CN111371626B - Bandwidth prediction method based on neural network - Google Patents

Bandwidth prediction method based on neural network Download PDF

Info

Publication number
CN111371626B
CN111371626B CN202010204155.2A CN202010204155A CN111371626B CN 111371626 B CN111371626 B CN 111371626B CN 202010204155 A CN202010204155 A CN 202010204155A CN 111371626 B CN111371626 B CN 111371626B
Authority
CN
China
Prior art keywords
bandwidth
neural network
time
future
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
CN202010204155.2A
Other languages
Chinese (zh)
Other versions
CN111371626A (en
Inventor
张旭
张欣宇
薛雨
马展
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.)
Chengdu Yunge Zhili Technology Co ltd
Nanjing University
Original Assignee
Chengdu Yunge Zhili Technology Co ltd
Nanjing 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 Chengdu Yunge Zhili Technology Co ltd, Nanjing University filed Critical Chengdu Yunge Zhili Technology Co ltd
Priority to CN202010204155.2A priority Critical patent/CN111371626B/en
Publication of CN111371626A publication Critical patent/CN111371626A/en
Application granted granted Critical
Publication of CN111371626B publication Critical patent/CN111371626B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level
    • H04L43/0894—Packet rate

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Environmental & Geological Engineering (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses a bandwidth prediction method based on a neural network. The method comprises the following steps: (1) establishing a neural network model and training, wherein the input of the neural network model is the bandwidth change characteristic at the historical moment, and the output is the bandwidth predicted value at the future moment; (2) recording real-time bandwidth data of a user and sequencing according to time; sampling the real-time bandwidth data according to a set sampling interval; (3) extracting time sequence characteristics of the bandwidth; (4) inputting the time sequence characteristics of the bandwidth into the trained neural network model, and calculating the bandwidth quantization factor at the future moment; (5) and converting the calculated bandwidth quantization factor into the predicted bandwidth at the future moment. The invention provides a method for quantitatively reflecting future network state changes, which predicts future bandwidth on the basis of bandwidth measurement, makes up for the defect of long time consumption of bandwidth measurement and can effectively guide internet application to deal with network changes.

Description

Bandwidth prediction method based on neural network
Technical Field
The invention relates to the technical field of network communication, in particular to a network condition prediction method, and specifically relates to a bandwidth prediction method based on a neural network.
Background
With the gradual popularization of the internet, the network communication technology is continuously improved, the demand of users on high-quality networks is gradually improved, and many daily internet applications depend on the stability of bandwidth. In order to improve the actual experience of the user, the real-time available bandwidth of the user needs to be known. For example, when a user watches a video, if the available bandwidth is smaller than the size of each second of video frame, the video will be jammed, and the user experience will be reduced. When a developer develops live video software, the fluctuation of bandwidth is considered, a video code rate and resolution adjustment strategy is added, and when a user network is not good, the bandwidth requirement is reduced, and the aim of smoothly playing videos is fulfilled.
Currently, mainstream methods for measuring network bandwidth include a single packet algorithm and the like, and the measured bandwidth is a user bandwidth at the current time or in a period of time before the current time. However, the bandwidth in the network changes constantly, the bandwidth obtained by direct measurement cannot reflect the network state in a period of time in the future, and the currently mainstream method for measuring the network bandwidth is long in time consumption, complex in calculation and incapable of supporting high concurrency, so that the requirement of low measurement interval cannot be met.
The bandwidth measurement interval is long because the bandwidth changes continuously with time. In practical application, the historical bandwidth measurement value may be greatly different from the bandwidth measurement value at the current moment, and the guiding significance is small. Compared with the bandwidth data directly measured, the bandwidth prediction method based on the neural network can improve the accuracy on the whole, but how to use the neural network to predict the bandwidth and how to improve the prediction effect are all technical problems to be solved.
Disclosure of Invention
In order to overcome the technical defects that the existing bandwidth measurement interval is long, and the measured bandwidth cannot reflect the future bandwidth change trend, the invention provides a bandwidth prediction method based on a neural network.
The technical scheme adopted by the invention is as follows:
a bandwidth prediction method based on a neural network comprises the following specific steps:
step 1, establishing a neural network model and training, wherein the input of the neural network model is the bandwidth change characteristic at the historical moment, the output of the neural network model is the bandwidth predicted value at the future moment, MSE loss is used as a loss function, and a self-adaptive moment estimation optimizer is used for optimizing the neural network model;
step 2, recording the real-time bandwidth data of the user, and sequencing according to time; sampling the real-time bandwidth data according to a set sampling interval;
step 3, extracting the time sequence characteristics of the bandwidth:
Figure BDA0002419999350000011
wherein O represents bandwidth data before filtering, and M represents bandwidth data after filtering;
and 4, inputting the time sequence characteristics of the bandwidth into the trained neural network model, and calculating the bandwidth quantization factor at the future moment:
Figure BDA0002419999350000021
wherein n represents the order of the bandwidth data;
step 5, converting the calculated bandwidth quantization factor into a predicted future time bandwidth:
Figure BDA0002419999350000022
further, the neural network model in step 1 adopts an MSE loss function, and an adaptive moment estimation optimizer is used for optimizing the training process.
On the basis of measuring the user bandwidth, the bandwidth change characteristics are used as input, and the bandwidth in a period of time in the future is predicted through calculation of the neural network. Compared with the prior art, the method for predicting the bandwidth by the neural network reduces the error between the predicted bandwidth and the actual bandwidth, and overcomes the defect that the predicted bandwidth is not accurate enough due to long time consumption of the conventional bandwidth measurement. According to the predicted bandwidth output by the method, the aim of guiding the internet to apply and adjust the network strategy can be achieved.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a diagram of a four-layer neural network structure in the embodiment of the present invention.
Detailed Description
The present invention is further described in detail with reference to the following examples, which should not be construed as limiting the scope of the present invention, but rather as embodying the invention in such insubstantial modifications and variations thereof as would be apparent to those skilled in the art based on the teachings set forth herein.
FIG. 1 is a flow chart of the bandwidth prediction method based on neural network of the present invention. The method comprises the following specific steps:
step 1, establishing a four-layer neural network model, wherein the network structure is shown in fig. 2.
The first layer of network acts on input data, utilizes an LSTM layer (Long-Short-Memory) to extract characteristics of an input time sequence, and stores a part of Long-term and Short-term characteristics into the network, wherein the length of the input time sequence of the first layer is 5, the number of input characteristics is 1, the number of hidden layer nodes is 12, the forgetting gate threshold value is 0.5, and the activation function is leak _ relu (0.3).
The second layer is a fully connected layer with hidden layer node number of 64, and is used for mapping the low-dimensional data features into a high-dimensional space domain.
The third layer is a dropout layer with a dropout probability of 0.5, and is used for randomly inactivating hidden layer neurons and preventing overfitting of a neural network.
The fourth layer is a full-connection layer with the hidden layer node number of 1 and is used for converting high-order characteristics into predicted future network bandwidth.
The training of the neural network adopts a supervised learning mode, and the step length is 0.001. Wherein, 30 groups of data are taken as a batch during training and sent to the model for training. The initial learning rate of the adaptive moment estimation optimizer is 0.001, every 100 batches of data, the learning rate becomes the previous 0.98, and the minimum learning rate does not exceed 0.00001. The parameters of the adaptive moment estimation optimizer adopt default values in the tensoflow framework, namely beta1Coefficient of 0.9, beta2The coefficient is 0.999, and the epsilon coefficient is 1 e-8. The input of the neural network is historical network state characteristics obtained after specific processing, the label is a bandwidth factor obtained after the specific processing, and the following steps are required for obtaining the historical network state characteristics and the label:
step 11, performing median filtering with kernel of 3 on the data of the sampling frequency of 3 minutes to obtain filtered data which is recorded as (M)0,M1,M2,M3,M4…Mn). The bandwidth data before filtering is noted as (O)0,O1,O2,O3,O4…On). Calculating a relative difference value for each corresponding point
Figure BDA0002419999350000031
i is 0 … … n, n indicates the order of the bandwidth data.
Step 12, traversing the sampling points, and taking the relative difference (sigma) of the nearest 5 sampling pointst，σt-1，σt-2，σt-3，σt-4) As input I to a neural network at time tt。
Step 13, taking the bandwidth quantization factor of the next sampling point as the output L of the neural network when the time is tt，
Figure BDA0002419999350000032
The verification training result adopts a cross verification mode, namely, a set of input and labels is divided into two parts, namely a training set for neural network training and a test set for neural network effect testing. When the performance of the neural network model on the test set reaches the requirement, training is finished and the neural network model is saved.
And 2, sequencing the measured bandwidth records according to time, sampling at intervals of 5 minutes, and outputting the sampled data.
Step 3, performing median filtering with kernel of 3 on the sampled data obtained in the step 2 to obtain filtered data which is recorded as (M)0,M1,M2,M3,M4…Mn). The bandwidth data before filtering is noted as (O)0,O1,O2,O3,O4…On). Calculating a relative difference value for each corresponding point
Figure BDA0002419999350000033
I.e., the characteristics of the historical bandwidth data, as output from step 3.
Step 4, taking the relative difference (sigma) of the nearest 5 sampling pointsn，σn-1，σn-2，σn-3，σn-4) As input to the step 1 neural network. After the operation of the neural network, the quantization factor p of the bandwidth at the future moment is obtained and output, and the mathematical meaning of the p is
Figure BDA0002419999350000034
Step 5, the quantization factor p of the bandwidth obtained in step 4 is calculated according to a formula
Figure BDA0002419999350000035
Calculating to obtain a bandwidth predicted value at the next moment,
it should be noted that: in step 1, the neural network may be any form of neural network, and is not limited to the four-layer neural network given in this embodiment. In step 2, the sampling interval may be any sampling interval, including 1 minute, 2 minutes, 3 minutes, 15 minutes, and the like, but theoretically, the smaller the sampling interval is, the more accurate the predicted bandwidth data at the future time is, and the more instructive is. The bandwidth characteristics extracted in step 3 may be replaced by any form of time sequence characteristics that can be obtained through mathematical operations, including a single or a series of numbers after processing such as smoothing and filtering, for example, data characteristics such as first order difference, second order difference, and smooth value of bandwidth data, and combinations thereof. The number of the sampling points in the step 4 can be changed arbitrarily according to the actual effect, and the meaning represented by the quantization factor should be changed correspondingly with the meaning of the label during the neural network training, for example, when the first-order difference value is taken as the training label, i.e., Li＝Oi+1-OiIn practical application, the output quantization factor should also mean p ═ Oi+1-OiIn which O isi+1Is the bandwidth value of the next time, OiIs the bandwidth value, L, at the current timeiThe training label is the training label at the current moment, and p is the output of the neural network at the current moment, namely the bandwidth quantization factor calculated at the current moment. The formula for converting the quantization factor into the prediction bandwidth in step 5 should be consistent with the meaning of the quantization factor.

Claims (3)

1. A bandwidth prediction method based on a neural network is characterized by comprising the following specific steps:
step 1, establishing a neural network model and training, wherein the input of the neural network model is the bandwidth change characteristic at the historical moment, and the output is the bandwidth predicted value at the future moment;
step 2, recording the real-time bandwidth data of the user, and sequencing according to time; sampling the real-time bandwidth data according to a set sampling interval;
step 3, extracting the time sequence characteristics of the bandwidth:
Figure FDA0002906107610000011
wherein O represents bandwidth data before filtering, and M represents bandwidth data after filtering;
and 4, inputting the time sequence characteristics of the bandwidth into the trained neural network model, and calculating the bandwidth quantization factor at the future moment:
Figure FDA0002906107610000012
wherein n represents the order of the bandwidth data;
and 5, converting the calculated bandwidth quantization factor into the predicted future time bandwidth.
2. The method as claimed in claim 1, wherein in step 1, the neural network model adopts an MSE loss function, and an optimizer is used to optimize the training process.
3. The method according to claim 1, wherein in the step 5, the predicted future time bandwidth is:
Figure FDA0002906107610000013
CN202010204155.2A 2020-03-20 2020-03-20 Bandwidth prediction method based on neural network Active CN111371626B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010204155.2A CN111371626B (en) 2020-03-20 2020-03-20 Bandwidth prediction method based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010204155.2A CN111371626B (en) 2020-03-20 2020-03-20 Bandwidth prediction method based on neural network

Publications (2)

Publication Number Publication Date
CN111371626A CN111371626A (en) 2020-07-03
CN111371626B true CN111371626B (en) 2021-05-28

Family

ID=71211915

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010204155.2A Active CN111371626B (en) 2020-03-20 2020-03-20 Bandwidth prediction method based on neural network

Country Status (1)

Country Link
CN (1) CN111371626B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113179175B (en) * 2021-03-16 2022-12-20 国家电网有限公司信息通信分公司 Real-time bandwidth prediction method and device for power communication network service
CN113033783B (en) * 2021-04-07 2024-04-26 苏州瑞立思科技有限公司 Bandwidth adjustment method based on BP neural network
CN113645487B (en) * 2021-07-14 2022-05-17 上海交通大学 Code rate self-adaptive distribution method
CN113709582B (en) * 2021-08-25 2024-02-06 三星电子（中国）研发中心 Bandwidth measurement method and device
CN114465915A (en) * 2021-12-22 2022-05-10 天翼云科技有限公司 CDN bandwidth prediction method, device and system and electronic equipment
CN114389975B (en) * 2022-02-08 2024-03-08 北京字节跳动网络技术有限公司 Network bandwidth estimation method, device and system, electronic equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108234496A (en) * 2018-01-05 2018-06-29 宝牧科技（天津）有限公司 A kind of method for predicting based on neural network
CN110648030A (en) * 2019-10-31 2020-01-03 吉林大学 Method and device for predicting seawater temperature

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11023353B2 (en) * 2018-08-24 2021-06-01 Vmware, Inc. Processes and systems for forecasting metric data and anomaly detection in a distributed computing system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108234496A (en) * 2018-01-05 2018-06-29 宝牧科技（天津）有限公司 A kind of method for predicting based on neural network
CN110648030A (en) * 2019-10-31 2020-01-03 吉林大学 Method and device for predicting seawater temperature

Also Published As

Publication number Publication date
CN111371626A (en) 2020-07-03

Similar Documents

Publication Publication Date Title
CN111371626B (en) Bandwidth prediction method based on neural network
CN110210658B (en) Prophet and Gaussian process user network flow prediction method based on wavelet transformation
CN108062561B (en) Short-time data flow prediction method based on long-time and short-time memory network model
CN109544197B (en) User loss prediction method and device
CN112468326B (en) Access flow prediction method based on time convolution neural network
CN111563706A (en) Multivariable logistics freight volume prediction method based on LSTM network
WO2021129086A1 (en) Traffic prediction method, device, and storage medium
WO2012149705A1 (en) Long-term prediction method and apparatus of network traffic
CN114637911B (en) Method for recommending next interest point of attention fusion perception network
CN112732738B (en) Adaptive network data acquisition method based on multi-objective optimization and related equipment
CN114694379B (en) Traffic flow prediction method and system based on self-adaptive dynamic graph convolution
CN111680786A (en) Time sequence prediction method based on improved weight gating unit
CN110633859A (en) Hydrological sequence prediction method for two-stage decomposition integration
CN113205223A (en) Electric quantity prediction system and prediction method thereof
CN115308558A (en) Method and device for predicting service life of CMOS (complementary Metal oxide semiconductor) device, electronic equipment and medium
CN117296061A (en) Diffusion model with improved accuracy and reduced computing resource consumption
CN109065176B (en) Blood glucose prediction method, device, terminal and storage medium
CN111404751B (en) RTT (round trip time) prediction method based on RNN (neural network)
CN111191113A (en) Data resource demand prediction and adjustment method based on edge computing environment
CN114581119A (en) Flow prediction method and device
CN115713044B (en) Method and device for analyzing residual life of electromechanical equipment under multi-condition switching
CN112307434A (en) Actually measured flow intelligent correction method based on Gaussian filtering
CN108134687B (en) Gray model local area network peak flow prediction method based on Markov chain
CN116308450A (en) Information generation and model training method, electronic equipment and storage medium
EP4205393A1 (en) Rate control machine learning models with feedback control for video encoding

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