Edge node access method, device, equipment and computer program product

Abstract

The present invention discloses an edge node access method, apparatus, device, and computer program product. The edge node access method comprises: obtaining access requirement data for a newly created edge node and obtaining topology data of a current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent network connection relationships between switches, and the topology data is the topology data of the switches; based on a trained decision model, the access requirement data, and the topology data, performing access category prediction on several switch nodes in the current bearer network topology to obtain access category prediction results, and selecting a switch to be connected to the newly created edge node based on the access category prediction results. The present invention improves the efficiency and accuracy of newly created edge nodes accessing bearer network switches.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular to an edge node access method, apparatus, device, and computer program product. Background Art

[0002] With the rapid development of communication technology, users have higher and higher requirements for communication speed. To reduce the bandwidth and latency loss caused by network transmission and multi-stage forwarding, edge nodes are usually built at the edge of the network close to users.

[0003] Currently, the selection of switches for newly built edge nodes to access the bearer network relies on expert experience, that is, the operator's operations and maintenance personnel use their experience to select the optimal access switch. However, due to the complex bearer network configuration, the performance differences between switches, and the diverse access requirements of edge nodes, manual experience is inefficient and has low accuracy when selecting access switches for newly built edge nodes. Summary of the Invention

[0004] The main purpose of the present invention is to provide an edge node access method, device, equipment and computer program product, aiming to solve the technical problem that the method of selecting access switches for new edge nodes based on manual experience is inefficient and has low accuracy.

[0005] To achieve the above object, the present invention provides an edge node access method, which includes the following steps:

[0006] Obtain access demand data for the newly created edge node and obtain topology data of the current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent network connection relationships between switches, and the topology data is the topology data of the switches;

[0007] Based on the trained decision model, the access demand data and the topology data, access category prediction is performed on several switch nodes in the current bearer network topology to obtain access category prediction results, so as to select the switch to which the newly created edge node is connected based on the access category prediction results.

[0008] Optionally, the decision model includes a first feature extractor, a second feature extractor, and a classifier, and the step of performing access category prediction on a plurality of switch nodes in the current bearer network topology based on the trained decision model, the access demand data, and the topology data to obtain the access category prediction result includes:

[0009] Based on the first feature extractor, feature extraction is performed on the topological data to obtain a topological feature vector;

[0010] performing feature extraction on the access demand data based on the second feature extractor to obtain an access demand feature vector;

[0011] Aggregating the topology feature vector and the access requirement feature vector to obtain an aggregated feature vector;

[0012] Based on the classifier and the aggregated feature vector, the access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result.

[0013] Optionally, the first feature extractor is composed of a first preset number of graph convolutional layers, and the step of performing feature extraction on the topological data based on the first feature extractor to obtain a topological feature vector includes:

[0014] Performing transformation processing on the topological data to obtain a characteristic matrix and an adjacency matrix;

[0015] Performing data preprocessing on the feature matrix;

[0016] Based on the first feature extractor, feature extraction is performed on the adjacency matrix and the feature matrix after data preprocessing to obtain a topological feature vector.

[0017] Optionally, the feature matrix includes the layers to which several switch nodes belong, the device performance of several switch nodes, the number of edge nodes connected to several switch nodes, and the geographical locations of several switch nodes. The step of performing data preprocessing on the feature matrix includes:

[0018] Performing one-hot encoding on the layers to which the plurality of switch nodes belong and the geographical locations of the plurality of switch nodes;

[0019] The performance of the switch nodes and the number of edge nodes connected to the switch nodes are standardized.

[0020] Optionally, the second feature extractor is composed of a second preset number of fully connected layers and dropout layers alternately arranged, and the step of performing feature extraction on the access demand data based on the second feature extractor to obtain the access demand feature vector includes:

[0021] performing data preprocessing on the access demand data to obtain an access demand vector;

[0022] Based on the second feature extractor, feature extraction is performed on the access demand vector to obtain an access demand feature vector.

[0023] Optionally, the decision model further includes a merging layer, and the step of aggregating the topology feature vector and the access demand feature vector to obtain an aggregated feature vector includes:

[0024] Based on the merging layer, aggregating the topology feature vector and the access demand feature vector to obtain an aggregated feature vector;

[0025] The classifier includes a first fully connected layer, and the step of predicting the access categories of several switch nodes in the current bearer network topology based on the classifier and the aggregated feature vector to obtain the access category prediction result includes:

[0026] Based on the first fully connected layer and the aggregated feature vector, access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result.

[0027] Optionally, before the step of performing access category prediction on a plurality of switch nodes in the current bearer network topology based on the trained decision model, the access demand data, and the topology data to obtain the access category prediction result, the step further includes:

[0028] Acquire access requirement training data of edge nodes and topology training data of bearer network topology, and label the access requirement training data and the topology training data with labels specific to switch access categories to obtain switch access category label data;

[0029] Acquire a model to be trained, and select training sample data from the access demand training data, the topology training data, and the switch access category label data;

[0030] Based on the training sample data, the model to be trained is iteratively trained to obtain the decision model.

[0031] In addition, to achieve the above-mentioned object, the present invention further provides an edge node access device, the edge node access device comprising:

[0032] An acquisition module, configured to acquire access demand data of a newly created edge node and acquire topology data of a current bearer network topology, wherein the topology nodes of the current bearer network topology are switches, and the edges of the current bearer network topology are network connections between switches;

[0033] A prediction module is used to predict the access category of several switch nodes in the current bearer network topology based on the trained decision model, the access demand data and the topology data, and obtain an access category prediction result for selecting the switch to which the newly created edge node is connected based on the access category prediction result.

[0034] In addition, to achieve the above-mentioned purpose, the present invention also provides an edge node access device, which includes: a memory, a processor, and an edge node access program stored on the memory and runnable on the processor. When the edge node access program is executed by the processor, the steps of the edge node access method described above are implemented.

[0035] In addition, to achieve the above-mentioned object, the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the edge node access method described above are implemented.

[0036] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium, on which an edge node access program is stored. When the edge node access program is executed by a processor, the steps of the edge node access method described above are implemented.

[0037] The present invention provides an edge node access method, apparatus, device, and computer program product, which obtain access demand data of a newly built edge node and obtain topology data of the current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent the network connection relationship between switches, and the topology data is the topology data of the switches; based on a trained decision model, access demand data, and topology data, access categories are predicted for several switch nodes in the current bearer network topology, and access category prediction results are obtained for selecting the switch to be connected to the newly built edge node based on the access category prediction results. Through the above method, based on the established and trained decision model, the bearer network switch to be connected to the newly built edge node is automatically selected, and because the decision model is obtained through iterative training, the efficiency and accuracy of the newly built edge node accessing the bearer network switch are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present invention;

[0039] Figure 2 This is a flow chart of a first embodiment of an edge node access method according to the present invention;

[0040] Figure 3 This is a flow chart of a second embodiment of the edge node access method of the present invention;

[0041] Figure 4 A schematic diagram of topological data involved in an embodiment of the present invention;

[0042] Figure 5 A schematic diagram of a decision model involved in an embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the functional modules of the first embodiment of the edge node access device of the present invention.

[0044] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0045] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] The main solution of the embodiment of the present invention is: obtaining the access demand data of the newly built edge node and obtaining the topology data of the current bearer network topology, wherein the topology node of the current bearer network topology is a switch, and the edge of the current bearer network topology is the network connection between switches; based on the trained decision model, access demand data and topology data, performing access category prediction on several switch nodes in the current bearer network topology, obtaining access category prediction results, and selecting the switch to be connected to the newly built edge node based on the access category prediction results. In this way, based on the established and trained decision model, the bearer network switch to be connected to the newly built edge node is automatically selected, and because the decision model is obtained through iterative training, the efficiency and accuracy of the newly built edge node accessing the bearer network switch are improved.

[0047] Technical terms involved in the embodiments of the present invention:

[0048] Mobile Edge Computing (MEC) is an architecture based on the evolution of 5G. In the 5G era, everything is connected, and massive IoT devices are expanding. Cloud computing models will face bottlenecks in data processing, cost, and energy consumption. At the same time, a superior user experience requires cloud content to reach users. Therefore, the rapid development of MEC is an inevitable technological evolution. By providing flexible network access capabilities and edge computing services at the edge of mobile networks, MEC reduces network transmission and service delivery latency, improves data security, and empowers the development of vertical industries. Edge computing nodes can be deployed in tiered locations, including core data centers, key aggregation centers, general aggregation centers, and access campus data centers, based on industry customer needs. MEC is a technology that deeply integrates mobile access networks with internet services, bringing computing and processing capabilities down to the edge, closest to the services. MEC improves user experience, conserves bandwidth resources, reduces congestion and burden on core and transport networks, alleviates network bandwidth pressure, and achieves low latency. Furthermore, by delegating computing power to mobile edge nodes, it facilitates third-party application integration, opens up unlimited possibilities for service innovation at the mobile edge, enabling rapid response to user requests and improving service quality.

[0049] Edge nodes are business platforms built at the edge of the network, close to users. They provide storage, computing, and networking resources, enabling them to move critical business applications to the edge of the access network to reduce bandwidth and latency losses associated with network transmission and multi-level forwarding. These nodes are characterized by their overall multi-dimensional performance KPIs (including overall load, request latency, and request success rate) at the current moment.

[0050] The embodiments of the present invention take into account that, in existing solutions, the selection of switches for edge node access to the bearer network currently relies on expert experience, using the experience of operator operations and maintenance personnel to select the optimal access switch. However, due to the complex bearer network configuration, the performance differences between switches, and the diverse access requirements of edge nodes, high requirements are placed on the accuracy and reliability of manual decision-making. The existing method of selecting access switches for newly built edge nodes based on manual experience is inefficient, unreliable, and has a low degree of automation, which is not conducive to the rapid development of edge computing.

[0051] Therefore, an embodiment of the present invention proposes a solution that automatically selects the bearer network switch to which the newly built edge node is connected based on the established and trained decision model. Moreover, since the decision model is obtained through iterative training, the efficiency and accuracy of the newly built edge node accessing the bearer network switch are improved.

[0052] Reference Figure 1 , Figure 1 This is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiment of the present invention.

[0053] The terminal in the embodiment of the present invention is an edge node access device, which can be a terminal device with processing function, such as a PC (personal computer), a microcomputer, a notebook computer, or a server.

[0054] like Figure 1As shown, the terminal may include: a processor 1001, such as a CPU (Central Processing Unit), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the user interface 1003 may optionally include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.

[0055] Those skilled in the art will understand that Figure 1 The terminal structure shown in the figure does not constitute a limitation to the terminal, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0056] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an edge node access program.

[0057] exist Figure 1 In the terminal shown, the processor 1001 can be used to call the edge node access program stored in the memory 1005 and execute various embodiments of the following edge node access method.

[0058] Based on the above hardware structure, various embodiments of the edge node access method of the present invention are proposed.

[0059] The present invention provides an edge node access method.

[0060] Reference Figure 2 , Figure 2 FIG. 1 is a flow chart of a first embodiment of an edge node access method according to the present invention.

[0061] In this embodiment, the edge node access method includes:

[0062] Step S10: Acquire access requirement data of the newly created edge node and acquire topology data of the current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent network connection relationships between switches, and the topology data is the topology data of the switches;

[0063] In this embodiment, the edge node access method can be applied to an edge node access device, which can be an edge computing management platform (MEPM), which can be a 5G (5th Generation Mobile Communication Technology) edge computing management platform, which can be managed by an operator; it can also be applied to an edge node access system composed of the edge computing management platform and other related devices, which is subordinate to the above-mentioned Figure 1 edge node access device.

[0064] In this embodiment, access requirement data for a newly created edge node is obtained, along with topology data for the current bearer network topology. The topology nodes of the current bearer network topology represent switches, and the edges of the current bearer network topology represent network connectivity between switches. This topology data is switch topology data. The access requirement data may include at least one of the following: bandwidth, latency, and rate. This access requirement data can be set based on actual needs and is not limited here.

[0065] It should be noted that the bearer network topology is constructed using several switches in the current bearer network as topological nodes and the network connections between these switches as edges to obtain the topological data of the bearer network topology. The topological data of the bearer network topology is converted and processed to obtain an adjacency matrix and a feature matrix. The adjacency matrix is used to characterize the connection relationship between several switches, and the feature matrix is used to characterize the attributes contained in several switch nodes. For example, the feature matrix includes the layer to which several switch nodes belong, the device performance of several switch nodes, the number of edge nodes connected to several switch nodes, and the geographical location of several switch nodes. The number of switches is determined based on the networking situation of the current bearer network and is not limited here.

[0066] In one embodiment, before step S10, the edge node access method further includes:

[0067] Obtaining networking information of the current bearer network and, based on the networking information, determining a number of switches in the current bearer network; and constructing a topology of the current bearer network using the switches as topology nodes and the network connections between the switches as edges. The networking information includes layout information, network connection relationships, and network performance data of the switches in the current bearer network.

[0068] Step S20: Based on the trained decision model, the access demand data, and the topology data, access category prediction is performed on several switch nodes in the current bearer network topology to obtain access category prediction results, so as to select the switch to which the newly created edge node is connected based on the access category prediction results.

[0069] In this embodiment, based on the trained decision model, access demand data, and topology data, access category prediction is performed on several switch nodes in the current bearer network topology to obtain access category prediction results, so as to select switches to be connected to the newly created edge node based on the access category prediction results. The number of the several switches is determined according to the networking conditions of the current bearer network and is not limited here. The access category prediction results are used to indicate whether several switches in the current bearer network are selected as the best access switches. It can be understood that one or more of the several switches can be selected. If one is selected, the selected switch is the best access switch, and the newly created edge node can directly connect to the selected switch. If multiple switches are selected, the selected multiple switches can be provided for further selection by the user, or one switch can be selected from them according to a preset strategy as the switch to be connected to the newly created edge node.

[0070] Specifically, based on the feature extractors in the trained decision model, the access demand feature information of the access demand data and the topology feature information of the topology data of the current bearer network topology are extracted, and then based on the access demand feature information, the topology feature information and the classifier in the decision model, the access demand feature information and the topology feature information of the topology data of the current bearer network topology are subjected to binary classification prediction to obtain a binary classification prediction result, and based on the binary classification prediction result, an access category prediction result is obtained.

[0071] The classifier's specific execution process involves obtaining a classification probability vector and then determining the access category prediction result (binary classification prediction result) corresponding to the maximum classification probability value in the classification probability vector. The access category prediction result includes whether the switch is selected or not. For example, a value of 1 in the access category prediction result indicates that the switch is selected, while a value of 0 indicates that the switch is not selected.

[0072] In another embodiment, the decision model is composed of a bearer network topology extraction module, a new edge node access demand extraction module, and an optimal access switch prediction module. Step S20 includes:

[0073] Based on the bearer network topology extraction module, feature extraction is performed on the topology data to obtain a topology feature vector; based on the newly built edge node access demand extraction module, feature extraction is performed on the access demand data to obtain an access demand feature vector; based on the optimal access switch prediction module, the topology feature vector and the access demand feature vector are aggregated to obtain an aggregated feature vector; based on the optimal access switch prediction module and the aggregated feature vector, the access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result. Specifically, the bearer network topology extraction module extracts the spatial features of the current bearer network topology and projects the connections between the switch nodes and switches in the bearer network topology into a low-dimensional vector space; the newly built edge node access demand extraction module maps the multi-dimensional attributes of the edge node access demand to a vector of fixed dimension; the optimal access switch prediction module merges and concatenates the output results of the two branches to form a new vector. Finally, by learning the encoded current bearer network topology vector representation, the mapping relationship between the access switch topology features, the newly built edge node access demand features, and the optimal access switch selection is discovered, thereby obtaining an access category prediction result.

[0074] Optionally, the bearer network topology extraction module is a graph convolutional neural network, and the step of performing feature extraction on the topology data based on the bearer network topology extraction module to obtain a topology feature vector includes:

[0075] Based on the graph convolutional neural network, the topology data is feature extracted to obtain a topology feature vector. Specifically, the bearer network topology extraction module uses the graph convolutional neural network to extract the spatial features of the current bearer network topology and project the switch nodes and connections between switches in the bearer network topology into a low-dimensional vector space.

[0076] Optionally, the newly-created edge node access demand extraction module is composed of a preset number of fully connected layers and discarded layers arranged alternately. The step of performing feature extraction on the access demand data based on the newly-created edge node access demand extraction module to obtain an access demand feature vector includes:

[0077] Based on the preset number of fully connected layers and discarded layers, feature extraction is performed on the access demand data to obtain an access demand feature vector. Specifically, the newly created edge node access demand extraction module uses the preset number of fully connected layers and discarded layers to map the multi-dimensional attributes of the edge node access demand to a vector of fixed dimension.

[0078] Optionally, the optimal access switch prediction module is composed of a merging layer and a fully connected layer. The step of aggregating the topology feature vector and the access demand feature vector based on the optimal access switch prediction module to obtain an aggregated feature vector includes:

[0079] Based on the merging layer, the topology feature vector and the access requirement feature vector are aggregated to obtain an aggregated feature vector. Specifically, the optimal access switch prediction module uses the merging layer to merge and splice the output results of the two branches to form a new vector.

[0080] The step of predicting the access categories of several switch nodes in the current bearer network topology based on the optimal access switch prediction module and the aggregated feature vector to obtain the access category prediction result includes:

[0081] Based on the fully connected layer and the aggregated feature vector, the access categories of several switch nodes in the current bearer network topology are predicted to obtain access category prediction results. Specifically, the fully connected layer is used to learn the encoded vector representation of the current bearer network topology to discover the mapping relationship between the access switch topology characteristics, the access requirement characteristics of the newly created edge node, and the optimal access switch selection, thereby obtaining the access category prediction results.

[0082] Furthermore, before the above step S20, the edge node access method further includes:

[0083] The topology data is transformed to obtain a feature matrix and an adjacency matrix; data preprocessing is performed on the feature matrix; and data preprocessing is performed on the access demand data to obtain an access demand vector. The adjacency matrix is used to characterize the connection relationship between a number of switches, and the feature matrix is used to characterize the attributes contained in a number of switch nodes. Data preprocessing includes one-hot encoding and normalization. For example, the feature matrix includes the level to which a number of switch nodes belong, the device performance of a number of switch nodes, the number of edge nodes connected to a number of switch nodes, and the geographical locations of a number of switch nodes. The level to which the number of switch nodes belong and the geographical locations of the number of switch nodes are one-hot encoded; and the device performance of the number of switch nodes and the number of edge nodes connected to the number of switch nodes are normalized.

[0084] Accordingly, the above step S20 includes:

[0085] Based on the trained decision model, the access demand vector, the feature matrix after data preprocessing and the adjacency matrix, access category prediction is performed on several switch nodes in the current bearer network topology to obtain access category prediction results, so as to select the switch to which the newly created edge node is connected based on the access category prediction results.

[0086] In one embodiment, the access demand vector, the feature matrix after data preprocessing, and the adjacency matrix are input into the decision model through each input layer of the decision model to improve the processing performance of the decision model.

[0087] An embodiment of the present invention provides an edge node access method, which obtains access demand data of a newly built edge node and obtains topology data of the current bearer network topology, wherein the topology nodes of the current bearer network topology are switches, and the edges of the current bearer network topology are network connections between switches; based on a trained decision model, access demand data, and topology data, an access category prediction is performed on several switch nodes in the current bearer network topology to obtain an access category prediction result, so as to select the switch to be connected to the newly built edge node based on the access category prediction result. In this way, based on the established and trained decision model, the bearer network switch to be connected to the newly built edge node is automatically selected, and because the decision model is obtained through iterative training, the efficiency and accuracy of the newly built edge node accessing the bearer network switch are improved.

[0088] Furthermore, based on the above first embodiment, a second embodiment of the edge node access method of the present invention is proposed.

[0089] Reference Figure 3 , Figure 3 FIG. 4 is a flow chart of a second embodiment of an edge node access method according to the present invention.

[0090] In this embodiment, the decision model includes a first feature extractor, a second feature extractor, and a classifier, and the above step S20 includes:

[0091] Step S21: extracting features from the topological data based on the first feature extractor to obtain a topological feature vector;

[0092] In this embodiment, the topological data is subjected to feature extraction based on the first feature extractor to obtain a topological feature vector. Specifically, the topological data feature information of the topological data is extracted based on the first feature extractor in the trained decision model to obtain a topological feature vector.

[0093] In one embodiment, the first feature extractor is a graph convolutional neural network model, and step S21 includes:

[0094] Based on the graph convolutional neural network, the topology data is feature extracted to obtain a topology feature vector. Specifically, the bearer network topology extraction module uses the graph convolutional neural network to extract the spatial features of the current bearer network topology and project the switch nodes and connections between switches in the bearer network topology into a low-dimensional vector space.

[0095] Optionally, the first feature extractor is composed of a preset number of graph convolutional layers, and the above step S21 includes:

[0096] Based on the preset number of graph convolution layers, feature extraction is performed on the topological data to obtain a topological feature vector.

[0097] In a specific embodiment, the first feature extractor is composed of a first preset number of graph convolutional layers, and the above step S21 includes:

[0098] Step A211, transforming the topological data to obtain a feature matrix and an adjacency matrix;

[0099] In this embodiment, the topology data of the current bearer network topology is converted to obtain a feature matrix and an adjacency matrix. The adjacency matrix is used to represent the connection relationships between multiple switches, and the feature matrix is used to represent the attributes of multiple switch nodes. For example, the feature matrix includes the layer to which multiple switch nodes belong, the device performance of multiple switch nodes, the number of edge nodes connected to multiple switch nodes, and the geographical location of multiple switch nodes. The number of multiple switches is determined based on the current bearer network configuration and is not limited here.

[0100] Among them, the levels to which the switch nodes belong may include the core computer room of a city or prefecture, important aggregation computer room, ordinary aggregation computer room, access campus computer room, etc.; the performance of the switch node equipment may include the total number of ports, the maximum number of concurrent connections supported, the maximum number of forwarding connections supported, the maximum bandwidth supported, the lowest latency, etc.

[0101] It should be noted that the current bearer network topology can be represented as a graph G = (V, E), where V is a set of switch nodes V = {V1, V2, V3, ..., V N}, E is the set of edges in the current bearer network topology, if the switch node V i and switch node V j If there is a direct connection between ij =1, otherwise e ij =0; X is the set of switch nodes in the current bearer network topology, x ij Represents the jth attribute of the i-th switch node. Based on this, we can get the adjacency matrix A and the feature matrix X. For details, please refer to Figure 4 , Figure 4 This is a schematic diagram of topological data involved in an embodiment of the present invention.

[0102] Step A212, performing data preprocessing on the feature matrix;

[0103] In this embodiment, data preprocessing is performed on the feature matrix, wherein the data preprocessing includes at least one of the following: one-hot encoding, standardization, and normalization.

[0104] In one embodiment, the feature matrix includes the layers to which the switch nodes belong, the device performance of the switch nodes, the number of edge nodes connected to the switch nodes, and the geographical locations of the switch nodes. Step A212 includes:

[0105] Step A2121: One-hot encode the layers to which the switch nodes belong and the geographical locations of the switch nodes.

[0106] In this embodiment, the layers to which several switch nodes belong and the geographical locations of the several switch nodes are uniquely encoded to obtain layer vectors to which the several switch nodes belong and geographical location vectors to which the several switch nodes are located, so as to perform feature extraction on the layer vectors to which the several switch nodes belong and the geographical location vectors to which the several switch nodes are located based on the first feature extractor to obtain a topological feature vector.

[0107] Step A2122: Standardize the performance of the switch nodes and the number of edge nodes connected to the switch nodes.

[0108] In this embodiment, the performance of several switch nodes and the number of edge nodes connected to the several switch nodes are standardized.

[0109] In one embodiment, the formula for normalization processing may be:

[0110] (X-mean) / std;

[0111] Where X represents an attribute in the performance of several switch nodes, mean represents the mean of all attributes in the performance of several switch nodes, and std represents the variance of all attributes in the performance of several switch nodes. Alternatively, X represents an attribute in the number of edge nodes connected to several switch nodes, mean represents the mean of all attributes in the number of edge nodes connected to several switch nodes, and std represents the variance of all attributes in the number of edge nodes connected to several switch nodes. The calculation is performed separately for each attribute (dimension): the data is subtracted from its mean (per column) and divided by its variance.

[0112] It can be understood that the standardized performance of several switch node devices and the number of edge nodes connected to several switch nodes can improve the convergence speed of the decision-making model, which makes it easier to train the model. At the same time, the standardized performance of several switch node devices and the number of edge nodes connected to several switch nodes improve the accuracy of the decision-making model compared to the original data, thereby improving the accuracy of the newly built edge nodes accessing the bearer network switch.

[0113] Step A213: Based on the first feature extractor, feature extraction is performed on the adjacency matrix and the feature matrix after data preprocessing to obtain a topological feature vector.

[0114] In this embodiment, based on the first feature extractor, feature extraction is performed on the adjacency matrix and the feature matrix after data preprocessing to obtain a topological feature vector. Specifically, based on a first preset number of graph convolution layers, topological data feature information of the topological data is extracted to obtain a topological feature vector.

[0115] In one embodiment, the first preset number is 3, the first feature extractor includes a first graph convolution layer, a second graph convolution layer, and a third graph convolution layer, and step A213 includes:

[0116] Based on the first graph convolution layer, the topological data feature information of the topological data is extracted to obtain a first topological feature vector; based on the second graph convolution layer, the feature information of the first topological feature vector is extracted to obtain a second topological feature vector; based on the third graph convolution layer, the feature information of the second topological feature vector is extracted to obtain a third topological feature vector (i.e., a topological feature vector).

[0117] Among them, the number of convolution kernels of the first graph convolution layer can be 256, and the activation function can be set to "relu"; the number of convolution kernels of the second graph convolution layer can be 256, and the activation function can be set to "relu"; the number of convolution kernels of the third graph convolution layer can be 128, and the activation function can be set to "lamda".

[0118] Furthermore, the first feature extractor further includes a flattening layer. After the above step A213, the edge node access method further includes:

[0119] Based on the flattening layer, the topological feature vector is flattened. Specifically, based on the flattening layer, the topological feature vector is "flattened", that is, the three-dimensional input is converted into two dimensions.

[0120] Step S22: extracting features from the access demand data based on the second feature extractor to obtain an access demand feature vector;

[0121] In this embodiment, the access demand data is subjected to feature extraction based on the second feature extractor to obtain an access demand feature vector. Specifically, the second feature extractor in the trained decision model extracts feature information of the access demand data to obtain an access demand feature vector.

[0122] In one embodiment, the second feature extractor is composed of a second preset number of fully connected layers and dropout layers arranged alternately, and the step S22 includes:

[0123] Step A221: pre-process the access demand data to obtain an access demand vector;

[0124] In this embodiment, data preprocessing is performed on the access demand data to obtain an access demand vector, wherein the data preprocessing includes standardization or normalization.

[0125] In one embodiment, the formula for normalization processing may be:

[0126] (X-mean) / std;

[0127] Where X represents an attribute in the access demand data, mean represents the mean of all attributes in the access demand data, and std represents the variance of all attributes in the access demand data. The calculation is performed separately for each attribute (dimension): the mean is subtracted from the attribute (per column) and the variance is divided.

[0128] It can be understood that standardized access demand data can improve the convergence speed of the decision-making model, which makes model training easier. At the same time, compared with the original data, the standardized access demand data improves the accuracy of the decision-making model, thereby improving the accuracy of the newly built edge node access bearer network switch.

[0129] Step A222: Based on the second feature extractor, perform feature extraction on the access demand vector to obtain an access demand feature vector.

[0130] In this embodiment, the access demand vector is subjected to feature extraction based on the second feature extractor to obtain the access demand feature vector. Specifically, feature information of the access demand vector is extracted based on a second preset number of fully connected layers and discarded layers to obtain the access demand feature vector.

[0131] In one embodiment, the second preset number is 2, the second feature extractor includes a first fully connected layer, a first dropout layer, a second fully connected layer, and a second dropout layer, and step A222 includes:

[0132] Based on the first fully connected layer, the feature information of the access demand vector is extracted to obtain a first access demand feature vector; based on the first discarding layer, the first access demand feature vector is discarded to obtain a second access demand feature vector; based on the second fully connected layer, the feature information of the second access demand feature vector is extracted to obtain a third access demand feature vector; based on the second discarding layer, the third access demand feature vector is discarded to obtain a fourth access demand feature vector (i.e., the access demand feature vector).

[0133] Among them, the number of fully connected neurons in the first fully connected layer can be set to 64, and the activation function can be set to "relu"; the number of fully connected neurons in the second fully connected layer can be set to 32, and the activation function can be set to "relu"; the discard probability of the first discard layer can be set to 0.2, that is, 20% of the neurons are randomly ignored and made invalid; the discard probability of the second discard layer can be set to 0.2, that is, 20% of the neurons are randomly ignored and made invalid.

[0134] Step S23, performing aggregation processing on the topology feature vector and the access demand feature vector to obtain an aggregated feature vector;

[0135] In this embodiment, the topology feature vector and the access requirement feature vector are aggregated to obtain an aggregated feature vector, wherein the aggregation process includes merging or weighting.

[0136] In one embodiment, the decision model further includes a merging layer, and step S23 includes:

[0137] Step A231: Based on the merging layer, aggregate the topology feature vector and the access demand feature vector to obtain an aggregated feature vector;

[0138] In this embodiment, the topology feature vector and the access requirement feature vector are aggregated based on the merging layer to obtain an aggregated feature vector. Specifically, the topology feature vector and the access requirement feature vector are merged and concatenated to form the aggregated feature vector.

[0139] In another embodiment, the above step S23 includes:

[0140] Based on a preset aggregation rule, the topology feature vector and the access demand feature vector are aggregated to obtain aggregated feature data, wherein the preset aggregation rule includes weighted summation and weighted average, etc. For example, assuming that the eigenvalue corresponding to the topology feature vector is 0.2, and the eigenvalue corresponding to the access demand feature vector is 200, the preset aggregation rule is weighted summation, and then the eigenvalue corresponding to the aggregated feature vector is 0.2+200*0.01=0.3, and then normalized to 1.

[0141] Step S24 : predicting access categories of several switch nodes in the current bearer network topology based on the classifier and the aggregated feature vector to obtain an access category prediction result.

[0142] In this embodiment, the access categories of several switch nodes in the current bearer network topology are predicted based on the classifier and the aggregated feature vector, obtaining an access category prediction result. Specifically, based on the aggregated feature vector and the classifier in the decision model, a binary classification prediction is performed on the aggregated feature vector to obtain a binary classification prediction result, and based on this binary classification prediction result, an access category prediction result is obtained.

[0143] In one embodiment, the classifier includes a first fully connected layer, and step S24 includes:

[0144] Step A241 : Based on the first fully connected layer and the aggregated feature vector, predict the access categories of several switch nodes in the current bearer network topology to obtain an access category prediction result.

[0145] In this embodiment, based on the first fully connected layer and the aggregated feature vector, the access categories of several switch nodes in the current bearer network topology are predicted to obtain access category prediction results. The number of fully connected neurons in the first fully connected layer can be 128, and the activation function can be set to "ReLU".

[0146] Furthermore, the classifier further includes an output layer, and the above step A241 includes:

[0147] Based on the first fully connected layer, the output layer, and the aggregated feature vector, the access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result. The output layer may be composed of a fully connected layer, and the number of fully connected neurons in the output layer may be set to N, where N is the number of switch nodes in the current bearer network topology, and the activation function may be set to "sigmoid."

[0148] In one embodiment, referring to Figure 5 , Figure 5It is a schematic diagram of the decision model involved in the embodiment of the present invention. Among them, the edge computing management platform is the executive body of the embodiment of the present invention. The bearer network topology feature matrix X is a feature matrix in the topology data of the current bearer network topology, and the feature matrix includes the layers to which several switch nodes belong (the layers to which each node in the figure belongs), the performance of several switch node devices (the performance of each node device in the figure), the number of edge nodes that have been hung under several switch nodes (the number of edge nodes that have been hung under each node in the figure) and the geographical locations of several switch nodes (the geographical locations of the nodes in the figure). The bearer network topology adjacency matrix A is the adjacency matrix in the topology data of the current bearer network topology. The access requirements of edge node i are the access requirements of the newly created edge node. For the bearer network topology extraction part, the first layer is the input layer (the input layer on the left in the figure): it inputs the bearer network topology adjacency matrix A and feature matrix X; the second layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 256, and the activation function is set to "relu". This layer uses the graph convolution layer to extract the bearer network topology features; the third layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 256, and the activation function is set to "relu"; the fourth layer is the graph convolution layer (Graph Conv): the number of convolution kernels is 128, and the activation function is set to "lamda"; the fifth layer is the flatten layer: this layer is used to "flatten" the input, converting the three-dimensional input into two-dimensional. For extracting new edge node access requirements, the first layer is the input layer (the input layer on the right in the figure): it inputs the multidimensional attributes of the new edge node access requirements. The second layer is the fully connected (Dense) layer: the number of fully connected neurons is set to 64, and the activation function is set to "ReLU". The third layer is the dropout layer: the dropout probability is set to 0.2, randomly ignoring 20% of the neurons and rendering them ineffective. The fourth layer is the fully connected (Dense) layer: the number of fully connected neurons is set to 32, and the activation function is set to "ReLU". The fifth layer is the dropout layer: the dropout probability is set to 0.2, randomly ignoring 20% of the neurons and rendering them ineffective. For the optimal access switch prediction, the first layer is a concatenation layer: it combines the outputs of the two branches to form a new vector. The second layer is a fully connected (Dense) layer with 128 neurons and the activation function set to "ReLU." The third layer is the output layer, consisting of a fully connected (Dense) layer with N neurons (N is the number of switch nodes in the bearer network topology) and the activation function set to "Sigmoid." Finally, the access prediction results for N switch nodes in the bearer network topology are obtained.

[0149] In this embodiment, feature extraction is performed on the topology data of the current bearer network topology to obtain a topology feature vector, and feature extraction is performed on the access demand data of the newly built edge node to obtain an access demand feature vector. Then, the topology feature vector and the access demand feature vector are aggregated to obtain an aggregated feature vector. Finally, based on the classifier and the aggregated feature vector of the decision model, the access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result, thereby automatically selecting the bearer network switch to which the newly built edge node accesses, thereby improving the efficiency and accuracy of the newly built edge node accessing the bearer network.

[0150] Furthermore, based on the above first embodiment, a third embodiment of the edge node access method of the present invention is proposed.

[0151] In this embodiment, before step S20, the edge node access method further includes:

[0152] Step A30: Acquire access requirement training data of edge nodes and topology training data of bearer network topology, and label the access requirement training data and the topology training data with labels specific to switch access categories to obtain switch access category label data.

[0153] In this embodiment, access requirement training data of edge nodes and topology training data of the bearer network topology are obtained, and the access requirement training data and topology training data are labeled with labels specific to the switch access category to obtain switch access category label data. Specifically, each access requirement representation value in the access requirement training data and each topology data representation value in the topology training data are extracted. Based on each access requirement representation value and each topology data representation value, corresponding switch access categories are matched to the access requirement training data and the topology training data to obtain the switch access category label data.

[0154] In one embodiment, the access requirements of each edge node are obtained, and then the access requirements are used as access requirement training data. In other embodiments, the access requirement training data can be set manually or by other means. The access requirement training data includes at least one access requirement.

[0155] In one embodiment, a historical bearer network topology is obtained, and then topology data in each bearer network topology is obtained, and all topology data is used as topology training data. In other embodiments, the topology training data can be set manually or through other methods. The topology training data includes at least one topology data.

[0156] Furthermore, after the step of obtaining access requirement training data of the edge node and topology training data of the bearer network topology, the edge node access method further includes:

[0157] The topology training data is transformed to obtain a feature matrix and an adjacency matrix; the feature matrix is subjected to data preprocessing; and the access demand training data is subjected to data preprocessing to obtain an access demand vector. The adjacency matrix is used to characterize the connection relationship between a number of switches, and the feature matrix is used to characterize the attributes contained in a number of switch nodes. Data preprocessing includes one-hot encoding and normalization. For example, the feature matrix includes the level to which a number of switch nodes belong, the device performance of a number of switch nodes, the number of edge nodes connected to a number of switch nodes, and the geographical locations of a number of switch nodes. The level to which the number of switch nodes belong and the geographical locations of the number of switch nodes are one-hot encoded; the device performance of the number of switch nodes and the number of edge nodes connected to the number of switch nodes are normalized. Data preprocessing includes standardization or normalization.

[0158] In one embodiment, the formula for normalization processing may be:

[0159] (X-mean) / std;

[0160] Where X represents an attribute in the performance of several switch node devices, mean represents the mean of all attributes in the performance of several switch node devices, and std represents the variance of all attributes in the performance of several switch node devices. Alternatively, X represents an attribute in the number of edge nodes connected to several switch nodes, mean represents the mean of all attributes in the number of edge nodes connected to several switch nodes, and std represents the variance of all attributes in the number of edge nodes connected to several switch nodes. Alternatively, X represents an attribute in the access demand data, mean represents the mean of all attributes in the access demand data, and std represents the variance of all attributes in the access demand data. The specific calculation is performed separately for each attribute (dimension), subtracting the mean from the data by attribute (per column), and dividing by the variance.

[0161] It can be understood that the standardized performance of several switch node devices and the number of edge nodes connected to several switch nodes can accelerate the convergence of the decision model, facilitating model training. Furthermore, compared to the original data, the standardized performance of several switch node devices and the number of edge nodes connected to several switch nodes improve the accuracy of the decision model, thereby improving the accuracy of newly added edge nodes accessing the bearer network switches. Standardized access demand data can accelerate the convergence of the decision model, facilitating model training. Furthermore, compared to the original data, the standardized access demand data improves the accuracy of the decision model, thereby improving the accuracy of newly added edge nodes accessing the bearer network switches.

[0162] Step A40: obtaining a model to be trained, and selecting training sample data from the access requirement training data, the topology training data, and the switch access category label data;

[0163] In this embodiment, a model to be trained is obtained, and training sample data is selected from the access requirement training data, the topology training data, and the switch access category label data. The training sample data includes at least one training sample, each training sample including access requirement data from the access requirement training data, topology data from the topology training data, and switch access category data from the switch access category label data.

[0164] Furthermore, the training sample data is divided into a training set and a test set, for example, 80% of the training sample data is divided into a training set and 20% of the training sample data is divided into a test set. The training set is used to train the model, and the test set is used to test the model.

[0165] Step A50: Iteratively train the model to be trained based on the training sample data to obtain the decision model.

[0166] In this embodiment, the model to be trained is iteratively trained based on training sample data to obtain a decision model. Specifically, a training sample is selected from the training sample data, and the access requirement data and topology data corresponding to the training sample are input into the model to be trained. Model prediction is performed to obtain a model output label. The difference between the model output label and the switch access category data corresponding to the training sample is then calculated to obtain a model loss. Based on the model loss, the model to be trained is then updated until the number of iterations of the model to be trained reaches a preset number of iterations.

[0167] It should be noted that the preset number of iterations can be set to 1500 (epochs = 1500), and the gradient descent optimization algorithm selects the adam optimizer to improve the learning speed of traditional gradient descent. It can be understood that through gradient descent, the optimal weight value that minimizes the objective function can be found, and the weight value will be learned autonomously through training. Use the training set to train so that the objective function is as small as possible, and use the test set to evaluate and verify the model after each round of training until the model converges and the weight of the model is derived.

[0168] Among them, the objective function can select the binary crossentropy loss function, which is described as follows:

[0169]

[0170] Among them, y i The switch access category data corresponding to the model output label (i.e., the real label), The model output label (i.e., predicted label) of the model to be trained, where n is the number of dimensions of the access demand data and topology data.

[0171] In this embodiment, access requirement training data for edge nodes and topology training data for the bearer network topology are obtained, and the access requirement training data and topology training data are labeled with labels specific to the switch access category to obtain switch access category label data. A model to be trained is obtained, and training sample data is selected from the access requirement training data, topology training data, and switch access category label data. Based on the training sample data, the model to be trained is iteratively trained to obtain a decision model. In this manner, based on the established and trained decision model, the bearer network switch to be accessed by the newly created edge node is automatically selected. Furthermore, because the decision model is obtained through iterative training, the efficiency and accuracy of the newly created edge node accessing the bearer network switch are improved.

[0172] The present invention also provides an edge node access device.

[0173] Reference Figure 6 , Figure 6 This is a schematic diagram of the functional modules of the first embodiment of the edge node access device of the present invention.

[0174] In this embodiment, the edge node access device includes:

[0175] An acquisition module 10 is configured to acquire access requirement data of a newly created edge node and acquire topology data of a current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent network connection relationships between switches, and the topology data is topology data of the switches;

[0176] The prediction module 20 is used to predict the access category of several switch nodes in the current bearer network topology based on the trained decision model, the access demand data and the topology data, and obtain the access category prediction results for selecting the switch to which the newly created edge node is connected based on the access category prediction results.

[0177] Among them, each virtual function module of the above-mentioned edge node access device is stored in Figure 1 The memory 1005 of the edge node access device is used to implement all functions of the edge node access program; when each module is executed by the processor 1001, the edge node access function can be implemented.

[0178] Furthermore, the decision model includes a first feature extractor, a second feature extractor and a classifier, and the prediction module 20 includes:

[0179] A first feature extraction unit is configured to perform feature extraction on the topological data based on the first feature extractor to obtain a topological feature vector;

[0180] a second feature extraction unit, configured to perform feature extraction on the access demand data based on the second feature extractor to obtain an access demand feature vector;

[0181] a vector aggregation unit, configured to aggregate the topology feature vector and the access demand feature vector to obtain an aggregated feature vector;

[0182] A prediction unit is configured to predict access categories of several switch nodes in the current bearer network topology based on the classifier and the aggregated feature vector, and obtain an access category prediction result.

[0183] Furthermore, the first feature extractor is composed of a first preset number of graph convolutional layers, and the first feature extraction unit includes:

[0184] A data conversion subunit, configured to convert the topological data to obtain a feature matrix and an adjacency matrix;

[0185] A matrix preprocessing subunit, configured to perform data preprocessing on the feature matrix;

[0186] The first feature extraction subunit is configured to perform feature extraction on the adjacency matrix and the feature matrix after data preprocessing based on the first feature extractor to obtain a topological feature vector.

[0187] Furthermore, the feature matrix includes the layers to which the switch nodes belong, the device performance of the switch nodes, the number of edge nodes connected to the switch nodes, and the geographical locations of the switch nodes. The matrix preprocessing subunit is further used to perform one-hot encoding on the layers to which the switch nodes belong and the geographical locations of the switch nodes; and to perform normalization on the device performance of the switch nodes and the number of edge nodes connected to the switch nodes.

[0188] Furthermore, the second feature extractor is composed of a second preset number of fully connected layers and discarded layers arranged alternately, and the second feature extraction unit includes:

[0189] a data preprocessing subunit, configured to perform data preprocessing on the access demand data to obtain an access demand vector;

[0190] The second feature extraction subunit is configured to perform feature extraction on the access demand vector based on the second feature extractor to obtain an access demand feature vector.

[0191] Furthermore, the decision model further includes a merging layer, and the vector aggregation unit includes:

[0192] A vector aggregation subunit is configured to aggregate the topology feature vector and the access requirement feature vector based on the merging layer to obtain an aggregated feature vector.

[0193] Furthermore, the classifier includes a first fully connected layer, and the prediction unit includes:

[0194] A prediction subunit is configured to predict access categories of several switch nodes in the current bearer network topology based on the first fully connected layer and the aggregated feature vector, and obtain an access category prediction result.

[0195] Furthermore, the edge node access device further includes:

[0196] a label tagging module, configured to obtain access requirement training data of edge nodes and topology training data of bearer network topology, and tag the access requirement training data and the topology training data with labels specific to switch access categories to obtain switch access category label data;

[0197] A sample selection module is used to obtain a model to be trained and select training sample data from the access demand training data, the topology training data and the switch access category label data;

[0198] The model training module is used to iteratively train the model to be trained based on the training sample data to obtain the decision model.

[0199] Among them, the functional implementation of each module in the above-mentioned edge node access device corresponds to each step in the above-mentioned edge node access method embodiment, and its functions and implementation processes are no longer repeated here.

[0200] The present invention also provides a computer program product, which includes a computer program. When the computer program product is executed by a processor, it implements the steps of the edge node access method as described in any of the above embodiments.

[0201] The specific embodiments of the computer program product of the present invention are basically the same as the embodiments of the above-mentioned edge node access method, and are not described in detail here.

[0202] The present invention also provides a computer-readable storage medium storing an edge node access program. When the edge node access program is executed by a processor, the steps of the edge node access method described in any of the above embodiments are implemented.

[0203] The specific embodiments of the computer-readable storage medium of the present invention are substantially the same as the embodiments of the above-mentioned edge node access method, and are not described in detail here.

[0204] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0205] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0206] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0207] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for accessing an edge node, characterized in that: The edge node access method comprises the following steps: Obtain access demand data for the newly created edge node and obtain topology data of the current bearer network topology, wherein the topology nodes of the current bearer network topology are used to represent switches, the edges of the current bearer network topology are used to represent network connection relationships between switches, and the topology data is the topology data of the switches; Based on the trained decision model, the access demand data, and the topology data, performing access category prediction on a plurality of switch nodes in the current bearer network topology to obtain an access category prediction result, so as to select a switch to be connected to the newly created edge node based on the access category prediction result; The decision model includes a first feature extractor, a second feature extractor, and a classifier, wherein the first feature extractor is composed of a first preset number of graph convolutional layers, and the second feature extractor is composed of a second preset number of fully connected layers and dropout layers alternately arranged; The step of performing access category prediction on a plurality of switch nodes in the current bearer network topology based on the trained decision model, the access demand data, and the topology data to obtain an access category prediction result includes: Based on the first feature extractor, feature extraction is performed on the topological data to obtain a topological feature vector; performing feature extraction on the access demand data based on the second feature extractor to obtain an access demand feature vector; Aggregating the topology feature vector and the access requirement feature vector to obtain an aggregated feature vector; Based on the classifier and the aggregated feature vector, access categories of several switch nodes in the current bearer network topology are predicted to obtain access category prediction results, where the access category prediction results include whether a switch is selected or not selected.

2. The edge node access method according to claim 1, wherein: The step of performing feature extraction on the topological data based on the first feature extractor to obtain a topological feature vector comprises: Performing transformation processing on the topological data to obtain a characteristic matrix and an adjacency matrix; Performing data preprocessing on the feature matrix; Based on the first feature extractor, feature extraction is performed on the adjacency matrix and the feature matrix after data preprocessing to obtain a topological feature vector.

3. The edge node access method according to claim 2, wherein: The feature matrix includes the layers to which the switch nodes belong, the device performance of the switch nodes, the number of edge nodes connected to the switch nodes, and the geographical locations of the switch nodes. The step of performing data preprocessing on the feature matrix includes: Performing one-hot encoding on the layers to which the plurality of switch nodes belong and the geographical locations of the plurality of switch nodes; The performance of the switch nodes and the number of edge nodes connected to the switch nodes are standardized.

4. The edge node access method according to claim 1, wherein: The step of extracting features from the access demand data based on the second feature extractor to obtain an access demand feature vector includes: performing data preprocessing on the access demand data to obtain an access demand vector; Based on the second feature extractor, feature extraction is performed on the access demand vector to obtain an access demand feature vector.

5. The edge node access method according to claim 1, wherein: The decision model further includes a merging layer, and the step of aggregating the topology feature vector and the access demand feature vector to obtain an aggregated feature vector includes: Based on the merging layer, aggregating the topology feature vector and the access demand feature vector to obtain an aggregated feature vector; The classifier includes a first fully connected layer, and the step of predicting the access categories of several switch nodes in the current bearer network topology based on the classifier and the aggregated feature vector to obtain the access category prediction result includes: Based on the first fully connected layer and the aggregated feature vector, access categories of several switch nodes in the current bearer network topology are predicted to obtain an access category prediction result.

6. The edge node access method according to any one of claims 1 to 5, wherein: Before the step of performing access category prediction on a plurality of switch nodes in the current bearer network topology based on the trained decision model, the access demand data, and the topology data to obtain an access category prediction result, the method further includes: Acquire access requirement training data of edge nodes and topology training data of bearer network topology, and label the access requirement training data and the topology training data with labels specific to switch access categories to obtain switch access category label data; Acquire a model to be trained, and select training sample data from the access demand training data, the topology training data, and the switch access category label data; Based on the training sample data, the model to be trained is iteratively trained to obtain the decision model.

7. An edge node access device, characterized in that: The edge node access device includes: An acquisition module, configured to acquire access demand data of a newly created edge node and acquire topology data of a current bearer network topology, wherein the topology nodes of the current bearer network topology are switches, and the edges of the current bearer network topology are network connections between switches; A prediction module is configured to perform access category prediction for a number of switch nodes in the current bearer network topology based on the trained decision model, the access demand data, and the topology data, to obtain access category prediction results for selecting a switch to be connected to the newly created edge node based on the access category prediction results; the decision model includes a first feature extractor, a second feature extractor, and a classifier, the first feature extractor being composed of a first preset number of graph convolutional layers, and the second feature extractor being composed of a second preset number of fully connected layers and dropout layers arranged alternately; The prediction module is also used to perform feature extraction on the topology data based on the first feature extractor to obtain a topology feature vector; perform feature extraction on the access demand data based on the second feature extractor to obtain an access demand feature vector; aggregate the topology feature vector and the access demand feature vector to obtain an aggregated feature vector; and predict the access categories of several switch nodes in the current bearer network topology based on the classifier and the aggregated feature vector to obtain an access category prediction result, wherein the access category prediction result includes whether the switch is selected and whether the switch is not selected.

8. An edge node access device, characterized in that: The edge node access device includes: a memory, a processor, and an edge node access program stored in the memory and executable on the processor. When the edge node access program is executed by the processor, the steps of the edge node access method according to any one of claims 1 to 6 are implemented.

9. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the edge node access method according to any one of claims 1 to 6 are implemented.

Images