Method and apparatus for evaluating access of new edge node, terminal device and product

Abstract

The application discloses a kind of new edge node access evaluation method, device, terminal equipment and product, the method includes: edge computing management platform receives the access request sent by new edge node;Obtain the bandwidth time series data used by each accessed edge node in first preset time, input to the accessed edge node bandwidth use predictor that pre-training is completed, and the predicted value of total bandwidth consumed by each accessed edge node in future second preset time is predicted;The total amount of bandwidth of bearer network is reduced by the predicted value of total bandwidth consumed that is predicted, and the remaining available total bandwidth time series of bearer network in future second preset time is obtained;The performance requirement text of new edge node is input to the new edge node access evaluator that pre-training is completed, and the remaining available total bandwidth time series are evaluated to obtain the feasibility evaluation result that new edge node accesses bearer network, and the accuracy and efficiency of new edge node access evaluation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of communication technology, in particular to a new edge node access evaluation method and device, terminal equipment and product. BACKGROUND

[0002] With the rapid development of mobile communication technology, 5G technology emerges as the times require. In the 5G era, everything is interconnected, and a large number of Internet of Things devices extend upwards. Cloud computing mode data processing and cost energy consumption will produce bottlenecks, and the user experience also requires the extension of cloud content to users. Therefore, the rapid development of MEC (Edge Computing) technology will be inevitable for technology evolution.

[0003] MEC technology provides flexible network access capability and edge computing services at the edge of the mobile network, reduces network transmission and service delivery delay, improves data security, and gives new development momentum to vertical industries. Edge computing nodes can be hierarchically deployed in city core machine rooms, important aggregation machine rooms, ordinary aggregation machine rooms, and access park machine rooms according to industry customer needs.

[0004] Currently, the evaluation of new 5G edge node access to the bearer network is mainly realized by expert experience. However, edge nodes serve industry customers and face thousands of industries, with characteristics such as demand diversity and demand burstiness. Evaluating the feasibility of new 5G edge node access to the bearer network through manual experience has the problems of low efficiency, time-consuming and laborious, and inaccurate evaluation. SUMMARY

[0005] The main purpose of the present application is to provide a new edge node access evaluation method, device, terminal equipment and product, which aims to improve the accuracy and efficiency of new edge node access evaluation.

[0006] To achieve the above purpose, the embodiment of the present application provides a new edge node access evaluation method, which is applied to an edge computing management platform, and the method comprises the following steps:

[0007] The edge computing management platform receives an access request for accessing a bearer network sent by a new edge node, and the access request carries performance requirement text of the new edge node;

[0008] According to the access request, time series data of the used bandwidth of each accessed edge node within a first preset time is obtained, and the used bandwidth time series data is input into a pre-trained accessed edge node bandwidth usage predictor to predict a predicted value of the total consumed bandwidth of each accessed edge node within a second preset time;

[0009] subtracting the predicted value of the total bandwidth consumed by each accessed edge node in the future second preset time from the total bandwidth of the bearer network to obtain a residual available total bandwidth time sequence of the bearer network in the future second preset time;

[0010] inputting the performance requirement text of the new edge node and the residual available total bandwidth time sequence of the bearer network into the pre-trained new edge node access evaluator to obtain a feasibility evaluation result of the new edge node accessing the bearer network.

[0011] Optionally, the step of predicting the total bandwidth consumed by each accessed edge node in the future second preset time comprises:

[0012] extracting, by the accessed edge node bandwidth usage predictor, time dynamic features of the used bandwidth time sequence data by using a long short-term memory layer;

[0013] encoding the time dynamic features into a plurality of fixed-length feature vectors and merging the fixed-length feature vectors, decoding the merged feature vectors, and mapping, by using the long short-term memory layer, the predicted value of the total bandwidth consumed by each accessed edge node in the future second preset time.

[0014] Optionally, the step of obtaining the feasibility evaluation result of the new edge node accessing the bearer network comprises:

[0015] extracting, by the new edge node access evaluator, text features of the performance requirement text and time sequence dynamic rules of the residual available total bandwidth time sequence of the bearer network by using a long short-term memory layer;

[0016] encoding the text features and the time sequence dynamic rules into two fixed-length feature vectors respectively and merging the two fixed-length feature vectors, decoding the merged feature vectors by using a fully connected neural network, and obtaining the feasibility evaluation result of the new edge node accessing the bearer network.

[0017] Optionally, before the step of receiving, by the edge computing management platform, an access request of a new edge node to access a bearer network, the method further comprises:

[0018] training the accessed edge node bandwidth usage predictor, and the training specifically comprises:

[0019] collecting a used bandwidth time sequence data set of each accessed edge node in a first preset time, labeling real data of used bandwidth of each accessed edge node in a next second preset time for the used bandwidth time sequence data of the accessed edge node in the first preset time, and obtaining a data set of the accessed edge node bandwidth usage predictor model.

[0020] Preprocessing data in a data set of the accessed edge node bandwidth usage predictor model;

[0021] Dividing the preprocessed data set of the accessed edge node bandwidth usage predictor model into a predictor training set and a predictor test set;

[0022] Training the accessed edge node bandwidth usage predictor model through the predictor training set;

[0023] Verifying the trained accessed edge node bandwidth usage predictor model through the predictor test set, and obtaining the final trained accessed edge node bandwidth usage predictor after the model converges.

[0024] Optionally, the edge computing management platform further includes, before the step of receiving the access request of the access bearer network sent by the newly-built edge node:

[0025] Training the newly-built edge node access evaluator, specifically including:

[0026] Collecting a performance requirement set of historical newly-built edge nodes;

[0027] Subtracting the real data of the bandwidth used by the accessed edge node in the next second preset time from the total bandwidth of the bearer network to obtain a remaining available total bandwidth time series set of the bearer network in the future second preset time;

[0028] Obtaining a set of feasibility evaluation results of the newly-built edge node accessing the bearer network;

[0029] Taking the performance requirement set, the remaining available total bandwidth time series set, and the set of feasibility evaluation results as a data set of the newly-built edge node access evaluator model;

[0030] Preprocessing data in the data set of the newly-built edge node access evaluator model;

[0031] Dividing the preprocessed data set of the newly-built edge node access evaluator model into an evaluator training set and an evaluator test set;

[0032] Training the newly-built edge node access evaluator model through the evaluator training set;

[0033] Verifying the trained newly-built edge node access evaluator model through the evaluator test set, and obtaining the final trained newly-built edge node access evaluator after the model converges.

[0034] Optionally, the accessed edge node bandwidth usage predictor and the newly-built edge node access evaluator are both composed of a codec neural network.

[0035] Optionally, before the step of inputting the performance requirement text of the new edge node and the remaining available total bandwidth time sequence of the bearer network into the pre-trained new edge node access evaluator, the method further comprises the steps that:

[0036] preprocessing the performance requirement text of the new edge node and the remaining available total bandwidth time sequence of the bearer network; and / or

[0037] After the step of obtaining the used bandwidth time sequence data of each accessed edge node within the first preset time according to the access request, the method further comprises the steps that:

[0038] preprocessing the used bandwidth time sequence data.

[0039] Embodiments of the present application also provide a new edge node access evaluation device, which comprises:

[0040] a request receiving module, configured to receive an access request of a new edge node for accessing a bearer network, wherein the access request carries a performance requirement text of the new edge node;

[0041] a prediction module, configured to obtain used bandwidth time sequence data of each accessed edge node within a first preset time according to the access request, input the used bandwidth time sequence data into a pre-trained accessed edge node bandwidth usage predictor, and predict a predicted value of total bandwidth consumed by each accessed edge node within a second preset time in the future; subtract the predicted value of total bandwidth consumed by each accessed edge node within the second preset time in the future from a total amount of bandwidth of the bearer network to obtain a remaining available total bandwidth time sequence of the bearer network within the second preset time in the future;

[0042] an evaluation module, configured to input the performance requirement text of the new edge node and the remaining available total bandwidth time sequence of the bearer network into a pre-trained new edge node access evaluator, and obtain a feasibility evaluation result of the new edge node accessing the bearer network.

[0043] Embodiments of the present application also provide a terminal device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is executed by the processor to implement the new edge node access evaluation method.

[0044] Embodiments of the present application also provide a computer program product, which comprises a computer program, wherein the computer program is executed by a processor to implement the new edge node access evaluation method.

[0045] The proposed methods, apparatus, terminal devices, and products for evaluating the access of newly established edge nodes in this invention receive access requests from newly established edge nodes to access the bearer network through an edge computing management platform. These access requests carry performance requirement text for the newly established edge nodes. Based on the access requests, the method acquires the bandwidth time-series data used by each already connected edge node within a first preset time period. This bandwidth time-series data is then input into a pre-trained bandwidth usage predictor for already connected edge nodes to predict the total bandwidth consumed by each already connected edge node within a second preset time period. Finally, the method subtracts the predicted second preset time period from the total bandwidth of the bearer network. The predicted total bandwidth consumed by each connected edge node within a given time period is used to obtain the time series of the remaining available total bandwidth of the bearer network within the next second preset time period. The performance requirements text of the newly established edge node and the time series of the remaining available total bandwidth of the bearer network are input into the pre-trained newly established edge node access evaluator to evaluate the feasibility assessment result of the newly established edge node accessing the bearer network. Thus, by combining the bandwidth usage predictor of the connected edge nodes with the newly established edge node access evaluator, the feasibility of the newly established edge node accessing the bearer network is evaluated, improving the accuracy and efficiency of the operation and maintenance personnel's evaluation of the newly established edge node access. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the structure of an LSTM neuron in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram of the functional modules of the terminal equipment to which the newly built edge node access evaluation device of this invention belongs;

[0048] Figure 3 This is a flowchart illustrating an exemplary embodiment of the edge node access evaluation method of the present invention;

[0049] Figure 4 This is a flowchart illustrating another exemplary embodiment of the edge node access evaluation method of the present invention;

[0050] Figure 5 This is a schematic diagram of the bandwidth usage predictor model for connected edge nodes in an embodiment of the present invention;

[0051] Figure 6 This is a schematic diagram of a newly created edge node access evaluator model in an embodiment of the present invention.

[0052] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0053] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0054] The main solution of the embodiment of the present application is that an edge computing management platform receives an access request of an access bearer network sent by a newly-built edge node, the access request carrying performance requirement text of the newly-built edge node; according to the access request, time series data of a used bandwidth of each accessed edge node within a first preset time is obtained, the used bandwidth time series data being input into a pre-trained accessed edge node bandwidth usage predictor to predict a predicted value of total consumed bandwidth of each accessed edge node within a second preset future time; the total bandwidth of the bearer network is subtracted by the predicted value of the total consumed bandwidth of each accessed edge node within the second preset future time to obtain a time series of a remaining total available bandwidth of the bearer network within the second preset future time; the performance requirement text of the newly-built edge node and the time series of the remaining total available bandwidth of the bearer network are input into a pre-trained newly-built edge node access evaluator to evaluate a feasibility evaluation result of the newly-built edge node accessing the bearer network, thereby, the feasibility of the newly-built edge node accessing the bearer network is evaluated by the accessed edge node bandwidth usage predictor in combination with the newly-built edge node access evaluator, and the accuracy and efficiency of the operation and maintenance personnel for the newly-built edge node access evaluation are improved.

[0055] The technical terms involved in the embodiment of the present application are as follows:

[0056] MEC (Mobile Edge Computing), mobile edge computing.

[0057] The coding and decoding neural network is a way of organizing a recurrent neural network, mainly used to solve sequence prediction problems containing multiple inputs or multiple outputs, and includes an encoder and a decoder. The encoder is responsible for encoding the input sequence word by word into a fixed-length vector, i.e. a context vector; the decoder is responsible for reading the context vector output by the encoder and generating an output sequence. In the embodiment of the present application, the encoder and the decoder are both composed of long short-term memory neurons.

[0058] Among them, long short-term memory (LSTM, long short-term memory) is a special type of recurrent neural network that can remember long-term information by controlling the time value stored in the cache, and is suitable for time series prediction. Each neuron has four inputs and one output, and each neuron has a Cell to store memory values. The LSTM neuron is as shown in Figure 1 The calculation formula involved is as follows:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] Y t = sigma(W'h t ) (7)

[0066] Wherein, each LSTM neuron contains three gates: forget gate, input gate, output gate. Formula (1) represents the forget gate, in formula (2), (3), new information is added, formula (4) fuses new information and old information, formula (5), (6) output the information that the current LSTM unit has learned about the next timestamp.

[0067] The long short-term memory neural network has good effect on long time sequence learning, each connection line in the LSTM unit contains corresponding weight, wherein, x t represents an input vector, h t represents a hidden state, C t represents the state of a neuron at t time, Y t represents the output of the neuron, W is a trainable weight matrix, b is a bias vector, sigma represents a sigmoid activation function, o t represents the output result of the output gate, f t represents

[0068] ^

[0069] The output result of the forget gate, i t represents the output result of the input gate, C t respectively represent the temporary state at the current moment.

[0070] The embodiment of the application considers that in the prior related scheme, the feasibility of the newly-built 5G edge node accessing the bearer network is evaluated in an artificial experience manner, and the related scheme has problems of low efficiency, time-consuming and laborious, inaccurate evaluation and the like.

[0071] Therefore, the embodiment of the present application proposes a solution, by building two different codec neural networks to respectively constitute an accessed edge node bandwidth usage predictor and a newly-built edge node access evaluator, the feasibility of the newly-built edge node, especially the 5G edge node, accessing the bearer network can be automatically evaluated, and the accuracy and efficiency of the operation and maintenance personnel in the access evaluation of the newly-built edge node are improved. The embodiment of the present application relates to communication core networks and artificial intelligence technologies, in particular to deep learning and 5G edge computing.

[0072] In the embodiment of the present application, the accessed edge node bandwidth usage predictor and the newly-built edge node access evaluator are configured, wherein the accessed edge node bandwidth usage predictor is composed of a codec neural network, the predictor extracts the time dynamic characteristics of the time series data of the bandwidth used by each accessed edge node in the last T time by using a long short-term memory layer, encodes the time series data into N fixed-length feature vectors by using an encoder, decodes the fixed-length feature vectors after merging, and finally maps the prediction value of the total bandwidth consumed by each accessed edge node in the future M time by using a long short-term memory layer of a decoder.

[0073] The newly-built edge node access evaluator is composed of a codec neural network, the evaluator extracts the text features of the performance requirements of the newly-built edge node i by using a long short-term memory layer, and extracts the time series dynamic rules of the remaining total bandwidth sequence of the bearer network in the future M time output by the last predictor, encodes the two fixed-length feature vectors into one feature vector after merging, decodes the feature vector by using a fully connected neural network, and finally outputs the feasibility evaluation result of the newly-built edge node i accessing the bearer network. The result is fed back to the edge computing management platform, thereby improving the accuracy and efficiency of the operation and maintenance personnel in the access evaluation of the newly-built edge node.

[0074] Specifically, referring to Figure 2 , Figure 2 is a functional module schematic diagram of a terminal device to which the newly-built edge node access evaluation device of the present application belongs. The newly-built edge node access evaluation device can be a device independent of the terminal device and capable of realizing the access evaluation of the newly-built edge node, which can be carried on the terminal device in the form of hardware or software. The terminal device can be a smart mobile terminal such as a mobile phone or a tablet computer, and can also be a network device such as a server.

[0075] In the embodiment, the terminal device to which the newly-built edge node access evaluation device belongs at least includes an output module 110, a processor 120, a memory 130, and a communication module 140.

[0076] The memory 130 stores an operating system and a new edge node access evaluation program; the output module 110 can be a display screen, a loudspeaker, etc. The communication module 140 can include a WIFI module, a mobile communication module, a Bluetooth module, etc., and communicates with external devices or servers through the communication module 140.

[0077] As an embodiment, the new edge node access evaluation program in the memory 130 is executed by the processor to implement the following steps:

[0078] An access request for accessing a bearing network is received from a new edge node through an edge computing management platform, and the access request carries performance requirement text of the new edge node;

[0079] According to the access request, time series data of used bandwidth of each accessed edge node within a first preset time is obtained, and the time series data of used bandwidth is input into a pre-trained accessed edge node bandwidth usage predictor to predict a predicted value of total consumed bandwidth of each accessed edge node within a second preset future time;

[0080] The total bandwidth of the bearing network is subtracted from the predicted value of the total consumed bandwidth of each accessed edge node within the second preset future time to obtain a time series of remaining available total bandwidth of the bearing network within the second preset future time;

[0081] The performance requirement text of the new edge node and the time series of remaining available total bandwidth of the bearing network are input into a pre-trained new edge node access evaluator to obtain a feasibility evaluation result of the new edge node accessing the bearing network.

[0082] Further, the new edge node access evaluation program in the memory 130 is executed by the processor to further implement the following steps:

[0083] The time dynamic characteristics of the time series data of used bandwidth are extracted by the long short-term memory layer through the accessed edge node bandwidth usage predictor;

[0084] The time dynamic characteristics are encoded into a plurality of fixed-length feature vectors and merged, the merged feature vectors are decoded, and the predicted value of total consumed bandwidth of each accessed edge node within the second preset future time is mapped through the long short-term memory layer.

[0085] Further, the new edge node access evaluation program in the memory 130 is executed by the processor to further implement the following steps:

[0086] The text features and the timing dynamic law are respectively encoded to obtain two fixed-length feature vectors and are merged, and the merged feature vector is decoded by a fully connected neural network to obtain a feasibility evaluation result of the new edge node accessing the bearer network.

[0087] The text features and the timing dynamic law are respectively encoded to obtain two fixed-length feature vectors and are merged, and the merged feature vector is decoded by a fully connected neural network to obtain a feasibility evaluation result of the new edge node accessing the bearer network.

[0088] Further, the new edge node access evaluation program in the memory 130 is also implemented when executed by the processor to implement the following steps:

[0089] The trained edge node bandwidth usage predictor is specifically obtained by:

[0090] The historical bandwidth time series data set of each edge node accessed in the first preset time is collected, the used bandwidth time series data of each edge node in the first preset time is labeled with the real data of the used bandwidth of the next second preset time of the edge node, and a data set of the edge node bandwidth usage predictor model is obtained;

[0091] The data in the data set of the edge node bandwidth usage predictor model is preprocessed;

[0092] The preprocessed data set of the edge node bandwidth usage predictor model is divided into a predictor training set and a predictor test set;

[0093] The edge node bandwidth usage predictor model is trained through the predictor training set;

[0094] The trained edge node bandwidth usage predictor model is verified through the predictor test set, and after the model converges, the finally trained edge node bandwidth usage predictor is obtained.

[0095] Further, the new edge node access evaluation program in the memory 130 is also implemented when executed by the processor to implement the following steps:

[0096] The trained edge node bandwidth usage predictor is specifically obtained by:

[0097] The performance requirement set of the historical new edge node is collected;

[0098] The total bandwidth of the bearer network is subtracted from the real data of the used bandwidth of the next second preset time of the edge node accessed, and a residual available total bandwidth time series set of the bearer network in the future second preset time is obtained;

[0099] obtaining a set of feasibility evaluation results of the new edge node accessing the bearer network;

[0100] taking the set of performance requirements, the set of remaining available total bandwidth time series, and the set of feasibility evaluation results as a data set of the new edge node access evaluator model;

[0101] preprocessing data in the data set of the new edge node access evaluator model;

[0102] dividing the data set of the new edge node access evaluator model after preprocessing into an evaluator training set and an evaluator test set;

[0103] training the new edge node access evaluator model through the evaluator training set;

[0104] verifying the trained new edge node access evaluator model through the evaluator test set, and obtaining a final trained new edge node access evaluator after model convergence.

[0105] Further, the new edge node access evaluation program in the memory 130, when executed by the processor, also implements the following steps:

[0106] preprocessing the performance requirement text of the new edge node and the remaining available total bandwidth time series of the bearer network; and / or

[0107] The step of obtaining the used bandwidth time series data of each edge node accessed within a first preset time according to the access request further comprises:

[0108] preprocessing the used bandwidth time series data;

[0109] The embodiment receives an access request of an access bearer network sent by a newly-built edge node through the edge computing management platform, the access request carrying a performance requirement text of the newly-built edge node; according to the access request, time sequence data of a used bandwidth of each accessed edge node within a first preset time is acquired, the used bandwidth time sequence data is input into a pre-trained accessed edge node bandwidth usage predictor, and a predicted value of total consumed bandwidth of each accessed edge node within a future second preset time is predicted; the total bandwidth of the bearer network is subtracted by the predicted value of the total consumed bandwidth of each accessed edge node within the future second preset time, and a residual available total bandwidth time sequence of the bearer network within the future second preset time is obtained; the performance requirement text of the newly-built edge node and the residual available total bandwidth time sequence of the bearer network are input into a pre-trained newly-built edge node access evaluator, and a feasibility evaluation result of the newly-built edge node accessing the bearer network is evaluated, thereby evaluating the feasibility of the newly-built edge node accessing the bearer network through the accessed edge node bandwidth usage predictor and the newly-built edge node access evaluator, and improving the accuracy and efficiency of the operation and maintenance personnel for the newly-built edge node access evaluation.

[0110] Based on the terminal device architecture but not limited to the above-mentioned architecture, the method embodiment of the present application is proposed.

[0111] Reference Figure 3 , Figure 3 The flowchart of an exemplary embodiment of the newly-built edge node access evaluation method of the present application is shown. The method is applied to an edge computing management platform, and the method comprises the following steps:

[0112] In step S101, the edge computing management platform receives an access request of an access bearer network sent by a newly-built edge node, the access request carrying a performance requirement text of the newly-built edge node;

[0113] Specifically, in the embodiment, the feasibility of the newly-built edge node accessing the bearer network is evaluated through the edge computing management platform, and in other embodiments, a corresponding network device, server or network platform can also be used to evaluate the feasibility of the newly-built edge node accessing the bearer network, and the embodiment is exemplified by the edge computing management platform, but this does not constitute a limitation.

[0114] In specific implementation, when the newly-built edge node accesses the bearer network, an access request of the access bearer network is submitted to the edge computing management platform, and the performance requirement text of the newly-built edge node is carried in the request message, wherein the performance requirement text includes bandwidth, delay, application scenario, application industry, SLA (Service-Level Agreement) level, etc.

[0115] In the embodiment, the performance requirement text is used for subsequent input into the pre-trained newly-built edge node access evaluator to evaluate the feasibility evaluation result of the newly-built edge node accessing the bearer network.

[0116] In step S102, the used bandwidth time series data of each accessed edge node in the first preset time is obtained according to the access request, and the used bandwidth time series data is input into the pre-trained accessed edge node bandwidth usage predictor to predict the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time.

[0117] After the edge computing management platform receives the access request of the newly-built edge node accessing the bearer network, the used bandwidth time series data of each accessed edge node in the recent first preset time is obtained, the used bandwidth time series data in the recent first preset time is input into the pre-trained accessed edge node bandwidth usage predictor, and the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time is predicted by the accessed edge node bandwidth usage predictor.

[0118] Specifically, as an implementation manner, after the edge computing management platform receives the access request of the newly-built edge node accessing the bearer network, the used bandwidth time series data of each accessed edge node in the recent first preset time is obtained, the used bandwidth time series data in the recent first preset time is input into the pre-trained accessed edge node bandwidth usage predictor, and the time dynamic characteristics of the used bandwidth time series data are extracted by the long short-term memory layer; the time dynamic characteristics are encoded into a plurality of fixed-length feature vectors and combined, the combined feature vectors are decoded, and the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time is mapped by the long short-term memory layer.

[0119] More specifically, as another implementation manner, after the edge computing management platform receives the access request of the newly-built edge node accessing the bearer network and obtains the used bandwidth time series data of each accessed edge node in the first preset time, the used bandwidth time series data of each accessed edge node in the first preset time can also be preprocessed.

[0120] The preprocessing can be standardization processing, and each dimension of the data can be processed respectively during processing calculation, and the processing process includes: subtracting the mean value of the data according to the attribute (according to the column) and dividing by the variance. By standardizing the used bandwidth time series data, the accuracy of the newly-built edge node access feasibility evaluation can be improved.

[0121] The specific implementation process can be as follows:

[0122] After receiving the access request of the access bearer network sent by the newly-built edge node, the edge computing management platform acquires the time series data of the bandwidth used by each accessed edge node in the last T time (first preset time) and sends the data to a pre-configured data preprocessing module for standardization processing.

[0123] The data after the standardization processing is input into an accessed edge node bandwidth usage predictor composed of an encoding-decoding neural network which has been pre-trained.

[0124] The accessed edge node bandwidth usage predictor extracts the time dynamic characteristics of the time series data of the bandwidth used by each accessed edge node in the last T time (first preset time) by using a long short-term memory layer, encodes the characteristics into N (where N is a positive integer) fixed-length feature vectors, merges the obtained feature vectors, decodes the merged feature vectors, and finally maps out the predicted values of the total bandwidth consumed by each accessed edge node in the future M time (second preset time) by the decoder through the long short-term memory layer.

[0125] In step S103, the total bandwidth of the bearer network is reduced by the predicted values of the total bandwidth consumed by each accessed edge node in the future second preset time to obtain the residual available total bandwidth time series of the bearer network in the future second preset time.

[0126] The accessed edge node bandwidth usage predictor reduces the total bandwidth of the bearer network by the predicted values of the total bandwidth consumed by each accessed edge node in the future second preset time to obtain the residual available total bandwidth time series of the bearer network in the future second preset time, and provides the residual available total bandwidth time series to the edge computing management platform, so as to evaluate the feasibility of the newly-built edge node accessing the bearer network by combining the residual available total bandwidth time series of the bearer network with the performance requirement text of the newly-built edge node, and obtain the feasibility evaluation result.

[0127] In one embodiment, the accessed edge node bandwidth usage predictor can also provide the total bandwidth of the bearer network to the edge computing management platform, and the edge computing management platform reduces the total bandwidth of the bearer network by the predicted values of the total bandwidth consumed by each accessed edge node in the future second preset time to obtain the residual available total bandwidth time series of the bearer network in the future second preset time.

[0128] In step S104, the performance requirement text of the newly-built edge node and the residual available total bandwidth time series of the bearer network are input into a pre-trained newly-built edge node access evaluator to evaluate the feasibility of the newly-built edge node accessing the bearer network and obtain the feasibility evaluation result.

[0129] Specifically, as an implementation, the edge computing management platform extracts the performance requirement text of the new edge node carried by the access request, inputs the performance requirement text of the new edge node and the time sequence of the remaining total bandwidth of the bearer network into a pre-trained new edge node access evaluator.

[0130] Through the new edge node access evaluator, the text features of the performance requirement text and the time sequence dynamic rules of the time sequence of the remaining total bandwidth of the bearer network are extracted by using a long short-term memory layer; the text features and the time sequence dynamic rules are respectively encoded to obtain two fixed-length feature vectors and are merged, and the merged feature vectors are decoded by a fully connected neural network to obtain the feasibility evaluation result of the new edge node accessing the bearer network.

[0131] Further, before the performance requirement text of the new edge node and the time sequence of the remaining total bandwidth of the bearer network are input into the pre-trained new edge node access evaluator, the performance requirement text of the new edge node and the time sequence of the remaining total bandwidth of the bearer network can also be preprocessed.

[0132] The preprocessing of the performance requirement text of the new edge node includes:

[0133] The performance requirement text of the new edge node is text cleaned and sequenced. All punctuation marks are removed, if the text is Chinese, the text is segmented, if the text is English, the letters are unified to lowercase, each word is tokenized, so that each piece of text is converted into a sequence of index numbers, and the sequence that does not reach the maximum text length is zero-padded.

[0134] More specifically, as another implementation, the performance requirement text of the new edge node i is obtained from the edge computing management platform, and the text is processed by the data preprocessing module to be an integer sequence.

[0135] Meanwhile, the total bandwidth of the bearer network is respectively reduced by the predicted values of the total bandwidth consumed by each accessed edge node in the future M time (second preset time) output by the bandwidth usage predictor of the accessed edge node, to obtain the time sequence of the remaining total bandwidth of the bearer network in the future M time.

[0136] The time sequence of the remaining total bandwidth of the bearer network in the future M time is preprocessed, and the preprocessed performance requirement text and the time sequence of the remaining total bandwidth of the bearer network in the future M time are input into the pre-trained new edge node access evaluator composed of an encoding and decoding neural network.

[0137] The newly-built edge node access evaluator extracts the text features of the performance requirement text of the newly-built edge node i by using a long short-term memory layer, and extracts the time sequence dynamic law of the total residual available bandwidth time sequence of the bearing network in the future M time output by the accessed edge node bandwidth usage predictor, and encodes the two fixed length feature vectors into one feature vector, and then decodes the feature vector through a fully connected neural network; finally, the feasibility evaluation result y of the newly-built edge node i accessing the bearing network is output i , wherein y i = 1 represents accessibility, and y i = 0 represents inaccessibility.

[0138] The newly-built edge node access evaluator feeds back the feasibility evaluation result to the edge computing management platform.

[0139] In the embodiment, the edge computing management platform receives the access request of the newly-built edge node accessing the bearing network, and the access request carries the performance requirement text of the newly-built edge node. According to the access request, the time sequence data of the used bandwidth of each accessed edge node in the first preset time is obtained, and the time sequence data of the used bandwidth is input into the pre-trained accessed edge node bandwidth usage predictor to predict the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time. The total bandwidth of the bearing network is subtracted from the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time to obtain the residual available total bandwidth time sequence of the bearing network in the future second preset time. The performance requirement text of the newly-built edge node and the residual available total bandwidth time sequence of the bearing network are input into the pre-trained newly-built edge node access evaluator to evaluate the feasibility evaluation result of the newly-built edge node accessing the bearing network. Thus, the feasibility of the newly-built edge node accessing the bearing network is evaluated by using two different coding and decoding neural networks to form the accessed edge node bandwidth usage predictor and the newly-built edge node access evaluator, thereby improving the accuracy and efficiency of the operation and maintenance personnel for the newly-built edge node access evaluation.

[0140] Referring to Figure 4 , Figure 4 is a flowchart of another exemplary embodiment of the newly-built edge node access evaluation method of the application. In the embodiment, before the edge computing management platform receives the access request of the newly-built edge node accessing the bearing network in step S101, the following steps are further included:

[0141] Step S1001: training the accessed edge node bandwidth usage predictor;

[0142] Step S1002: training the newly-built edge node access evaluator.

[0143] In the embodiment, the accessed edge node bandwidth usage predictor and the new edge node access evaluator are constituted by a codec neural network.

[0144] As an implementation form, the accessed edge node bandwidth usage predictor is trained, and specifically can include:

[0145] A data set of a used bandwidth time series of each accessed edge node in a first preset time is collected, and a real data of a used bandwidth of each accessed edge node in a next second preset time is labeled to a used bandwidth time series of the accessed edge node in the first preset time, to obtain a data set of an accessed edge node bandwidth usage predictor model;

[0146] The data in the data set of the accessed edge node bandwidth usage predictor model is preprocessed;

[0147] The preprocessed data set of the accessed edge node bandwidth usage predictor model is divided into a predictor training set and a predictor test set;

[0148] The accessed edge node bandwidth usage predictor model is trained through the predictor training set;

[0149] The trained accessed edge node bandwidth usage predictor model is verified through the predictor test set, and after the model converges, a final trained accessed edge node bandwidth usage predictor is obtained.

[0150] Thus, through the above scheme, the accessed edge node bandwidth usage predictor is trained, so that in the process of evaluating the feasibility of the new edge node access to the bearer network, the predicted value of the total consumed bandwidth of each accessed edge node in the future second preset time is predicted through the accessed edge node bandwidth usage predictor.

[0151] As an implementation form, the new edge node access evaluator is trained, and specifically can include:

[0152] A performance requirement set of a historical new edge node is collected;

[0153] The total bandwidth of the bearer network is subtracted from the labeled real data of the used bandwidth of the accessed edge node in the next second preset time, to obtain a residual available total bandwidth time series of the bearer network in the future second preset time;

[0154] An evaluation result set of the feasibility of the new edge node to access the bearer network is obtained;

[0155] The performance requirement set, the remaining available total bandwidth time sequence set, and the feasibility evaluation result set are taken as a data set of the new edge node access evaluator model;

[0156] Data in the data set of the new edge node access evaluator model is preprocessed.

[0157] The preprocessed data set of the new edge node access evaluator model is divided into an evaluator training set and an evaluator test set.

[0158] The new edge node access evaluator model is trained through the evaluator training set.

[0159] The trained new edge node access evaluator model is verified through the evaluator test set, and after the model converges, a final trained new edge node access evaluator is obtained.

[0160] Thus, through the above scheme, the new edge node access evaluator is trained, so that the new edge node access evaluator is used to evaluate the feasibility of the new edge node accessing the bearer network.

[0161] More specifically, the process of training the already-accessed edge node bandwidth usage predictor and the new edge node access evaluator in the embodiment includes a data collection and preprocessing process and a model building and training process. Wherein:

[0162] The data collection and preprocessing process can include:

[0163] The time sequence data set of the bandwidth used by each already-accessed edge node in the last T time is collected from the edge computing management platform (MEPM), and the real data of the bandwidth used by the already-accessed edge node in the next M time is labeled for each already-accessed edge node in the T time, as a data set of the already-accessed edge node bandwidth usage predictor model.

[0164] At the same time, the performance requirement text set of the historical new edge node is collected from the edge computing management platform, and the real data of the bandwidth used by the already-accessed edge node in the next M time is labeled by the last model (the already-accessed edge node bandwidth usage predictor model), to form a remaining available total bandwidth time sequence set of the bearer network in the future M time.

[0165] Then, the feasibility evaluation result of the new edge node accessing the bearer network is artificially labeled as a data set of the new edge node access evaluator model.

[0166] The standardization processing is performed on the time series data set of the bandwidth used by each accessed edge node in the recent T time: (X-mean) / std. The data is subtracted by the mean value and divided by the variance according to each dimension during the calculation. Through the standardization processing of the data, the convergence speed of the model can be improved, and the accuracy of the model can be improved.

[0167] Then, the performance requirement text set of the newly built edge node is text cleaned and serialized. All punctuation marks are removed, if the text is Chinese, the text is segmented, if the text is English, the letters are unified to lowercase, and each word is tokenized at the same time, so that each piece of text is converted into a piece of index number, and the sequence that does not reach the maximum text length is zero-padded. Take the longest length L as the index sequence length, and take the dictionary size as requirement_vocab_size.

[0168] Wherein, the artificial labeling of the feasibility evaluation result y of the newly built edge node accessing the bearer network i , y i = 1 represents accessibility, y i = 0 represents inaccessibility.

[0169] Finally, the total data set obtained is divided into a training set and a test set. As an implementation manner, 90% of the total data set can be divided into a training set, and 10% of the total data set can be divided into a test set. The training set is used for training the model, and the test set is used for testing the model.

[0170] For the model building and training process, the following scheme can be adopted:

[0171] The embodiment scheme needs to build and train two models: an accessed edge node bandwidth usage predictor and a newly built edge node access evaluator.

[0172] For the accessed edge node bandwidth usage predictor, the accessed edge node bandwidth usage predictor is composed of a coding and decoding neural network. The accessed edge node bandwidth usage predictor extracts the time dynamic characteristics of the time series data of the bandwidth used by each accessed edge node in the recent T time by using a long short-term memory layer, encodes the time series data into N fixed length feature vectors, and decodes the N fixed length feature vectors after merging. The decoder finally maps out the prediction value of the total bandwidth consumed by each edge node in the future M time through the long short-term memory layer. The built accessed edge node bandwidth usage predictor model is as shown in Figure 5 .

[0173] Wherein, the encoding part contains N branches, N is the number of accessed edge nodes, and each branch contains two layers:

[0174] The first layer is an input layer: the indexed accessed edge nodes 1-N are inputted into the input layer, and the time series data of the bandwidth used in the last T time is inputted into the input layer, and thus the shape of the output data of the layer is (None, T) respectively;

[0175] The second layer is an LSTM encoding layer: each layer contains 128 LSTM neurons, and the activation function is set to "relu", and the shape of the output data of the layer is (None, T, 128) respectively, and the encoding is a fixed-length context vector;

[0176] The third layer is a merge layer (concatenate): N fixed-length context vectors are spliced and merged into one fixed-length context vector h in the column dimension;

[0177] The fourth layer is an LSTM decoding layer: it contains 128 LSTM neurons, and the activation function is set to "relu";

[0178] The fifth layer is a fully connected layer (Dense): the number of neurons is set to M, that is, the predicted value of the total bandwidth consumed by each edge node in the future M time is outputted, represented as {z1, z2,..., zM}, and the activation function is set to "relu".

[0179] The model is trained for 1000 epochs (epochs = 1000), the batch size is set to 32 (batch_size = 32), and the mean squared error MSE (Mean Squared Error) is selected as the loss function, that is, the objective function (loss ='mean_squared_error'):

[0180]

[0181] Wherein, loss represents the loss function, y i represents the true value.

[0182] Wherein, the gradient descent optimization algorithm selects the adam optimizer to improve the learning speed of the traditional gradient descent (optimizer = 'adam'). The neural network can find the optimal weight value that minimizes the objective function through gradient descent, and the neural network will automatically learn the weight value through training. The training set is used for training, so that the objective function is as small as possible, and the test set is used to evaluate and verify the model after each training. After the model converges, the weight of the model is derived, and the model training is completed.

[0183] For the newly-built edge node access evaluator, the newly-built edge node access evaluator is composed of a codec neural network, which extracts text features of the performance requirement text of the newly-built edge node i by using a long short-term memory layer, and extracts a time sequence dynamic rule of a total bandwidth sequence of the residual carrying network in the next M time output by the last predictor, and encodes the two fixed-length feature vectors into one feature vector, and then decodes the feature vector through a full-connection neural network, and finally outputs a feasibility evaluation result of the newly-built edge node i accessing the carrying network.

[0184] As shown in the following table, the newly-built edge node access evaluator model includes: Figure 6

[0185] Branch 1:

[0186] The first layer is an embedding layer: input the performance requirement text sequence {s1 i , s2 i ,..., s L i} of the newly-built edge node i, convert each word into a vector by using word embedding, the input data dimension is requirement_vocab_size, the output is set to convert the word into a 128-dimensional spatial vector, the input sequence length is L, and therefore the shape of the output data of this layer is (None, L, 128). The role of this layer is to map the input word into a vector, and convert the index of each word into a fixed shape vector of 128 dimensions;

[0187] The second layer is an LSTM encoding layer: containing 128 LSTM neurons, the activation function is set to "relu", and the shape of the output data of this layer is (None, L, 128) and (None, n, 128) respectively, and is encoded into a fixed-length context vector;

[0188] Branch 2:

[0189] The first layer is an input layer: input the standardized total bandwidth sequence {b-z1, b-z2,..., b-z M} of the residual carrying network in the next M time;

[0190] The second layer is an LSTM encoding layer: containing two parallel LSTM layers, each layer containing 128 LSTM neurons, and the activation function is set to "relu", and the shape of the output data of this layer is (None, L, 128) and (None, n, 128) respectively, and is encoded into two fixed-length context vectors;

[0191] ​The third layer is a concatenate layer: two fixed-length context vectors output by the above two branches are spliced and merged into one fixed-length context vector h according to the column dimension;

[0192] The fourth layer is an LSTM decoding layer: containing 128 LSTM neurons, and the activation function is set to "relu";

[0193] The fifth layer is a full connection (Dense) layer (output layer): the number of neurons is set to 1, and the output is the feasibility evaluation result y of the newly built edge node i accessing the bearer network i , y i = 1 represents accessibility, and y i = 0 represents inaccessibility. The activation function is set to "sigmoid".

[0194] The error between the prediction result and the correct result of the above model is calculated, and the training target is to minimize the error. The objective function is selected as a binary_crossentropy two-class logarithmic loss function:

[0195]

[0196] Where, loss represents the loss function, and y i represents the true value.

[0197] The number of training rounds is set to 1500 (epochs = 1500), and the gradient descent optimization algorithm selects the adam optimizer to improve the learning speed of the traditional gradient descent (optimizer = 'adam'). The neural network can find the optimal weight value that minimizes the objective function through gradient descent, and the neural network will automatically learn the weight value through training. The training set is used for training, so that the objective function is as small as possible, and the test set is used to evaluate and verify the model after each training. After the model converges, the weight of the model is derived, and the model training is completed.

[0198] Compared with the prior art, the embodiment of the application builds two different codec neural networks to respectively constitute an accessed edge node bandwidth usage predictor and a newly-built edge node access evaluator, and trains them, wherein the accessed edge node bandwidth usage predictor extracts time dynamic characteristics of time series data of bandwidth used by each accessed edge node in the last T time by using a long short-term memory layer, encodes the time dynamic characteristics into N fixed-length feature vectors, merges the feature vectors, and decodes the merged feature vectors, and the decoder finally maps out predicted values of total bandwidth consumed by each edge node in the future M time through the long short-term memory layer; wherein the newly-built edge node access evaluator extracts text features of performance requirements of the newly-built edge node i and extracts time series dynamic rules of a time series of total bandwidth remaining available in the future M time output by the last predictor, encodes the text features and the time series dynamic rules into two fixed-length feature vectors respectively, merges the two feature vectors into one feature vector, decodes the feature vector through a fully connected neural network, and finally outputs a feasibility evaluation result of the newly-built edge node i accessing the bearer network, and feeds back the result to an edge computing management platform, thereby improving the accuracy and efficiency of the newly-built edge node access evaluation by an operation and maintenance personnel.

[0199] In addition, the embodiment of the application also provides a newly-built edge node access evaluation device, which comprises:

[0200] A request receiving module is configured to receive an access request of a newly-built edge node accessing a bearer network, wherein the access request carries a performance requirement text of the newly-built edge node.

[0201] A prediction module is configured to acquire time series data of bandwidth used by each accessed edge node in a first preset time according to the access request, input the time series data of bandwidth used into a pre-trained accessed edge node bandwidth usage predictor, and predict to obtain predicted values of total bandwidth consumed by each accessed edge node in a future second preset time; subtract the predicted values of total bandwidth consumed by each accessed edge node in the future second preset time from a total amount of bandwidth of the bearer network to obtain a time series of total bandwidth remaining available in the future second preset time of the bearer network.

[0202] An evaluation module is configured to input the performance requirement text of the newly-built edge node and the time series of total bandwidth remaining available of the bearer network into a pre-trained newly-built edge node access evaluator, and evaluate to obtain a feasibility evaluation result of the newly-built edge node accessing the bearer network.

[0203] The principle and implementation process of the newly-built edge node access evaluation are realized in the embodiment, and please refer to each embodiment described above, which will not be repeated here.

[0204] In addition, the embodiment of the present application further provides a terminal device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program is executed by the processor to realize the new edge node access evaluation method as described in the above embodiment.

[0205] Since the new edge node access evaluation program is executed by the processor, all the technical solutions of all the above embodiments are adopted, and at least all the beneficial effects brought by all the technical solutions of all the above embodiments are obtained, which will not be repeated here.

[0206] In addition, the embodiment of the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by the processor to realize the new edge node access evaluation method as described in the above embodiment.

[0207] Since the new edge node access evaluation program is executed by the processor, all the technical solutions of all the above embodiments are adopted, and at least all the beneficial effects brought by all the technical solutions of all the above embodiments are obtained, which will not be repeated here.

[0208] Compared with the prior art, the new edge node access evaluation method, device, terminal device and product provided by the embodiment of the present application receive the access request for accessing the bearing network sent by the new edge node through the edge computing management platform, and the access request carries the performance requirement text of the new edge node; according to the access request, the used bandwidth time series data of each accessed edge node within a first preset time is obtained, the used bandwidth time series data is input into the pre-trained accessed edge node bandwidth usage predictor, and the prediction value of the total consumed bandwidth of each accessed edge node within a second preset future time is obtained; the total bandwidth of the bearing network is subtracted from the prediction value of the total consumed bandwidth of each accessed edge node within the second preset future time, and the residual available total bandwidth time series of the bearing network within the second preset future time is obtained; the performance requirement text of the new edge node and the residual available total bandwidth time series of the bearing network are input into the pre-trained new edge node access evaluator, and the feasibility evaluation result of the new edge node accessing the bearing network is obtained. Therefore, the feasibility of the new edge node accessing the bearing network is evaluated through the accessed edge node bandwidth usage predictor and in combination with the new edge node access evaluator, and the accuracy and efficiency of the operation and maintenance personnel for the new edge node access evaluation are improved.

[0209] It should be noted that, in this document, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0210] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0211] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and the necessary general hardware platform, of course, they can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a number of instructions for making a terminal device (which can be a mobile phone, a computer, a server, a controlled terminal, or a network device) execute the method of each embodiment of the present application.

[0212] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which is made by using the content of the specification and drawings of the present application, is also included in the patent protection scope of the present application.

Claims

1. A method for new edge node access evaluation, characterized in that, The method is applied to an edge computing management platform, and the method comprises the following steps: The edge computing management platform receives an access request for an access bearer network sent by a newly-built edge node, and the access request carries performance requirement text of the newly-built edge node; According to the access request, time series data of used bandwidth of each accessed edge node within a first preset time is obtained, and the time series data of used bandwidth is input into a pre-trained accessed edge node bandwidth usage predictor to obtain a predicted value of total consumed bandwidth of each accessed edge node within a second preset future time; The total bandwidth of the bearer network is reduced by the predicted value of the total consumed bandwidth of each accessed edge node within the second preset future time to obtain a time series of remaining total available bandwidth of the bearer network within the second preset future time; The performance requirement text of the newly-built edge node and the time series of remaining total available bandwidth of the bearer network are input into a pre-trained newly-built edge node access evaluator to obtain a feasibility evaluation result of the newly-built edge node accessing the bearer network.

2. The method of claim 1, wherein, The step of obtaining the predicted value of total consumed bandwidth of each accessed edge node within the second preset future time comprises: The time dynamic characteristics of the time series data of used bandwidth are extracted by using a long short-term memory layer through the accessed edge node bandwidth usage predictor; The time dynamic characteristics are encoded into a plurality of fixed-length feature vectors and combined, the combined feature vectors are decoded, and the predicted value of total consumed bandwidth of each accessed edge node within the second preset future time is mapped through a long short-term memory layer.

3. The method of claim 1, wherein, The step of obtaining the feasibility evaluation result of the newly-built edge node accessing the bearer network comprises: Text features of the performance requirement text and time series dynamic rules of the time series of remaining total available bandwidth of the bearer network are extracted by using a long short-term memory layer through the newly-built edge node access evaluator; The text features and the time series dynamic rules are respectively encoded into two fixed-length feature vectors and combined, and the combined feature vectors are decoded through a fully connected neural network to obtain the feasibility evaluation result of the newly-built edge node accessing the bearer network.

4. The method of claim 1, wherein, Before the step of receiving, by the edge computing management platform, an access request for an access bearer network sent by a newly-built edge node, the method further comprises the following steps: The accessed edge node bandwidth usage predictor is trained, and the training specifically comprises: A time series data set of used bandwidth of each historically accessed edge node within a first preset time is collected, and the time series data of used bandwidth of each accessed edge node within the first preset time is labeled with real data of used bandwidth of the accessed edge node within a next second preset time to obtain a data set of an accessed edge node bandwidth usage predictor model; Data in the data set of the accessed edge node bandwidth usage predictor model is preprocessed; The preprocessed data set of the accessed edge node bandwidth usage predictor model is divided into a predictor training set and a predictor test set; The accessed edge node bandwidth usage predictor model is trained through the predictor training set; The trained accessed edge node bandwidth usage predictor model is verified through the predictor test set, and after the model converges, a final trained accessed edge node bandwidth usage predictor is obtained.

5. The method of claim 4, wherein, The edge computing management platform further comprises the following steps before the step of receiving the access request for accessing the bearer network sent by the new edge node: The accessed edge node bandwidth usage predictor and the new edge node access evaluator are trained. A performance requirement set of historical new edge nodes is collected. The total bandwidth of the bearer network is reduced by the real data of the used bandwidth of the accessed edge node within the next second preset time, and a remaining available total bandwidth time series set of the bearer network within the future second preset time is obtained. A feasibility evaluation result set of the new edge node accessing the bearer network is obtained. The performance requirement set, the remaining available total bandwidth time series set, and the feasibility evaluation result set are used as a data set of the new edge node access evaluator model. The data in the data set of the new edge node access evaluator model is preprocessed. The preprocessed data set of the new edge node access evaluator model is divided into an evaluator training set and an evaluator test set. The new edge node access evaluator model is trained through the evaluator training set. The trained new edge node access evaluator model is verified through the evaluator test set, and after the model converges, a final trained new edge node access evaluator is obtained.

6. The method of new edge node access evaluation according to any one of claims 1-5, characterized in that, The accessed edge node bandwidth usage predictor and the new edge node access evaluator are both composed of a coding and decoding neural network.

7. The method of new edge node access evaluation according to any one of claims 1-5, characterized in that, Before the step of inputting the performance requirement text of the new edge node and the remaining available total bandwidth time series of the bearer network into the pre-trained new edge node access evaluator, the following steps are further included: The performance requirement text of the new edge node and the remaining available total bandwidth time series of the bearer network are preprocessed; and / or After the step of obtaining the used bandwidth time series data of each accessed edge node within the first preset time according to the access request, the following steps are further included: The used bandwidth time series data is preprocessed.

8. A new edge node access evaluation apparatus, characterized by, The new edge node access evaluation device comprises: A request receiving module is configured to receive an access request for accessing a bearer network sent by a new edge node, wherein the access request carries a performance requirement text of the new edge node. A prediction module is configured to obtain used bandwidth time series data of each accessed edge node within a first preset time according to the access request, input the used bandwidth time series data into a pre-trained accessed edge node bandwidth usage predictor, and predict a predicted value of total consumed bandwidth of each accessed edge node within a future second preset time; and obtain a remaining available total bandwidth time series of the bearer network within the future second preset time by reducing the total bandwidth of the bearer network by the predicted value of the total consumed bandwidth of each accessed edge node within the future second preset time. An evaluation module is configured to input the performance requirement text of the new edge node and the time series of the remaining total bandwidth of the bearer network into a pre-trained new edge node access evaluator to obtain a feasibility evaluation result of the new edge node accessing the bearer network.

9. A terminal device, comprising: The terminal device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program, when executed by the processor, implements the method for evaluating access of a new edge node as claimed in any one of claims 1-7.

10. A computer program product, characterised in that, The computer program product includes a computer program, and the computer program, when executed by a processor, implements the method for evaluating access of a new edge node as claimed in any one of claims 1-7.

Images