Bandwidth control method and device, computer equipment and readable storage medium

Abstract

The invention relates to the technical field of communication, in particular to a bandwidth control method and device, computer equipment and a readable storage medium. Comprising the steps of obtaining a historical bandwidth utilization rate of a target link; performing multi-scale feature analysis on the historical bandwidth utilization rate to obtain a bandwidth utilization feature corresponding to each scale; constructing a multi-scale feature map according to the bandwidth utilization rate features corresponding to the scales; processing the multi-scale feature map by using a map convolution operation to obtain an update utilization rate feature corresponding to each scale; determining a predicted bandwidth utilization rate of the target link according to the updated utilization rate characteristics corresponding to the scales; and controlling the bandwidth of the target link according to the predicted bandwidth utilization rate.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a bandwidth control method, device, computer equipment and readable storage medium. Background Art

[0002] In the network planning process, predicting and monitoring bandwidth utilization is crucial to ensure efficient allocation of network resources and optimize network performance. Accurate bandwidth utilization prediction can help network operators avoid resource waste and reduce service interruptions caused by bandwidth overload or shortage, while also providing decision support for network expansion and upgrade.

[0003] However, traditional methods rely on single-scale data sources or modes, are easily affected by random noise, and have certain limitations, resulting in low accuracy of the final prediction results, which urgently needs to be improved. Summary of the invention

[0004] Based on this, it is necessary to provide a bandwidth control method, apparatus, computer equipment and readable storage medium that can improve the accuracy of bandwidth control in response to the above technical problems.

[0005] In a first aspect, the present application provides a bandwidth control method, comprising:

[0006] Obtain the historical bandwidth utilization of the target link;

[0007] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0008] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0009] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0010] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0011] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0012] In one embodiment, a multi-scale feature analysis is performed on the historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale, including:

[0013] The historical bandwidth utilization is processed using observation windows of different scales to obtain the utilization time series of each scale;

[0014] Feature extraction is performed on the utilization time series of each scale respectively to obtain bandwidth utilization features corresponding to each scale; wherein the bandwidth utilization features include utilization trend features and / or utilization seasonal features.

[0015] In one embodiment, a multi-scale feature map is constructed according to the bandwidth utilization characteristics corresponding to each scale, including:

[0016] The bandwidth utilization features of each scale are used as nodes in the initial feature graph;

[0017] Determine the correlation strength between the bandwidth utilization features corresponding to each scale;

[0018] The correlation strength between the bandwidth utilization features corresponding to each scale is used as the weight of the connecting edge between each node in the initial feature graph to obtain a multi-scale feature graph.

[0019] In one embodiment, a graph convolution operation is used to process a multi-scale feature map to obtain update utilization rate features corresponding to each scale, including:

[0020] For any node in the multi-scale feature map, determine the adjacent node corresponding to the node in the multi-scale feature map;

[0021] According to the weight of the connection edge between the node and the adjacent nodes, the bandwidth utilization feature of the node is fused with the bandwidth utilization feature of the adjacent nodes to obtain the fused utilization feature corresponding to the node;

[0022] A nonlinear activation function is used to activate the fused utilization rate features corresponding to the nodes to obtain updated utilization rate features of the corresponding scales of the nodes.

[0023] In one embodiment, determining the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale includes:

[0024] The update utilization rate features corresponding to each scale are input into the prediction model to obtain the scale prediction utilization rate corresponding to each scale output by the prediction model; wherein the loss function of the prediction model is determined according to the bulldozing Wasserstein distance between the prediction result of the prediction model and the training sample;

[0025] The predicted bandwidth utilization of the target link is determined according to the scale predicted utilization corresponding to each scale.

[0026] In one embodiment, determining the predicted bandwidth utilization of the target link according to the scale predicted utilization corresponding to each scale includes:

[0027] Use the self-attention mechanism to determine the attention score corresponding to each scale;

[0028] According to the attention scores corresponding to each scale and the scale predicted utilization, the predicted bandwidth utilization of the target link is determined.

[0029] In one embodiment, controlling the bandwidth of the target link according to the predicted bandwidth utilization includes:

[0030] Determine a utilization difference between a current bandwidth utilization and a predicted bandwidth utilization of the target link;

[0031] If the utilization difference is greater than the alarm threshold, an alarm process and / or capacity expansion / contraction process is performed on the bandwidth of the target link.

[0032] In a second aspect, the present application further provides a bandwidth control device, including:

[0033] An acquisition module is used to obtain the historical bandwidth utilization of the target link;

[0034] The multi-scale extraction module is used to perform multi-scale feature analysis on the historical bandwidth utilization to obtain the bandwidth utilization features corresponding to each scale;

[0035] A feature map construction module is used to construct a multi-scale feature map according to the bandwidth utilization characteristics corresponding to each scale;

[0036] A convolution processing module is used to process multi-scale feature maps using graph convolution operations to obtain update utilization rate features corresponding to each scale;

[0037] A prediction module, used to determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0038] The bandwidth control module is used to control the bandwidth of the target link according to the predicted bandwidth utilization.

[0039] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0040] Obtain the historical bandwidth utilization of the target link;

[0041] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0042] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0043] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0044] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0045] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0047] Obtain the historical bandwidth utilization of the target link;

[0048] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0049] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0050] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0051] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0052] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0053] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0054] Obtain the historical bandwidth utilization of the target link;

[0055] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0056] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0057] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0058] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0059] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0060] The above-mentioned bandwidth control method, device, computer equipment and readable storage medium perform multi-scale feature analysis on the historical bandwidth utilization in the time dimension to obtain bandwidth utilization features of each scale. The bandwidth utilization features of multiple different scales contain information of historical bandwidth utilization at different levels, thereby realizing a comprehensive analysis of historical bandwidth utilization; the bandwidth utilization features of each scale in the multi-scale feature graph are fused through a graph convolution operation to obtain updated utilization features corresponding to each scale containing the correlation between the bandwidth utilization features of each scale; the dependency and interaction between the bandwidth utilization features of each scale are analyzed through a graph convolution operation, with the aim of obtaining updated utilization features corresponding to each scale containing more useful information, thereby making the predicted bandwidth utilization obtained based on the updated utilization features corresponding to each scale more accurate, thereby improving the control accuracy of bandwidth control of the target link. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 A schematic diagram of a bandwidth control method according to an embodiment;

[0063] Figure 2 A schematic flow chart of steps for obtaining the historical utilization rate characteristics of bandwidth benefits corresponding to each scale in one embodiment;

[0064] Figure 3 A schematic diagram of a process for constructing a multi-scale feature map in one embodiment;

[0065] Figure 4 A schematic flow chart of the steps of obtaining update utilization rate characteristics corresponding to each scale in one embodiment;

[0066] Figure 5 A schematic diagram of a flow chart of a step of determining a predicted bandwidth utilization rate of a target link in one embodiment;

[0067] Figure 6 A schematic diagram of a flow chart of steps for controlling the bandwidth of a target link in one embodiment;

[0068] Figure 7 is a flow chart of a bandwidth control method in another embodiment;

[0069] Figure 8 is a structural block diagram of a bandwidth control device in one embodiment;

[0070] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] In an exemplary embodiment, Figure 1 As shown, a bandwidth control method is provided, and the method is described by taking the application of the method to a computer device as an example. The method includes:

[0073] S101, obtaining historical bandwidth utilization of a target link.

[0074] Specifically, traffic data of a target link in an Internet Protocol (IP) network is collected, and a historical bandwidth utilization rate of the target link is calculated based on the traffic data of the target link.

[0075] The traffic data of the target link refers to the statistical information of the data packets transmitted on a specific link (physical connection or logical path) in the IP network. The statistical information may include the number of data packets, the number of bytes, and the timestamp.

[0076] Optionally, the flow data of the target link can be collected through network monitoring tools and flow data mirroring technology. After the flow data is collected, the flow data is preprocessed, and the preprocessing operation can include anomaly detection, denoising and smoothing and other preprocessing operations.

[0077] Then, the pre-processed traffic data is analyzed. For each time period (such as every hour or every day), the total number of bytes passing through the target link in the time period is calculated, and then the average bandwidth actually used by the target link in each time period is obtained; for each time period, the historical bandwidth utilization corresponding to the time period = (actual average bandwidth used / total available bandwidth) * 100%. Accordingly, the bandwidth utilization time series formed by the historical bandwidth utilization corresponding to each time period in chronological order is the historical bandwidth utilization of the target link.

[0078] S102, performing multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale.

[0079] Traditional methods are often limited to a single time scale when analyzing network traffic, and fail to comprehensively consider multi-dimensional time characteristics such as short-term fluctuations, medium-term trends, and long-term seasonality. This single-perspective analysis ignores the complexity and diversity of traffic data, limits the comprehensive capture and in-depth understanding of bandwidth utilization changes, and thus affects the accuracy and comprehensiveness of bandwidth utilization predictions.

[0080] In order to solve the above problem, in this embodiment, in one implementable manner, observation windows of different time scales can be set, such as 5 minutes, 30 minutes, 1 hour, 1 day, etc., and observation windows of different time scales are used to perform multi-scale decomposition and representation on the above bandwidth utilization time series, and the bandwidth utilization time series is converted into multiple sub-sequences of different time granularities through sliding window segmentation and aggregation operations. In another implementable manner, the bandwidth utilization time series can also be subjected to multi-level pooling processing to convert the bandwidth utilization time series into multiple sub-sequences of different time granularities.

[0081] Optionally, the bandwidth utilization feature corresponding to any scale may include the local trend estimation value and utilization seasonal feature corresponding to the scale. In addition, the correlation modeling based on the self-attention mechanism may be further adopted, and the self-attention mechanism is introduced into the local trend estimation value and utilization seasonal feature corresponding to each scale to learn the intrinsic correlation pattern of the traffic data, and obtain the key feature, and the key feature is used as the bandwidth utilization feature.

[0082] S103: construct a multi-scale feature map according to the bandwidth utilization characteristics corresponding to each scale.

[0083] It is understandable that in deep learning, especially in convolutional neural networks (CNN), feature maps are features extracted from input images through operations such as convolutional layers and pooling layers, while multi-scale feature maps mean that feature representations of multiple scales are extracted from the same input image, which can provide different levels of information about the image content.

[0084] In this embodiment, a convolutional neural network may be used to analyze the bandwidth utilization characteristics corresponding to each scale to obtain a multi-scale feature map in this embodiment.

[0085] S104, using a graph convolution operation to process the multi-scale feature map to obtain update utilization rate features corresponding to each scale.

[0086] It can be understood that the specific process of graph convolution operation can include:

[0087] 1) Transmission: Each node transforms its own feature information (such as through linear transformation) and sends it to its neighboring nodes. This step is to extract and transform the node's feature information for subsequent information aggregation.

[0088] 2) Receiving: Each node receives feature information from its neighbor nodes. This information may have undergone some form of aggregation operation (such as summation, averaging, or max pooling) to combine information from multiple neighbors into a unified representation.

[0089] 3) Transformation: The received neighbor node information and the node’s own information (if any) are combined and further nonlinearly transformed (such as through activation functions). The purpose of this step is to increase the expressive power of the model so that the model can learn more complex patterns.

[0090] For the above-mentioned multi-scale feature graphs, the graph convolution operation can directly utilize the node connection information in the graph structure to capture the dependencies and interactions between nodes, automatically extract useful feature representations from the original data, and realize in-depth feature extraction.

[0091] S105 , determining the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale.

[0092] It can be understood that the updated utilization feature obtained based on the above graph convolution operation not only contains the local information of the node, but also contains the global structural information through the aggregation of neighboring nodes, and the feature dimension is more comprehensive. Therefore, the predicted bandwidth utilization of the target link obtained based on the above more comprehensive updated utilization feature analysis is more accurate.

[0093] Specifically, a prediction model may be used to analyze update utilization characteristics corresponding to each scale, and the output of the prediction model is the predicted bandwidth utilization of the target link.

[0094] S106: Control the bandwidth of the target link according to the predicted bandwidth utilization.

[0095] Optionally, set reasonable bandwidth utilization thresholds based on network needs and performance requirements. These thresholds can be used as the basis for triggering bandwidth control operations. Based on the prediction results and the set thresholds, formulate specific bandwidth control policies. For example, when it is predicted that bandwidth utilization will exceed the threshold, measures such as increasing bandwidth resources, optimizing traffic distribution, and limiting non-critical application bandwidth can be taken.

[0096] In the above bandwidth control method, a multi-scale feature analysis is performed on the historical bandwidth utilization in the time dimension to obtain bandwidth utilization features of each scale. The bandwidth utilization features of multiple different scales contain information of historical bandwidth utilization at different levels, thereby realizing a comprehensive analysis of historical bandwidth utilization. The bandwidth utilization features of each scale in the multi-scale feature graph are fused through a graph convolution operation to obtain updated utilization features corresponding to each scale containing the correlation between the bandwidth utilization features of each scale. The dependency and interaction between the bandwidth utilization features of each scale are analyzed through a graph convolution operation, with the aim of obtaining updated utilization features corresponding to each scale containing more useful information, thereby making the predicted bandwidth utilization obtained based on the updated utilization features corresponding to each scale more accurate, thereby improving the control accuracy of the bandwidth control of the target link.

[0097] In an exemplary embodiment, Figure 2 As shown in the figure, a multi-scale feature analysis is performed on the historical bandwidth utilization to obtain the historical bandwidth utilization features corresponding to each scale, including:

[0098] S201, using observation windows of different scales to process historical bandwidth utilization to obtain utilization time series of each scale.

[0099] Optionally, a series of different observation windows (or time windows) can be set, which can be of fixed or variable length, depending on the needs of the analysis and the characteristics of the data.

[0100] Among them, the fixed window width segmentation method is to set the same subsequence length to ensure that each subsequence contains the same sample data; the variable window width time series segmentation uses a series of different time windows to make each subsequence statistically uniform.

[0101] For each scale, the historical bandwidth utilization is divided according to the observation window corresponding to the scale. The data in each window constitutes a subsequence, and the subsequences corresponding to the scale form the utilization time series of the scale. By changing the size of the window, subsequences of different scales can be obtained. These subsequences reflect the performance of the original time series at different time granularities.

[0102] S202, extracting features from the utilization time series of each scale respectively to obtain bandwidth utilization features corresponding to each scale.

[0103] The bandwidth utilization characteristics include utilization trend characteristics and / or utilization season characteristics.

[0104] It is understandable that the utilization trend feature mainly describes the change trend of bandwidth utilization in a longer period of time (such as month, quarter, year, etc.). The utilization seasonal feature focuses on describing the regular changes in bandwidth utilization in a specific season or time period (such as summer and winter vacations, holidays, etc.).

[0105] Specifically, on the utilization time series corresponding to each scale, the STL (Seasonal-Trend Decomposition using LOESS) algorithm can be used to extract the seasonal, trend and random components of the corresponding scale to characterize the multi-scale periodic law and change trend of the traffic. The specific process is: for any utilization time series among multiple utilization time series, the utilization time series is subjected to local weighted regression processing to obtain the corresponding local trend estimation values. Based on the local trend estimation values, the utilization trend characteristics corresponding to the utilization time series are generated. The local trend estimation values ​​are deleted from the utilization time series to obtain the local seasonal estimation values. Based on the local seasonal estimation values, the utilization seasonal characteristics corresponding to the utilization time series are generated.

[0106] If the utilization trend feature and the utilization seasonal feature are further analyzed to obtain the key feature, the specific process is as follows: for any utilization time series in the multi-scale utilization sequence, the utilization trend feature and the utilization seasonal feature corresponding to the utilization time series are fused to obtain the utilization fusion utilization feature corresponding to the utilization time series; the utilization fusion utilization feature is linearly transformed to obtain each attention matrix; according to the attention weight of each attention matrix, each attention matrix is ​​weighted summed to obtain the first key feature corresponding to the utilization time series; the first key feature corresponding to the utilization time series is dimensionally upgraded to form a multi-dimensional feature space corresponding to the utilization time series; a feature similarity matrix of the multi-dimensional feature space is constructed; according to the feature similarity matrix, the feature similarity between different dimensional features in the multi-dimensional feature space is determined; according to the feature similarity matrix, the second key feature is extracted from the first key feature. In this case, the bandwidth utilization feature corresponding to each scale may include the above-mentioned first key feature and the second key feature.

[0107] In this embodiment, the trend characteristics and seasonal characteristics of bandwidth utilization jointly reflect the dynamic changes in network bandwidth usage. Understanding these characteristics helps network managers to better plan and manage network resources and ensure efficient and stable operation of the network. At the same time, network managers can also take corresponding optimization measures based on the utilization trend and seasonal fluctuations, such as adjusting bandwidth allocation strategies, upgrading network equipment, optimizing network architecture, etc., to further improve bandwidth utilization and network performance.

[0108] In an exemplary embodiment, Figure 3 As shown in the figure, a multi-scale feature map is constructed according to the bandwidth utilization characteristics corresponding to each scale, including:

[0109] S301, taking bandwidth utilization features of each scale as nodes in an initial feature graph.

[0110] The initial feature map refers to a feature map whose content is a null value.

[0111] Optionally, the multi-scale feature graph in this embodiment includes each node and the connecting edges between each node. Each node represents the bandwidth utilization characteristics of each scale; the connecting edges are the lines between the nodes in the multi-scale feature graph, which are used to represent the relationship or interaction between the nodes, which can be spatial proximity, semantic similarity, functional relevance, etc.

[0112] S302: Determine the correlation strength between the bandwidth utilization characteristics corresponding to each scale.

[0113] The correlation strength between the bandwidth utilization features corresponding to any two scales may refer to the cosine similarity between the bandwidth utilization features corresponding to the any two scales;

[0114] The bandwidth utilization characteristic corresponding to one of the scales is , the bandwidth utilization characteristic corresponding to another scale is Take this as an example to illustrate;

[0115] and The strength of the association between The following formula (1) is used to calculate:

[0116] (1)

[0117] S303, using the correlation strength between the bandwidth utilization features corresponding to each scale as the weight of the connecting edge between each node in the initial feature graph, to obtain a multi-scale feature graph.

[0118] For example, the bandwidth utilization characteristics corresponding to each scale time series (n scale time series) are divided into , construct a multi-scale feature map, which is a graph model , where the node Corresponding to feature sets of different scales, is the edge set of connected edges, A set of attributes that represent the strength of the association.

[0119] Specifically, Figure 4As shown in the figure, the multi-scale feature map is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale, including:

[0120] S401: For any node in the multi-scale feature graph, determine an adjacent node corresponding to the node in the multi-scale feature graph.

[0121] Among them, neighboring nodes refer to nodes that are directly connected to a given node through connecting edges. These nodes share at least one edge with the given node in the feature graph, thus forming a direct interaction or relationship. In order to determine the neighboring nodes of a given node, it is necessary to analyze all the connecting edges connected to the node in the feature graph. Each connecting edge points to a neighboring node, so by traversing these edges, all neighboring nodes can be identified.

[0122] S402, based on the weight of the connection edge between the node and the adjacent nodes, feature fusion is performed on the bandwidth utilization feature of the node and the bandwidth utilization feature of the adjacent nodes to obtain a fused utilization feature corresponding to the node.

[0123] Specifically, according to the weight of the connection edge between the node and the adjacent nodes, the bandwidth utilization feature of the node and the bandwidth utilization feature of the adjacent nodes are fused to obtain the fused utilization feature corresponding to the node.

[0124] Specifically, the bandwidth utilization feature of a node and the bandwidth utilization feature of an adjacent node are fused using the following formula (2):

[0125] (2)

[0126] in, It is The node in The feature representation of the layer,

[0127] is the bias term;

[0128] yes and The strength of the association between

[0129] .

[0130] S403, using a nonlinear activation function, activates the fused utilization feature corresponding to the node to obtain an updated utilization feature of the scale corresponding to the node.

[0131] For example, for each scale Node Apply graph convolution, according to the node, obtain the fused utilization feature corresponding to the node and use nonlinear activation function to activate the fused utilization feature corresponding to the node, and obtain the updated utilization feature of the node corresponding scale, which is calculated by the following formula (3):

[0132] (3)

[0133] in, That is The update utilization characteristics of the corresponding scale of each node, is a non-linear activation function such as ReLU.

[0134] In this embodiment, the stacking of multiple layers of graph convolution can effectively capture the high-order spatial correlation of traffic and learn a more global and robust multi-scale fusion utilization feature representation.

[0135] In an exemplary embodiment, Figure 5 As shown, according to the updated utilization characteristics corresponding to each scale, the predicted bandwidth utilization of the target link is determined, including:

[0136] S501, inputting the update utilization rate features corresponding to each scale into the prediction model to obtain the scale prediction utilization rate corresponding to each scale output by the prediction model.

[0137] The prediction model refers to a model that analyzes and predicts the bandwidth utilization of the target link in the future period based on the bandwidth utilization characteristics of the target link in the historical period. The prediction model can be a model based on a time series analysis algorithm, a machine learning algorithm, or a decision tree and integration method.

[0138] The prediction models in traditional methods often show poor generalization ability when dealing with noisy data sets or facing abnormal traffic events. The prediction models are sensitive to outliers and outliers in the data and lack effective robustness control mechanisms. This leads to the prediction accuracy of the prediction models being easily disturbed in practical applications, and the stability and reliability are insufficient, which limits the applicability and effectiveness of the prediction models in diverse network environments.

[0139] In this embodiment, in order to solve the above technical problems, the loss function of the prediction model is determined according to the bulldozing Wasserstein distance between the prediction results of the prediction model and the training samples.

[0140] Specifically, the process of training the prediction model using the loss function includes the following steps:

[0141] 1) Determine the prediction results: The prediction model uses model parameters Predict bandwidth utilization and determine predicted distribution As a prediction result;

[0142] 2) Determine the training samples: Determine the true distribution corresponding to the training samples ;

[0143] 3) Wasserstein distance calculation: Use the discrete Wasserstein distance estimation method, as shown in the following formula (4):

[0144] (4)

[0145] in, is from arrive transfer scheme, It's the cost.

[0146] 4) Regularized loss function: Add Wasserstein distance as a regularization term to the loss function, as shown in formula (4):

[0147] (5)

[0148] It can be understood that in order to improve the generalization and robustness of the model, this embodiment introduces Wasserstein distance as a regularization term to constrain the parameter distribution of the prediction model. Wasserstein distance, also known as bulldozer distance or EarthMover's Distance, measures the minimum cost required to transform one distribution into another. By minimizing the Wasserstein distance between the predicted distribution of the prediction model and the actual traffic distribution, the prediction model parameters are forced to learn smoother and more regular feature representations, reducing the risk of overfitting and improving the adaptability and generalization of the prediction model. At the same time, the Wasserstein distance is more robust to outliers and outlier samples, giving the prediction model tolerance to noisy data.

[0149] S502: Determine a predicted bandwidth utilization of a target link according to the scale predicted utilization corresponding to each scale.

[0150] Optionally, a fusion strategy may be adopted to fuse the scale predicted utilizations corresponding to each scale to obtain the predicted bandwidth utilization of the target link.

[0151] Specifically, the predicted bandwidth utilization of the target link is determined according to the scale predicted utilization corresponding to each scale, including: using a self-attention mechanism to determine the attention score corresponding to each scale; and determining the predicted bandwidth utilization of the target link according to the attention score corresponding to each scale and the scale predicted utilization.

[0152] First, for each scale Update utilization features obtained using graph convolution operations (i.e. the above ), input into the prediction model, and obtain the scale prediction utilization rate corresponding to the scale .

[0153] Then, the self-attention mechanism is used to calculate the attention score of each scale , specifically using the following formula (6): (6)

[0154] Furthermore, the softmax function is used to convert the attention score into attention weight , specifically using the following formula (7):

[0155] (7)

[0156] Finally, the prediction results of different scales are weighted and summed according to the attention weights to get the final prediction:

[0157] In this embodiment, an adaptive weighted fusion strategy is used to integrate the predicted utilization rates of different scales. By introducing the attention mechanism for automatic learning, a higher weight is given to the scale with a good prediction effect, and a relatively lower weight is given to the scale with a poor prediction effect. Finally, the weighted predicted utilization rates of each scale are added together to obtain the overall trend prediction of the network bandwidth utilization in the future period, that is, the predicted bandwidth utilization rate of the target link.

[0158] In an exemplary embodiment, Figure 6 As shown, according to the predicted bandwidth utilization, the bandwidth of the target link is controlled, including:

[0159] S601: Determine a utilization difference between a current bandwidth utilization and a predicted bandwidth utilization of a target link.

[0160] Specifically, an absolute value of a difference between a current bandwidth utilization rate and a predicted bandwidth utilization rate of a target link is determined as a utilization difference between the current bandwidth utilization rate and the predicted bandwidth utilization rate of the target link.

[0161] S602: If the utilization difference is greater than the alarm threshold, an alarm process and / or a capacity expansion / contraction process is performed on the bandwidth of the target link.

[0162] Specifically, this embodiment sets an alarm threshold T. When the difference between the current bandwidth utilization and the predicted bandwidth utilization of the target link exceeds the threshold, an alarm is triggered.

[0163] Subsequently, based on real-time monitoring data and historical trend analysis, a capacity expansion and contraction plan is automatically generated, which will dynamically adjust resource allocation to meet current load requirements. In addition, the network status will be automatically changed to complete the expansion and contraction of link bandwidth, ensuring smooth and efficient network traffic, thereby maximizing cost-effectiveness and optimizing service quality.

[0164] In an exemplary embodiment, Figure 7 As shown, the bandwidth control method in this embodiment includes:

[0165] S701, obtaining the historical bandwidth utilization of the target link.

[0166] S702, performing multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale.

[0167] The bandwidth utilization characteristics include utilization trend characteristics and / or utilization season characteristics.

[0168] S703: Use the bandwidth utilization characteristics of each scale as nodes in the initial feature graph.

[0169] S704: Determine the correlation strength between the bandwidth utilization characteristics corresponding to each scale.

[0170] S705 , using the correlation strength between the bandwidth utilization features corresponding to each scale as the weight of the connecting edge between each node in the initial feature graph, to obtain a multi-scale feature graph.

[0171] S706, using a graph convolution operation to process the multi-scale feature graph to obtain update utilization rate features corresponding to each scale.

[0172] S707, inputting the update utilization rate features corresponding to each scale into the prediction model to obtain the scale prediction utilization rate corresponding to each scale output by the prediction model.

[0173] Among them, the loss function of the prediction model is determined according to the bulldozer Wasserstein distance between the prediction results of the prediction model and the training samples.

[0174] S708, using the self-attention mechanism to determine the attention score corresponding to each scale.

[0175] S709: Determine the predicted bandwidth utilization of the target link according to the attention score corresponding to each scale and the predicted utilization of the scale.

[0176] S710: Control the bandwidth of the target link according to the predicted bandwidth utilization.

[0177] In this embodiment, in terms of feature mining, a multi-scale feature map is constructed according to the bandwidth utilization characteristics corresponding to each scale; the multi-scale feature map is processed by using a graph convolution operation to obtain updated utilization characteristics corresponding to each scale. This deep feature representation and association modeling method not only improves the ability to capture the complexity of bandwidth utilization data, but also more accurately models the inherent laws and patterns of bandwidth utilization data by learning long-term dependencies in the data; it enhances the model's understanding of the deep-level characteristics of bandwidth utilization data, and by capturing the periodicity and correlation in the data, provides a more accurate modeling method for predicting bandwidth utilization.

[0178] In terms of the prediction model, introducing Wasserstein distance as a regularization term into the prediction model not only provides a new generalization constraint for the model, but also improves the prediction model's adaptability to changes in the actual network environment by minimizing the difference between the predicted distribution and the true distribution, thereby enhancing the robustness of the prediction model.

[0179] In determining the predicted bandwidth utilization of the target link, the adaptive weighted fusion strategy is adopted to achieve effective integration of traffic features at different time scales by dynamically adjusting the weights of predicted utilization at different scales. This novel fusion strategy not only improves the prediction accuracy, but also brings innovative weight adjustment and feature integration methods to the field of network traffic prediction by automatically learning and giving higher weights to scales with good prediction effects.

[0180] The practicality of the bandwidth control method provided in this embodiment is first reflected in the ability to provide strong support for the optimal configuration of network resources. Through deep feature representation and association modeling, the model can accurately predict bandwidth demand in different time periods, helping operators and enterprises to reasonably allocate network resources and avoid resource waste. Improve network service quality: By utilizing the enhanced generalization and robustness of this embodiment, network service providers can maintain the stability and reliability of services in the face of network traffic fluctuations. This helps to improve user satisfaction and reduce potential losses caused by service interruptions or delays. Network anomaly detection and prevention: The adaptive feature integration and dynamic weight adjustment strategy of this embodiment can quickly respond to abnormal changes in network traffic, adjust resource allocation in a timely manner, and achieve early detection and prevention of network anomalies, thereby ensuring network security and data integrity.

[0181] In a cloud computing environment, this embodiment can be applied to intelligent cloud service management to achieve dynamic allocation and optimization of cloud resources to adapt to changing user needs and traffic patterns.

[0182] In the Internet of Things (IoT) traffic control environment, as the number of IoT devices increases, this embodiment can be applied to the IoT environment to effectively predict and manage the network traffic of a large number of devices to ensure the efficiency and stability of data transmission.

[0183] In the future network technology environment of the fifth generation mobile communication technology (5G), the method of this embodiment is applicable to 5G and future network technologies, can cope with the network traffic requirements of higher speed and larger capacity, and provide a powerful traffic prediction and management tool for the new generation of communication networks.

[0184] Through the practicality of this embodiment, operators and enterprises can reduce unnecessary network resource investment, reduce operating costs, and improve resource utilization efficiency, thereby bringing direct economic benefits. The accurate traffic prediction and management capabilities provided by this embodiment can help service providers gain a competitive advantage in the market and attract and retain users by providing higher quality services. Innovation in the field of network traffic prediction not only has practical application value, but also provides new ideas and methods for the development of related technical fields, promoting technological progress and innovation in the entire industry.

[0185] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0186] Based on the same inventive concept, the embodiment of the present application also provides a bandwidth control device for implementing the bandwidth control method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more bandwidth control device embodiments provided below can refer to the limitations on the bandwidth control method above, and will not be repeated here.

[0187] In an exemplary embodiment, Figure 8As shown, a bandwidth control device is provided, comprising: an acquisition module 11, a multi-scale extraction module 12, a feature map construction module 13, a convolution processing module 14, a prediction module 15 and a bandwidth control module 16, wherein:

[0188] An acquisition module 11 is used to acquire the historical bandwidth utilization of a target link;

[0189] A multi-scale extraction module 12 is used to perform multi-scale feature analysis on the historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0190] A feature map construction module 13 is used to construct a multi-scale feature map according to the bandwidth utilization characteristics corresponding to each scale;

[0191] A convolution processing module 14 is used to process the multi-scale feature graph using a graph convolution operation to obtain update utilization rate features corresponding to each scale;

[0192] A prediction module 15, configured to determine a predicted bandwidth utilization of a target link according to update utilization characteristics corresponding to each scale;

[0193] The bandwidth control module 16 is used to control the bandwidth of the target link according to the predicted bandwidth utilization.

[0194] In one embodiment, the multi-scale extraction module 12 is used to: process the historical bandwidth utilization using observation windows of different scales to obtain utilization time series of each scale;

[0195] Feature extraction is performed on the utilization time series of each scale respectively to obtain bandwidth utilization features corresponding to each scale; wherein the bandwidth utilization features include utilization trend features and / or utilization seasonal features.

[0196] In one embodiment, the feature graph construction module 13 is used to: use the bandwidth utilization features of each scale as nodes in the initial feature graph;

[0197] Determine the correlation strength between the bandwidth utilization features corresponding to each scale;

[0198] The correlation strength between the bandwidth utilization features corresponding to each scale is used as the weight of the connecting edge between each node in the initial feature graph to obtain a multi-scale feature graph.

[0199] In one embodiment, the convolution processing module 14 is used to: for any node in the multi-scale feature map, determine an adjacent node corresponding to the node in the multi-scale feature map;

[0200] According to the weight of the connection edge between the node and the adjacent nodes, the bandwidth utilization feature of the node is fused with the bandwidth utilization feature of the adjacent nodes to obtain the fused utilization feature corresponding to the node;

[0201] A nonlinear activation function is used to activate the fused utilization rate features corresponding to the nodes to obtain updated utilization rate features of the corresponding scales of the nodes.

[0202] In one embodiment, the prediction module 15 is used to: input the update utilization rate characteristics corresponding to each scale into the prediction model to obtain the scale prediction utilization rate corresponding to each scale output by the prediction model; wherein the loss function of the prediction model is determined according to the bulldozing Wasserstein distance between the prediction result of the prediction model and the training sample;

[0203] The predicted bandwidth utilization of the target link is determined according to the scale predicted utilization corresponding to each scale.

[0204] In one embodiment, the bandwidth control module 16 is used to: use a self-attention mechanism to determine the attention score corresponding to each scale;

[0205] According to the attention scores corresponding to each scale and the scale predicted utilization, the predicted bandwidth utilization of the target link is determined.

[0206] In one embodiment, the bandwidth control module 16 is used to: determine a utilization difference between a current bandwidth utilization and a predicted bandwidth utilization of a target link;

[0207] If the utilization difference is greater than the alarm threshold, an alarm process and / or capacity expansion / contraction process is performed on the bandwidth of the target link.

[0208] Each module in the above bandwidth control device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0209] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig. 9As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data of the bandwidth control method. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a bandwidth control method is implemented.

[0210] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0211] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0212] Obtain the historical bandwidth utilization of the target link;

[0213] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0214] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0215] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0216] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0217] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0218] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0219] Obtain the historical bandwidth utilization of the target link;

[0220] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0221] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0222] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0223] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0224] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0225] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0226] Obtain the historical bandwidth utilization of the target link;

[0227] Perform multi-scale feature analysis on historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale;

[0228] Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale;

[0229] The multi-scale feature graph is processed by graph convolution operation to obtain the update utilization rate features corresponding to each scale;

[0230] Determine the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale;

[0231] The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

[0232] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0233] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0234] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A bandwidth control method, characterized in that: The method comprises: Obtain the historical bandwidth utilization of the target link; Performing multi-scale feature analysis on the historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale; Construct a multi-scale feature map based on the bandwidth utilization characteristics corresponding to each scale; Using a graph convolution operation to process the multi-scale feature map, and obtaining an update utilization rate feature corresponding to each scale; Determining the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale; The bandwidth of the target link is controlled according to the predicted bandwidth utilization.

2. The method according to claim 1, characterized in that The performing multi-scale feature analysis on the historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale includes: The historical bandwidth utilization is processed using observation windows of different scales to obtain utilization time series of each scale; Feature extraction is performed on the utilization time series of each scale respectively to obtain bandwidth utilization features corresponding to each scale; wherein the bandwidth utilization features include utilization trend features and / or utilization seasonal features.

3. The method according to claim 1, characterized in that The step of constructing a multi-scale feature map according to the bandwidth utilization characteristics corresponding to each scale includes: The bandwidth utilization features of each scale are used as nodes in the initial feature graph; Determine the correlation strength between the bandwidth utilization features corresponding to each scale; The correlation strength between the bandwidth utilization features corresponding to each scale is used as the weight of the connecting edge between each node in the initial feature graph to obtain a multi-scale feature graph.

4. The method according to claim 3, characterized in that The multi-scale feature map is processed by using a graph convolution operation to obtain update utilization rate features corresponding to each scale, including: For any node in the multi-scale feature map, determining an adjacent node corresponding to the node in the multi-scale feature map; According to the weight of the connection edge between the node and the adjacent node, the bandwidth utilization feature of the node and the bandwidth utilization feature of the adjacent node are subjected to feature fusion to obtain the fused utilization feature corresponding to the node; A nonlinear activation function is used to activate the fused utilization rate feature corresponding to the node to obtain an updated utilization rate feature of the scale corresponding to the node.

5. The method according to claim 1, characterized in that The step of determining the predicted bandwidth utilization of the target link according to the updated utilization characteristics corresponding to each scale includes: Inputting the update utilization rate features corresponding to each scale into the prediction model to obtain the scale prediction utilization rate corresponding to each scale output by the prediction model; wherein the loss function of the prediction model is determined according to the bulldozing Wasserstein distance between the prediction result of the prediction model and the training sample; The predicted bandwidth utilization of the target link is determined according to the scale predicted utilization corresponding to each scale.

6. The method according to claim 1, characterized in that The step of determining the predicted bandwidth utilization of the target link according to the scale predicted utilization corresponding to each scale includes: Use the self-attention mechanism to determine the attention score corresponding to each scale; The predicted bandwidth utilization of the target link is determined according to the attention scores corresponding to each scale and the predicted utilization of the scale.

7. The method according to claim 1, characterized in that The controlling the bandwidth of the target link according to the predicted bandwidth utilization includes: Determine a utilization difference between a current bandwidth utilization of the target link and the predicted bandwidth utilization; If the utilization difference is greater than an alarm threshold, an alarm process and / or a capacity expansion / contraction process is performed on the bandwidth of the target link.

8. A bandwidth control device, characterized in that: The device comprises: Acquisition module, for the historical bandwidth utilization of the target link; A multi-scale extraction module, used to perform multi-scale feature analysis on the historical bandwidth utilization to obtain bandwidth utilization features corresponding to each scale; A feature map construction module is used to construct a multi-scale feature map according to the bandwidth utilization characteristics corresponding to each scale; A convolution processing module, used to process the multi-scale feature map by using a graph convolution operation to obtain an update utilization rate feature corresponding to each scale; A prediction module, used to determine the predicted bandwidth utilization of the target link according to the update utilization characteristics corresponding to each scale; The bandwidth control module is used to control the bandwidth of the target link according to the predicted bandwidth utilization.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images