, and computing power allocation standards). Subsequently, through a precise indicator matching mechanism, the hard QoS indicators of each terminal are compared and adapted with the configuration requirements of the corresponding network slice. Finally, the precise allocation and access of terminals to the target network slice is completed, thereby ensuring resource guarantees for different types of terminals within dedicated network slices. This ensures that the hard QoS indicators of various terminals are fully met, thus supporting the stable and reliable operation of scenarios such as industrial control, remote medical care, and autonomous driving. However, in complex and diverse scenarios such as industrial control, telemedicine, and autonomous driving, if terminal access adaptation is only achieved by prioritizing network slicing, the dynamic fluctuations in terminal access demand at different times, sudden changes in traffic of various scenarios, and the mismatch between slicing standards and actual terminal needs will lead to congestion problems such as too many terminals in some network slices, resulting in resource contention and inability to guarantee QoS indicators, or too few terminals in some slices, resulting in idle and wasted network resources. This will further lead to a chain of problems such as an imbalance in the overall network resource utilization efficiency and a decline in the stability of critical business transmission. Furthermore, in the process of matching terminals with corresponding slices, terminals in different scenarios not only carry hard QoS indicators such as latency, bandwidth, and packet loss rate, but also have corresponding business attribute parameters in terms of service type (such as equipment control command transmission in industrial control, image data interaction in telemedicine, and environmental perception data upload in autonomous driving), data transmission cycle, and service priority level. Traditional slice matching mechanisms often cannot fully take into account the coordinated adaptation of hard QoS indicators and business attribute parameters, which will further lead to insufficient matching accuracy between network slices and terminals, resulting in problems such as terminal mismatch with slices and increased interference in services within slices, ultimately affecting the reliability of network services and scenario adaptability. In view of this, we propose a multi-terminal collaborative access resource allocation system under 5G network slicing. Summary of the Invention

[0003] The purpose of this invention is to solve the problems in the existing 5G network slicing resource allocation, which only relies on hard QoS indicators to match terminals and slices without fully taking into account service attribute parameters. This results in insufficient accuracy in matching slices with terminal needs, resource contention or idle waste, and poor stability of critical service transmission.

[0004] To achieve the above objectives, the present invention provides a multi-terminal collaborative access resource allocation system under 5G network slicing, comprising: The terminal requirement parameter standardization processing module collects the hard QoS indicators corresponding to all terminals; and uses the analytic hierarchy process to analyze the service attribute parameters of each terminal in turn, and standardizes the hard QoS indicators and service attribute parameters respectively. The standardized hard QoS indicators and service attribute parameters are sorted in a fixed order to obtain the requirement parameters of each terminal. The terminal clustering and cluster partitioning module uses the requirement parameters of each terminal as the core clustering basis and employs the K-means algorithm to cluster all terminals, resulting in multiple terminal clusters. The number of terminal clusters is equal to the number of network slices. When clustering terminals using the K-means algorithm, the similarity between different requirement parameters is calculated for initial clustering. After the initial clustering, the corresponding intra-cluster sum of squared errors is calculated. A clustering threshold is set. If the intra-cluster sum of squared errors of any terminal cluster is greater than the clustering threshold, the process is iteratively updated. If, after the iterative update, there are still terminal clusters with intra-cluster sum of squared errors greater than the clustering threshold, the preset number of clusters is incrementally added until the intra-cluster sum of squared errors of all terminal clusters is less than or equal to the clustering threshold.

[0005] The cluster configuration requirement extraction module sets a number of network slices corresponding to each terminal cluster in the terminal clustering and cluster partitioning module; and extracts the common requirements and core QoS constraints of each terminal cluster to form the configuration requirements corresponding to each terminal cluster. The slice resource quantification and allocation module converts the configuration requirements in the cluster configuration requirement extraction module into slice resources for the corresponding network slices. Specifically, it sets the redundancy coefficient for each requirement parameter in each core QoS constraint by retrieving the common service requirements in the configuration requirements, and calculates the slice resources of the network slice using the redundancy coefficient and the requirement parameters.

[0006] Preferably, the service attribute parameters in the terminal demand parameter standardization processing module include interruption loss level and QoS rigidity. The hierarchical analysis method first constructs a hierarchical structure corresponding to the target layer, criterion layer, and scheme layer. For indicators at the same level in the hierarchical structure, based on the indicators of the previous level, a 1-9 scaling method is used to perform pairwise comparisons to construct a judgment matrix A. Then, the sum-product method is used to calculate the weight of the level indicator relative to the indicator of the previous level, which is the initial feature vector. The product of the judgment matrix and the initial feature vector is calculated to solve for the largest eigenvalue. The consistency of the judgment matrix A is judged by the consistency ratio. Then, the weights of the criterion layer and the scheme layer are combined to obtain the total weight of each level of the scheme layer. The level with the largest total weight is selected as the result of the target layer.

[0007] Preferably, the analytic hierarchy process (AHP) uses the interruption loss level and QoS rigidity of the terminal service as the target layer, and the key dimensions corresponding to the interruption loss level and QoS rigidity as the criterion layers; the calibration result and QoS rigidity are used as the scheme layers; and a 1-9 scaling method is used for pairwise comparisons to construct a judgment matrix. When, determine the matrix middle For the element of the i-th indicator relative to the j-th indicator, satisfying the element... ,element ,element ; Preferably, the sum-product method is used to determine the matrix. Based on this, normalize the judgment matrix column by column. ; Column-normalized judgment matrix The column normalized matrix is ​​then obtained. The specific expression is ,in To determine the matrix elements, To determine the matrix The sum of the elements in the j-th column; And sum and normalize the matrix by row. Get rows and vectors ,in Normalize the rows and vectors again. The rows and results are converted into weighted form to obtain the initial feature vector. The i-th element And at this time, the initial feature vector This refers to the weight of each indicator at the current level relative to the corresponding indicator at the previous level.

[0008] Preferably, the initial feature vector is obtained. Then, based on the definitions of matrix eigenvalues ​​and eigenvectors, multiplication is used to determine the matrix. With the initial feature vector The product vector is obtained. The largest eigenvalue is then obtained through element-by-element calculation and mean averaging. Specifically: ,in Represents matrix A and weight vector The i-th element after multiplication.

[0009] Preferably, the consistency ratio in the terminal demand parameter standardization processing module is... ,in As a consistency indicator, The random consistency index is used; and a test threshold is set. If the consistency ratio is... If the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable; if the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable. If the value is greater than or equal to the test threshold, adjust the scale value of the judgment matrix and recalculate. The sum of the weights of the criterion layer and the weights of the scheme layer relative to the criterion layer: Receives the weight vector of the criterion layer relative to the target layer. The weight vector of the scheme layer for the k-th indicator of the criterion layer is: Then the total weight of the i-th level of the scheme layer ; Preferably, in the terminal clustering and cluster partitioning module, the K-means algorithm receives the demand parameters corresponding to each terminal from the terminal demand parameter standardization processing module to form a sample set. The i-th requirement parameter is ,in Let n be the required parameter for each terminal; The preset number of clusters is K, forming k terminal clusters; and a sample set is arbitrarily selected. We select k distinct demand parameters from the sample set as initial cluster centers for the terminal clusters; we then calculate the similarity between each demand parameter in the sample set and all the initial cluster centers, and assign the demand parameters accordingly. To the terminal cluster corresponding to the initial cluster center with the lowest similarity; In the sample set After all the demand parameters are assigned to the terminal clusters, i.e. after the initial clustering: calculate the mean of all demand parameters in each terminal cluster as the new cluster center; and calculate the sum of squared errors within each cluster corresponding to each terminal cluster. Set a clustering threshold, and compare the sum of squared errors within each cluster with the clustering threshold in turn; if the sum of squared errors within any terminal cluster is greater than the clustering threshold, then perform iterative updates again, that is, perform the following steps again: calculate the similarity between each required parameter and the current cluster center → assign the required parameters to the corresponding clusters → update the cluster centers → calculate the sum of squared errors within the clusters.

[0010] Preferably, the cluster configuration requirement extraction module sequentially retrieves all terminals corresponding to each terminal cluster, and retrieves the hard QoS indicators and service attribute parameters of each terminal in the terminal requirement parameter standardization processing module; then, it extracts the core QoS constraints and common service requirements corresponding to the terminal cluster through the hard QoS indicators and service attribute parameters of all terminals in each terminal cluster, and the core QoS constraints and common service requirements together constitute the configuration requirements of the terminal cluster.

[0011] Preferably, the core QoS constraints in the cluster configuration requirement extraction module include both inverse and forward indicators. Specifically, the core QoS constraints are analyzed as follows: the constraint threshold for the inverse indicator in the core QoS constraints is... ,in For a certain terminal cluster, the actual values ​​of the reverse indicator corresponding to N terminals; The constraint threshold for positive indicators is ,in For a given terminal cluster, the actual values ​​of the positive metrics corresponding to N terminals; Preferably, the cluster configuration requirement extraction module analyzes the common service requirements corresponding to the terminal cluster: a set of receive interruption loss levels. Level within a certain terminal cluster The corresponding number of terminals is The common business requirements corresponding to the interruption loss levels are: , where N is the total number of terminals in a certain terminal cluster.

[0012] Preferably, the common business requirement in the slice resource quantization allocation module is the sum of the quantization scores corresponding to the interruption loss level and QoS rigidity after standardization processing in the terminal requirement parameter standardization processing module. The slice resource quantification and allocation module presets a mapping table, which includes redundancy coefficients for different requirement parameters. When setting redundancy coefficients for common business requirements, it matches the redundancy coefficients corresponding to each requirement parameter in the mapping table, and then multiplies the redundancy coefficients by the corresponding requirement parameters to obtain the slice resources for the requirement parameters.

[0013] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall module of the present invention.

[0015] The meanings of the labels in the diagram are as follows: 100. Terminal requirement parameter standardization processing module; 200. Terminal clustering and cluster partitioning module; 300. Cluster configuration requirement extraction module; 400. Slice resource quantitative allocation module. Detailed Implementation

[0016] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] refer to Figure 1 As shown, a multi-terminal collaborative access resource allocation system under 5G network slicing includes a terminal requirement parameter standardization processing module 100, a terminal clustering and cluster partitioning module 200, a cluster configuration requirement extraction module 300, and a slice resource quantification and allocation module 400. Specifically, to accurately analyze the requirement parameters corresponding to each terminal cluster and avoid network slice partitioning imbalances, unreasonable resource allocation, and substandard service quality caused by incomplete terminal requirement identification and inaccurate parameter matching, the terminal requirement parameter standardization processing module 100 collects all hard QoS indicators (latency ≤10ms / 50ms, bandwidth ≥100Mbps / 1Gbps, etc.) corresponding to all terminals. It then uses the analytic hierarchy process (AHP) to sequentially analyze the service attribute parameters of each terminal. Specifically, the service attribute parameters include interruption penalty level (extremely high / medium / low) and QoS rigidity (QoS rigidity "non-compromising / medium rigidity / compromising"). The hard QoS indicators and service attribute parameters are standardized separately, and then sorted in a fixed order to obtain the requirement parameters for each terminal. The working principle of the hierarchical analysis method in the terminal demand parameter standardization processing module 100 is as follows: construct a hierarchical structure corresponding to the target layer, criterion layer, and scheme layer. Specifically, the target layer is the interruption loss level and QoS rigidity of the terminal service; the criterion layer is the key dimensions corresponding to the interruption loss level and QoS rigidity; and the scheme layer is the result of the level to be calibrated (specifically, the interruption loss level (extremely high, high, medium, low) and the QoS rigidity (non-compromising, medium rigidity, compromising)). Key dimensions of interruption loss levels include impact indicators such as economic loss impact, security risk impact, and reputation damage impact; key dimensions of QoS rigidity include core indicators such as sensitivity, business fault tolerance rate, and scenario necessity. For indicators at the same level in the hierarchical structure, based on the indicators of the level above, a pairwise comparison is performed using the 1-9 scale method to construct a judgment matrix. ,in For the element of the i-th indicator relative to the j-th indicator, satisfying the element... ,element ,element ; The meaning of the 1-9 scale is as follows: The aforementioned indicators at the same level specifically refer to a set of similar evaluation dimensions existing at the same analytical level within a hierarchical structure to support the goals / indicators of the next higher level. Their core characteristics are that they belong to the same analytical level and serve the same higher-level object. Specifically: If we focus on the criteria layer, the economic loss impact, security risk impact, and reputation damage impact of the interruption loss level belong to the same level of indicators (all are in the criteria layer corresponding to the interruption loss level, and together serve to quantify the weight of terminal business attributes at the target layer). When focusing on the solution layer, the "extremely high", "high", "medium" and "low" interruption loss levels belong to the same level of indicators (all are in the solution layer corresponding to the interruption loss level, serving the interruption loss level dimension of the criterion layer), and the "uncompromising", "medium rigid" and "compromising" QoS rigidity also belong to the same level of indicators (all are in the solution layer corresponding to the QoS rigidity, serving the QoS rigidity dimension of the criterion layer).

[0018] Interruption loss level and QoS rigidity are indicators at the same level—both are at the criteria level and jointly serve the target level to quantify the weight ratio of terminal service attribute parameters. They are the core dimensions reflecting the degree to which services cannot compromise on network services. If we focus on the solution layer, the extremely high / medium / low level of interruption loss belongs to the same level of indicators (all are in the "solution layer corresponding to the interruption loss level"), and the uncompromising / medium rigid / compromising degree of QoS rigidity also belongs to the same level of indicators (all are in the solution layer corresponding to the degree of QoS rigidity) - the two respectively serve the "interruption loss level" and "QoS rigidity" of the criterion layer, and are the concrete manifestation of the criteria layer indicators. Construct a judgment matrix Then, the weight of the hierarchical indicator relative to the indicator of the previous level is calculated using the sum-product method. Specifically: based on the judgment matrix... Based on this, normalize the judgment matrix column by column. Eliminate the judgment matrix The differences in the dimensions of different indicators provide a unified benchmark for comparing the importance of each column of indicators; the judgment matrix is ​​normalized by column. The column normalized matrix is ​​then obtained. The specific expression is ,in To determine the matrix elements, To determine the matrix The sum of the elements in the j-th column; To integrate the normalized results of the same indicator across all columns, the normalized matrix is ​​summed row by row. Get rows and vectors ,in Normalize the rows and vectors again. The rows and results are converted into weighted form (satisfying the constraint that the sum of the weights is 1), resulting in the initial feature vector. The i-th element And at this time, the initial feature vector This refers to the weight of each indicator at the current level relative to the corresponding indicator at the previous level.

[0019] Obtain the initial feature vector Then, according to the definition of matrix eigenvalues ​​and eigenvectors (if the product vectors...) ,but To determine the matrix eigenvalues, (as the corresponding eigenvectors), therefore, multiplication is used to determine the matrix. With the initial feature vector The product vector is obtained. The largest eigenvalue is then obtained through element-by-element calculation and mean averaging. Specifically: ,in Represents matrix A and weight vector The i-th element after multiplication; Consistency check: Calculate the consistency ratio ,in It serves as a consistency indicator (reflecting the degree to which the judgment matrix deviates from consistency). The random consistency index is the average consistency index calculated by randomly generating a judgment matrix, and it is related to the matrix order n. Set a test threshold; if the consistency ratio... If the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable; if the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable. If the value is greater than or equal to the test threshold, the scale value of the judgment matrix needs to be adjusted and recalculated.

[0020] The calculation of each level of the scheme layer is specifically the sum of the weighted products of the criterion layer weights and the weights of the scheme layer relative to the criterion layer: The weight vector of the criterion layer relative to the target layer is received. The weight vector of the scheme layer for the k-th indicator of the criterion layer is: Then the total weight of the i-th level of the scheme layer ; Select the scheme layer level with the highest total weight and define it as the corresponding level of the target layer (i.e., the interruption loss level or the QoS rigidity).

[0021] The working principle of the terminal requirement parameter standardization processing module 100 in standardizing the hard QoS indicators corresponding to each terminal is as follows: Hard QoS metrics include both positive and negative metrics. Positive metrics are those where a larger value indicates better service quality, such as bandwidth (a larger value means a faster data transmission rate) and throughput (a larger value means more data can be transmitted per unit time). The standardized expression for positive metrics is: the actual value of the hard QoS metric perceived by the i-th terminal is... The minimum value of hard QoS indicators is The maximum value of the hard QoS indicator is Standardized hard QoS indicators ; Inverse metrics, on the other hand, are those where "the smaller the value, the better the service quality." Specific examples include latency (the smaller the value, the shorter the data transmission delay), packet loss rate (the smaller the value, the higher the data transmission integrity), and jitter (the smaller the value, the stronger the data transmission stability). The standardized expression for inverse metrics is: the actual value of the perceived hard QoS metric for the i-th terminal is... The minimum value of hard QoS indicators is The maximum value of the hard QoS indicator is Standardized hard QoS indicators ; The terminal requirement parameter standardization processing module 100 standardizes business attribute parameters: mapping business attribute parameters to quantified scores (e.g., interruption loss levels "extremely high = 4, high = 3, medium = 2, low = 1"); standardizing the quantified scores using a positive indicator approach, resulting in standardized business attribute parameters. , where is the quantized score of the i-th terminal service attribute parameter. , The minimum score. This represents the maximum score.

[0022] By analyzing the service attribute parameters of each terminal (including the level of interruption loss and the rigidity of QoS), we can accurately uncover the deep-seated needs and priority differences of terminal services, providing dimensional support beyond hard QoS indicators for subsequent network slice matching and resource allocation. Specifically, by clarifying the severity of the loss after the interruption of terminal services and the compromise range of QoS indicators, we can clearly distinguish the level of service importance of different terminals, avoiding the adaptation deviation caused by the subsequent cluster configuration requirement extraction module 300 dividing network slices based solely on hard QoS indicators. This example avoids the problem of traditional fixed-number network slices, which rely solely on a single hard QoS metric for slice allocation. This approach fails to adapt to the differentiated needs of various terminals in terms of service attributes (such as differences in interruption penalty levels and QoS rigidity), and struggles to cope with dynamic fluctuations in terminal access volume and sudden changes in service traffic. This leads to resource mismatch issues such as conflicting terminal demands, resource contention, or wasted resources within slices. Furthermore, traditional slice matching mechanisms lack comprehensive consideration of terminal service attribute parameters, easily resulting in mismatched slices and increased service interference within slices. Therefore, the terminal clustering and grouping module 200 uses terminal demand parameters that integrate hard QoS metrics (latency, bandwidth, packet loss rate, etc.) and service attribute parameters (interruption penalty levels, QoS rigidity, etc.) as the core clustering basis. Then, the K-means algorithm is used to cluster all terminals, resulting in multiple terminal clusters. This dynamically determines the subsequent number of network slices. The specific working principle is as follows: The receiving terminal requirement parameter standardization processing module 100 receives the requirement parameters corresponding to each terminal to form a sample set. The i-th requirement parameter is ,in Let n be the required parameter for each terminal; The preset number of clusters is K (to be determined based on the business scenario, such as categorizing into 3 types according to requirements), forming k terminal clusters; and arbitrarily selecting a sample set. k different requirement parameters are used as the initial cluster centers for the terminal cluster; Calculate the similarity between each demand parameter in the sample set and all initial cluster centers in turn, and assign the demand parameters accordingly. To the terminal cluster corresponding to the initial cluster center with the lowest similarity; where the requirement parameters are... With cluster center Similarity between ; In the sample set After all the demand parameters are allocated to the terminal clusters, the mean of all demand parameters in each terminal cluster is calculated as the new cluster center; and the sum of squared errors within each cluster is calculated to measure the clustering effect. Set a clustering threshold, and compare the sum of squared errors within each cluster with the clustering threshold in turn; if the sum of squared errors within any terminal cluster is greater than the clustering threshold, then perform iterative updates again, that is, perform the following steps again: similarity between each required parameter and the current cluster center → assign the required parameters to the corresponding cluster → update the cluster center → calculate the sum of squared errors within the cluster. If, after iterative updates, there are still cases where the sum of squared errors within a cluster of terminals exceeds the clustering threshold, then the number of preset clusters is increased incrementally (with each addition being 1 cluster) until the sum of squared errors within a cluster of all terminals is less than or equal to the clustering threshold. At this point, it is determined that the clustering effect has reached the preset standard, and the iteration is stopped, with the final terminal cluster partitioning result output.

[0023] The sum of squared errors within a terminal cluster intuitively illustrates the dispersion of the required parameters of all terminals within the same cluster. A larger value indicates more significant differences in hard QoS indicators (latency, bandwidth, packet loss rate, etc.) and service attribute parameters (interruption penalty level, QoS rigidity, etc.) among terminals within the cluster, resulting in lower requirement matching. Conversely, a smaller value indicates a high degree of convergence in terminal requirements within the cluster. Therefore, when the terminal clustering and cluster partitioning module 200 compares the sum of squared errors within a terminal cluster with the clustering threshold, if the sum of squared errors within a cluster exceeds the clustering threshold, it indicates that the differences in terminal requirements under the current cluster size have exceeded the range that network slicing can adapt to. If the cluster is not split... Clustering can lead to subsequent slice configurations failing to meet the diverse needs of all terminals simultaneously. Therefore, by increasing the number of clusters to create more dedicated slices, terminals with significantly different needs can be assigned to different clusters, making the needs of terminals within each cluster more uniform. This allows for dynamic optimization of the precision of terminal cluster partitioning, guided by comparisons between intra-class error sum of squares and clustering thresholds. This ensures a high degree of consistency in terminal needs within each cluster, laying the foundation for the subsequent cluster configuration requirement extraction module 300 to extract precise cluster configuration requirements and partition suitable network slices. This fundamentally avoids the resource mismatch and substandard service quality issues caused by traditional fixed cluster partitioning. Furthermore, while existing methods, by relying on 5G core network (5GC), access network (RAN) standardization technologies and cooperative communication protocols to divide multiple network slices and matching them based on terminal hard QoS indicators, can prioritize the division of multiple network slices to provide dedicated logical network support for different types of terminals and meet the basic transmission needs of terminals to a certain extent, when a new terminal accesses the network, matching hard QoS indicators and the configuration requirements corresponding to each network slice can not clearly identify the network slice to which the new terminal belongs. Therefore, in this example, to avoid the inefficiency of the terminal clustering and cluster division module 200 having to re-cluster all terminals to determine the corresponding network slice when a new terminal is present, the cluster configuration requirement extraction module 300 sets a number of network slices corresponding to each terminal cluster in the terminal clustering and cluster division module 200; and extracts the common requirements and core QoS constraints of each terminal cluster (such as "latency ≤ 12ms, packet loss rate ≤ 10⁻"). 5 "Bandwidth ≥ 800Mbps, latency ≤ 50ms"), forming the configuration requirements for each terminal cluster. These requirements provide a unified and comprehensive adaptation standard for new terminal access. New terminals only need to compare their own requirements with the configuration requirements of each cluster to quickly match to the corresponding slice, without needing to restart full clustering. This improves access efficiency, avoids conflicts between new terminals and existing terminals within the slice, and provides a clear basis for slice resource scheduling. The specific working principle is as follows: The system sequentially retrieves all terminals in each terminal cluster and retrieves the hard QoS indicators and service attribute parameters of each terminal in the terminal requirement parameter standardization processing module 100. Then, it extracts the core QoS constraints and common service requirements of the terminal cluster through the hard QoS indicators and service attribute parameters of all terminals in each terminal cluster. The core QoS constraints and common service requirements together form the configuration requirements of the terminal cluster. Since the terminal clustering and cluster partitioning module 200 divides similar terminals into the same terminal cluster, the sum of squared errors within each cluster is less than or equal to the clustering threshold. This indicates that all terminals within the same terminal cluster are highly similar in terms of hard QoS indicators and service attribute parameters. Based on this high similarity, the cluster configuration requirement extraction module 300 extracts the hard QoS indicators (numerical parameters such as latency, bandwidth, and packet loss rate) corresponding to each terminal cluster, calculates boundary values ​​to determine the QoS threshold that can cover all terminals within the terminal cluster, and forms the core QoS constraints. It also extracts service attribute parameters (categorical parameters such as interruption loss level and QoS rigidity) and determines the common service requirements of the terminal cluster by statistically analyzing the highest percentage of these parameters. After obtaining the core QoS constraints and common service requirements for each terminal cluster, the core QoS constraints and common service requirements are integrated to obtain the configuration requirement for each terminal cluster = {core QoS constraints, common service requirements}. The core QoS constraints in the cluster configuration requirement extraction module 300 include inverse indicators (latency, packet loss rate) and positive indicators (bandwidth). A detailed analysis of the working principle of these core QoS constraints is as follows: The constraint thresholds for reverse metrics (latency, packet loss rate) in core QoS constraints are: ,in For a certain terminal cluster, the actual values ​​of the reverse indicator corresponding to N terminals; The constraint threshold for the positive metric (bandwidth) is ,in For a given terminal cluster, the actual values ​​of the positive metrics corresponding to N terminals; The cluster configuration requirements extraction module 300 analyzes the common business requirements corresponding to the terminal cluster. The specific working principle is: set of receive interruption loss levels. Level within a certain terminal cluster The corresponding number of terminals is The common business requirements corresponding to the interruption loss levels are: , where N is the total number of terminals in a certain terminal cluster; Core QoS constraints refer to the boundary thresholds extracted from the hard QoS indicators (such as latency, bandwidth, and packet loss rate) of all terminals in the terminal cluster. These are the technical red lines to ensure the normal operation of the cluster services, such as "latency ≤ 12ms, bandwidth ≥ 800Mbps, packet loss rate ≤ 10⁻". 5 "Through core QoS constraints, the lower or upper limits of the performance indicators that must be met when dividing network slices can be further clarified; Common business requirements refer to the business characteristics with the highest proportion extracted from the business attribute parameters (interruption loss level, QoS rigidity) of all terminals in the terminal cluster. They reflect the core business attributes of all terminals in the terminal cluster, such as "extremely high interruption loss level, uncompromising QoS rigidity". Based on this, the types and priorities of services that the subsequent network slices need to carry can be clearly defined. This example further considers that after the cluster configuration requirement extraction module 300 analyzes the configuration requirements of each network slice, since there are significant differences between the common service requirements and core QoS constraints of different terminal clusters, resources need to be matched in a targeted manner to ensure the stability of service operation. At the same time, in order to avoid insufficient resource allocation leading to a decline in the service quality of high-rigidity services or excessive resource allocation causing waste, the slice resource quantification and allocation module 400 converts the configuration requirements in the cluster configuration requirement extraction module 300 into the slice resources of the corresponding network slice to clarify the slice resources of each network slice. Specifically, by calling up the common service requirements in the configuration requirements (the common service requirements are the sum of the quantified scores corresponding to the interruption loss level and QoS rigidity after standardization processing in the terminal requirement parameter standardization processing module 100), the redundancy coefficient corresponding to each requirement parameter in each core QoS constraint is set, and the slice resources of the network slice are calculated using the redundancy coefficient and the requirement parameters. A preset mapping table is provided, which includes redundancy coefficients for different requirement parameters, such as latency redundancy coefficient (α1), bandwidth redundancy coefficient (α2), and packet loss rate redundancy coefficient (α3). When setting redundancy coefficients for common business requirements, the redundancy coefficients corresponding to each requirement parameter in the mapping table are matched, and then the redundancy coefficients are multiplied by the corresponding requirement parameters to obtain the slice resources of the requirement parameters. Therefore, the slice resource quantification and allocation module 400 can not only accurately match the redundancy coefficient of core QoS parameters through common business requirements based on the terminal cluster configuration requirements extracted by the step cluster configuration requirement extraction module 300, but also quickly calculate the appropriate network slice resources by combining the quantification formula, realizing the efficient and accurate conversion from configuration requirements to slice resources. Moreover, it can also deeply match the requirement parameters of the terminal cluster (interruption loss level, QoS rigidity), so that the slice resources corresponding to each network slice not only meet the stringent requirements of low latency and high reliability for high rigidity services such as industrial control and remote medical care, but also configure appropriate redundancy for ordinary services to avoid resource waste. At the same time, it ensures the dynamic adaptability of network slice resources and terminal cluster requirements, avoiding situations such as mismatch between resource allocation and actual service priority, service quality degradation of high interruption loss services due to resource crowding, and low network resource utilization due to over-allocation of ordinary service resources when dividing network slices based solely on hard QoS indicators. Furthermore, the preset mapping table can be obtained through machine learning. During the construction of the preset mapping table, the dynamic changes in demand when a new terminal is added to each network slice, the fluctuation of demand similarity within the cluster, and the need for resource adaptation adjustment are taken into consideration. Therefore, the specific working principle of the above machine learning method is as follows: Based on the terminal hard QoS indicators after forward / reverse standardization processing in the terminal demand parameter standardization processing module 100 and the quantified standardized business attribute parameters, combined with the terminal cluster data after clustering by the terminal clustering and cluster division module 200 (including the sum of squared errors within the cluster, cluster size, and initial threshold of core QoS constraints), the common business requirements extracted by the cluster configuration requirement extraction module 300 (quantified interruption loss level + QoS rigidity), and the resource utilization rate and QoS compliance rate in historical network operation data are used as sample datasets. The common business requirements are used as the input features of the model, and the optimal redundancy coefficient of each core QoS parameter (latency, bandwidth, packet loss rate) (based on the redundancy configuration labeling during stable operation of services in historical data) is used as the output label to construct a gradient boosting tree or random forest regression prediction model. During the model training phase, the sample data is first divided, and the model hyperparameters are optimized by cross-validation. The L1 regularization mechanism is introduced to avoid overfitting. At the same time, the demand parameters of newly added terminals and the corresponding actual operation feedback data (such as resource adaptation effect and QoS compliance status) are added to the sample set in real time for incremental training, and the model parameters are dynamically updated to adapt to the dynamic changes in terminal demand. After training, inputting the common service requirements of any terminal cluster will output the corresponding latency redundancy coefficient (α1), bandwidth redundancy coefficient (α2), and packet loss rate redundancy coefficient (α3), generating a preset mapping table adapted to different scenarios. When a new terminal is added to the network slice, the standardized requirement parameters of the new terminal are automatically extracted to determine its impact on the common service requirements of the cluster. If the impact reaches the preset threshold, the model is triggered to re-predict and update the redundancy coefficient of the corresponding cluster in the mapping table to ensure that the mapping table always matches the actual needs of the cluster.

[0024] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-terminal collaborative access resource allocation system under 5G network slicing, characterized in that, include: The terminal requirement parameter standardization processing module (100) collects the hard QoS indicators corresponding to all terminals; and uses the analytic hierarchy process to analyze the service attribute parameters of each terminal in turn, and standardizes the hard QoS indicators and service attribute parameters respectively. The standardized hard QoS indicators and service attribute parameters are sorted in a fixed order to obtain the requirement parameters of each terminal. The terminal clustering and cluster partitioning module (200) uses the requirement parameters of each terminal as the core clustering basis and uses the K-means algorithm to cluster all terminals to obtain multiple terminal clusters. The number of terminal clusters is the number of network slices. When the K-means algorithm clusters terminals, it calculates the similarity between different requirement parameters to perform initial clustering. After the initial clustering, it calculates the corresponding sum of squared errors within each cluster. Set a clustering threshold. If the sum of squared errors within any terminal cluster is greater than the clustering threshold, then perform another iterative update. If, after iterative updates, there are still cases where the sum of squared errors within a terminal cluster is greater than the clustering threshold, then the preset number of clusters is added incrementally until the sum of squared errors within a cluster for all terminal clusters is less than or equal to the clustering threshold. The cluster configuration requirement extraction module (300) sets the number of network slices corresponding to each terminal cluster in the terminal clustering and cluster partitioning module (200); and extracts the common requirements and core QoS constraints of each terminal cluster to form the configuration requirements corresponding to each terminal cluster. The slice resource quantization and allocation module (400) converts the configuration requirements in the cluster configuration requirement extraction module (300) into slice resources for the corresponding network slice. Specifically, it sets the redundancy coefficient corresponding to each requirement parameter in each core QoS constraint by calling up the common service requirements in the configuration requirements, and calculates the slice resources of the network slice using the redundancy coefficient and the requirement parameters.

2. The multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 1, characterized in that: The service attribute parameters in the terminal requirement parameter standardization processing module (100) include interruption loss level and QoS rigidity. The hierarchical analysis method first constructs a hierarchical structure corresponding to the target layer, criterion layer, and scheme layer. For the indicators of the same level in the hierarchical structure, the indicators of the next higher level are used as the basis, and then the 1-9 scale method is used to compare them pairwise to construct the judgment matrix A. Then, the sum-product method is used to calculate the weight of the hierarchical indicator relative to the indicator of the next higher level, which is the initial feature vector. The product of the judgment matrix and the initial feature vector is calculated to solve for the maximum eigenvalue. The consistency of matrix A is determined by the consistency ratio. Then, the weights of the criterion layer and the scheme layer are combined and weighted to obtain the total weight of each level of the scheme layer. The level with the largest total weight is selected as the result of the target layer.

3. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 2, characterized in that: The analytic hierarchy process (AHP) uses the interruption loss level and QoS rigidity of terminal services as the target layer, and the key dimensions corresponding to the interruption loss level and QoS rigidity as the criterion layers; the calibration result and QoS rigidity are used as the scheme layers; and a 1-9 scaling method is used for pairwise comparisons to construct a judgment matrix. When, the judgment matrix middle For the element of the i-th indicator relative to the j-th indicator, satisfying the element... ,element ,element ; A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 1, characterized in that: the sum-product method is used to determine the matrix. Based on this, normalize the judgment matrix column by column. ; Column-normalized judgment matrix The column normalized matrix is ​​then obtained. The specific expression is ,in To determine the matrix elements, To determine the matrix The sum of the elements in the j-th column; And sum and normalize the matrix by row. Get rows and vectors ,in Normalize the rows and vectors again. The rows and results are converted into weighted form to obtain the initial feature vector. The i-th element And at this time, the initial feature vector This refers to the weight of each indicator at the current level relative to the corresponding indicator at the previous level.

4. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 3, characterized in that: The initial feature vector is obtained. Then, based on the definitions of matrix eigenvalues ​​and eigenvectors, multiplication is used to determine the matrix. With the initial feature vector The product vector is obtained. The largest eigenvalue is then obtained through element-by-element calculation and mean averaging. Specifically: ,in Represents matrix A and weight vector The i-th element after multiplication.

5. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 4, characterized in that: The consistency ratio in the terminal demand parameter standardization processing module (100) ,in As a consistency indicator, The random consistency index is used; and a test threshold is set. If the consistency ratio is... If the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable; if the consistency ratio is less than the test threshold, then the consistency of the judgment matrix is ​​considered acceptable. If the value is greater than or equal to the test threshold, adjust the scale value of the judgment matrix and recalculate. The sum of the weights of the criterion layer and the weights of the scheme layer relative to the criterion layer: Receives the weight vector of the criterion layer relative to the target layer. The weight vector of the scheme layer for the k-th indicator of the criterion layer is: Then the total weight of the i-th level of the scheme layer .

6. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 3, characterized in that: The K-means algorithm in the terminal clustering and cluster partitioning module (200) receives the demand parameters corresponding to each terminal from the terminal demand parameter standardization processing module (100) to form a sample set. The i-th requirement parameter is ,in Let n be the required parameter for each terminal; The preset number of clusters is K, forming k terminal clusters; and a sample set is arbitrarily selected. We select k distinct demand parameters from the sample set as initial cluster centers for the terminal clusters; we then calculate the similarity between each demand parameter in the sample set and all the initial cluster centers, and assign the demand parameters accordingly. To the terminal cluster corresponding to the initial cluster center with the lowest similarity; And in the sample set After all the demand parameters are allocated to the terminal clusters, i.e. after the initial clustering: calculate the mean of all demand parameters in each terminal cluster as the new cluster center; And calculate the sum of squared errors within each class corresponding to each terminal cluster; Set a clustering threshold, and compare the sum of squared errors within each cluster with the clustering threshold in turn; if the sum of squared errors within any terminal cluster is greater than the clustering threshold, then perform iterative updates again, that is, perform the following steps again: calculate the similarity between each required parameter and the current cluster center → assign the required parameters to the corresponding clusters → update the cluster centers → calculate the sum of squared errors within the clusters.

7. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 6, characterized in that: The cluster configuration requirement extraction module (300) sequentially retrieves all terminals in each terminal cluster and retrieves the hard QoS indicators and service attribute parameters of each terminal in the terminal requirement parameter standardization processing module (100). Then, it extracts the core QoS constraints and common service requirements corresponding to the terminal cluster through the hard QoS indicators and service attribute parameters of all terminals in each terminal cluster. The core QoS constraints and common service requirements together constitute the configuration requirements of the terminal cluster.

8. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 7, characterized in that: The core QoS constraints in the cluster configuration requirement extraction module (300) include reverse indicators and forward indicators. A specific analysis of the core QoS constraints is as follows: the constraint threshold for the reverse indicator in the core QoS constraints is... ,in For a certain terminal cluster, the actual values ​​of the reverse indicator corresponding to N terminals; The constraint threshold for positive indicators is ,in This represents the actual values ​​of positive indicators for N terminals within a certain terminal cluster.

9. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 3, characterized in that: The cluster configuration requirement extraction module (300) analyzes the common service requirements corresponding to the terminal cluster: set of receive interruption loss levels. Level within a certain terminal cluster The corresponding number of terminals is The common business requirements corresponding to the interruption loss levels are: , where N is the total number of terminals in a certain terminal cluster.

10. A multi-terminal collaborative access resource allocation system under 5G network slicing according to claim 9, characterized in that: The common service requirements in the slice resource quantization allocation module (400) are the sum of the quantization scores corresponding to the interruption loss level and QoS rigidity after standardization processing in the terminal requirement parameter standardization processing module (100); The slice resource quantification and allocation module (400) presets a mapping table, which includes redundancy coefficients for different requirement parameters. When setting redundancy coefficients for common business requirements, it matches the redundancy coefficients corresponding to each requirement parameter in the mapping table, and then multiplies the redundancy coefficients by the corresponding requirement parameters to obtain the slice resources for the requirement parameters.

Images