A method and system for dynamic optimization of service function chaining of a 5g core network

Abstract

The application discloses a kind of 5G core network service function chain dynamic optimization method and system, it is related to mobile communication relevant technical field, the method includes: by three-dimensional perception module, real-time collection network service data;Service function chain optimization problem is decomposed into multi-stage decision model;Decomposed into long-term UPF deployment main problem, short-term traffic scheduling sub-problem, obtain global optimal solution by solving;Dynamically determine UPF deployment location and quantity;Dynamically select the traffic forwarding path that satisfies service level;In core layer and edge layer, construct double node deployment framework, and execute differentiated service function chain link control strategy.The technical problems that UPF deployment and traffic scheduling lack dynamic collaborative optimization in prior art, service function chain static optimization leads to low resource utilization, high energy consumption, poor service adaptability, to improve edge server resource utilization, reduce network energy consumption, optimize service quality and guarantee the technical effects of user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mobile communication, and particularly relates to a service function chain dynamic optimization method and system of a 5G core network. BACKGROUND

[0002] The 5G core network needs to support diversified scenarios such as ultra-low latency, ultra-high reliability and massive connection, and puts forward higher requirements for traditional static service function chain management. The traditional SFC deployment usually adopts a fixed strategy, which is difficult to adapt to the dynamically changing network environment and differentiated business requirements, resulting in problems such as low resource utilization, high energy consumption, and uneven user experience. How to realize the dynamic optimization of the service function chain has become the key to the intelligent evolution of the 5G core network. The deployment and traffic scheduling of the user plane function UPF directly affect the network performance and user experience. The deployment strategy of the UPF involves the optimal allocation of edge computing resources, and the traffic scheduling needs to meet the service level agreement of different businesses. The existing methods mostly adopt static or semi-static optimization, and fail to fully consider the time coupling relationship between the UPF deployment and the traffic scheduling, resulting in the difficulty in collaborative optimization of long-term resource planning and short-term business demand. In addition, the centralized optimization has high computational complexity and is difficult to adapt to the real-time decision-making requirements in a large-scale network environment.

[0003] Therefore, in the related art at present, there are technical problems of lack of dynamic collaborative optimization of UPF deployment and traffic scheduling, low resource utilization, high energy consumption, and poor business adaptability caused by static optimization of service function chain. SUMMARY

[0004] The present application provides a service function chain dynamic optimization method and system of a 5G core network, which solves the technical problems of lack of dynamic collaborative optimization of UPF deployment and traffic scheduling, low resource utilization, high energy consumption, and poor business adaptability caused by static optimization of service function chain in the prior art, and achieves the technical effects of improving the resource utilization of edge servers, reducing network energy consumption, optimizing business service quality, and guaranteeing user experience.

[0005] The application provides a service function chain dynamic optimization method of a 5G core network, which comprises the following steps: collecting network service data in real time by a three-dimensional perception module, including user attribute data, network state data and service experience data, and constructing an input data set for service function chain optimization; based on the input data set, decomposing the service function chain optimization problem into a multi-stage decision model, which has a time coupling relationship between UPF deployment and traffic scheduling; decomposing the multi-stage decision model into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem by a decomposition algorithm, and obtaining a global optimal solution by alternately iterating the main and sub-problems; dynamically determining the UPF deployment position and quantity according to the solution result of the long-term UPF deployment main problem, maximizing the edge server resource utilization rate and reducing the energy consumption; dynamically selecting a traffic forwarding path that meets the service level according to the solution result of the short-term traffic scheduling sub-problem, in combination with real-time network state and service level information, to ensure user experience; and based on a user grading strategy, constructing a double-node deployment framework at the core layer and the edge layer, and executing a differentiated service function chain control strategy according to different service levels.

[0006] In possible implementation manners, the service function chain dynamic optimization method of the 5G core network further performs the following processing: the three-dimensional perception module comprises a data fusion engine, and the configuration of the data fusion engine comprises the following steps: adding a privacy desensitization label to the user attribute data to obtain the user attribute data; normalizing the obtained network state data by a time sliding window; and dynamically adjusting the collection frequency according to the service type to obtain the service experience data.

[0007] In possible implementation manners, the service function chain dynamic optimization method of the 5G core network further performs the following processing: performing time and space distribution analysis on regional service traffic based on historical data, and constructing a service density heat map; calculating the optimal coverage radius of the network service according to the regional service intensity and the distance relationship between regions in the service density heat map; deploying delay constraints of each region based on the optimal coverage radius and the service distribution characteristics, determining the edge node distribution, and establishing the edge node.

[0008] In possible implementation manners, the service function chain dynamic optimization method of the 5G core network further performs the following processing: based on the service density heat map, setting the network resource reservation amount according to the peak load; defining node intimacy based on the network service coverage heat relationship of the edge node, and grouping the nodes into a cooperative node group based on the node intimacy; performing joint resource configuration on the edge node and the cooperative node group based on the network resource reservation amount and the node cooperation relationship, constructing a cooperative elastic resource pool, and using the cooperative elastic resource pool for dynamic reconstruction or elastic deployment in service function scheduling.

[0009] In a possible implementation, the method further performs the following processing: based on local differential, injecting noise into original user information, uploading the desensitized feature vector to each edge node; mapping the feature vector to a temporary encrypted ID; training a local model using desensitized data, uploading the encrypted model gradient to a federated learning aggregator deployed in the core layer, generating a global gradient through homomorphic summation aggregation; the federated learning aggregator distributes the global model to the edge node, outputs the de-identified service experience rule label, and obtains the user attribute data.

[0010] In a possible implementation, the method further performs the following processing: the de-identified service experience rule label is a privacy desensitization label, and does not contain identifiable personal information.

[0011] In a possible implementation, the method further performs the following processing: dividing an optimization period into a plurality of continuous time slots, defining UPF deployment strategy variables and traffic scheduling strategy variables in each time slot; based on the evolution relationship of the UPF deployment state between adjacent time slots, establishing a UPF deployment state transmission constraint between adjacent time slots; introducing UPF transmission migration cost and deployment switching cost in the objective function, and constructing the multi-stage decision model including time dependence, which is used to quantify the time coupling relationship between UPF deployment and traffic scheduling.

[0012] In a possible implementation, the method further performs the following processing: defining the UPF deployment state variable as a main problem decision variable, which is used to determine the UPF deployment location and quantity of each time slot edge node in the optimization period; defining the service flow path selection variable and the resource scheduling variable as sub-problem decision variables, and limiting the sub-problem decision variables to be optimized within the deployed UPF node range output by the main problem through constraints; establishing a dependent constraint relationship between the main problem and the sub-problem, so that the feasible solution space of the sub-problem is constrained by the deployment result of the main problem, wherein a proxy variable is introduced in the objective function of the main problem to represent the scheduling cost fed back by the sub-problem; in each iteration, the main problem is solved first to obtain a set of UPF deployment schemes, which are input to the sub-problem, and the sub-problem is independently solved based on this to obtain the optimal traffic scheduling path; the solving result of the sub-problem is fed back to the main problem through a cutting plane, a cutting constraint for limiting the deployment combination is constructed, and the main problem model is added; through the alternating iteration optimization of the main-sub problem, the combination scheme of the UPF deployment and traffic scheduling strategy is corrected until the convergence condition is met, and the global optimal solution is obtained.

[0013] In a possible implementation, the method further performs the following processing: establishing a user hierarchical policy, including policy mapping rules between user attributes, user categories, service priorities, and delay constraints, wherein the user categories include high-service users and ordinary-service users; constructing a dual-node deployment framework, wherein a UPF resource with large capacity and high availability is deployed at a core layer to carry out basic network communication of the ordinary-service users; a UPF instance with low latency and high response is deployed at an edge layer to provide a fast access path for the high-service users; and a mapping relationship of a service function chain control policy is established between the user hierarchical policy and the dual-node deployment framework according to the user categories, and a differentiated service function chain control policy is executed.

[0014] The application further provides a system for dynamically optimizing a service function chain of a 5G core network, which includes: a network service data acquisition unit configured to acquire network service data in real time through a three-dimensional perception module, including user attribute data, network state data, and service experience data, and to construct an input data set for service function chain optimization; a multi-stage decision model obtaining unit configured to decompose a service function chain optimization problem into a multi-stage decision model based on the input data set, the multi-stage decision model having a time coupling relationship between UPF deployment and traffic scheduling; a globally optimal solution obtaining unit configured to decompose the multi-stage decision model into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem through a decomposition algorithm, and to obtain a globally optimal solution through iterative solving of the main and sub-problems; a UPF deployment determining unit configured to dynamically determine a UPF deployment location and quantity based on a solving result of the long-term UPF deployment main problem, to maximize edge server resource utilization and reduce energy consumption; a traffic forwarding path selecting unit configured to dynamically select a traffic forwarding path that meets a service level based on a solving result of the short-term traffic scheduling sub-problem, in combination with real-time network state and service level information, to ensure user experience; and a link control policy executing unit configured to construct a dual-node deployment framework at a core layer and an edge layer based on a user hierarchical policy, and to execute a differentiated service function chain link control policy according to different service levels.

[0015] The application provides a 5G core network service function chain dynamic optimization method and system. A three-dimensional perception module is used to collect network service data in real time. A service function chain optimization problem is decomposed into a multi-stage decision model. The problem is decomposed into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem, and a global optimal solution is obtained. The deployment position and quantity of the UPF are dynamically determined. Real-time network state and service level information are combined to dynamically select a traffic forwarding path that meets the service level. A double-node deployment framework is constructed at the core layer and the edge layer, and a differentiated service function chain link control strategy is executed according to different service levels. The technical problems of lack of dynamic collaborative optimization of UPF deployment and traffic scheduling, low resource utilization, high energy consumption and poor service adaptability caused by static optimization of the service function chain are solved, and the technical effects of improving the resource utilization of the edge server, reducing the network energy consumption, optimizing the service quality and guaranteeing the user experience are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. The flowchart is used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, according to the needs, various steps can be processed in reverse order or at the same time. Meanwhile, other operations can be added to these processes, or one or more steps of the operations can be removed from these processes.

[0017] Figure 1 A 5G core network service function chain dynamic optimization method flowchart is provided for the embodiments of the present application.

[0018] Figure 2 A 5G core network service function chain dynamic optimization system structure schematic diagram is provided for the embodiments of the present application.

[0019] Explanation of reference signs: network service data collection unit 10, multi-stage decision model obtaining unit 20, global optimal solution obtaining unit 30, UPF deployment determining unit 40, traffic forwarding path selecting unit 50, and link control strategy executing unit 60. DETAILED DESCRIPTION

[0020] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0021] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limitations to the present application. All other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The term "first\second" referred to is only to distinguish similar objects, and does not represent a specific order of the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide a method for dynamically optimizing a service function chain of a 5G core network, as shown in Figure 1 The method comprises the following steps.

[0024] In step S100, a three-dimensional perception module is used to collect network service data in real time, including user attribute data, network state data and service experience data, and to construct an input data set for service function chain optimization.

[0025] Preferably, the three-dimensional perception module is a functional component responsible for collecting multi-dimensional network service data in real time. Data perception is performed from three dimensions of user attributes, network states and service experiences to construct a complete input data set, provide data support for dynamic optimization of the service function chain (SFC), and realize fine and intelligent SFC optimization decisions. The network service data includes user attribute data, network state data and service experience data. Specifically, the user attribute data refers to static and dynamic characteristic information related to terminal users or service flows, supports differentiated service policies, for example, high-priority users or critical services can be preferentially allocated to low-latency paths, including but not limited to user identity information such as user ID, subscription package type, etc.; service types such as video streaming, real-time gaming, industrial control, IoT data transmission, etc.; mobility characteristics such as user location, moving speed, handover frequency, etc.; historical behavior data such as traffic usage patterns, service access preferences, etc.

[0026] Preferably, the network state data refers to dynamic indicators reflecting the current network operating conditions, providing the basis for dynamic traffic scheduling and load balancing, such as automatically adjusting traffic to low-load nodes when a certain UPF node is overloaded to avoid congestion, which can include resource utilization, such as CPU, memory, storage occupancy rate of UPF node, link bandwidth usage, etc.; topology and load information, such as edge server distribution, network congestion degree, inter-node delay and packet loss rate, etc.; network function virtualization (NFV) status, such as virtualized network function instance running status, elastic scaling capacity, etc.; wireless side information, such as base station load, wireless channel quality, air interface delay, etc. Service experience data refers to service quality indicators that directly affect user perception, used for real-time optimization of SFC path, such as dynamically switching to a better path when detecting that the delay of a certain link exceeds the standard to ensure service experience, which can include end-to-end delay, the total delay of service flow from the sender to the receiver; throughput, such as whether the code rate of video stream meets the standard, whether the file download rate meets the requirement; reliability indicators, such as packet loss rate, error rate, service availability; user subjective feedback, such as mean opinion score (MOS).

[0027] Preferably, the three-dimensional perception module obtains network state and user session information from core network elements such as AMF, SMF, UPF, collects service quality indicators from terminals or service servers, and combines controller and network probe to monitor link quality, and then cleans and normalizes the obtained multi-source heterogeneous data, and further integrates it into an input data set for service function chain optimization, i.e. a multi-dimensional, real-time updated network situation image, providing data-driven basis for UPF deployment optimization and dynamic traffic scheduling, which can more accurately adapt to network changes, and thus improve resource utilization and user experience.

[0028] Further, step S100 further comprises step S101 of performing time and space distribution analysis of regional business traffic based on historical data to construct a business density heat map; step S102 of calculating the optimal coverage radius of network service according to the regional business intensity in the business density heat map and the distance relationship between regions; and step S103 of determining the distribution of edge nodes based on the delay constraints of each region according to the optimal coverage radius and business distribution characteristics, and establishing edge nodes.

[0029] Preferably, through historical traffic analysis, a business density heat map is constructed, and then the optimal network coverage radius is calculated, and the deployment position of the edge node, such as UPF / MEC, is optimized to reduce the latency and improve the resource utilization. Specifically, historical data of user traffic, business request distribution, mobility trajectory, etc. are collected, and the time-space distribution of the business traffic in each region is analyzed, i.e. statistical analysis is performed in the time and space dimensions to analyze the business traffic characteristics of each region, for example, the traffic of high-density business areas such as commercial centers, sports venues, and transportation hubs increases sharply in specific time periods, the traffic of low-density business areas such as suburbs and rural areas is relatively stable, and the periodic business mode, such as the high traffic of enterprise parks on weekdays and the increase of video traffic in residential areas on weekends. Then, based on the historical data, a business density heat map is generated by using spatial interpolation or machine learning clustering such as DBSCAN, which visually displays the business intensity and time-space change trend, wherein the color depth represents the traffic load, such as red for high load and blue for low load; the time-space change trend dynamically displays the business distribution change in different time periods.

[0030] Preferably, based on the business density heat map, the optimal coverage radius of the edge node is calculated in combination with the regional business intensity and the distance relationship between regions. Specifically, the business heat map is mapped to the geographical space, the service area is divided by the Voronoi algorithm, so that each edge node covers a cell, and then the size of the cell is adjusted, so that the cell in the high-density area is smaller, i.e. densely deployed, and the cell in the low-density area is larger, i.e. widely covered. Further, the coverage radius of the high-business-density area is small, more edge nodes are deployed, and the latency is reduced; the coverage radius of the low-business-density area is large, the number of nodes is reduced, and the cost is saved; finally, the optimal coverage radius of the network service and the theoretically optimal distribution position of the edge node are output, such as 500m in urban areas and 2km in suburban areas.

[0031] Preferably, according to the optimal coverage radius and the business distribution characteristics, the actual deployment position of the edge node is determined and the delay constraint is deployed. Usually, the network edge facilities close to the user side are selected as the edge nodes and the edge node distribution is determined, for example, high-density urban base stations are deployed in urban centers, business areas, and large event sites; industrial park MEC nodes are deployed in areas with high demand for enterprise private networks, such as factories, ports, and logistics centers, to support industrial Internet of Things, automatic guided vehicles AGV, and other low-latency applications; dedicated enterprise sites are deployed in industries with high requirements for data localization, such as banks, hospitals, and energy industries, to support network slicing and allocate exclusive service function chains SFCs for enterprises. At the same time, differential latency constraints are set according to the type of business, such as video stream latency < 50ms and industrial control latency < 10ms; in combination with the real-time network state, such as congestion and node failure, the edge node load is dynamically adjusted, including load balancing and elastic scaling.

[0032] Further, step S100 further comprises step S104, setting network resource reservation according to peak load based on the service density heat map; step S105, defining node intimacy based on network service coverage heat relationship of the edge node, and grouping collaborative nodes based on the node intimacy; step S106, jointly configuring resources of the edge node and the collaborative node group based on the network resource reservation and node collaborative relationship, and constructing a collaborative elastic resource pool for dynamic reconstruction or elastic deployment in service function scheduling.

[0033] Preferably, through intelligent resource reservation and node collaboration mechanism, an elastic resource pool is constructed to realize dynamic optimization of edge network resources to cope with service fluctuations and improve the flexibility of service function chain. Specifically, according to the service density heat map, the peak load of each region is identified, such as peak load during rush hours and during large events, and then the network resource reservation is set according to the peak load, that is, the calculation, storage and bandwidth resources are reserved by 120%-150% of the load demand to avoid service degradation caused by sudden traffic, and the network resource reservation is dynamically adjusted based on the time series prediction model. Based on the network service coverage heat relationship of the edge node, the collaborative potential between nodes is quantified, and then the node intimacy is defined according to geographical distance, business overlap and resource complementarity, wherein the smaller the distance between adjacent nodes, the higher the intimacy, and the more suitable for collaboration; the business overlap degree indicates that the user mobility in the coverage area is high, and the nodes are tightly collaborative; the resource complementarity, such as surplus computing power of node A and surplus storage of node B, forms a complementary group. Then, the high-intimacy nodes are divided into collaborative node groups according to hierarchical clustering, wherein the master node has sufficient resources and is responsible for task distribution within the group, and the standby node has low load at ordinary times and takes over overflow traffic at peak times.

[0034] Preferably, based on the network resource reservation and the node collaborative relationship, the edge node and the collaborative node group are jointly configured, specifically, the CPU, memory, bandwidth and other resources of the nodes in the collaborative group are virtualized into a unified resource pool, that is, a collaborative elastic resource pool, and then the resource state is synchronized in real time through a controller or a distributed database (such as Redis), and is used for dynamic reconstruction or elastic deployment in service function scheduling, including load balancing, fault recovery and elastic scaling, wherein when the load of a node is >80%, part of the traffic is automatically scheduled to a low-load node in the collaborative group; when a node fault is detected, other nodes in the collaborative group take over its business; during business peak, more VNF instances are automatically allocated from the resource pool to dynamically adjust the resource configuration of a single VNF. According to the resource pool state, the SFC path is optimized in real time, such as migrating the video analysis function from the overloaded node to the idle node in the collaborative group, and high-priority business can exclusively occupy the dedicated link in the resource pool to avoid low-priority business from occupying resources.

[0035] Further, step S100 further comprises that the three-dimensional perception module comprises a data fusion engine, and the configuration of the data fusion engine comprises: step S110, adding a privacy desensitization label to user attribute data to obtain the user attribute data; step S120, collecting network state data through time sliding window normalization; and step S130, acquiring the service experience data by dynamically adjusting the collection frequency according to the business type.

[0036] Preferably, the three-dimensional perception module processes multi-source heterogeneous network service data through the data fusion engine to standardize and secure the data, and constructs a high-quality input data set to support dynamic optimization of the service function chain. The configuration of the fusion engine comprises: performing label-based privacy desensitization on user attribute data, specifically, performing hash encryption or replacing a direct identifier such as a user ID and IMSI with a temporary anonymous identifier, generalizing an indirect identifier such as a geographic location to a regional level, and adding a data label to limit unauthorized business analysis, thereby obtaining the user attribute data for business classification and resource allocation.

[0037] Preferably, the network state data is normalized through a time sliding window, a time window is defined, data in the window is aggregated according to an event sequence, a sliding step is configured to ensure real-time data update, an index is scaled to the [0, 1] interval through Min-Max normalization to eliminate transient fluctuation interference and reflect stable network performance. The collection frequency is dynamically adjusted according to the business type, that is, for industrial control business, high-frequency collection is performed to ensure sub-millisecond latency; for video stream business, medium-frequency collection is performed to monitor code rate stability; and for sensor reporting business, low-frequency collection is performed to focus on packet loss rate. At the same time, the collection frequency is automatically increased to a fault troubleshooting mode when an anomaly such as a sudden increase in latency is detected, and edge computing is used to perform preprocessing close to the user side to reduce core network data backhaul pressure, thereby obtaining the service experience data, and finally generating a standardized, secure and compliant input data set to ensure efficient use of resources.

[0038] Further, step S110 further comprises: step S111, injecting noise into original user information based on local differential, uploading the desensitized feature vector to each edge node; step S112, mapping the feature vector to a temporary encrypted ID; step S113, training a local model using the desensitized data, uploading the encrypted model gradient to a federated learning aggregator deployed in the core layer, generating a global gradient through homomorphic summation aggregation; and step S114, the federated learning aggregator distributes the global model to the edge node, outputs a de-identified service experience rule label, and obtains the user attribute data.

[0039] Preferably, under the premise of protecting user privacy, through federated learning and differential privacy, joint data modeling across edge nodes is carried out to realize privacy protection and distributed collaborative modeling of user data, and finally de-identified business rule labels are generated to optimize the decision-making ability of the network service function chain. Specifically, according to local differential, controllable noise such as Laplace noise or Gaussian noise is added to the original user information at each edge node, and then the desensitized discrete data is converted into a numerical feature vector, which is uploaded through the edge node. The original user information cannot be inferred from the noisy features, and the risk of information leakage is controlled through the privacy budget. Then the feature vector is mapped to a temporary encrypted ID, i.e. a lightweight encryption is used to generate a temporary encrypted ID for each user or business session, such as a hash-based pseudonym or homomorphic encryption token, which replaces the real user identifier. The key is kept locally by the edge node. The ID life cycle is limited to be updated every 30 minutes, and an intelligent contract is deployed in the core layer to record the ID-node mapping relationship to prevent forgery.

[0040] Preferably, a lightweight local model is constructed based on a decision tree, logistic regression or a small neural network, and desensitized data is used to train it to predict business experience labels. Through a homomorphic encryption algorithm, encrypted model gradients are output, and then uploaded to the federated learning aggregator deployed in the core layer. The federated learning aggregator deployed in the core layer realizes homomorphic summation of gradients through secure multi-party computation to generate global gradients and update the global model. The weights are dynamically allocated according to the data volume. Then the updated global model is distributed to the edge nodes through an encrypted channel. The edge nodes synchronize the model version regularly. Finally, the de-identified business experience rule labels are output, i.e. the global model clusters user features and outputs non-sensitive labels, which are completely decoupled from the original user information, obtaining user attribute data, and thus realizing efficient mining of data value and optimization of network services, and ensuring improvement of resource utilization.

[0041] Further, step S114 further comprises that the de-identified business experience rule label is a privacy desensitized label and does not contain identifiable personal identity information.

[0042] Preferably, the de-identified business experience rule label is a privacy desensitized label, which is a business feature abstraction output by a federated learning model after analyzing desensitized data. It belongs to a non-sensitive business feature identifier, completely strips the user's personal identity information, and only retains abstract features at the business level for network optimization decisions. Each business experience rule label corresponds to at least K users, where K is a positive integer greater than 1, such as K = 300, so that an attacker cannot locate an individual. Random noise is added during the label generation stage, so that the data of a single user has negligible impact on the result.

[0043] At step S200, based on the input dataset, the service function chain optimization problem is decomposed into a multi-stage decision model having a time coupling relationship between UPF deployment and traffic scheduling.

[0044] Preferably, in the 5G core network, the service function chain optimization needs to solve two key problems simultaneously, including long-term planning, deployment location of user plane function UPF and calculation, storage resources and other resource capacities, considering edge node construction cost, energy consumption, etc.; short-term scheduling, dynamic routing of business traffic, such as selecting UPF path, needs to meet real-time network state and business delay, bandwidth, etc. Specifically, according to the input dataset, the service function chain optimization problem is decomposed into a time-coupled multi-stage decision model, including a long-term stage decision model for solving the UPF deployment main problem, and a short-term stage decision model for solving the traffic scheduling sub-problem, wherein the decision variables of the long-term stage decision model are the deployment location, resource reservation amount, and activation / sleeping strategy of the UPF; the target is to minimize the long-term cost, such as edge server energy consumption and deployment overhead, while reserving elastic resources to cope with future traffic demand; the input basis includes historical business density heat map and predicted future load trend. The decision variables of the short-term stage decision model are the path selection and resource allocation of the business flow, such as bandwidth allocation ratio; the target is to minimize the real-time delay / packet loss rate and meet the business SLA; the input basis is the current network state and business priority label. The multi-stage decision model has a time coupling relationship between UPF deployment and traffic scheduling, that is, long-term UPF deployment provides resource basis for short-term scheduling, and the feedback of short-term scheduling in turn guides long-term deployment adjustment, and by alternately iterating to solve the main problem and the sub-problem, the global resource and traffic are adapted. Thus, the time and space coupling problem of UPF deployment and traffic scheduling is solved, the efficient utilization of 5G core network resources and the global optimization of business experience are realized, thereby ensuring to improve resource efficiency, reduce energy consumption and protect user business experience.

[0045] Step S200 further includes step S210 of dividing an optimization period into a plurality of consecutive time slots, and defining UPF deployment strategy variables and traffic scheduling strategy variables in each time slot, respectively; step S220 of establishing UPF deployment state transmission constraints between adjacent time slots based on the evolution relationship of UPF deployment state between adjacent time slots; and step S230 of introducing UPF transmission migration cost and deployment switching cost in the objective function, and constructing the multi-stage decision model including time dependence, for quantifying the time coupling relationship between UPF deployment and traffic scheduling.

[0046] Preferably, the optimization period is divided into a plurality of discrete continuous time slots, two types of decision variables are defined in each time slot, including UPF deployment strategy variables for long-term decision and traffic scheduling strategy variables for short-term decision, i.e., the deployment state of the UPF and the scheduling path of the traffic are described in each time slot, respectively; then, based on the evolution relationship of the UPF deployment state between adjacent time slots, a transfer constraint of the deployment state is established between adjacent time slots for constraining the state change between adjacent time slots to ensure the time continuity of the UPF deployment, wherein the deployment state of the adjacent time slots satisfies the UPF activation / sleep delay, and the resource allocation change of the adjacent time slots satisfies the resource adjustment amplitude limit. The change cost of the UPF deployment is modeled as a migration cost term added to the optimization objective, and the service flow needs to be rerouted during the migration, increasing the delay penalty term, thereby establishing a time coupling dependency model of the UPF deployment and the traffic scheduling in the service function chain, i.e., the multi-stage decision model including time dependency is constructed, containing the service cost of the resource cost, the migration cost and the delay / packet loss rate penalty of the traffic scheduling, for quantifying the time coupling relationship of the UPF deployment and the traffic scheduling, the long-term UPF deployment determines the feasible solution space of the short-term traffic scheduling, the short-term scheduling result triggers the long-term deployment adjustment, and then the cross-time multi-stage collaborative optimization allocation is realized.

[0047] Step S300, the multi-stage decision model is decomposed into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem through a decomposition algorithm, and a global optimal solution is obtained through the iterative solution of the main and sub-problems.

[0048] Step S300 further comprises the following steps: S310, defining the UPF deployment state variable as the decision variable of the main problem for determining the UPF deployment location and quantity of the edge node in each time slot within the optimization period; S320, defining the traffic flow path selection variable and the resource scheduling variable as the decision variable of the sub-problem, and limiting the optimization of the decision variable of the sub-problem within the deployed UPF node range output by the main problem through constraints; S330, establishing a dependency constraint relationship between the main problem and the sub-problem, so that the feasible solution space of the sub-problem is constrained by the deployment result of the main problem, wherein an agent variable is introduced in the objective function of the main problem for representing the scheduling cost fed back by the sub-problem; S340, in each iteration, the main problem is solved to obtain a set of UPF deployment schemes, which are input to the sub-problem, and the sub-problem is independently solved to obtain the optimal traffic scheduling path based on the input; S350, the solving result of the sub-problem is fed back to the main problem through a cutting plane to build a cutting constraint for limiting the deployment combination, and the cutting constraint is added to the main problem model; S360, through the iterative optimization of the main and sub-problems, the combined scheme of the UPF deployment and the traffic scheduling strategy is corrected until the convergence condition is met, and the global optimal solution is obtained.

[0049] Preferably, the complex service function chain dynamic optimization problem is decomposed into two levels using a hierarchical optimization framework. Specifically, the UPF deployment state variable is defined as the main problem decision variable, responsible for long-term decision making, used to determine the UPF deployment location, number and computing, storage and other resource capacities of the edge node in each time slot within the optimization period, belonging to the infrastructure layer planning; the service flow path selection variable and the resource scheduling variable are defined as the sub-problem decision variable, responsible for short-term decision making, and the sub-problem decision variable is limited within the deployed UPF node range output by the main problem through constraints, optimizing the path selection and resource allocation of the service flow, belonging to the service layer scheduling; by decoupling long-term deployment and short-term scheduling, the problem complexity is reduced, and the iterative feedback mechanism is used to ensure the collaborative optimization of the two.

[0050] Preferably, a dependent constraint relationship is established between the main problem and the sub-problem, so that the feasible solution space of the sub-problem is constrained by the deployment result of the main problem, wherein the UPF deployment state variable identifies whether a UPF is deployed at a certain edge node in a specific time slot, and defines the resource allocation amount of the UPF, such as the number of CPU cores and the size of memory; the proxy variable is used to represent the optimization result of the sub-problem, such as the total time delay of traffic scheduling and resource utilization, as the cost feedback of the main problem. The service flow path selection variable defines the UPF node sequence through which the service flow passes, and the resource scheduling variable is used to allocate bandwidth, computing resources and the like. The dependent constraint relationship between the main problem and the sub-problem includes a deployment range constraint, the path selection of the sub-problem is limited to the UPF nodes deployed by the main problem, ensuring that the traffic is routed to the active UPF; a resource capacity constraint, the total amount of resources allocated by the sub-problem should not exceed the reserved amount by the main problem.

[0051] Preferably, the historical business data, network topology, and agent variable feedback of the sub-problem are inputted to solve the main problem with the goal of minimizing the cost, and a set of UPF deployment schemes is outputted. The UPF deployment scheme outputted by the main problem and the real-time network state are inputted to solve the sub-problem, and the optimal traffic path and resource allocation scheme, and the corresponding scheduling cost, such as the total delay, are outputted. The service flow must pass through the deployed UPF node and meet the SLA requirement. Then, the solving result of the sub-problem is fed back to the main problem through the cutting plane, that is, the sub-problem converts the performance defects of the current UPF deployment scheme, such as high delay caused by overload of a certain node, into linear inequalities to build the cutting constraints for limiting the deployment combination, which is used to eliminate the invalid deployment scheme in the main problem and is added to the main problem model. Then, the combination scheme of the UPF deployment and the traffic scheduling strategy is corrected through the alternating iteration optimization of the main-sub problem until the convergence condition is met. Specifically, the main problem sets the initial UPF deployment scheme, such as uniform distribution, solves the output of the current optimal deployment scheme, fixes the deployment scheme to solve the sub-problem, that is, optimizes the traffic scheduling path, calculates the actual scheduling cost, limits the deployment combination according to the cutting constraints, excludes the current inefficient deployment, terminates the iteration when the gap between the target value of the main problem and the actual cost of the sub-problem is less than the threshold, and finally obtains the global optimal solution, thereby guaranteeing the joint optimization of the UPF deployment and the traffic scheduling and the calculation efficiency of the problem solving.

[0052] In step S400, the long-term UPF deployment main problem is solved, and the UPF deployment position and quantity are dynamically determined to maximize the edge server resource utilization and reduce the energy consumption.

[0053] Preferably, based on the solving result of the main problem of long-term UPF deployment, the physical deployment location and the number of instances of the user plane function are dynamically adjusted to maximize the resource utilization rate of the edge server and reduce energy consumption, that is, to avoid resource idling or overload, improve hardware efficiency, and reduce the total power consumption of the edge data center through intelligent start and stop of UPF instances and resource scaling. Specifically, based on the deployment scheme output by the main problem and the business density heat map, UPFs are densely deployed in high-load areas and the coverage of UPFs is combined in low-load areas; at the same time, the number of UPF instances is dynamically increased or decreased according to the business load, when the resource utilization rate is continuously >80%, new instances are triggered to expand, when the utilization rate is <30%, redundant instances are closed, and the CPU and memory allocation of a single UPF instance is adjusted. Maximizing the resource utilization rate of the edge server means that the business traffic is evenly distributed to the deployed UPFs through load balancing to avoid hot nodes, and the traffic distribution is optimized based on reinforcement learning, and the computing resources of UPF instances are pooled through network function virtualization; reducing energy consumption means that idle UPF instances are put into low-power mode or completely shut down, UPFs are preferentially deployed on edge nodes powered by renewable energy, and UPF load is intelligently scheduled to reduce cooling energy consumption in combination with a data center temperature model. Further, an intelligent UPF management mode balancing resources and energy efficiency is realized to ensure efficient resource utilization and reduce energy consumption.

[0054] Step S500, according to the solving result of the short-term traffic scheduling sub-problem, combining real-time network state and business level information, dynamically selecting a traffic forwarding path that meets the service level to ensure user experience.

[0055] Preferably, based on the solving result of the short-term traffic scheduling sub-problem, combining real-time network state and business level information, the optimal traffic forwarding path is dynamically selected to ensure that high-priority businesses such as remote medical care and industrial control always meet strict service level agreements, and low-priority businesses such as background downloads reasonably use remaining resources to maximize global network efficiency, avoid local congestion, and ensure user experience. Specifically, the delay, packet loss rate, available bandwidth prediction value of each candidate path output by the short-term scheduling sub-problem, UPF node load, link congestion degree, wireless side signal quality, and business level information are used as inputs to dynamically select a traffic forwarding path that meets the service level, including forcing key businesses to select the shortest delay path even if the resource cost is high; ordinary businesses select the path with the lowest load under the premise of meeting the basic SLA; background businesses are allowed to use idle resources, and the path selection aims to save energy; at the same time, link utilization is continuously monitored for congestion avoidance and fault recovery.

[0056] Step S600, based on the user classification strategy, a dual-node deployment framework is constructed at the core layer and the edge layer, and a differentiated service function link control strategy is executed according to different business levels.

[0057] The step S600 further comprises a step S610 of establishing a user hierarchical policy including policy mapping rules between user attributes, user categories, service priorities, and delay constraints, wherein the user categories include premium service users and ordinary service users; a step S620 of constructing a dual-node deployment framework, wherein UPF resources with large capacity and high availability are deployed at a core layer for carrying basic network communications of ordinary service users, and low-latency and high-response UPF instances are deployed at an edge layer for providing fast access paths for premium service users; and a step S630 of establishing a mapping relationship of service function chain control policies between the user hierarchical policy and the dual-node deployment framework according to the user categories, and executing differentiated service function chain control policies.

[0058] Preferably, the user categories and corresponding service rules are defined by multi-dimensional attributes, including policy mapping rules between user attributes, user categories, service priorities, and delay constraints, wherein the user categories include premium service users and ordinary service users, the premium users are bound to edge low-latency UPFs, and path selection is forced to meet SLA; the ordinary users are dynamically allocated to core or edge UPFs based on a load balancing strategy, thereby forming a user hierarchical policy, as shown in Table 1:

[0059] Table 1 User hierarchical policy data table

[0060]

[0061] Preferably, a dual-node deployment framework of the core layer and the edge layer is constructed, wherein UPF resources with large capacity and high availability are deployed at the core layer for carrying basic network communications of ordinary service users, including web browsing and video streaming of ordinary users, and the UPF resources are deployed in regional data centers; low-latency and high-response UPF instances are deployed at the edge layer for providing fast access paths for premium service users, including industrial AR / VR, automatic driving V2X communication, and remote medical image transmission, and the UPF instances are deployed at base stations or in park MECs. Then, a mapping relationship of service function chain control policies is established between the user hierarchical policy and the dual-node deployment framework according to the user categories, that is, premium service users preferentially select edge UPF paths, and ordinary service users default to access core UPF paths; when network load mutates or path quality decreases, the function chain control policies are dynamically adjusted, for example, when edge UPFs are overloaded, the resources of ordinary users are preferentially preempted, and a pre-empty resource pool mechanism is used to always reserve 20% of edge resources for burst demand; when the core layer is congested, part of the traffic is offloaded to idle edge nodes, and a degradation mode is enabled to temporarily relax SLA, so as to realize network service upgrading, ensure service continuity and level guarantee, thereby improving resource utilization and reducing energy consumption.

[0062] In the foregoing, reference is made toFigure 1 A method for dynamically optimizing a service function chain of a 5G core network according to an embodiment of the present application is described in detail. Next, a system for dynamically optimizing a service function chain of a 5G core network according to an embodiment of the present application will be described with reference to Figure 2 A system for dynamically optimizing a service function chain of a 5G core network according to an embodiment of the present application is described.

[0063] The system for dynamically optimizing a service function chain of a 5G core network according to the embodiment of the present application is used to solve the technical problems of lack of dynamic collaborative optimization of UPF deployment and traffic scheduling, low resource utilization, high energy consumption, and poor service adaptability caused by static optimization of service function chains in the prior art, and achieves the technical effects of improving edge server resource utilization, reducing network energy consumption, optimizing service quality, and ensuring user experience. As shown in Figure 2 The system for dynamically optimizing a service function chain of a 5G core network includes a network service data acquisition unit 10, a multi-stage decision model obtaining unit 20, a global optimal solution obtaining unit 30, a UPF deployment determining unit 40, a traffic forwarding path selection unit 50, and a link control strategy execution unit 60.

[0064] The network service data acquisition unit 10 is used to acquire network service data in real time through a three-dimensional perception module, including user attribute data, network state data, and service experience data, and construct an input data set for service function chain optimization. The multi-stage decision model obtaining unit 20 is used to decompose the service function chain optimization problem into a multi-stage decision model based on the input data set, and the multi-stage decision model has a time coupling relationship between UPF deployment and traffic scheduling. The global optimal solution obtaining unit 30 is used to decompose the multi-stage decision model into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem through a decomposition algorithm, and obtain a global optimal solution through the iterative solution of the main and sub-problems. The UPF deployment determining unit 40 is used to dynamically determine the location and number of UPF deployment based on the solution result of the long-term UPF deployment main problem, maximize edge server resource utilization, and reduce energy consumption. The traffic forwarding path selection unit 50 is used to dynamically select a traffic forwarding path that meets the service level based on the solution result of the short-term traffic scheduling sub-problem, combined with real-time network state and service level information, to ensure user experience. The link control strategy execution unit 60 is used to construct a dual-node deployment framework at the core layer and the edge layer based on user classification strategies, and execute differentiated service function chain link control strategies according to different service levels.

[0065] Below, the specific configuration of the network service data collection unit 10 will be described in detail. The network service data collection unit 10 further comprises: a three-dimensional perception module comprising a data fusion engine, the configuration of the data fusion engine comprising: adding a privacy desensitization label to the user attribute data to obtain the user attribute data; collecting the obtained network state data through time sliding window normalization; and acquiring the service experience data according to the dynamic adjustment of the collection frequency according to the business type.

[0066] Below, the specific configuration of the network service data collection unit 10 will be described in detail. The network service data collection unit 10 further comprises: performing regional business traffic spatiotemporal distribution analysis based on historical data to construct a business density heat map; calculating the optimal coverage radius of the network service according to the regional business intensity in the business density heat map and the distance relationship between regions; deploying delay constraints for each region based on the optimal coverage radius and the business distribution characteristics to determine the distribution of edge nodes and establish edge nodes.

[0067] Below, the specific configuration of the network service data collection unit 10 will be described in detail. The network service data collection unit 10 further comprises: setting the network resource reservation amount according to the peak load based on the business density heat map; defining node intimacy based on the network service coverage heat relationship of the edge nodes and forming a collaborative node group based on the node intimacy; performing joint resource configuration on the edge nodes and the collaborative node group based on the network resource reservation amount and the node collaboration relationship to construct a collaborative elastic resource pool for dynamic reconstruction or elastic deployment in service function scheduling.

[0068] Below, the specific configuration of the network service data collection unit 10 will be described in detail. The network service data collection unit 10 further comprises: injecting noise into the original user information based on local difference to upload the desensitized feature vectors on each edge node; mapping the feature vectors to temporary encrypted IDs; training a local model using the desensitized data, uploading the encrypted model gradient to a federated learning aggregator deployed in the core layer, generating a global gradient through homomorphic summation aggregation; the federated learning aggregator issues a global model to the edge nodes, outputs a de-identified service experience rule label, and obtains the user attribute data.

[0069] Below, the specific configuration of the network service data collection unit 10 will be described in detail. The network service data collection unit 10 further comprises: the de-identified service experience rule label is a privacy desensitization label and does not contain information that can identify personal identity.

[0070] In the following, the specific configuration of the multi-stage decision model obtaining unit 20 will be described in detail. The multi-stage decision model obtaining unit 20 further comprises: dividing the optimization period into a plurality of consecutive time slots, defining UPF deployment strategy variables and traffic scheduling strategy variables in each time slot respectively; establishing UPF deployment state transmission constraints between adjacent time slots based on the evolution relationship of UPF deployment state between adjacent time slots; introducing UPF transmission migration cost and deployment switching cost in the objective function, and constructing the multi-stage decision model including time dependence, which is used to quantify the time coupling relationship between UPF deployment and traffic scheduling.

[0071] In the following, the specific configuration of the global optimal solution obtaining unit 30 will be described in detail. The global optimal solution obtaining unit 30 further comprises: defining the UPF deployment state variable as the main problem decision variable, which is used to determine the UPF deployment location and quantity of the edge node in each time slot within the optimization period; defining the traffic flow path selection variable and the resource scheduling variable as the sub-problem decision variable, and limiting the optimization of the sub-problem decision variable within the deployed UPF node range output by the main problem through constraints; establishing a dependent constraint relationship between the main problem and the sub-problem, so that the feasible solution space of the sub-problem is constrained by the deployment result of the main problem, wherein an agent variable is introduced in the objective function of the main problem to represent the scheduling cost fed back by the sub-problem; in each iteration, the main problem is solved first to obtain a set of UPF deployment schemes, which are input to the sub-problem, and the sub-problem is independently solved based on this to obtain the optimal traffic scheduling path; the solving result of the sub-problem is fed back to the main problem through the cutting plane, a cutting constraint for limiting the deployment combination is constructed, and is added to the main problem model; through the alternating iteration optimization of the main-sub problem, the combined scheme of UPF deployment and traffic scheduling strategy is corrected until the convergence condition is met, and the global optimal solution is obtained.

[0072] In the following, the specific configuration of the link control strategy execution unit 60 will be described in detail. The link control strategy execution unit 60 further comprises: establishing a user hierarchical strategy, including the mapping rules between user attributes, user categories, service priorities, and delay constraints, wherein the user categories include high-level service users and ordinary service users; constructing a dual-node deployment framework, wherein UPF resources with large capacity and high availability are deployed in the core layer to carry the basic network communication of ordinary service users; low-latency and high-response UPF instances are deployed in the edge layer to provide fast access paths for high-level service users; according to the user categories, a mapping relationship between the user hierarchical strategy and the dual-node deployment framework is established to execute the differentiated service function link control strategy.

[0073] The system for dynamically optimizing a service function chain of a 5G core network provided by the embodiment of the application can execute the method for dynamically optimizing a service function chain of a 5G core network provided by any embodiment of the application, and has the function modules and beneficial effects corresponding to the execution method.

[0074] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual differentiation, and do not limit the protection scope of the present application.

[0075] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1.A method for dynamic optimization of service function chaining of a 5G core network, characterized in that, Comprise: Through the three-dimensional perception module, real-time acquisition of network service data, including user attribute data, network state data, service experience data, construct the input data set for service function chain optimization; Based on the input data set, the service function chain optimization problem is decomposed into a multi-stage decision model, which has a time coupling relationship between UPF deployment and traffic scheduling; The multi-stage decision model is decomposed into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem by a decomposition algorithm, and the global optimal solution is obtained by iterative solving of the main and sub-problems; According to the solving result of the long-term UPF deployment main problem, dynamically determine the UPF deployment location and quantity, maximize the edge server resource utilization and reduce the energy consumption; According to the solving result of the short-term traffic scheduling sub-problem, combined with real-time network state and service level information, dynamically select the traffic forwarding path that meets the service level to ensure user experience; Based on the user classification strategy, a double-node deployment framework is constructed in the core layer and the edge layer, and a differentiated service function chain control strategy is executed according to different service levels. 2.The method of claim 1, wherein, Through the three-dimensional perception module, real-time acquisition of network service data, wherein the three-dimensional perception module comprises a data fusion engine, and the configuration of the data fusion engine comprises: Add privacy desensitization label to user attribute data to obtain the user attribute data; Through time sliding window normalization, the obtained network state data is collected; According to the business type, dynamically adjust the collection frequency to obtain the business experience data. 3.The method of claim 2, wherein, Through the three-dimensional perception module, real-time acquisition of network service data, previously also includes: Based on historical data, analyze the temporal and spatial distribution of regional business traffic, and construct a business density heat map; According to the regional business intensity and the distance relationship between regions in the business density heat map, calculate the optimal coverage radius of the network service; Based on the optimal coverage radius and the business distribution characteristics, deploy the delay constraints of each region to determine the edge node distribution and establish the edge node. 4.The method of claim 3, wherein, After establishing the edge node, it further comprises: Based on the business density heat map, set the network resource reservation amount according to the peak load; According to the network service coverage heat relationship of the edge node, define the node intimacy, and form a cooperative node group based on the node intimacy; Based on the network resource reservation amount and the node cooperation relationship, perform joint resource configuration on the edge node and the cooperative node group to construct a cooperative elastic resource pool for dynamic reconstruction or elastic deployment in service function scheduling. 5.The method of claim 3, wherein, The user attribute data is added with a privacy desensitization label to obtain the user attribute data, comprising: Based on local difference, inject noise into the original user information, and upload the desensitized feature vector on each edge node; Map the feature vector to a temporary encrypted ID; Use the desensitized data to train a local model, upload the encrypted model gradient to the federated learning aggregator deployed in the core layer, aggregate through homomorphic summation to generate a global gradient; The federated learning aggregator issues a global model to the edge node, outputs a de-identified business experience rule label, and obtains the user attribute data. 6.The method of claim 5, wherein, The de-identified service experience rule label is a privacy desensitization label, and does not contain identifiable personal identity information. 7.The method of claim 1, wherein, The service function chain optimization problem is decomposed into a multi-stage decision model having a time coupling relationship between UPF deployment and traffic scheduling, including: An optimization period is divided into a plurality of consecutive time slots, and UPF deployment strategy variables and traffic scheduling strategy variables are respectively defined in each time slot. UPF deployment state transmission constraints between adjacent time slots are established based on the evolution relationship of the UPF deployment state between adjacent time slots. UPF transmission migration costs and deployment switching costs are introduced into the objective function to construct the multi-stage decision model including time dependence, which is used to quantify the time coupling relationship between UPF deployment and traffic scheduling. 8.The method of claim 7, wherein, The multi-stage decision model is decomposed into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem through a decomposition algorithm, and a global optimal solution is obtained through the iterative solution of the main and sub-problems, including: The UPF deployment state variable is defined as a main problem decision variable for determining the UPF deployment location and number of edge nodes in each time slot within the optimization period; The service flow path selection variable and the resource scheduling variable are defined as sub-problem decision variables, and the sub-problem decision variables are optimized within the deployed UPF node range output by the main problem through constraints; A dependent constraint relationship between the main problem and the sub-problem is established, so that the feasible solution space of the sub-problem is constrained by the deployment results of the main problem, wherein an agent variable is introduced into the objective function of the main problem to represent the scheduling cost feedback of the sub-problem; In each iteration, the main problem is solved to obtain a set of UPF deployment schemes, which are input to the sub-problem, and the sub-problem is independently solved based on the deployment schemes to obtain the optimal traffic scheduling path; The solution of the sub-problem is fed back to the main problem through a cutting plane to construct a cutting constraint for limiting the deployment combination, and the cutting constraint is added to the main problem model; Through the iterative optimization of the main and sub-problems, the combined scheme of UPF deployment and traffic scheduling strategy is modified until the convergence condition is met, and the global optimal solution is obtained. 9.The method of claim 8, wherein, Based on the user classification strategy, a dual-node deployment framework is constructed at the core layer and the edge layer, and a differentiated service function chain link control strategy is executed according to different service levels, including: A user classification strategy is established, including the mapping rules between user attributes, user categories, service priorities, and delay constraints, wherein the user categories include high-level service users and ordinary service users; A dual-node deployment framework is constructed, wherein UPF resources with large capacity and high availability are deployed at the core layer to carry out basic network communication for ordinary service users, and low-latency and high-response UPF instances are deployed at the edge layer to provide fast access paths for high-level service users; According to the user categories, a mapping relationship between the user classification strategy and the dual-node deployment framework is established for the service function chain control strategy, and a differentiated service function chain link control strategy is executed. 10.A system for dynamic optimization of service function chaining of a 5G core network, characterized in that, The system is used to implement the 5G core network service function chain dynamic optimization method of any one of claims 1 to 9, and the system comprises: The network service data collection unit is configured to collect network service data in real time through the three-dimensional perception module, including user attribute data, network state data, and service experience data, and construct an input data set for service function chain optimization. The multi-stage decision model obtaining unit is configured to decompose the service function chain optimization problem into a multi-stage decision model based on the input data set, and the multi-stage decision model has a time coupling relationship between UPF deployment and traffic scheduling. The global optimal solution obtaining unit is configured to decompose the multi-stage decision model into a long-term UPF deployment main problem and a short-term traffic scheduling sub-problem through a decomposition algorithm, and obtain a global optimal solution through iterative solving of the main and sub-problems. The UPF deployment determining unit is configured to dynamically determine the UPF deployment location and quantity according to the solving result of the long-term UPF deployment main problem, maximize the edge server resource utilization rate, and reduce the energy consumption. The traffic forwarding path selection unit is configured to dynamically select a traffic forwarding path that meets the service level according to the solving result of the short-term traffic scheduling sub-problem, in combination with real-time network state and service level information, to ensure user experience. The link control strategy execution unit is configured to construct a double-node deployment framework in the core layer and the edge layer based on user classification strategies, and execute differentiated service function chain link control strategies according to different service levels.

Images