Low-delay slice resource intelligent allocation method for 5G network

Abstract

The invention discloses a low-delay slice resource intelligent allocation method for a 5G network, and the method comprises the steps: collecting the connection state and performance index of a network node in real time, dynamically adjusting the collection frequency according to the fluctuation of a network load, and generating structured topological data; calculating a node degree distribution entropy value based on the topological data, generating a virtual resource topological graph, and predicting resource requirements of network nodes; constructing a dynamic confidence interval by using the virtual resource topological graph and the entropy change rate, and designing a resource pre-allocation strategy; network burst traffic is detected, resource allocation is rapidly adjusted through incremental updating of the graph neural network, and it is ensured that the reconfiguration delay is lower than 1 millisecond; according to the invention, through topology perception driven by the graph neural network, prediction guided by entropy and an adaptive allocation strategy, low-delay resource allocation of the uRLLC slice is realized, resource waste is avoided, and strict performance requirements of a 5G network are met.

Description

Technical Field

[0001] The present invention relates to the technical field of network slice resource management, and particularly to an intelligent allocation method for low-latency slice resources for 5G networks. Background Art

[0002] With the wide deployment of the fifth-generation mobile communication (5G) network, network slicing technology has become a core technology to support diverse service scenarios (such as enhanced mobile broadband eMBB, ultra-reliable low-latency communication uRLLC, massive machine communication mMTC). Network slicing realizes dynamic allocation of resources through software-defined network (SDN) and network function virtualization (NFV), providing customized quality of service (QoS) for different applications. In recent years, with the rapid development of artificial intelligence technology, it has shown significant potential in network topology modeling and resource allocation optimization. By integrating communication technology with artificial intelligence technology, positive progress has been made in improving network efficiency and supporting diverse services.

[0003] However, there are still some deficiencies in the existing technologies, which significantly limit their application in the low-latency scenario of uRLLC slices. First, traditional resource allocation methods (mainly focusing on resource allocation in static or semi-static scenarios) lack the ability to perceive real-time network topology changes, resulting in high reconfiguration latency when facing burst traffic or topology mutations (such as base station handover, VNF migration), and it is difficult to meet the strict latency requirements of uRLLC slices. Second, although existing machine learning-based methods can handle complex network topologies, most of them rely on historical data prediction, thus ignoring the real-time quantification of network dynamic complexity, leading to the inability to accurately predict resource requirements or quickly respond to traffic mutations. Third, existing methods lack adaptability in resource pre-allocation and fail to dynamically adjust the allocation strategy according to the network state, resulting in low resource utilization and large latency fluctuations. Summary of the Invention

[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides an intelligent allocation method for low-latency slice resources for 5G networks to solve the problems raised in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solution: An intelligent allocation method for low-latency slice resources for 5G networks, comprising:

[0007] Collect the connection status and performance metrics of network nodes in real time, dynamically adjust the collection frequency according to network load fluctuations, and generate structured topology data;

[0008] According to the topology data, calculate the degree distribution entropy value of network nodes, generate a virtual resource topology graph, and predict the resource requirements of network nodes;

[0009] Based on the virtual resource topology graph and the entropy value change rate, construct a dynamic confidence interval and design a resource pre-allocation strategy;

[0010] Detect the burst traffic of the network, and quickly adjust the resource allocation by incrementally updating the graph neural network, and reduce the reconfiguration delay.

[0011] As a preferred scheme of the low-latency slice resource intelligent allocation method for 5G networks according to the present invention, wherein: collecting the connection status and performance metrics of network nodes in real time, and dynamically adjusting the collection frequency according to network load fluctuations, generating structured topology data, including:

[0012] Collect the connection status data of network nodes through distributed network probes to form an adjacency matrix of the connection strength between nodes;

[0013] Collect the performance metrics of each node, including bandwidth utilization, computing load, and latency, to form a node feature matrix;

[0014] Calculate the degree of network load fluctuation according to the volatility of the bandwidth utilization rate, and dynamically adjust the sampling frequency of the distributed network probes;

[0015] Denoise and normalize the adjacency matrix and the node feature matrix to generate structured topology data.

[0016] As a preferred scheme of the low-latency slice resource intelligent allocation method for 5G networks according to the present invention, wherein: calculating the degree distribution entropy value of network nodes according to the topology data, including:

[0017] According to the adjacency matrix, calculate the degree of each node, which is the sum of the connection strengths between each node and its neighbor nodes;

[0018] Based on the degree of the nodes, construct a smooth degree distribution function of the nodes and calculate the entropy value of the structured topology data;

[0019] Fuse the entropy value of the current structured topology data with its historical entropy value in a time-weighted manner to generate a smooth effective entropy value;

[0020] Calculate the change rate of the effective entropy value, and store the effective entropy value and the change rate. When the effective entropy value exceeds the preset threshold of the effective entropy value, trigger the generation operation of the virtual resource topology graph.

[0021] As a preferred solution of the intelligent low-latency slice resource allocation method for 5G networks according to the present invention, wherein: when the effective entropy value exceeds the preset threshold of the effective entropy value, the generation operation of the virtual resource topology graph is triggered, including:

[0022] The preset threshold of the effective entropy value is obtained by calculating the mean and standard deviation of the effective entropy value and the network topology change coefficient.

[0023] As a preferred solution of the intelligent low-latency slice resource allocation method for 5G networks according to the present invention, wherein: the generation of the virtual resource topology graph includes:

[0024] Based on the structured topology data, construct a network topology graph, where the nodes represent network nodes, the edges represent connection relationships, and the node features are performance indicators;

[0025] Apply a graph neural network to process the network topology graph through a message passing mechanism to generate a virtual resource topology graph, and represent the virtual resource topology graph as a resource demand matrix, where the resource demand matrix includes the bandwidth and computing resource requirements of each node.

[0026] As a preferred solution of the intelligent low-latency slice resource allocation method for 5G networks according to the present invention, wherein: construct a dynamic confidence interval and design a resource pre-allocation strategy, including:

[0027] According to the resource demand matrix, calculate the predicted resource demand of each node and its degree of fluctuation;

[0028] Based on the predicted resource demand and degree of fluctuation, dynamically adjust the confidence level in combination with the change rate of the effective entropy value to construct a resource demand confidence interval for each node;

[0029] Design a resource pre-allocation strategy to preferentially meet the boundary requirements of the resource demand confidence interval, and at the same time constrain the total resource demand not to exceed the network available resources;

[0030] Store the allocation result obtained after executing the resource pre-allocation strategy and update the current network resource status.

[0031] As a preferred solution of the intelligent low-latency slice resource allocation method for 5G networks according to the present invention, wherein: detect the burst traffic of the network, and quickly adjust the resource allocation and reduce the reconfiguration delay by incrementally updating the graph neural network, including:

[0032] According to the temporal change of the bandwidth utilization rate, calculate the traffic mutation index, determine whether there is burst traffic, and when the traffic mutation index exceeds the preset threshold of the traffic mutation index, identify the affected subgraph;

[0033] By incrementally updating the graph neural network, recalculate the nodes for the affected subgraphs to generate a new resource demand matrix;

[0034] Combined with the resource demand confidence interval, optimize resource reallocation so that while meeting the high-priority node requirements of the uRLLC slice, the reallocation delay is less than 1 millisecond;

[0035] Feed back the resource reallocation result to the current network resource status.

[0036] As a preferred solution of the intelligent low-latency slice resource allocation method for 5G networks described in the present invention, where: when the traffic mutation index exceeds the preset threshold of the traffic mutation index, it includes:

[0037] The preset threshold of the traffic mutation index is obtained by calculating the mean and twice the standard deviation of the traffic mutation index within the time window.

[0038] Compared with the prior art, the beneficial effects of the invention are:

[0039] 1. Through the real-time topology awareness and incremental update mechanism driven by the graph neural network, the present invention can quickly respond to sudden traffic and network topology changes, ensure that the resource reallocation delay is less than 1 millisecond, meet the strict delay requirements of 5G ultra-reliable low-latency communication (uRLLC) slices, and overcome the limitation of the relatively high reallocation delay of traditional static or semi-static resource allocation methods in dynamic scenarios;

[0040] 2. Use the node degree distribution entropy value to quantify the network topology complexity, and combine with the dynamic confidence interval strategy to predict and pre-allocate resources in real time, give priority to meeting the needs of high-priority nodes (such as uRLLC slice VNFs), improve the accuracy of resource allocation, and adapt to the dynamic complexity of 5G networks;

[0041] 3. Through adaptive sampling frequency adjustment and entropy-guided graph neural network resource prediction, the present invention ensures low latency while dynamically restricting the total resource allocation not to exceed the network available resources, avoiding resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. Among them:

[0043] Figure 1 It is the overall flowchart of the intelligent low-latency slice resource allocation method for 5G networks described in an embodiment of the present invention. Detailed Embodiments

[0044] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0045] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0046] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0047] The present invention is described in detail in conjunction with schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views showing the device structure will be enlarged locally in a non-general proportion, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, in actual production, three-dimensional spatial dimensions including length, width, and depth should be included.

[0048] At the same time, in the description of the present invention, it should be noted that the orientation or positional relationships indicated by terms such as "upper, lower, inner, and outer" are based on the orientation or positional relationships shown in the drawings. This is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0049] Unless otherwise clearly defined and limited in the present invention, the terms "mounted, connected, and coupled" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can also be a mechanical connection, an electrical connection, or a direct connection, or can be indirectly connected through an intermediate medium, or can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0050] Embodiment 1

[0051] Refer toFigure 1 , which is the first embodiment of the present invention. This embodiment provides an intelligent allocation method for low-latency slice resources for 5G networks, including:

[0052] S1. Collect the connection status and performance metrics of network nodes in real time, and dynamically adjust the collection frequency according to network load fluctuations to generate structured topology data;

[0053] Furthermore, deploy distributed network probes (such as sensors based on the SNMP protocol), integrate them into the network function virtualization (NFV) infrastructure of the 5G core network, and collect node connection status and node performance metrics every time step Δt (the initial value is 100 ms);

[0054] Furthermore, form an adjacency matrix according to the collected node connection status (logical connections between nodes) N is the number of nodes;

[0055] Specifically, the elements in this adjacency matrix include the connection strength between nodes:

[0056]

[0057] Among them, A ij (t) represents the connection strength between nodes i and j; B ij (t) represents the link bandwidth utilization rate; B max represents the maximum bandwidth; L ij (t) represents the link delay; L ref is the reference delay (default is 1 ms). To support the uRLLC slice, the weight of the low-latency link is preferentially enhanced (by reducing L ref to 0.5 ms);

[0058] Furthermore, store the collected node performance metrics as a node feature vector. Among them, the node performance metrics include bandwidth utilization rate, computing load, and delay. Through this node feature vector, form a node feature matrix 3 represents the dimension of the node feature vector;

[0059] Specifically, the node feature vector x i (t) is expressed as:

[0060] x i (t) = [B i (t), C i (t), D i (t)]

[0061] Among them, B i (t) represents the bandwidth utilization rate of each node, C i(t) represents the computing load (specifically CPU / GPU utilization rate), D i (t) represents the latency;

[0062] Furthermore, calculate the degree of network load fluctuation according to the volatility of bandwidth utilization, and dynamically adjust the sampling frequency of the distributed network probe;

[0063] Specifically, the formula for calculating the degree of network load fluctuation is expressed as:

[0064]

[0065] Among them, represents the volatility of bandwidth utilization, which is obtained from the bandwidth utilization of each node and can be expressed as ∈ t represents the degree of network load fluctuation. When it exceeds the upper limit of the current network load fluctuation (determined according to the deployment environment of the 5G core network), the acquisition frequency of the distributed network probe is shortened to Δt′ = Δt / (1 + ∈ t ), otherwise it is extended to Δt′ = min(1.5Δt, 500ms);

[0066] Furthermore, denoise and normalize the adjacency matrix and node feature matrix to generate structured topology data;

[0067] Specifically, wavelet transform method (select Daubechies wavelet, decomposition level 4) is used for denoising to filter out high-frequency noise and retain the change trend of structured topology data;

[0068] Specifically, normalize A ij (t) to the interval [0, 1], and based on the time window [t - 10Δt, t], standardize each dimension of x i (t) to zero mean and unit variance;

[0069] It should be noted that since denoising and normalization are conventional operations of data processing and are not related to the innovation direction of the present invention, they will not be elaborated here;

[0070] Specifically, store the data after denoising and normalization (such as Redis) to support fast query;

[0071] It should be noted that the data acquisition operation of the dynamic distributed network probe according to the degree of network load fluctuation can effectively reduce the amount of redundant data and enhance the accuracy of the structured topology data representation;

[0072] S2. According to the topology data, calculate the degree distribution entropy value of network nodes, generate a virtual resource topology map, and predict the resource requirements of network nodes;

[0073] Further, according to the adjacency matrix in the structured topology data, calculate the degree of each node as the sum of the connection strengths between each node and its neighbor nodes;

[0074] Specifically, assuming that node j is a neighbor node of node i, then the degree d i (t) of each node is:

[0075]

[0076] Further, based on the degree of the nodes, construct a degree distribution function for smoothing the nodes and calculate the entropy value of the structured topology data;

[0077] Specifically, construct the node degree distribution and use kernel density estimation (using a Gaussian kernel function, and the kernel density estimation bandwidth is expressed as h = 0.1·std(d), that is, 0.1 times the standard deviation of the node degree, which is an empirical selection) to smooth the distribution:

[0078]

[0079] Among them, P(d,t) represents the degree distribution function of the smoothed nodes, and in the uRLLC slice scenario, it can reflect the changes in the structured topology data (such as base station handover or VNF migration); K is the Gaussian kernel function;

[0080] It should be noted that using the smoothing property of the Gaussian kernel function helps to filter the short-term noise of the structured topology data (such as instantaneous connection disconnection) in the uRLLC scenario to improve the robustness of the entropy value calculation;

[0081] Further, fuse the entropy value of the current structured topology data with its historical entropy value through a time-weighted method to generate a smoothed effective entropy value;

[0082] Specifically, calculate the entropy value of the current structured topology data as:

[0083]

[0084] It should be noted that based on the Shannon entropy formula, the information contribution of each node's degree value d is quantified through logP(d,t) to show the changes in the network topology complexity, providing support for the resource demand prediction of uRLLC slices;

[0085] Specifically, introduce time weighting to generate a smoothed effective entropy value H eff (t):

[0086] H eff (t) = (1 - α)H(t) + αH(t - 1)

[0087] Among them, α is the time weighting factor, usually taking a value of 0.7. By adjusting the time weighting factor, the smoothing degree is controlled to reduce the influence of short-term fluctuations on the entropy value and improve the reliability of triggering the generation of the virtual resource topology map; H(t - 1) refers to the smoothed entropy value at the previous time step, which is incorporated into the current calculation as a representative of the historical entropy value.

[0088] Furthermore, calculate the change rate of the effective entropy value, and store the effective entropy value and the change rate. When the effective entropy value exceeds the preset threshold of the effective entropy value, trigger the generation operation of the virtual resource topology map.

[0089] Specifically, the change rate of the effective entropy value is obtained from the current effective entropy value and the effective entropy value at the previous time step:

[0090] ΔH(t) = H eff (t) - H eff (t - 1)

[0091] Among them, ΔH(t) represents the change rate of the effective entropy value.

[0092] It should be noted that obtaining the change rate of the effective entropy value is mainly used for adjusting the dynamic confidence interval, thereby connecting the entropy value with the uRLLC slice resource allocation.

[0093] Specifically, the preset threshold of the effective entropy value is obtained by statistically analyzing the time series distribution of the entropy value of the current structured topology data, calculating the mean and standard deviation of the effective entropy value, and the network topology change coefficient (k = 1.5):

[0094]

[0095] Among them, ΔH th (t) represents the preset threshold of the effective entropy value, represents the mean of the effective entropy value, represents the standard deviation of the effective entropy value;

[0096] Specifically, statistically analyzing the time series distribution of the entropy value of the current structured topology data means that when the entropy value of the current structured topology data changes significantly in the network topology (such as node connection disconnection / new addition, traffic pattern mutation);

[0097] It should be noted that because the uRLLC slice is sensitive to changes in the network topology, it is necessary to preset the threshold of the effective entropy value to ensure that when the network topology complexity increases significantly, resource prediction is triggered in a timely manner to meet the requirements of low latency.

[0098] S3. Based on the virtual resource topology map and the entropy value change rate, construct a dynamic confidence interval and design a resource pre-allocation strategy.

[0099] Furthermore, based on the structured topology data, a network topology graph is constructed, where nodes represent network nodes, edges represent connection relationships, and node features are performance metrics;

[0100] Specifically, the network topology graph can be expressed by the formula:

[0101] G(t) = (V, E, X)

[0102] where V is the set of nodes, E is the set of edges representing the connection relationships between nodes, defined by the adjacency matrix A(t), and X is the node feature, obtained from the node feature matrix;

[0103] Furthermore, applying a graph neural network, the network topology graph is processed through a message passing mechanism to generate a virtual resource topology graph, and the virtual resource topology graph is represented as a resource demand matrix, where the resource demand matrix contains the bandwidth and computing resource demands of each node;

[0104] Specifically, the applied graph neural network (GNN) consists of two layers of graph convolutional networks (GCN), with the output dimension of each layer being 64 (F′ = 64). First, the node features, adjacency matrix, and effective entropy value in the input network topology graph are used to initialize the GNN, and the pre-trained GNN model is loaded using [t - 10Δt, t] to optimize the weight matrix. Then, the message passing mechanism is used to perform the forward propagation of the GCN to update the node embeddings. The first layer of GCN aggregates the neighbor node information, and the second layer of GCN receives the aggregation result of the first layer of GCN for secondary aggregation to obtain the final node embeddings. Finally, the node embeddings are transformed into resource demands through a fully connected layer and output by the output layer to generate the virtual resource topology graph;

[0105] Specifically, the update of the node embeddings is expressed by the formula:

[0106]

[0107] where, represents the embedding vector of node i in the (l + 1)-th layer, with the initial value of N i represents the neighbor set of node i, obtained based on A(t); represents the weight matrix of the l-th layer; ω j (t) is the attention weight, used to dynamically adjust the contribution of neighbor node j. Based on the degree and effective entropy value of the node, it can enhance the resource prediction priority of nodes in the high entropy region and can be further expressed as:

[0108]

[0109] Specifically, the fully connected layer transforms the node embedding into a resource demand:

[0110]

[0111] Among them, R i (t) represents the resource demand vector of node i, including bandwidth and computing resources (dimension is 2); W out represents the weight matrix of the output layer; b out represents the bias vector;

[0112] Specifically, the virtual resource topology graph generated by the GNN model is represented as a resource demand matrix and is directly used for resource pre-allocation of uRLLC slices to improve the priority of low-latency requirements;

[0113] Furthermore, according to the resource demand matrix, calculate the predicted resource demand of each node and its degree of fluctuation;

[0114] Specifically, the predicted resource demand is expressed as:

[0115]

[0116] Specifically, the degree of fluctuation of the predicted resource demand ∈ R (t) is the same as the aforementioned network load fluctuation and is expressed as:

[0117]

[0118] Among them, represents the degree of fluctuation of the predicted resource demand, which is obtained from the resource demand vector of each node and can be expressed as

[0119] Furthermore, based on the predicted resource demand and the degree of fluctuation, dynamically adjust the confidence level in combination with the change rate of the effective entropy value, and construct the resource demand confidence interval for each node;

[0120] Specifically, the constructed resource demand confidence interval CI i (t) is expressed as:

[0121]

[0122] It should be noted that in the uRLLC scenario, the resource demand confidence interval needs to ensure that the resource demands of high-priority nodes (such as VNFs) are met to avoid increased latency caused by prediction resource demand errors;

[0123] Specifically, z(t) represents the dynamic confidence level and is adjusted in combination with the change rate of the effective entropy value:

[0124] z(t) = z0 + γ·tanh(κ·|ΔH(t)|)

[0125] where z0 is the initial confidence level, with a value of 1.96 (corresponding to a 95% confidence level, based on the standard normal distribution); γ is used to adjust the change range of the confidence level, with a value of 0.5; κ is used to control the sensitivity of the entropy value change rate, with a value of 2;

[0126] It should be noted that when the network topology complexity changes greatly (|ΔH(t)| is high), z(t) increases, making the confidence interval of resource requirements wider, thus enhancing the conservatism of slice resource allocation (i.e., allocating more resources to cover the boundary requirements and reducing the risk of resource shortage caused by prediction resource requirement errors); when the network topology is stable, z(t) is close to z0, making the confidence interval of resource requirements narrower, thus making the resource requirement prediction more accurate. At this time, through an optimization algorithm (such as a linear programming solver), it can be closer to the predicted value thus reducing resource waste;

[0127] Furthermore, design a resource pre-allocation strategy to preferentially meet the boundary requirements of the resource requirement confidence interval, while restricting the total resource demand not to exceed the network available resources;

[0128] Specifically, input the confidence interval into the optimization algorithm to establish an optimization objective:

[0129]

[0130] where represents the actual resource allocation, obtained by the optimization algorithm, which is directly applied to the SDN / NFV architecture of the 5G network, and the resource configuration of the virtual network function (VNF) is adjusted through the management and orchestration (MANO) system, thereby realizing resource allocation;

[0131] Specifically, establish the constraint conditions of the optimization objective:

[0132] (The total resource demand does not exceed the network available resources)

[0133] (The resource allocation result is within the resource requirement confidence interval)

[0134] where R total represents the total resource demand;

[0135] Specifically, store the allocation result obtained after executing the resource pre-allocation strategy and update the current network resource status;

[0136] S4. Detect the burst traffic of the network, and quickly adjust the resource allocation and reduce the reconfiguration delay by incrementally updating the graph neural network;

[0137] Further, according to the temporal variation of the bandwidth utilization rate, calculate the traffic mutation index to determine whether there is burst traffic. When the traffic mutation index exceeds the preset threshold of the traffic mutation index, identify the affected subgraphs;

[0138] Specifically, the formula for calculating the traffic mutation index is:

[0139]

[0140] where S(t) represents the traffic mutation index;

[0141] It should be noted that reconfiguration is only triggered when the traffic mutation index exceeds the preset threshold of the traffic mutation index;

[0142] Furthermore, the preset threshold of the traffic mutation index is obtained by calculating the mean and twice the standard deviation of the traffic mutation index within the time window:

[0143] ΔS th (t) = μ S + 2σ S

[0144] where ΔS th (t) represents the preset threshold of the traffic mutation index, μ S represents the mean of the traffic mutation index, and σ S represents the standard deviation of the traffic mutation index;

[0145] Further, by incrementally updating the graph neural network, recalculate the nodes of the affected subgraphs to generate a new resource demand matrix;

[0146] Specifically, incrementally update the node embeddings in the graph neural network:

[0147]

[0148] where h′ i (t) represents the updated node embedding, represents the affected neighbor set;

[0149] Specifically, in combination with the resource demand confidence interval, optimize the resource reconfiguration to meet the high-priority node requirements of the uRLLC slice while the reconfiguration delay is less than 1 millisecond, and feedback the resource reconfiguration result to the current network resource status;

[0150] It should be noted that through incremental GNN updates, the limitation of high reconfiguration delay of traditional static or semi-static resource allocation methods in dynamic scenarios is overcome.

[0151] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0152] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more flows and / or Figure 1 blocks.

[0153] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more flows and / or Figure 1 blocks.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in Figure 1 one or more flows and / or Figure 1 blocks.

[0155] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0156] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to include these changes and modifications.

Claims

1. An intelligent allocation method for low-latency slice resources for 5G networks, characterized in that, Including: Real-time collect the connection status and performance metrics of network nodes, dynamically adjust the collection frequency according to network load fluctuations, and generate structured topology data; According to the topology data, calculate the degree distribution entropy value of network nodes, generate a virtual resource topology graph, and predict the resource requirements of network nodes; Based on the virtual resource topology graph and the entropy value change rate, construct a dynamic confidence interval and design a resource pre-allocation strategy; Detect the burst traffic of the network, and quickly adjust the resource allocation by incrementally updating the graph neural network, and reduce the reconfiguration delay.

2. The intelligent allocation method for low-latency slice resources for 5G networks according to claim 1, wherein Real-time collect the connection status and performance metrics of network nodes, dynamically adjust the collection frequency according to network load fluctuations, and generate structured topology data, including: Collect the connection status data of network nodes through distributed network probes to form an adjacency matrix of the connection strength between nodes; Collect the performance metrics of each node, including bandwidth utilization, computing load, and latency, to form a node feature matrix; Calculate the degree of network load fluctuation according to the volatility of the bandwidth utilization rate, and dynamically adjust the sampling frequency of the distributed network probes; Denoise and normalize the adjacency matrix and the node feature matrix to generate structured topology data.

3. The intelligent allocation method for low-latency slice resources for 5G networks according to claim 2, wherein, According to the topology data, calculate the degree distribution entropy value of network nodes, including: According to the adjacency matrix, calculate the degree of each node, which is the sum of the connection strengths between each node and its neighbor nodes; Based on the degree of the nodes, construct a smooth node degree distribution function and calculate the entropy value of the structured topology data; Fuse the entropy value of the current structured topology data with its historical entropy value through time weighting to generate a smooth effective entropy value; Calculate the change rate of the effective entropy value, and store the effective entropy value and the change rate. When the effective entropy value exceeds the preset threshold of the effective entropy value, trigger the generation operation of the virtual resource topology graph.

4. The intelligent allocation method for low-latency slice resources for 5G networks according to claim 3, wherein When the effective entropy value exceeds the preset threshold of the effective entropy value, trigger the generation operation of the virtual resource topology graph, including: The preset threshold of the effective entropy value is obtained by calculating the mean and standard deviation of the effective entropy value and the network topology change coefficient.

5. The intelligent allocation method for low-latency slice resources for 5G networks according to claim 2 or 3, characterized in that, The generation of the virtual resource topology graph includes: Based on the structured topology data, construct a network topology graph, where nodes represent network nodes, edges represent connection relationships, and node features are performance metrics; Apply a graph neural network to process the network topology graph through a message passing mechanism to generate a virtual resource topology graph, and represent the virtual resource topology graph as a resource demand matrix, and the resource demand matrix contains the bandwidth and computing resource requirements of each node.

6. The intelligent allocation method of low-latency slice resources for 5G networks according to claim 5, wherein Construct a dynamic confidence interval and design a resource pre-allocation strategy, including: According to the resource demand matrix, calculate the predicted resource requirements of each node and its degree of fluctuation; Based on the predicted resource requirements and the degree of fluctuation, dynamically adjust the confidence level in combination with the change rate of the effective entropy value, and construct a resource demand confidence interval for each node; Design a resource pre-allocation strategy to preferentially meet the boundary requirements of the resource demand confidence interval, and at the same time constrain the total resource demand not to exceed the network available resources; Store the allocation results obtained after executing the resource pre-allocation strategy and update the current network resource status.

7. The intelligent allocation method of low-latency slice resources for 5G networks according to claim 6, wherein, Detect the burst traffic of the network, and quickly adjust the resource allocation and reduce the reconfiguration delay by incrementally updating the graph neural network, including: Calculate the traffic mutation index according to the temporal change of the bandwidth utilization rate, judge whether there is burst traffic, and identify the affected subgraphs when the traffic mutation index exceeds the preset threshold of the traffic mutation index; Incrementally update the graph neural network to recalculate the nodes of the affected subgraphs and generate a new resource demand matrix; Combine the resource demand confidence interval to optimize the resource reconfiguration, so that while meeting the high-priority node requirements of the uRLLC slice, the reconfiguration delay is less than 1 millisecond; Feed back the resource reconfiguration result to the current network resource status.

8. The intelligent allocation method for low-latency slice resources for 5G networks according to claim 7, wherein, When the traffic mutation index exceeds the preset threshold of the traffic mutation index, including: The preset threshold of the traffic mutation index is obtained by calculating the mean and twice the standard deviation of the traffic mutation index within the time window.

Images