CN110995520B - Network flow prediction method and device, computer equipment and readable storage medium - Google Patents

Network flow prediction method and device, computer equipment and readable storage medium Download PDF

Info

Publication number
CN110995520B
CN110995520B CN202010126560.7A CN202010126560A CN110995520B CN 110995520 B CN110995520 B CN 110995520B CN 202010126560 A CN202010126560 A CN 202010126560A CN 110995520 B CN110995520 B CN 110995520B
Authority
CN
China
Prior art keywords
network
data
graph convolution
target
network model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010126560.7A
Other languages
Chinese (zh)
Other versions
CN110995520A (en
Inventor
王剑
杜军
王景璟
任勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202010126560.7A priority Critical patent/CN110995520B/en
Publication of CN110995520A publication Critical patent/CN110995520A/en
Application granted granted Critical
Publication of CN110995520B publication Critical patent/CN110995520B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14Network analysis or design
    • H04L41/147Network analysis or design for predicting network behaviour
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/25Fusion techniques
    • G06F18/253Fusion techniques of extracted features
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14Network analysis or design
    • H04L41/142Network analysis or design using statistical or mathematical methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Algebra (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The application provides a network flow prediction method, a device, a computer device and a readable storage medium, wherein the method comprises the following steps: carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result, carrying out feature extraction on the graph convolution result to obtain initial flow feature prediction data of a network node, and carrying out feature fusion processing on the initial flow feature prediction data to obtain target flow prediction data of the network node; the method can comprehensively consider the time characteristics and the network space characteristics of the network to predict the network flow, and can improve the accuracy of the prediction result.

Description

Network flow prediction method and device, computer equipment and readable storage medium
Technical Field
The present application relates to the field of network technologies, and in particular, to a method and an apparatus for predicting network traffic, a computer device, and a readable storage medium.
Background
With the rapid development of internet technology, the scale of the internet technology is continuously increased, the network structure is continuously complicated, and the service scale is increasingly large, so that the reasonable allocation of network resources is very important. In order to implement reasonable network resource allocation and guarantee service quality, and also to analyze the health condition of the network and evaluate the network carrying capacity and stability, it is necessary to effectively predict network traffic data.
In the traditional technology, a linear or nonlinear model is established based on network historical data to realize network traffic prediction. However, the conventional network traffic prediction method is only applicable to a fixed network with unchanged topology, and cannot adapt to dynamic network topology, so that the accuracy of the network traffic prediction result is low.
Disclosure of Invention
In view of the foregoing, it is desirable to provide a method, an apparatus, a computer device and a readable storage medium for predicting network traffic, which can improve accuracy of a network traffic prediction result.
The embodiment of the application provides a network flow prediction method, which comprises the following steps:
carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
extracting features of the graph convolution result to obtain initial flow feature prediction data of the network node;
and carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
In one embodiment, the graph convolution result includes a network topology feature; the method for performing graph convolution on network data to be processed by adopting the target graph convolution network model to obtain a graph convolution result comprises the following steps: and extracting data association characteristics of the network data to be processed to obtain network topology characteristics.
In one embodiment, the extracting data association features of the network data to be processed includes:
performing linear weighting operation on the network data to be processed to obtain data correlation characteristics;
performing activation operation on the data association characteristics by using an activation function to obtain hidden layer output;
and continuing to execute the linear weighting operation on the network data to be processed until the network topology characteristics are obtained.
In one embodiment, the initial flow characteristic prediction data comprises first flow prediction data and second flow prediction data; the feature extraction of the graph convolution result to obtain initial traffic feature prediction data of the network node includes:
adopting a target long-short term memory artificial neural network model to extract time series characteristics of the graph convolution result to obtain the first flow prediction data of the network node;
and extracting network space characteristics of the graph convolution result by adopting a target convolution neural network model to obtain the second flow prediction data of the network node.
In one embodiment, the performing feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node includes:
and performing feature fusion processing on the first flow prediction data and the second flow prediction data to obtain target flow prediction data of the network node.
In one embodiment, the performing feature fusion processing on the first traffic prediction data and the second traffic prediction data to obtain target traffic prediction data of the network node includes:
and performing feature fusion processing on the first flow prediction data and the second flow prediction data by adopting a target feature fusion network model to obtain the target flow prediction data.
In one embodiment, the method further comprises:
and preprocessing the initial network data to obtain the network data to be processed.
An embodiment of the present application provides a network traffic prediction apparatus, where the network traffic prediction apparatus includes:
the graph convolution module is used for performing graph convolution on the network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
the characteristic extraction module is used for extracting the characteristics of the graph convolution result to obtain initial flow characteristic prediction data of the network node;
and the characteristic fusion module is used for carrying out characteristic fusion processing on the initial flow characteristic prediction data to obtain target flow prediction data of the network node.
The embodiment of the application provides a computer device, which comprises a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, and the processor executes the computer program to realize the following steps:
carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
extracting features of the graph convolution result to obtain initial flow feature prediction data of the network node;
and carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
An embodiment of the application provides a readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the following steps:
carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
extracting features of the graph convolution result to obtain initial flow feature prediction data of the network node;
and carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
In the network traffic prediction method provided by this embodiment, a target graph convolution network model may be adopted to perform graph convolution on network data to be processed to obtain a graph convolution result, feature extraction is performed on the graph convolution result to obtain initial traffic feature prediction data of a network node, and feature fusion processing is performed on the initial traffic feature prediction data to obtain target traffic prediction data of the network node; the method can comprehensively consider the time characteristics and the network space characteristics of the network to predict the network flow, and can improve the accuracy of the prediction result.
Drawings
Fig. 1 is a schematic structural diagram of a network traffic prediction system according to an embodiment.
Fig. 2 is a flowchart illustrating a network traffic prediction method according to an embodiment.
Fig. 3 is a schematic structural diagram of network data to be processed according to another embodiment.
Fig. 4 is a schematic diagram of a layer of network data to be processed implementing graph convolution according to another embodiment.
Fig. 5 is a schematic diagram of a neuron structure of a long-short term memory artificial neural network model according to another embodiment.
Fig. 6 is a schematic diagram of an overall network model according to another embodiment.
Fig. 7 is a schematic structural diagram of a network traffic prediction apparatus according to an embodiment.
FIG. 8 is an internal block diagram of a computer device, provided in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The network traffic prediction method provided in the embodiment of the present application may be applied to a network traffic prediction system as shown in fig. 1, where the network traffic prediction system may include a network device and a network management server. Optionally, the network device may be a computer, a hub, a switch, a network bridge, a router, a network manager, a network interface card, a wireless access point, or the like; the method comprises the steps that an interface for acquiring network data is configured on network equipment and/or a network management server, the network data is acquired through the configured interface, the acquired network data is sent to the network management server, and the network management server carries out a series of processing on the network data, so that actual network flow is predicted. The network device and the network management server can communicate through wireless connection or wired connection. Optionally, the wireless connection mode may be Wi-Fi, mobile network or bluetooth connection. The method for realizing the network traffic prediction in the embodiment can be suitable for static networks, non-heterogeneous networks and time-varying heterogeneous networks. The main implementation of the following method embodiments is described by taking a network traffic prediction device as an example.
In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions in the embodiments of the present application are further described in detail by the following embodiments in combination with the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the invention.
Fig. 2 is a flowchart illustrating a network traffic prediction method according to an embodiment. The present embodiment relates to a process of how to predict network traffic. As shown in fig. 2, the method includes:
and S101, carrying out graph convolution on the network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result.
Specifically, the network data to be processed may be the acquired original network data in the historical time period and the network data obtained by preprocessing the acquired original network data. Optionally, the network data may include data indicating traffic characteristics, service characteristics, delay characteristics, packet loss rate characteristics, service characteristics, and the like of nodes in the network; wherein different characteristics of different network nodes are interrelated. In the present embodiment, the effect of graph convolution can be understood as decoupling. Wherein decoupling can be characterized as a process of associating network nodes with an overall network structure.
In this embodiment, the network management server may implement multiple graph convolutions on the acquired network data to be processed, that is, the network management server may decouple the network data to be processed from the entire network structure by using a target graph convolution network model. The above-mentioned network data structure to be processed may be represented by fig. 3, where fig. 3 represents a multi-layer graph structure data including a network topology structure, and different layers represent different types of network data to be processed, such as a traffic layer, a service layer, a delay layer, and a packet loss rate layer, but only 7 network node 3-layer data structures, that is, a traffic layer, a delay layer, and a packet loss rate layer, are shown in fig. 3. In practice, the network management server may use the target graph convolution network model to decouple (i.e. graph convolution) each layer of network data to be processed.
And S102, extracting the characteristics of the graph convolution result to obtain initial flow characteristic prediction data of the network node.
Specifically, the network management server may perform feature extraction on the graph convolution result by using a deep learning network to obtain initial traffic feature prediction data of the network node. Optionally, the feature extraction result may include a time series feature and a network space feature. Alternatively, the deep learning network may include a long-short term memory artificial neural network model, a convolutional neural network model, and the like. Optionally, the initial traffic characteristic prediction data of the network node may be traffic prediction mixed data of the network node.
And step S103, carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
Specifically, the network management server may perform feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node. Optionally, the target traffic prediction data may be predicted target traffic prediction data after combining temporal and spatial service characteristics, and may also be understood as traffic prediction data of each network node in the network at the next time. In this embodiment, the above method may be applicable not only to the time-varying heterogeneous network, but also to the static network and the non-heterogeneous network,
in the network traffic prediction method provided by this embodiment, a target graph convolution network model may be adopted to perform graph convolution on network data to be processed to obtain a graph convolution result, feature extraction is performed on the graph convolution result to obtain initial traffic feature prediction data of a network node, and feature fusion processing is performed on the initial traffic feature prediction data to obtain target traffic prediction data of the network node; the method can comprehensively consider the time characteristics and the network space characteristics of the network, so that the method is not only suitable for the time-varying heterogeneous network, but also suitable for the static network and the non-heterogeneous network, and the universality of the network flow prediction method on networks with different structures is improved; meanwhile, the network flow is predicted by comprehensively considering the time characteristics and the network space characteristics of the network, so that the accuracy of the prediction result can be improved.
As one embodiment, the graph convolution result includes a network topology characteristic. The process of performing graph convolution on the network data to be processed by using the target graph convolution network model in the step S101 to obtain a graph convolution result may include the following steps: and extracting data association characteristics of the network data to be processed to obtain network topology characteristics.
The step of extracting data association features of the network data to be processed to obtain network topology features may specifically include: performing linear weighting operation on the network data to be processed to obtain data correlation characteristics; performing activation operation on the data association characteristics by using an activation function to obtain hidden layer output; and continuing to execute the linear weighting operation on the network data to be processed until the network topology characteristics are obtained.
Specifically, the network management server may extract data associated features (i.e., linear weighted operation) from the network data to be processed, perform activation operation on the data associated features by using an activation function to obtain output of a hidden layer in the target graph convolutional network model, perform linear weighted operation on the output of a previous hidden layer by using a next hidden layer to obtain another data associated feature, perform activation operation on another data associated feature by using the activation function to obtain output of a current hidden layer, and so on, take the output of the previous hidden layer as the input of the next hidden layer, continuously and repeatedly perform linear weighted operation and activation operation, and take the output of the last hidden layer of the target graph convolutional network model as a graph convolution result, i.e., a network topology feature. In the present embodiment, the activation function used by the target graph convolution network model is a ReLu activation function. Alternatively, the data structure of the network topology features may be similar to the pending network data structure. For each node in different layers, the network management server can extract data association characteristics of the network node and adjacent network nodes by adopting a target graph convolutional network model, and a data association characteristic extraction formula can be represented as formula (1)
Figure 214712DEST_PATH_IMAGE001
(1);
Wherein,
Figure 764642DEST_PATH_IMAGE002
in order to associate the characteristics with the data,
Figure 442748DEST_PATH_IMAGE003
as a network nodevAt the feature representation of the current layer,
Figure 611693DEST_PATH_IMAGE004
to be connected with a network nodevThe characteristic representation of the connected edges is,
Figure 58855DEST_PATH_IMAGE005
as a network nodevIs a function of
Figure 514107DEST_PATH_IMAGE006
Representing a graph convolution kernel, which is actually a function that is linearly weighted according to the adjacency matrix, wherein the linear weighting coefficients need to be obtained by training. In this embodiment, if the network data to be processed is 3 layers, linear weighting operation (i.e., multiplying a coefficient by the network data and adding an offset) may be performed on the network data corresponding to the current network node and the neighboring network node each time the graph is convolved, so as to obtain data association characteristics, and then the data association characteristics are input into an activation function adopted by a target graph convolution network model, so as to obtain hidden layer output; the coefficients and the offsets are obtained through training, and actually, each time the graph is convolved, the adjacent matrix can be multiplied by an element to be multiplied by a coefficient, the element is multiplied by the output of the previous hidden layer, the operation result is summed and then added with an offset, and the operation result is input into an activation function to carry out activation operation. In the present embodiment, it is preferred that,
Figure 820454DEST_PATH_IMAGE004
the data in (1) can be represented by 0 or 1, 0 represents that two network nodes are not communicated, and 1 represents that two network nodes are communicated. Alternatively, the data in the adjacency matrix may be represented by 0 and 1, such as a network packetContaining 8 network nodes, the adjacency matrix A is a 0 and 1 matrix of 8 x 8, A [ [ alpha ] ]i][j]Equal to 1 denotesiAnd a firstjNetwork nodes are connected, and the value equal to 0 represents the secondiAnd a firstjThe individual network nodes are not in communication,iindicating the number of the current network node,jrepresenting other network node numbers. Optionally, the network topology characteristics may include network node characteristics of the current network node at the current time and in the historical time period, and network node characteristics of all network nodes in the entire network at the current time.
Illustratively, the input data for each layer is in different dimensions, e.g. the current layer inputs 8 × 5 data plus an adjacency matrix, then, the data representation is composed of 8 network nodes, each network node has 5 characteristics such as flow, delay and optional, the network management server can acquire the characteristic information of the current network node 1-hop neighbor network node through one-time graph convolution, the characteristic information of the current network node 2 hops and the neighbor network nodes within the 2-hop range can be obtained through two times of graph convolution, and so on, the characteristic information of the network nodes within the multi-hop range of the current network nodes can be obtained through multiple graph convolutions, and the finally obtained result is a multi-channel matrix with the length being the number of the network nodes, namely, the network topological characteristics of different network parameters are respectively extracted through the graph convolutions, so that different types of network data to be processed are all decoupled from the whole network structure. It can also be understood that the obtained information of a single network node can embody the performance of the whole network, so that the method can adapt to the topology variability during network traffic prediction and is suitable for dynamic network topology. In this embodiment, in a 3-layer graph convolution network structure, a network management server may obtain feature information of a network node within at least a 3-hop range, and at this time, the network topology feature obtained by graph convolution is already decoupled from the entire network structure to a certain extent.
As shown in fig. 4, an example of implementing graph convolution by using one network node in a certain layer in a network is shown, and for the network node 1, the network data of the network node 1 and the neighboring nodes 2, 3, 4, and 5 and the weights of the connected edges are subjected to graph convolution to obtain the output corresponding to the network node 1, that is, the network topology characteristic of the network node 1.
Optionally, before the process of step S101, the network traffic prediction method may further include: and preprocessing the initial network data to obtain the network data to be processed.
It should be noted that the initial network data may be original network data acquired by the network management server. The network management server can preprocess the obtained initial network data to obtain the network data to be processed.
In this embodiment, historical network data in a past period of time is used as initial network data, and the time interval of the collected initial network data may be 1 minute. Optionally, the network management server may calculate a mean value and a standard deviation of the initial network data, and divide the mean value by the standard deviation after subtracting the mean value from each initial network data to obtain the normalized to-be-processed network data. Optionally, the average value of the network data to be processed may be equal to 0, and the numerical ranges of the network data to be processed are all between-1 and 1.
According to the network traffic prediction method provided by the embodiment, the time characteristics and the network space characteristics of the network can be comprehensively considered, so that the method is not only suitable for the time-varying heterogeneous network, but also suitable for the static network and the non-heterogeneous network, and the universality of the network traffic prediction method on networks with different structures is improved; meanwhile, the network flow is predicted by comprehensively considering the time characteristics and the network space characteristics of the network, so that the accuracy of the prediction result can be improved.
As one example, the initial flow characteristic prediction data includes first flow prediction data and second flow prediction data; the step of extracting features of the graph convolution result in the step S102 to obtain initial traffic feature prediction data of the network node may include the following steps:
and S1021, adopting a target long-short term memory artificial neural network model to extract time series characteristics of the graph convolution result to obtain the first flow prediction data of the network node.
Specifically, the target long-short term memory artificial neural network model may be a network model obtained by training the initial long-short term memory artificial neural network model. Optionally, the network management server may perform two image convolutions on the network data to be processed through two parallel target image convolution network models, input a result of one image convolution to the target long-short term memory artificial neural network model, input a result of the other image convolution to the target convolution neural network model, and execute subsequent processing. Optionally, the two image convolution may be processed in parallel, and parameters of the target image convolution network model may be different when the two image convolution are performed, so that the two obtained image convolution results may also be different. In this embodiment, before the training of the network model, two initial graph convolutional networks, an initial long-short term memory artificial neural network model and an initial convolutional neural network model may be formed into two network model branches. The network model branch circuit is formed by cascading an initial graph convolution network and an initial long-short term memory artificial neural network model, the other network model branch circuit is formed by cascading an initial graph convolution network and an initial convolution neural network model, and the two network model branch circuits are in parallel structures. Optionally, after the network model branches are respectively trained, the initial graph convolution network model in the two network model branches is used as a target graph convolution network model, and the target graph convolution network model has no full connection layer; the parameters in the two target graph convolutional network models can be the same or different. Optionally, the network management server may train the initial long-short term memory artificial neural network model and the cascaded initial graph convolution model in the network model branch by using a time-based back propagation algorithm, so as to obtain the target long-short term memory artificial neural network model and the target graph convolution network model. Optionally, the network management server may obtain the target convolutional neural network model and another target graph convolutional network model by using an initial convolutional neural network model and a cascaded initial graph convolutional model in a network model branch trained based on a back propagation algorithm.
In this embodiment, the network management server may input the network node characteristics (i.e., single node characteristics) of the current network node at the current time and in the historical time period, which are obtained after the graph convolution, to the target long-short term memory artificial neural network model, and may output single-node traffic prediction data after the processing by the target long-short term memory artificial neural network model (the traffic prediction data only considers the time series of the network, but does not consider the spatial characteristics of the network).
Wherein, for the initial long-short term memory artificial neural network model, the neuron structure is shown in fig. 5. The network management server can adopt standard normal distribution random number initialization, and a convolution kernel, a hidden layer, a weight and an offset in an initial graph convolution network model corresponding to the connection of the target long-term and short-term memory artificial neural network model; in addition, the method adopts the standard normal distribution random number initialization, the weights and the offsets of an input gate, a forgetting gate and an output gate in the initial long-short term memory artificial neural network model, and the cell state of the initial long-short term memory artificial neural network model is initialized to be 0 vector. Optionally, the target long-short term memory artificial neural network model obtained after training mainly focuses on time correlation and is used for extracting time series features.
As shown in fig. 5, which is a neuron structure of the long-short term memory artificial neural network model, FC represents a full connection layer, and when training the initial long-short term memory artificial neural network model in the network model branch, the output value of each neuron in fig. 5 can be calculated in the forward direction, and the calculation formula is as follows:
Figure 652144DEST_PATH_IMAGE007
(2);
Figure 219392DEST_PATH_IMAGE008
(3);
Figure 845545DEST_PATH_IMAGE009
(4);
Figure 373609DEST_PATH_IMAGE010
(5);
Figure 8990DEST_PATH_IMAGE011
(6);
Figure 165165DEST_PATH_IMAGE012
(7);
wherein,
Figure 227799DEST_PATH_IMAGE013
representing a sigmod function;
Figure 240230DEST_PATH_IMAGE014
represents the output of the last neuron;
Figure 679301DEST_PATH_IMAGE015
an input representing a current neuron;
Figure 955562DEST_PATH_IMAGE016
an output representing a current neuron;
Figure 64463DEST_PATH_IMAGE017
Figure 426174DEST_PATH_IMAGE018
Figure 403358DEST_PATH_IMAGE019
Figure 534125DEST_PATH_IMAGE020
respectively representing multiplication coefficients of a forgetting gate, an input gate, an output gate and cell state updating;
Figure 813928DEST_PATH_IMAGE021
Figure 662935DEST_PATH_IMAGE022
Figure 178230DEST_PATH_IMAGE023
Figure 429083DEST_PATH_IMAGE024
respectively representing the bias of the forgetting gate, the input gate, the output gate and the updating of the cell state;
Figure 879787DEST_PATH_IMAGE025
Figure 216090DEST_PATH_IMAGE026
Figure 535076DEST_PATH_IMAGE027
respectively representing control information of a forgetting gate, an input gate and an output gate, wherein the value can be within the interval of 0-1, 0 represents complete forgetting, and 1 represents complete retention;
Figure 640435DEST_PATH_IMAGE028
representing candidate cell states;
Figure 262040DEST_PATH_IMAGE029
Figure 820061DEST_PATH_IMAGE030
respectively representing the cell state at the last moment and the cell state updated at the current moment;
Figure 942738DEST_PATH_IMAGE031
outputting the forgetting gate in the long-short term memory artificial neural network model; and continuously repeating the back propagation error updating parameters until the iteration number reaches a preset maximum value. Optionally, the network management server may train the initial long-short term memory artificial neural network model in the network model leg using a time-based back propagation algorithm. Optionally, the final full-connection layer in the trained initial long-short term memory artificial neural network model is removed to obtain the target long-short term memory artificial neural network model, and it can be understood that the pre-most full-connection layer in the trained initial long-short term memory artificial neural network model isAnd the output result of the later hidden layer is used as the first traffic prediction data of the network node. During the processing, the full connection layer in the long-short term memory artificial neural network model shown in fig. 5 is removed.
In this embodiment, the target long-short term memory artificial neural network model may input a graph convolution result corresponding to a single node in a period of time, and output a time series characteristic of the single node in the period of time. Optionally, the activation functions used by the input gate, the forgetting gate and the output gate in the target long-short term memory artificial neural network model are Sigmod functions, and the activation functions of state updating and output are tanh functions.
Step S1022, performing network spatial feature extraction on the graph convolution result by using a target convolutional neural network model, to obtain the second traffic prediction data of the network node.
Specifically, the network management server may train an initial graph convolution network model, which is cascaded with the initial convolution neural network model, in a network model branch by using a BP algorithm, to obtain a target convolution neural network model and a target graph convolution network model. In this embodiment, both the training set and the test set of the training network model may be historical network data. Optionally, the trained target convolutional neural network model may extract the overall spatial features of the network.
It should be noted that the network management server may input the network node characteristics of all network nodes in the entire network at the current time (i.e., all node characteristics at a certain time) obtained after the graph convolution to the target convolutional neural network model, and may output the traffic prediction data of all nodes at a certain time after the processing by the target convolutional neural network model (the traffic prediction data does not consider the time series of the network, but only considers the spatial characteristics of the network). Optionally, the final fully-connected layer in the trained initial convolutional neural network model is removed to obtain the target convolutional neural network model, and it can be understood that the output result of the last hidden layer before the fully-connected layer in the trained convolutional neural network model is used as the second traffic prediction data of the network node. In this embodiment, the target convolutional neural network model outputs a spatial correlation feature, that is, a network spatial feature, between network nodes at a certain time in the graph convolution result. Alternatively, the activation function used by the target convolutional neural network model may be a ReLu activation function.
Further, the process of performing feature fusion processing on the initial traffic feature prediction data in step S103 to obtain target traffic prediction data of the network node may be implemented by the following steps: and performing feature fusion processing on the first flow prediction data and the second flow prediction data to obtain target flow prediction data of the network node.
It should be noted that the network management server may perform feature fusion processing on the first traffic prediction data and the second traffic prediction data by using an algorithm based on a bayesian decision theory, an algorithm based on a sparse representation theory, and an algorithm based on a deep learning theory, so as to obtain target traffic prediction data of the network node.
The step of performing feature fusion processing on the first traffic prediction data and the second traffic prediction data to obtain target traffic prediction data of the network node may specifically include: and performing feature fusion processing on the first flow prediction data and the second flow prediction data by adopting a target feature fusion network model to obtain the target flow prediction data.
In this embodiment, the network management server may use a target feature fusion network model to perform feature fusion processing on the first traffic prediction data and the second traffic prediction data, so as to obtain target traffic prediction data of the network node. The output of the target long-short term memory artificial neural network model and the output of the target convolutional neural network model can be used as the input of the target characteristic fusion network model; that is, the input end of the target feature fusion network model is cascaded with the output ends of the target long-short term memory artificial neural network model and the target convolutional neural network model, and the structural schematic diagram of the overall network model in this embodiment is shown in fig. 6.
The target feature fusion network model can fuse the time series features and the network space features to obtain target traffic prediction data with high accuracy.
According to the network traffic prediction method provided by the embodiment, the time characteristics and the network space characteristics of the network can be comprehensively considered, so that the method is not only suitable for the time-varying heterogeneous network, but also suitable for the static network and the non-heterogeneous network, and the universality of the network traffic prediction method on networks with different structures is improved; meanwhile, the network flow is predicted by comprehensively considering the time characteristics and the network space characteristics of the network, so that the accuracy of the prediction result can be improved.
It should be understood that, although the steps in the flowchart of fig. 2 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 2 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
For specific limitations of the network traffic prediction apparatus, reference may be made to the above limitations of the network traffic prediction method, which is not described herein again. The modules in the network flow predicting device in the computer equipment can be wholly or partially implemented by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
Fig. 7 is a schematic structural diagram of a network traffic prediction apparatus according to an embodiment. As shown in fig. 7, the apparatus may include: a decoupling module 11, a feature extraction module 12 and a feature fusion module 13.
Specifically, the decoupling module 11 is configured to perform graph convolution on the network data to be processed by using a target graph convolution network model to obtain a graph convolution result;
the feature extraction module 12 is configured to perform feature extraction on the graph convolution result to obtain initial traffic feature prediction data of the network node;
the feature fusion module 13 is configured to perform feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the graph convolution result includes a network topology feature; the decoupling module 11 comprises an associated feature extraction unit.
Specifically, the associated feature extraction unit is configured to perform data associated feature extraction on the to-be-processed network data to obtain a network topology feature.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the associated feature extraction unit is specifically configured to perform linear weighting operation on the network data to be processed to obtain a data associated feature, and perform activation operation on the data associated feature by using an activation function to obtain hidden layer output; and continuing to execute the linear weighting operation on the network data to be processed until the network topology characteristics are obtained.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the initial flow characteristic prediction data comprises first flow prediction data and second flow prediction data; the feature extraction module 12 includes: a time series feature extraction unit and a spatial feature extraction unit.
Specifically, the time series feature extraction unit is configured to perform time series feature extraction on the graph convolution result by using a target long-term and short-term memory artificial neural network model to obtain the first traffic prediction data of the network node;
the spatial feature extraction unit is configured to perform network spatial feature extraction on the graph convolution result by using a target convolutional neural network model to obtain the second traffic prediction data of the network node.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the feature fusion module 13 includes: and a feature fusion unit.
Specifically, the feature fusion unit is configured to perform feature fusion processing on the first traffic prediction data and the second traffic prediction data to obtain target traffic prediction data of the network node.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the feature fusion unit is specifically configured to perform feature fusion processing on the first traffic prediction data and the second traffic prediction data by using a target feature fusion network model to obtain the target traffic prediction data.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, the network traffic prediction apparatus further includes: and a preprocessing module.
Specifically, the preprocessing module is configured to preprocess the initial network data to obtain the to-be-processed network data.
The network traffic prediction apparatus provided in this embodiment may implement the method embodiments described above, and the implementation principle and the technical effect are similar, which are not described herein again.
In one embodiment, a computer device is provided, the internal structure of which may be as shown in FIG. 8. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a network traffic prediction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 8 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:
carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
extracting features of the graph convolution result to obtain initial flow feature prediction data of the network node;
and carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
In one embodiment, a readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, performs the steps of:
carrying out graph convolution on network data to be processed by adopting a target graph convolution network model to obtain a graph convolution result;
extracting features of the graph convolution result to obtain initial flow feature prediction data of the network node;
and carrying out feature fusion processing on the initial traffic feature prediction data to obtain target traffic prediction data of the network node.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A network traffic prediction method is applicable to static networks, non-heterogeneous networks and time-varying heterogeneous networks in the Internet field, and comprises the following steps:
carrying out graph convolution on network data to be processed by adopting two target graph convolution network models respectively to obtain graph convolution results; the network data to be processed is data to be processed in an internet network, and the graph convolution refers to a process of associating a network node with an integral internet network structure;
adopting a target long-short term memory artificial neural network model to extract time series characteristics of the graph convolution result to obtain first flow prediction data of the network node; wherein the graph convolution result comprises a network topology feature;
extracting network space characteristics of the graph convolution result by adopting a target convolution neural network model to obtain second flow prediction data of the network node; one target graph convolution network model and the target long-short term memory artificial neural network model are cascaded to form one network model branch, the other target graph convolution network model and the target convolution neural network model are cascaded to form the other network model branch, and the two network model branches are in parallel relation;
and performing feature fusion processing on the first flow prediction data and the second flow prediction data by adopting a target feature fusion network model to obtain target flow prediction data of the network node.
2. The method of claim 1, wherein the performing graph convolution on the network data to be processed by using the target graph convolution network model to obtain a graph convolution result comprises: and extracting data association characteristics of the network data to be processed to obtain network topology characteristics.
3. The method of claim 2, wherein the performing data association feature extraction on the network data to be processed comprises:
performing linear weighting operation on the network data to be processed to obtain data correlation characteristics;
performing activation operation on the data association characteristics by using an activation function to obtain hidden layer output;
and continuing to execute the linear weighting operation on the network data to be processed until the network topology characteristics are obtained.
4. The method of claim 1, wherein the network data comprises traffic characteristics, delay characteristics, packet loss rate characteristics, and traffic characteristic data of the network nodes in the network.
5. The method of claim 4, wherein the activation function used by the target graph convolution network model is a ReLu activation function.
6. The method of claim 5, wherein the network topology characteristics comprise network node characteristics of current network nodes at the current time and in historical time periods, and network node characteristics of all network nodes in the entire network at the current time.
7. The method of claim 1, further comprising:
and preprocessing the initial network data to obtain the network data to be processed.
8. A network traffic prediction apparatus, wherein the apparatus is suitable for use in a static network, a non-heterogeneous network and a time-varying heterogeneous network in the internet domain, and the apparatus comprises:
the graph convolution module is used for performing graph convolution on the network data to be processed by adopting two target graph convolution network models respectively to obtain a graph convolution result; the network data to be processed is data to be processed in an internet network, and the graph convolution refers to a process of associating a network node with an integral internet network structure;
the characteristic extraction module is used for extracting time series characteristics of the graph convolution result by adopting a target long-short term memory artificial neural network model to obtain first flow prediction data of the network node, and extracting network space characteristics of the graph convolution result by adopting a target convolution neural network model to obtain second flow prediction data of the network node; the graph convolution result comprises network topology characteristics, a target graph convolution network model and the target long-short term memory artificial neural network model are cascaded to form a network model branch, the other target graph convolution network model and the target convolution neural network model are cascaded to form another network model branch, and the two network model branches are in parallel relation;
and the characteristic fusion module is used for performing characteristic fusion processing on the first flow prediction data and the second flow prediction data by adopting a target characteristic fusion network model to obtain target flow prediction data of the network node.
9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
10. A readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
CN202010126560.7A 2020-02-28 2020-02-28 Network flow prediction method and device, computer equipment and readable storage medium Active CN110995520B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010126560.7A CN110995520B (en) 2020-02-28 2020-02-28 Network flow prediction method and device, computer equipment and readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010126560.7A CN110995520B (en) 2020-02-28 2020-02-28 Network flow prediction method and device, computer equipment and readable storage medium

Publications (2)

Publication Number Publication Date
CN110995520A CN110995520A (en) 2020-04-10
CN110995520B true CN110995520B (en) 2020-06-30

Family

ID=70081459

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010126560.7A Active CN110995520B (en) 2020-02-28 2020-02-28 Network flow prediction method and device, computer equipment and readable storage medium

Country Status (1)

Country Link
CN (1) CN110995520B (en)

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111726243B (en) * 2020-05-14 2021-10-22 华为技术有限公司 Method and device for predicting node state
CN113988347A (en) * 2020-07-08 2022-01-28 阿里巴巴集团控股有限公司 Data processing method, device, equipment and storage medium
CN114448840A (en) * 2020-10-20 2022-05-06 南京中兴软件有限责任公司 Flow signal conversion method and device, electronic equipment and storage medium
CN112383516A (en) * 2020-10-29 2021-02-19 博雅正链(北京)科技有限公司 Graph neural network construction method and abnormal flow detection method based on graph neural network
CN112437022B (en) * 2020-11-11 2023-05-19 中国科学技术大学先进技术研究院 Network traffic identification method, device and computer storage medium
CN114531363A (en) * 2020-11-20 2022-05-24 中兴通讯股份有限公司 Information processing method, model generation and training method, electronic device, and medium
CN112491712B (en) * 2020-11-30 2021-08-17 复旦大学 Data packet routing algorithm based on multi-agent deep reinforcement learning
CN112766633B (en) * 2020-12-22 2023-10-24 国网浙江省电力有限公司绍兴供电公司 Flow balance-based power wireless heterogeneous network management method and device
CN114726745B (en) * 2021-01-05 2024-05-17 中国移动通信有限公司研究院 Network traffic prediction method, device and computer readable storage medium
CN112884020B (en) * 2021-01-29 2024-06-04 北京联合大学 Service quality prediction method based on multi-scale circular convolutional neural network
CN112910710B (en) * 2021-02-08 2022-04-01 清华大学 Network flow space-time prediction method and device, computer equipment and storage medium
CN113469415B (en) * 2021-06-07 2023-04-07 湖北工业大学 Network flow prediction method and computer equipment
CN113743594B (en) * 2021-09-30 2023-12-05 中国联合网络通信集团有限公司 Network traffic prediction model establishment method and device, electronic equipment and storage medium
CN114466369A (en) * 2021-12-20 2022-05-10 中兴通讯股份有限公司 Network resource processing method, storage medium and electronic device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109740785A (en) * 2018-10-22 2019-05-10 北京师范大学 The method of node state prediction based on figure convolutional neural networks
CN110264709A (en) * 2019-05-06 2019-09-20 北京交通大学 The prediction technique of the magnitude of traffic flow of road based on figure convolutional network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109887282B (en) * 2019-03-05 2022-01-21 中南大学 Road network traffic flow prediction method based on hierarchical timing diagram convolutional network
CN110188936B (en) * 2019-05-23 2021-11-05 浙江大学 Short-term traffic flow prediction method based on multi-factor space selection deep learning algorithm

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109740785A (en) * 2018-10-22 2019-05-10 北京师范大学 The method of node state prediction based on figure convolutional neural networks
CN110264709A (en) * 2019-05-06 2019-09-20 北京交通大学 The prediction technique of the magnitude of traffic flow of road based on figure convolutional network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Satellite Image Prediction Relying on GAN and LSTM Neural Networks;Zhan Xu 等;《ICC 2019-2019 IEEE International Conference on Communications (ICC)》;20190715;全文 *

Also Published As

Publication number Publication date
CN110995520A (en) 2020-04-10

Similar Documents

Publication Publication Date Title
CN110995520B (en) Network flow prediction method and device, computer equipment and readable storage medium
Ramakrishnan et al. Network traffic prediction using recurrent neural networks
US11223644B2 (en) Graphical structure model-based prevention and control of abnormal accounts
CN112910710B (en) Network flow space-time prediction method and device, computer equipment and storage medium
Hu et al. Individual-based optimal weight adaptation for heterogeneous epidemic spreading networks
CN113743594B (en) Network traffic prediction model establishment method and device, electronic equipment and storage medium
CN114372581B (en) Federal learning method and device based on block chain and computer equipment
WO2020237689A1 (en) Network structure search method and apparatus, computer storage medium, and computer program product
EP3537344B1 (en) Signal processing method and device based on spiking neural network
CN112734008A (en) Classification network construction method and classification method based on classification network
CN111314171A (en) Method, device and medium for predicting and optimizing SDN routing performance
CN114338416B (en) Space-time multi-index prediction method and device and storage medium
CN115801549B (en) Self-adaptive network recovery method, device and equipment based on key node identification
Theodoropoulos et al. Multi-service demand forecasting using graph neural networks
CN115984025A (en) Influence propagation estimation method and system based on deep learning graph network model
Li et al. Kernel adaptive filtering over complex networks
CN118511493A (en) Method and apparatus for managing network traffic through uncertainty
CN112396477B (en) Construction method and device of business prediction model
CN111368412B (en) Simulation model construction method and device for nursing demand prediction
CN113642592B (en) Training method of training model, scene recognition method and computer equipment
CN114880363A (en) Data center flow prediction system, training method and prediction method
CN113283589A (en) Updating method and device of event prediction system
CN112765491B (en) Link prediction method and device considering node local area link compactness
CN114297572B (en) Method and device for identifying node propagation influence in social network and computer equipment
CN115225140B (en) Satellite paging method and device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant