Network link state prediction method, model training method, device and apparatus

Abstract

The application relates to the technical field of software-defined networks, and provides a network link state prediction method, a training method and device of a model, and equipment. A target link adjacency matrix and a corresponding target link state matrix of a target network are acquired; the target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain a link state matrix corresponding to the target link adjacency matrix at the next moment; the link state prediction model is obtained by training a network prediction model using training sample data; the training sample data comprises a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix. The method can effectively improve network state monitoring efficiency.

Description

Technical Field

[0001] This application relates to the field of software-defined networking technology, and in particular to a network link state prediction method, a model training method, an apparatus, and a device. Background Technology

[0002] When network congestion occurs in an IP network, the overall network performance will degrade accordingly, severely impacting the operational efficiency of user services. Existing network status measurement technologies such as TWMAP and INT can help monitor network link performance, but network congestion events are sudden and real-time. Using these measurement methods to obtain network status and then make adjustments and responses results in only a delayed and passive response.

[0003] In related technologies, network status measurement techniques are typically used to monitor network link performance, which can easily lead to low efficiency in network status monitoring. Therefore, providing a method to improve the efficiency of network status monitoring is of significant practical importance. Summary of the Invention

[0004] This application provides a network link state prediction method, model training method, apparatus, and device that can predict the network link state at the next future moment. This can help network operators predict and respond to congestion events in advance, thereby effectively improving the efficiency of network state monitoring.

[0005] To achieve the above objectives, the technical solution of this application embodiment is implemented as follows:

[0006] In a first aspect, embodiments of this application provide a network link state prediction method, the method comprising:

[0007] Obtain the target link adjacency matrix and the corresponding target link state matrix of the target network; the target link adjacency matrix is ​​obtained based on the network topology diagram of the target network; the target link state matrix is ​​generated based on the link state data of the target network.

[0008] The target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training a network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition.

[0009] Secondly, embodiments of this application also provide a method for training a link state prediction model, used to generate the link state prediction model used in the first aspect, the training method comprising:

[0010] Obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix;

[0011] Based on the training sample data, batch training information is selected; the batch training information includes a portion of the link adjacency matrix and a corresponding portion of the historical link state matrix;

[0012] The batch training information is input into the network prediction model to be trained for training, and the recognition loss value of the network prediction model to be trained is determined.

[0013] Determine whether the identification loss value has converged to the preset target value. If the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain. If the determination result is yes, end the training and obtain the trained link state prediction model.

[0014] Thirdly, embodiments of this application also provide a network link state prediction device, the device comprising:

[0015] The link information acquisition unit is used to acquire the target link adjacency matrix and the corresponding target link state matrix of the target network; the target link adjacency matrix is ​​obtained based on the network topology diagram of the target network; the target link state matrix is ​​generated based on the link state data of the target network.

[0016] A network state prediction unit is used to input the target link adjacency matrix and the target link state matrix into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition.

[0017] In an optional embodiment, the system further includes a model training unit, which is used to train the trained link state prediction model in the following manner:

[0018] Obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix;

[0019] Based on the training sample data, batch training information is selected; the batch training information includes a portion of the link adjacency matrix and a corresponding portion of the historical link state matrix;

[0020] The batch training information is input into the network prediction model to be trained for training, and the recognition loss value of the network prediction model to be trained is determined.

[0021] Determine whether the identification loss value converges to the preset target value. If it does, end the training to obtain the trained link state prediction model. Otherwise, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain it.

[0022] In an optional embodiment, the network state prediction unit is further configured to: generate a link utilization state corresponding to the next moment of the target link adjacency matrix based on the link state matrix corresponding to the next moment of the target link adjacency matrix; the link utilization state characterizes the warning level of the link utilization of each link in the target link adjacency matrix.

[0023] Fourthly, embodiments of this application also provide a training apparatus for a link state prediction model, used to generate the link state prediction model used in the third aspect, the training apparatus comprising:

[0024] The training data acquisition unit is used to obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix.

[0025] The model training execution unit is used to select batch training information based on the training sample data; the batch training information includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix; input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained; determine whether the recognition loss value converges to a preset target value; if the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined recognition loss value and train again; if the determination result is yes, end the training and obtain the trained link state prediction model.

[0026] In an alternative embodiment, the network prediction model is obtained by changing the linear transformation layer in the recurrent neural network to two graph convolutional neural network layers.

[0027] In one optional embodiment, the model training execution unit is specifically used for:

[0028] The parameters of the network prediction model to be trained are adjusted according to the determined recognition loss value until the recognition loss value converges to the preset target value. The resulting network prediction model is then used as the trained link state prediction model.

[0029] Fifthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method described in either the first or second aspect.

[0030] In a sixth aspect, embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and when the computer program is executed by the processor, the processor causes the processor to implement the method described in either the first aspect or the second aspect.

[0031] The network link state prediction method in the above embodiments of this application obtains the target link adjacency matrix and the corresponding target link state matrix of the target network. The target link adjacency matrix is ​​obtained based on the network topology of the target network. The target link state matrix is ​​generated based on the link state data of the target network. The target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition. Therefore, this embodiment can obtain the spatiotemporal characteristics of network link state based on the target link adjacency matrix and the target link state matrix, and can accurately predict the network link state at the next moment. This can help network operators predict and respond to congestion events in advance, and can effectively improve the efficiency of network state monitoring. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A flowchart illustrating a network link state prediction method provided in an embodiment of this application;

[0034] Figure 2 A schematic diagram illustrating the process of obtaining a link state prediction model using a network link state prediction method provided in this application embodiment;

[0035] Figure 3 A schematic diagram illustrating the process of generating link utilization status using a network link state prediction method provided in this application embodiment;

[0036] Figure 4 A flowchart illustrating a training method for a link state prediction model provided in an embodiment of this application;

[0037] Figure 5 A flowchart illustrating another training method for a link state prediction model provided in an embodiment of this application;

[0038] Figure 6 This is one of the structural schematic diagrams of a network link state prediction device provided in an embodiment of this application;

[0039] Figure 7 This is a second schematic diagram of a network link state prediction device provided in an embodiment of this application;

[0040] Figure 8 A schematic diagram of the structure of a training device for a link state prediction model provided in an embodiment of this application;

[0041] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] It should be noted that the terms "comprising" and "having" and their variations used in this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0044] The following are explanations of some of the words that appear in the text:

[0045] (1) Recurrent Neural Network (RNN): A recurrent neural network is a type of recursive neural network that takes sequence data as input, recurses along the direction of sequence evolution, and connects all nodes (recurrent units) in a chain-like manner. RNNs are mainly used to process sequence data (such as natural language, audio, stock market price trends, etc.), and they have certain advantages in learning the non-linear features of sequences. The biggest difference between RNNs and traditional NN models is that the input of the hidden layer includes not only the output of the input layer, but also the output of the hidden layer in the previous time step.

[0046] (2) Graph Neural Network (GNN): Graph Neural Networks refer to a general term for algorithms that use neural networks to learn graph-structured data, extract and discover features and patterns in graph-structured data, and meet the needs of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. A Graph Neural Network (GNN) is a framework that uses deep learning to directly learn from graph-structured data. By formulating certain strategies on the nodes and edges in the graph, GNNs transform graph-structured data into a standardized representation, which is then input into various different neural networks for training. It exhibits excellent performance in tasks such as node classification, edge information propagation, and graph clustering.

[0047] (3) Multilayer Perceptron (MLP): Also known as an Artificial Neural Network (ANN), a multilayer perceptron has multiple hidden layers in addition to the input and output layers. The simplest MLP contains only one hidden layer, i.e., a three-layer structure. The layers of a multilayer perceptron are fully connected.

[0048] (4) sigmoid function: The sigmoid function is also called the Logistic function. The sigmoid function is used for the output of neurons in the hidden layer. The value range is (0,1). It can map a real number to the interval (0,1) and is used for binary classification.

[0049] When network congestion occurs in an IP network, the overall network performance will degrade accordingly, severely impacting the operational efficiency of user services. Existing network status measurement technologies such as TWMAP and INT can help monitor network link performance, but network congestion events are sudden and real-time. Using these measurement methods to obtain network status and then make adjustments and responses results in only a delayed and passive response.

[0050] In related technologies, network status measurement techniques are typically used to monitor network link performance, which can easily lead to low efficiency in network status monitoring. Therefore, providing a method to improve the efficiency of network status monitoring is of significant practical importance.

[0051] To address existing technical problems, embodiments of this application provide a network link state prediction method, a model training method, an apparatus, and a device. The network link state prediction method obtains a target link adjacency matrix and a corresponding target link state matrix of a target network. The target link adjacency matrix is ​​obtained based on the network topology of the target network. The target link state matrix is ​​generated based on the link state data of the target network. The target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is trained using training sample data, including the link adjacency matrix and the corresponding historical link state matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition. This method inputs the target link adjacency matrix and the corresponding target link state matrix of the target network into a trained link state prediction model to obtain the link state matrix for the next moment corresponding to the target link adjacency matrix. It can learn spatial features based on the target link adjacency matrix and extract temporal features based on the corresponding target link state matrix. This enables the model to obtain the spatiotemporal features of the network link state based on the target link adjacency matrix and the target link state matrix, and to accurately predict the network link state for the next moment. This can help network operators predict and respond to congestion events in advance, thereby effectively improving the efficiency of network state monitoring.

[0052] To make the inventive objectives, technical solutions, and advantages of the embodiments of this application clearer, the present application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] It should be noted that the application scenarios described in the following embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0054] Figure 1 This application illustrates a network link state prediction method provided by an embodiment of the present application, such as... Figure 1 As shown, the method includes the following steps:

[0055] Step S101: Obtain the target link adjacency matrix and the corresponding target link state matrix of the target network.

[0056] The target link adjacency matrix is ​​obtained based on the network topology of the target network; the target link state matrix is ​​generated based on the link state data of the target network.

[0057] In IP networks, routers transmit data through links. In some embodiments of this application, in order to focus the network status monitoring observation on the links, a link relationship network is modeled based on a preset link node mapping and transformation strategy to generate a link adjacency matrix, which can be input into a graph neural network for processing to achieve network link performance prediction based on spatial features.

[0058] In one optional embodiment, the link node mapping and conversion strategy constructs a link relationship network corresponding to the network, using links in the IP network as nodes in the topology and routers as the relationship links between different links. If two links have a common router endpoint in the IP network, then the corresponding nodes of the two links in the link relationship network are connected. For example, let L1, L2, and L3 represent links in the IP network, and R1, R2, and R3 represent router nodes in the IP network; assuming that L1 and L2 have a common routing endpoint R1, after implementing the link node mapping and conversion strategy, nodes L1 and L2 are directly connected in the link relationship network.

[0059] During implementation, historical link status data is continuously collected through network status measurement technology and preprocessed to generate a historical link status matrix. The historical link status data may include multiple link status information items. In the resulting link status matrix, each row represents the historical link status information, and each column represents a specific link status information item for each link.

[0060] In some optional embodiments of this application, the link state information items include latency, bandwidth, and packet loss rate.

[0061] Step S102: Input the target link adjacency matrix and the target link state matrix into the trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data. The training sample data includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment. The trained link state prediction model is obtained when the loss of the network prediction model converges to the preset target condition.

[0062] In practice, the graph convolution process of the graph neural network is pre-integrated into the recurrent neural network to construct a sequence graph neural network and obtain a network prediction model. The link network topology and historical link state data after link node mapping transformation are input into the trained link state prediction model. By capturing and learning the spatiotemporal features of the network link state, accurate link state prediction is achieved, and the link state matrix for the next moment is obtained. The trained link state prediction model is obtained by training the network prediction model using training sample data.

[0063] In some alternative embodiments, the network prediction model is obtained by replacing the linear transformation layer in the recurrent neural network with two graph convolutional neural network layers.

[0064] In some alternative embodiments, such as Figure 2 As shown, the link state prediction model is trained through the following steps:

[0065] Step S201: Obtain training sample data; the training sample data includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix.

[0066] In the embodiments of this application, historical link state data can be collected for a large number of link adjacency matrices, which are used as historical link state matrices corresponding to the link adjacency matrices, thereby obtaining training sample data.

[0067] Step S202: Based on the training sample data, select batch training information data; the batch training information data includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix.

[0068] In the embodiments of this application, the batch training information is obtained by randomly selecting from the training sample data. The batch training information includes a portion of the link adjacency matrix and a portion of the corresponding historical link state matrix.

[0069] Step S203: Input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained.

[0070] Specifically, the identification loss value of the network prediction model can be determined by a preset loss function. In some embodiments of this application, the loss function can be a mean squared error loss function; in other embodiments of this application, the loss function can also be a mean absolute value error loss function.

[0071] Step S204: Determine whether the identification loss value has converged to the preset target value. If so, end the training and obtain the trained link state prediction model. Otherwise, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain it.

[0072] In some embodiments of this application, the network prediction model is obtained by training through the following steps:

[0073] Step A01: Collect the network dataset; the network dataset includes a subset of the link adjacency matrix; the subset of the link state matrix includes sub-elements of the link state matrix.

[0074] In practice, a large amount of network dataset can be collected first; the network dataset includes a subset of the link adjacency matrix; the subset of the link state matrix includes sub-elements of the link state matrix.

[0075] Step A02: Based on the network dataset, perform unsupervised training on the preset network model to obtain the network prediction model.

[0076] In practice, after collecting the network dataset, the pre-defined network model is trained and learned in an unsupervised manner using the network dataset to obtain the network prediction model.

[0077] In some embodiments of this application, the link state prediction model captures and learns the spatiotemporal features of link state information in the IP network, including the link adjacency matrix and the corresponding target link state matrix. In addition to predicting the future link state, it can also provide accurate early warning of link utilization.

[0078] Figure 3 This application illustrates another network link state prediction method provided by an embodiment of the present application, such as... Figure 3 As shown, the method includes the following steps:

[0079] Step S301: Obtain the target link adjacency matrix and the corresponding target link state matrix of the target network.

[0080] The target link adjacency matrix is ​​obtained based on the network topology of the target network; the target link state matrix is ​​generated based on the link state data of the target network.

[0081] Step S302: Input the target link adjacency matrix and the target link state matrix into the trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data. The training sample data includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment. The trained link state prediction model is obtained when the loss of the network prediction model converges to the preset target condition.

[0082] Step S303: Based on the link state matrix corresponding to the next moment of the target link adjacency matrix, generate the link utilization status corresponding to the next moment of the target link adjacency matrix; the link utilization status represents the warning level of the link utilization of each link in the target link adjacency matrix.

[0083] In some embodiments of this application, a sequence graph neural network model is pre-constructed based on recurrent neural networks and graph neural networks to predict future network link states. Specifically, the sequence graph neural network model can replace the linear transformation layer in a basic recurrent neural network with a two-layer graph convolutional neural network to adapt to taking link state data (e.g., bandwidth, latency, and packet loss rate) and the link adjacency matrix as input. Through nonlinear feature learning, it captures the spatiotemporal variation characteristics of the link state and outputs the link state matrix for the next time step and a warning level of link utilization.

[0084] In the aforementioned sequence graph neural network model, features of adjacent link nodes are aggregated to the final features of the target link node through graph convolution. This process corresponds to the actual influence of links on their associated links and routers, effectively capturing and extracting the spatial features of link state changes. Subsequently, through recursive learning of the recurrent neural network in the sequence evolution direction, the sequence graph neural network model can simultaneously capture the temporal and spatial features of the link state. After completing the nonlinear feature learning of the input link state data and the link adjacency matrix, the sequence graph neural network model outputs a link state matrix containing predicted link state values, as well as the link utilization state obtained through MLP layer learning and sigmoid function normalization.

[0085] In some optional embodiments, the link utilization state can be a link utilization state vector. The link utilization state vector can include the link utilization state values ​​of each link. The link utilization state values ​​in the link utilization state vector calculated by the sigmoid function take values ​​in the range [0,1].

[0086] In some optional embodiments, the link utilization status value is divided into four alarm levels at 25% each, providing a fine-grained view of the link's operational status. The alarm levels can be found in Table 1.

[0087] Table 1

[0088] Level 1 >75% and ≤100% red Level 2 >50% and ≤75% orange color Level 3 >25% and ≤50% blue Level 4 >0% and ≤25% green

[0089] Referring to Table 1, in some embodiments of this application, the link utilization status value can be distinguished by different identification colors according to different ranges. For example, when the link utilization status value meets the range of (>75%, ≤100%), the corresponding link can be configured with a red identification color.

[0090] The network link state prediction method provided in the above embodiments obtains the target link adjacency matrix and the corresponding target link state matrix of the target network. The target link adjacency matrix is ​​obtained based on the network topology of the target network. The target link state matrix is ​​generated based on the link state data of the target network. The target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment. When the loss of the network prediction model converges to the preset target condition, the trained link state prediction model is obtained. This allows the model to obtain the spatiotemporal characteristics of the network link state based on the target link adjacency matrix and the target link state matrix, enabling accurate prediction of the network link state at the next moment. This can help network operators predict and respond to congestion events in advance, thereby effectively improving the efficiency of network state monitoring.

[0091] Furthermore, this application embodiment also provides a training method for a link state prediction model, used to train the link state prediction model used in the above-mentioned network link state prediction method. Figure 4 This application illustrates a method for training a link state prediction model, which can be executed by a server or a terminal device. This embodiment uses server-side execution of the training method as an example.

[0092] like Figure 4 As shown, the training method for this link state prediction model specifically includes the following steps:

[0093] Step S401: Obtain training sample data.

[0094] The training sample data includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix.

[0095] In practice, a training dataset can be pre-acquired, and training sample data can be extracted from the training dataset when training the model. The training sample data in the training dataset may include the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix.

[0096] In some optional embodiments, the historical link state matrix includes link state data; wherein the link state data includes latency, packet loss rate, and bandwidth.

[0097] Step S402: Based on the training sample data, select batch training information data; the batch training information data includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix.

[0098] In one embodiment of this application, the batch training information is obtained by randomly selecting from the link adjacency matrix and the corresponding historical link state matrix.

[0099] Step S403: Input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained.

[0100] In practice, the server inputs batch training information into the network prediction model to be trained for training and determines the recognition loss value of the network prediction model to be trained.

[0101] In one embodiment, the network prediction model is obtained by changing the linear transformation layer in the recurrent neural network to two graph convolutional neural network layers.

[0102] Step S404: Determine whether the recognition loss value has converged to the preset target value. If the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined recognition loss value and train it again. If the determination result is yes, end the training and obtain the trained link state prediction model.

[0103] In one embodiment, step S404, adjusting the parameters of the network prediction model to be trained based on the determined recognition loss value and retraining it, can specifically involve:

[0104] The parameters of the network prediction model to be trained are adjusted according to the determined recognition loss value until the recognition loss value converges to the preset target value. The resulting network prediction model is then used as the trained link state prediction model.

[0105] Figure 5This is a flowchart illustrating another training method for a link state prediction model provided in an embodiment of this application. Figure 5 As shown, the training method for this link state prediction model specifically includes the following steps:

[0106] Step S501: Obtain training sample data; the training sample data includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix.

[0107] Step S502: Based on the training sample data, select batch training information data; the batch training information data includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix.

[0108] Step S503: Input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained.

[0109] In one embodiment, the network prediction model in step S503 above can be obtained through the following steps:

[0110] Step B1: Collect the network dataset; the network dataset includes a subset of the link adjacency matrix; the subset of the link state matrix includes sub-elements of the link state matrix.

[0111] Step B2: Based on the network dataset, perform unsupervised training on the pre-set network model to obtain the network prediction model.

[0112] In one embodiment, the process of unsupervised training of a preset network model based on a network dataset to obtain a network prediction model in the above steps can be achieved through the following steps:

[0113] Step C1: Based on the network dataset, perform unsupervised training on the preset network model to obtain a pre-trained network model; the pre-trained network model is a sequence graph neural network model constructed based on recurrent neural networks and graph neural networks.

[0114] Step C2 involves reconstructing the structure of the pre-trained network model to obtain the network prediction model; the structural reconstruction involves adding a sigmoid normalization module to the pre-trained network model.

[0115] Step S504: Determine whether the recognition loss value has converged to the preset target value. If the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined recognition loss value until the recognition loss value converges to the preset target value. The obtained network prediction model is used as the trained link state prediction model. If the determination result is yes, end the training and obtain the trained link state prediction model.

[0116] The training method of the link state prediction model in this application embodiment enables the trained link state prediction model to obtain the spatiotemporal characteristics of the network link state based on the target link adjacency matrix and the target link state matrix, thereby achieving accurate prediction of the network link state at the next future moment. This can help network operators predict and respond to congestion events in advance, thereby effectively improving the efficiency of network state monitoring.

[0117] and Figure 1 The network link state prediction method shown is based on the same inventive concept. This application also provides a network link state prediction device, which can be deployed in a server or terminal device. Since this device corresponds to the network link state prediction method of this application, and the principle by which this device solves the problem is similar to that of the method, the implementation of this device can refer to the implementation of the above method; repeated details will not be elaborated further.

[0118] Figure 6 This paper illustrates a schematic diagram of a network link state prediction device provided in an embodiment of this application. Figure 6 As shown, the network link state prediction device includes a link information acquisition unit 601 and a network state prediction unit 602.

[0119] in,

[0120] The link information acquisition unit 601 is used to acquire the target link adjacency matrix and the corresponding target link state matrix of the target network; the target link adjacency matrix is ​​obtained based on the network topology diagram of the target network; the target link state matrix is ​​generated based on the link state data of the target network.

[0121] The network state prediction unit 602 is used to input the target link adjacency matrix and the target link state matrix into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition.

[0122] In an optional embodiment, the network link state prediction device, such as Figure 7 As shown, it also includes a model training unit 701, which is used to train the link state prediction model in the following manner:

[0123] Obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix;

[0124] Based on the training sample data, batch training information is selected; the batch training information includes a portion of the link adjacency matrix and a corresponding portion of the historical link state matrix;

[0125] The batch training information is input into the network prediction model to be trained for training, and the recognition loss value of the network prediction model to be trained is determined.

[0126] Determine whether the identification loss value converges to the preset target value. If it does, end the training to obtain the trained link state prediction model. Otherwise, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain it.

[0127] In an optional embodiment, the model training unit 701 is further configured to: collect a network dataset; the network dataset includes a subset of the link adjacency matrix; the subset of the link state matrix includes sub-elements of the link state matrix;

[0128] Based on the network dataset, an unsupervised training process is performed on a pre-defined network model to obtain a network prediction model.

[0129] In an optional embodiment, the network state prediction unit 602 is further configured to: generate a link utilization state corresponding to the next moment of the target link adjacency matrix based on the link state matrix corresponding to the next moment of the target link adjacency matrix; the link utilization state characterizes the warning level of the link utilization of each link in the target link adjacency matrix.

[0130] Based on the same inventive concept as the link state prediction model training method described above, this application also provides a link state prediction model training device. Similarly, this link state prediction model training device can be deployed in a server or terminal device. Since this device corresponds to the link state prediction model training method of this application, and the principle of solving the problem by this device is similar to that of the method, the implementation of this device can refer to the implementation of the above method, and repeated details will not be elaborated further.

[0131] Figure 8 This illustration shows a schematic diagram of a training device for a link-state prediction model provided in an embodiment of this application. This network link-state prediction device is used to generate the aforementioned link-state prediction model, such as... Figure 8 As shown, the training device for the network link state prediction model includes a training data acquisition unit 801 and a model training execution unit 802.

[0132] The training data acquisition unit 801 is used to acquire training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix.

[0133] The model training execution unit 802 is used to select batch training information based on the training sample data; the batch training information includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix; input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained; determine whether the recognition loss value converges to a preset target value; if the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined recognition loss value and train again; if the determination result is yes, end the training and obtain the trained link state prediction model.

[0134] In an alternative embodiment, the network prediction model is obtained by changing the linear transformation layer in the recurrent neural network to two graph convolutional neural network layers.

[0135] In one optional embodiment, the model training execution unit is specifically used for:

[0136] The parameters of the network prediction model to be trained are adjusted according to the determined recognition loss value until the recognition loss value converges to the preset target value. The resulting network prediction model is then used as the trained link state prediction model.

[0137] Based on the same inventive concept as the above-described method embodiments, this application also provides an electronic device. This electronic device can be used for network link-state prediction or for training a link-state prediction model. In one embodiment, the electronic device can be a server; in other embodiments, it can be a terminal device or other electronic device. In this embodiment, the structure of the electronic device can be as follows: Figure 9 As shown, it includes a memory 901, a communication module 903, and one or more processors 902.

[0138] The memory 901 is used to store computer programs executed by the processor 902. The memory 901 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0139] Memory 901 may be volatile memory, such as random-access memory (RAM); memory 901 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 901 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 901 may be a combination of the above-described memories.

[0140] The processor 902 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 902 is used to implement the aforementioned network link-state prediction method or link-state prediction model training method when calling the computer program stored in the memory 901.

[0141] The communication module 903 is used to communicate with terminal devices and other servers.

[0142] This application embodiment does not limit the specific connection medium between the memory 901, communication module 903, and processor 902 described above. This application embodiment... Figure 9 The memory 901 and the processor 902 are connected via a bus 904, which is in... Figure 9 The connections between other components are shown in thick lines and are for illustrative purposes only, not as limiting information. The 904 bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0143] The memory 901 stores a computer storage medium containing computer-executable instructions. These instructions are used to implement the network link state prediction method or the link state prediction model training method of the embodiments of this application. The processor 902 is used to execute the aforementioned network link state prediction method or link state prediction model training method.

[0144] In addition, this application also provides a computer-readable storage medium storing a computer program that causes a computer to perform the method described in any of the above embodiments.

[0145] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the network link state prediction method or link state prediction model training method described above. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0146] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for predicting network link states, characterized in that, The method includes: Obtain the target link adjacency matrix and the corresponding target link state matrix of the target network; the target link adjacency matrix is ​​obtained based on the network topology diagram of the target network; the target link state matrix is ​​generated based on the link state data of the target network; wherein, a link relationship network is modeled based on a preset link node mapping and transformation strategy to generate the link adjacency matrix; The target link adjacency matrix and the target link state matrix are input into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training a network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition. The link state prediction model achieves future link state prediction by capturing and learning the spatiotemporal features of link state information in the IP network, including the link adjacency matrix and the corresponding target link state matrix. The historical link state matrix is ​​obtained by collecting historical link state data for a large number of link adjacency matrices. The network prediction model is a sequence graph neural network model built based on recurrent neural networks and graph neural networks.

2. The method according to claim 1, characterized in that, The process of obtaining the trained link state prediction model includes the following steps: Obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix; Based on the training sample data, batch training information is selected; the batch training information includes a portion of the link adjacency matrix and a corresponding portion of the historical link state matrix; The batch training information is input into the network prediction model to be trained for training, and the recognition loss value of the network prediction model to be trained is determined. Determine whether the identification loss value converges to the preset target value. If it does, end the training to obtain the trained link state prediction model. Otherwise, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain it.

3. The method according to claim 1, characterized in that, After obtaining the link state matrix corresponding to the target link adjacency matrix at the next moment, the method further includes: Based on the link state matrix at the next moment corresponding to the target link adjacency matrix, the link utilization status at the next moment corresponding to the target link adjacency matrix is ​​generated; the link utilization status represents the warning level of the link utilization of each link in the target link adjacency matrix.

4. A training method for a link state prediction model, characterized in that, The training method for generating the link state prediction model used in claim 1 includes: Obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix; Based on the training sample data, batch training information is selected; the batch training information includes a portion of the link adjacency matrix and a corresponding portion of the historical link state matrix; The batch training information is input into the network prediction model to be trained for training, and the recognition loss value of the network prediction model to be trained is determined. Determine whether the identification loss value has converged to the preset target value. If the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined identification loss value and retrain. If the determination result is yes, end the training and obtain the trained link state prediction model.

5. The training method according to claim 4, characterized in that, The network prediction model is obtained by changing the linear transformation layer in the recurrent neural network to two graph convolutional neural network layers.

6. The training method according to claim 4, characterized in that, The step of adjusting the parameters of the network prediction model to be trained based on the determined recognition loss value and then retraining it includes: The parameters of the network prediction model to be trained are adjusted according to the determined recognition loss value until the recognition loss value converges to the preset target value. The resulting network prediction model is then used as the trained link state prediction model.

7. A network link state prediction device, characterized in that, The device includes: The link information acquisition unit is used to acquire the target link adjacency matrix and the corresponding target link state matrix of the target network; the target link adjacency matrix is ​​obtained based on the network topology diagram of the target network; the target link state matrix is ​​generated based on the link state data of the target network; wherein, the link relationship network is modeled based on a preset link node mapping and transformation strategy to generate the link adjacency matrix; A network state prediction unit is used to input the target link adjacency matrix and the target link state matrix into a trained link state prediction model to obtain the link state matrix corresponding to the target link adjacency matrix at the next moment. The link state prediction model is obtained by training the network prediction model using training sample data, which includes the link adjacency matrix and the historical link state matrix corresponding to the link adjacency matrix. The network prediction model is used to estimate the link state matrix corresponding to the input link adjacency matrix at the next moment, and the trained link state prediction model is obtained when the loss of the network prediction model converges to a preset target condition. The link state prediction model achieves future link state prediction by capturing and learning the spatiotemporal features of link state information in the IP network, including the link adjacency matrix and the corresponding target link state matrix. The historical link state matrix is ​​obtained by collecting historical link state data for a large number of link adjacency matrices. The network prediction model is a sequence graph neural network model built based on recurrent neural networks and graph neural networks.

8. A training device for a link state prediction model, characterized in that, The training apparatus for generating the link state prediction model used in claim 7 includes: The training data acquisition unit is used to obtain training sample data; the training sample data includes a link adjacency matrix and a historical link state matrix corresponding to the link adjacency matrix. The model training execution unit is used to select batch training information based on the training sample data; the batch training information includes a part of the link adjacency matrix and a part of the corresponding historical link state matrix; input the batch training information into the network prediction model to be trained for training, and determine the recognition loss value of the network prediction model to be trained; determine whether the recognition loss value converges to a preset target value; if the determination result is no, adjust the parameters of the network prediction model to be trained according to the determined recognition loss value and train again; if the determination result is yes, end the training and obtain the trained link state prediction model.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method of any one of claims 1 to 3 or the method of any one of claims 4 to 6.

10. An electronic device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and when the computer program is executed by the processor, causes the processor to implement the method of any one of claims 1 to 3 or the method of any one of claims 4 to 6.

Images