CN115567405A - Network flow gray prediction method based on self-adaptive feedback regulation mechanism - Google Patents

Network flow gray prediction method based on self-adaptive feedback regulation mechanism Download PDF

Info

Publication number
CN115567405A
CN115567405A CN202211199477.8A CN202211199477A CN115567405A CN 115567405 A CN115567405 A CN 115567405A CN 202211199477 A CN202211199477 A CN 202211199477A CN 115567405 A CN115567405 A CN 115567405A
Authority
CN
China
Prior art keywords
prediction
predicted
value
flow
sequence
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.)
Pending
Application number
CN202211199477.8A
Other languages
Chinese (zh)
Inventor
潘成胜
王英植
石怀峰
施建锋
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.)
Nanjing University of Information Science and Technology
Original Assignee
Nanjing University of Information Science and Technology
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 Nanjing University of Information Science and Technology filed Critical Nanjing University of Information Science and Technology
Priority to CN202211199477.8A priority Critical patent/CN115567405A/en
Publication of CN115567405A publication Critical patent/CN115567405A/en
Pending legal-status Critical Current

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
    • 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
    • H04L41/142—Network analysis or design using statistical or mathematical 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Algebra (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Pure & Applied Mathematics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses a network flow gray prediction method based on an adaptive feedback regulation mechanism, which is characterized in that a feedback correction function based on a correction factor is constructed and a feedback correction function is applied to each element in a normalized flow sequence based on a constructed target continuous flow sequence to be predicted of a network topology through normalization processing, then the flow sequence is predicted in a sliding window mode to obtain a predicted value, the window sliding predicted value of each moment is subtracted from the flow value corresponding to the next moment to calculate the error of the flow sequence, the magnitude relation between the three predicted values and a true value is obtained, the correction factor of the window sliding prediction of the next moment is updated according to the three relations, and finally, the inverse normalization processing is carried out to obtain the predicted flow sequence of the continuous flow sequence to be predicted.

Description

Network flow gray prediction method based on adaptive feedback regulation mechanism
Technical Field
The invention belongs to the field of information engineering, and particularly relates to a network flow gray prediction method based on a self-adaptive feedback regulation mechanism.
Background
With the rapid development of computer network technology, the number of network nodes and network applications is increasing explosively, and the scale of network traffic is becoming large and complex day by day, which prompts the network communication architecture to evolve from the traditional network of "fixed & planned" to the intelligent network of "perception & regulation". In an intelligent network communication architecture facing 'perception & regulation', key nodes have the functions of calculation and storage. Because of having the calculation function, the original large-capacity data is processed into small-capacity data, and the data content is compressed; due to the storage function, the burst characteristic of the network flow is weakened or enhanced, and the flow characteristic is changed along with the reduction or enhancement. The change affects the distribution rule and the statistical characteristic of flow data of the intelligent network in the heterogeneous link convergence process, so that the network flow characteristic description related to the change affects the intelligent network flow prediction precision. Currently, methods for predicting network traffic are mainly classified into linear prediction and nonlinear prediction.
(1) Linear prediction
The linear prediction method mainly comprises ARMA, ARIMA and FARIMA. An ARMA model-based combined prediction method is provided by the people of the Staloqiong and the like at the university of Sichuan, the combined prediction of network data by a plurality of linear models with different scales is realized, and a simulation result shows that the prediction mean error of the method is 10 -3 And the method has higher prediction accuracy. Shenyang industry university and field university provide a network flow prediction model of a Gaussian process regression model compensation autoregressive integrated sliding average (ARIMA) model, and simulation results show that the method has higher prediction accuracy and smaller prediction error. The Suqiang et al of Beijing university of transportation proposed a novel railway data network traffic prediction method based on a FARIMA model, and simulation results show that the method is more accurate than the traditional prediction method based on an ARMA model. The linear prediction method has a good prediction effect on a stationary sequence, but is difficult to capture implicit information of data in a non-stationary sequence.
(2) Non-linear prediction
The nonlinear prediction method mainly comprises prediction methods related to artificial intelligence, such as machine learning and deep learning. The Li Mingqi of Nanjing post and telecommunications university provides a method for realizing unsteady state video flow prediction based on a smooth auxiliary support vector machine (SSVM) model, and experimental results show that the provided SSVM model has great advantages in the aspects of prediction precision and statistical comparison. Hongsuk Yi of Korean science and technology information research institute proposes a traffic prediction deep learning model based on super-parameter search, and actual measurement shows that the model has a good traffic prediction effect. Although the method has high prediction accuracy, the overhead in the flow prediction process cannot be ignored due to the fact that mathematical modeling is complex and depends on large sample data.
Although the method can better realize the prediction of the network flow, the method cannot well balance the precision, the efficiency and the like. Aiming at the problems, the domestic scholars propose a method for realizing network traffic prediction according to a grey prediction theory. In order to improve the prediction accuracy of the gray model, a learner combines the gray prediction model with a Markov process, and obtains the rising, falling and stabilizing frequency of the flow at the next moment by counting the one-step transition probability of the current flow, so that the accuracy of flow prediction is improved. However, the gray traffic prediction algorithm of this method is more complex.
Disclosure of Invention
The invention aims to: the problem that the current flow prediction method cannot achieve good balance in the aspects of precision, efficiency and the like is solved. On the premise of ensuring higher flow prediction efficiency, the high-precision prediction of the network flow is realized.
In order to achieve the purpose, the invention provides the following technical scheme: a gray prediction method for network traffic based on adaptive feedback regulation mechanism is based on constructed network topology and target continuous traffic sequence to be predicted of the network topology
Figure 718910DEST_PATH_IMAGE001
Executing the following steps S1 to S8 to obtain the target continuous flow rate sequence to be predicted
Figure 130300DEST_PATH_IMAGE002
The predicted flow sequence of (1);
step S1, flow sequence
Figure 302655DEST_PATH_IMAGE003
Carrying out normalization processing to obtain a normalized flow sequence
Figure 723272DEST_PATH_IMAGE004
；
Step S2, constructing a correction factor
Figure 651301DEST_PATH_IMAGE005
Feedback correction function of
Figure 499172DEST_PATH_IMAGE006
；
Step S3, initializing a feedback correction function
Figure 893244DEST_PATH_IMAGE007
Of the feedback factor
Figure 851973DEST_PATH_IMAGE005
Is provided with
Figure 913469DEST_PATH_IMAGE005
Is initially of
Figure 666662DEST_PATH_IMAGE008
；
S4, according to the normalized flow sequence
Figure 797298DEST_PATH_IMAGE009
Feedback correction function
Figure 559717DEST_PATH_IMAGE010
And a feedback correction function
Figure 210142DEST_PATH_IMAGE011
Of the feedback factor
Figure 399814DEST_PATH_IMAGE005
For normalized flow series
Figure 768479DEST_PATH_IMAGE012
Applying a feedback modification function to each element in (1)
Figure 52699DEST_PATH_IMAGE010
Obtaining a sequence of preprocessed flows
Figure 88788DEST_PATH_IMAGE013
；
S5, adopting a window sliding one-step prediction mode pair
Figure 449362DEST_PATH_IMAGE014
Performing GM (1, 1) grey prediction, and sequentially reserving window sliding prediction values at each moment;
step S6, according to the window sliding predicted value at each moment in the step S5, respectively corresponding the window sliding predicted value at each moment to the window sliding predicted value at each moment
Figure 39743DEST_PATH_IMAGE015
Subtracting and comparing the flow values at the next moment to obtain the error of each flow sequence, and predicting the threshold error value based on the preset flow
Figure 143965DEST_PATH_IMAGE016
And obtaining the magnitude relation between three predicted values and the true value: the difference value between the predicted value and the true value is approximately equal to the preset range, the predicted value is greater than the true value, and the predicted value is smaller than the true value;
s7, when the difference value between the predicted value and the true value is approximately equal to the preset range, keeping the feedback factor of the window sliding prediction at the next time
Figure 768982DEST_PATH_IMAGE005
The size is not changed, i.e.
Figure 549725DEST_PATH_IMAGE017
Then, returning to execute the step S4;
when the predicted value is larger than the true value, the preset feedback factor is used for judging whether the predicted value is larger than the true value
Figure 892982DEST_PATH_IMAGE005
Modified gradient of
Figure 535316DEST_PATH_IMAGE018
Updating the correction factor of the window sliding prediction of the next time to be
Figure 14838DEST_PATH_IMAGE019
Then, returning to execute the step S4;
when the predicted value is smaller than the true value, the preset feedback factor is used
Figure 717215DEST_PATH_IMAGE005
Modified gradient of
Figure 282189DEST_PATH_IMAGE018
Updating the correction factor of the window sliding prediction of the next time to be
Figure 980411DEST_PATH_IMAGE020
Then, returning to execute the step S4;
and S8, performing inverse normalization on a series of flow prediction results obtained by traversing the flow sequence through the sliding window to obtain a predicted flow sequence.
Further, the step S1 specifically includes: to the flow sequence
Figure 314440DEST_PATH_IMAGE021
Normalization is carried out to obtain a normalized flow sequence
Figure 187718DEST_PATH_IMAGE004
，
Figure 239988DEST_PATH_IMAGE022
。
Further, in the aforementioned step S2, the correction factor-based structure is constructed as follows
Figure 489704DEST_PATH_IMAGE005
Feedback correction function of
Figure 943819DEST_PATH_IMAGE023
：
Figure 971686DEST_PATH_IMAGE024
Wherein,
Figure 776831DEST_PATH_IMAGE005
is a feedback factor.
Further, the step S4 specifically includes: generating a preconditioning flow order as follows
Figure 299080DEST_PATH_IMAGE025
:
Figure 873280DEST_PATH_IMAGE026
。
Further, the aforementioned step S5 includes the following sub-steps:
s5.1, flow sequence
Figure 88361DEST_PATH_IMAGE027
Initializing to obtain initialized flow sequence
Figure 364491DEST_PATH_IMAGE028
；
S5.2, presetting the sliding prediction window size asmWill be
Figure 690430DEST_PATH_IMAGE029
In a continuous neighborhood ofmOne flow sample as initialization sequence of each prediction
Figure 119137DEST_PATH_IMAGE028
：
Figure 505119DEST_PATH_IMAGE030
In which
Figure 19277DEST_PATH_IMAGE031
；
S5.3, calculating
Figure 883328DEST_PATH_IMAGE032
Generating a sequence by a single accumulation
Figure 681388DEST_PATH_IMAGE033
：
Figure 238272DEST_PATH_IMAGE034
；
Wherein,
Figure 974146DEST_PATH_IMAGE035
；
s5.4, calculating according to the following formula
Figure 907467DEST_PATH_IMAGE036
Is generated in the sequence of the close-proximity mean
Figure 310767DEST_PATH_IMAGE037
：
Figure 772972DEST_PATH_IMAGE038
，
Wherein,
Figure 513920DEST_PATH_IMAGE039
s5.5, constructed according to the following formula
Figure 985352DEST_PATH_IMAGE040
And
Figure 977579DEST_PATH_IMAGE041
gray differential equation of (c):
Figure 141844DEST_PATH_IMAGE042
wherein,
Figure 117890DEST_PATH_IMAGE043
the coefficient of development of the grey colour,
Figure 861856DEST_PATH_IMAGE044
the amount is gray effect;
s5.6, constructing the parameter vector of the gray differential equation of the step S5.5
Figure 489015DEST_PATH_IMAGE045
And solving the solution by using a least square method according to the following formula:
Figure 558602DEST_PATH_IMAGE046
，
wherein,
Figure 756365DEST_PATH_IMAGE047
；
s5.7, obtained by solving according to step S5.6
Figure 569600DEST_PATH_IMAGE048
、
Figure 801999DEST_PATH_IMAGE049
The time response function of the gray differential equation is calculated as follows
Figure 42487DEST_PATH_IMAGE050
：
Figure 976814DEST_PATH_IMAGE051
S5.8, solved according to step S5.7
Figure 593740DEST_PATH_IMAGE052
The predicted result is calculated as follows
Figure 415065DEST_PATH_IMAGE053
：
Figure 92034DEST_PATH_IMAGE054
。
Further, the magnitude relationship between the three predicted values and the true values obtained in the step S6 is specifically:
s6.1, judgment
Figure 264390DEST_PATH_IMAGE055
If the difference value is equal to the preset range, determining that the difference value between the predicted value and the true value is approximately equal to the preset range; otherwise, executing step S6.2;
s6.2, judging
Figure 153848DEST_PATH_IMAGE056
And if the judgment result is true, determining that the predicted value is greater than the true value, otherwise, determining that the predicted value is less than the true value.
Further, the foregoing network topology includes at least one traffic sending node and one traffic receiving node, where the traffic sending node and the traffic receiving node are connected in P2P communication.
Further, in the network traffic gray prediction method based on the adaptive feedback regulation mechanism, wireshark packet capturing software is adopted to continuously capture traffic to be predicted in the network topology.
Compared with the prior art, the network flow gray prediction method based on the self-adaptive feedback regulation mechanism has the following technical effects by adopting the technical scheme: accuracy of flow prediction
Figure 610106DEST_PATH_IMAGE057
Compared with GM (1, 1) model prediction
Figure 192397DEST_PATH_IMAGE058
The improvement is 6.49 percent compared with the prediction of an ewbogM (1, 1) model
Figure 852049DEST_PATH_IMAGE059
The improvement is 1.55 percent. Under an Intel (R) Core (TM) i7-9700 CPU @3GHz processor, 32GB memory and a 64-bit Win10 operating system, the average prediction time length of the algorithm is about 0.65 second. The result shows that the network flow gray prediction method based on the self-adaptive feedback regulation mechanism has advantages in the aspects of prediction precision and efficiency.
Drawings
FIG. 1 is an algorithmic flow chart of the method of the present invention.
FIG. 2 is a flow chart of GM (1, 1) Gray prediction.
Fig. 3 is a diagram of a window sliding one-step prediction principle.
Fig. 4 is a schematic diagram of an adaptive feedback adjustment mechanism.
Fig. 5 is a graph of the effect of the NS 3-based PyViz visualization network traffic simulation.
Fig. 6 is a diagram of self-similar network traffic data obtained by wireshark packet-grabbing software.
Fig. 7 is a comparison graph of predicted values and actual values of network traffic according to the method of the present invention.
FIG. 8 shows feedback factors during network traffic prediction
Figure 810778DEST_PATH_IMAGE060
The adjustment process of (2).
FIG. 9 is three initial
Figure 341116DEST_PATH_IMAGE061
Normalized absolute error contrast map.
FIG. 10 is a graph comparing predicted values and true values of network traffic based on a conventional GM (1, 1) model.
FIG. 11 is a comparison of predicted and true network traffic values based on the ewbogM (1, 1) model.
FIG. 12 is a graph of normalized absolute error of predicted versus true values for three models of flow.
Detailed Description
In order to better understand the technical content of the present invention, specific embodiments are described below with reference to the accompanying drawings.
In the present disclosure, aspects of the disclosure are described with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. Embodiments of the invention are not limited to those described in the figures. It is to be understood that the invention is capable of implementation in any of the numerous concepts and embodiments described hereinabove or described in the following detailed description, since the disclosed concepts and embodiments are not limited to any embodiment. Additionally, some aspects of the present disclosure may be used alone, or in any suitable combination with other aspects of the present disclosure.
Fig. 1 is an algorithm flowchart of the method of the present invention, which specifically includes the following steps:
s1, flow sequence
Figure 359888DEST_PATH_IMAGE062
Normalization is carried out to prevent the problem of insufficient prediction precision caused by overlarge absolute error, and the flow sequence after normalization is obtained
Figure 770751DEST_PATH_IMAGE063
Therein is that
Figure 267592DEST_PATH_IMAGE064
。
S2, constructing and based on correction factors
Figure 183595DEST_PATH_IMAGE065
Feedback correction function of
Figure 107689DEST_PATH_IMAGE066
(ii) a For most systems, the data sequence of the system is changed non-steadily, and particularly in an intelligent network, the network flow is more sudden, the system impact disturbance is more obvious, and the flow prediction precision is greatly reduced. To reduce the systemThe impact disturbance of the system, the flow prediction precision is improved, and a feedback correction function is designed
Figure 476353DEST_PATH_IMAGE067
，
Figure 776885DEST_PATH_IMAGE068
The method is composed of an exponential type weakening buffer operator and a feedback factor.
1) The de-emphasis buffer operator has the following theorem:
theorem 1: is provided with
Figure 62241DEST_PATH_IMAGE069
Is a non-negative system behavior data sequence, and
Figure 157236DEST_PATH_IMAGE070
let us order
Figure 13197DEST_PATH_IMAGE071
Then when
Figure 117419DEST_PATH_IMAGE072
In the case of a monotonically increasing sequence, a monotonically decreasing sequence, or a oscillating sequence,
Figure 476856DEST_PATH_IMAGE073
are all the de-emphasis buffer operators. Wherein
Figure 8332DEST_PATH_IMAGE074
Namely an exponential weakening buffer operator;
2) Based on theorem 1, a feedback correction function based on a feedback factor is constructed
Figure 600856DEST_PATH_IMAGE075
Is provided with
Figure 243190DEST_PATH_IMAGE076
Wherein,
Figure 988292DEST_PATH_IMAGE065
i.e. a feedback factor, the effect of which is to change by adjusting the size of itself
Figure 690669DEST_PATH_IMAGE077
The action strength of (1).
3) According to 2), proving
Figure 255642DEST_PATH_IMAGE078
Still belong to the exponential-type de-emphasis buffer operator. Since the flow sequence fluctuates with time and thus belongs to an oscillating sequence, based on theorem 1, the correlation proves to be as follows:
proof 1: when the temperature is higher than the set temperature
Figure 701667DEST_PATH_IMAGE079
When it is an oscillating sequence, it is set
Figure 550543DEST_PATH_IMAGE080
Due to the fact that
Figure 158242DEST_PATH_IMAGE081
，
Therefore, it is not only easy to use
Figure 476091DEST_PATH_IMAGE082
。
Due to the fact that
Figure 460227DEST_PATH_IMAGE083
Therefore, it is possible to
Figure 179922DEST_PATH_IMAGE084
。
Therefore, it is
Figure 958522DEST_PATH_IMAGE085
In the case of an oscillating sequence, the frequency of the oscillation,
Figure 15864DEST_PATH_IMAGE086
to weaken the buffer operator.
4) Discussion of the related Art
Figure 538112DEST_PATH_IMAGE087
And
Figure 112313DEST_PATH_IMAGE088
the relationship (c) in (c). Demonstration 1 shows that
Figure 327394DEST_PATH_IMAGE089
After that, the air conditioner is started to work,
Figure 88677DEST_PATH_IMAGE090
the buffer operator is still weakened and thus its buffer properties are not changed. According to the certification 1, there are
Figure 680195DEST_PATH_IMAGE091
At this time
Figure 358170DEST_PATH_IMAGE075
Satisfy a negative exponential function when
Figure 478572DEST_PATH_IMAGE065
When the number of the holes is increased, the size of the holes is increased,
Figure 258310DEST_PATH_IMAGE092
is enlarged and then makes
Figure 122360DEST_PATH_IMAGE075
Decreasing; in the same way, the method for preparing the composite material,
Figure 405574DEST_PATH_IMAGE065
will be reduced by
Figure 228037DEST_PATH_IMAGE075
And is increased.
Figure 213179DEST_PATH_IMAGE093
At this time
Figure 146500DEST_PATH_IMAGE075
Satisfy an exponential function when
Figure 284220DEST_PATH_IMAGE065
When the size of the pipe is increased, the pipe is enlarged,
Figure 12005DEST_PATH_IMAGE094
is also enlarged, and then
Figure 500755DEST_PATH_IMAGE095
Decreasing; in the same way, the method for preparing the composite material,
Figure 972188DEST_PATH_IMAGE065
will be reduced by
Figure 213682DEST_PATH_IMAGE096
And is increased. Finally, it can be concluded that
Figure 377947DEST_PATH_IMAGE097
In the case of an oscillating sequence, the frequency of the oscillation,
Figure 88414DEST_PATH_IMAGE098
with following
Figure 97959DEST_PATH_IMAGE099
Is increased and decreased, and is increased. The conclusion can be used in the flow prediction process
Figure 475850DEST_PATH_IMAGE065
The basis for the adjustment.
S3, initializing a feedback correction function
Figure 545437DEST_PATH_IMAGE100
Of the feedback factor
Figure 995398DEST_PATH_IMAGE101
Is provided with
Figure 808633DEST_PATH_IMAGE102
Is initially of
Figure 41031DEST_PATH_IMAGE103
Let us order
Figure 15941DEST_PATH_IMAGE104
；
S4, according to the normalized flow sequence
Figure 966579DEST_PATH_IMAGE105
Feedback correction function
Figure 567194DEST_PATH_IMAGE106
And a feedback correction function
Figure 654098DEST_PATH_IMAGE107
Of the feedback factor
Figure 331067DEST_PATH_IMAGE108
For normalized flow series
Figure 237843DEST_PATH_IMAGE109
Applying a feedback modification function to each element in (1)
Figure 392881DEST_PATH_IMAGE110
Generating a pre-treatment flow sequence according to the following formula
Figure 599872DEST_PATH_IMAGE111
:
Figure 165851DEST_PATH_IMAGE112
。
S5, adopting a window sliding one-step prediction mode pair
Figure 825502DEST_PATH_IMAGE113
Performing GM (1, 1) Gray prediction, whichThe grey prediction process is shown in fig. 2 and sequentially retains the window sliding prediction value at each moment; to ensure the accuracy of the flow prediction, the prediction window size is set to 5, i.e. the prediction window is set to be 5
Figure 784231DEST_PATH_IMAGE114
5 flow samples which are continuously adjacent are used as an initialization sequence of each prediction
Figure 845728DEST_PATH_IMAGE115
：
Figure 598920DEST_PATH_IMAGE116
。
The window sliding one-step prediction principle is shown in fig. 3, the prediction window continuously slides to the right side, and the continuous update of the traffic data is ensured, so that the sliding prediction of the whole traffic sequence is realized.
Computing
Figure 745868DEST_PATH_IMAGE117
Generating a sequence by a single accumulation
Figure 226397DEST_PATH_IMAGE118
Is provided with
Figure 142400DEST_PATH_IMAGE119
Wherein,
Figure 332073DEST_PATH_IMAGE120
computing
Figure 700738DEST_PATH_IMAGE121
To generate a sequence of closely adjacent means
Figure 1269DEST_PATH_IMAGE122
Is provided with
Figure 771779DEST_PATH_IMAGE123
Wherein,
Figure 118971DEST_PATH_IMAGE124
construction of
Figure 240511DEST_PATH_IMAGE125
And
Figure 813575DEST_PATH_IMAGE126
gray differential equation of
Figure 704170DEST_PATH_IMAGE127
Wherein,
Figure 235646DEST_PATH_IMAGE128
in order to be a coefficient of development of gray,
Figure 578902DEST_PATH_IMAGE129
the grey effect is indicated.
Constructing a parameter vector of a gray differential equation
Figure 470504DEST_PATH_IMAGE130
And solving it by using least square method, have
Figure 215606DEST_PATH_IMAGE131
Wherein,
Figure 917983DEST_PATH_IMAGE132
obtained by solving
Figure 482956DEST_PATH_IMAGE133
And
Figure 663402DEST_PATH_IMAGE134
calculating the time response function of the gray differential equation
Figure 263010DEST_PATH_IMAGE135
Is provided with
Figure 385556DEST_PATH_IMAGE136
Obtained by solving
Figure 703405DEST_PATH_IMAGE137
Calculating the predicted result
Figure 421962DEST_PATH_IMAGE138
Is provided with
Figure 141657DEST_PATH_IMAGE139
S6, classifying the possible situations of flow prediction according to the window sliding prediction value at each moment in the step S5, specifically: the window sliding predicted value at each moment is corresponding to the window sliding predicted value
Figure 185836DEST_PATH_IMAGE140
Subtracting and comparing the flow value at the next moment to obtain a flow sequence error, and predicting a threshold error value based on preset flow
Figure 725402DEST_PATH_IMAGE141
And obtaining the magnitude relation between three predicted values and real values: the difference value between the predicted value and the true value is approximately equal to the preset range, the predicted value is greater than the true value, and the predicted value is smaller than the true value;
is provided with
Figure 762497DEST_PATH_IMAGE142
Time pair
Figure 336698DEST_PATH_IMAGE143
The predicted value of the flow at the moment is
Figure 286199DEST_PATH_IMAGE144
，
Figure 313061DEST_PATH_IMAGE145
True value of time flow of
Figure 904579DEST_PATH_IMAGE146
，
S6.1, judgment
Figure 67707DEST_PATH_IMAGE147
If yes, the system is considered to be
Figure 694168DEST_PATH_IMAGE148
To indicate that
Figure 473905DEST_PATH_IMAGE149
Time of day
Figure 337956DEST_PATH_IMAGE150
Has moderate action intensity, so that
Figure 621170DEST_PATH_IMAGE151
Is not changed to maintain
Figure 443632DEST_PATH_IMAGE152
Time of day
Figure 445086DEST_PATH_IMAGE150
The action strength of (c); determining that the predicted value is approximately equal to the true value; otherwise, executing step S6.2;
s6.2, judging
Figure 96516DEST_PATH_IMAGE153
If yes, it indicates that
Figure 499816DEST_PATH_IMAGE154
Time of day
Figure 493179DEST_PATH_IMAGE150
Too small in action strength, and therefore needs to be reduced
Figure 450771DEST_PATH_IMAGE155
To increase
Figure 187783DEST_PATH_IMAGE150
Thereby enhancing
Figure 180010DEST_PATH_IMAGE156
Time of day
Figure 327963DEST_PATH_IMAGE150
Determining that the predicted value is greater than the true value, otherwise, determining that the predicted value is less than the true value, i.e. determining
Figure 304009DEST_PATH_IMAGE157
Indicates that
Figure 313554DEST_PATH_IMAGE158
Time of day
Figure 425866DEST_PATH_IMAGE150
Too strong, and therefore, needs to be increased
Figure 761033DEST_PATH_IMAGE155
To reduce
Figure 693217DEST_PATH_IMAGE150
Thereby weakening
Figure 21299DEST_PATH_IMAGE143
Time of day
Figure 253697DEST_PATH_IMAGE150
The action strength of (1).
And S7, setting up an adjusting standard of an adaptive feedback adjusting mechanism according to the magnitude relation between the three predicted values and the true value obtained in the step S6, wherein the feedback adjusting principle is shown in figure 4.
When the predicted value is approximately equal to the true valueKeeping the feedback factor of the window sliding prediction of the next time
Figure 228606DEST_PATH_IMAGE159
The size is not changed, i.e.
Figure 179245DEST_PATH_IMAGE160
Then, returning to execute the step S4;
when the predicted value is larger than the true value, the preset feedback factor is used for judging whether the predicted value is larger than the true value
Figure 796171DEST_PATH_IMAGE161
Modified gradient of
Figure 883075DEST_PATH_IMAGE162
Updating the correction factor for the window sliding prediction of the next time to be
Figure 281083DEST_PATH_IMAGE163
In which
Figure 453439DEST_PATH_IMAGE164
Then, returning to execute the step S4;
when the predicted value is smaller than the true value, the preset feedback factor is used for judging whether the predicted value is smaller than the true value
Figure 608476DEST_PATH_IMAGE165
Modified gradient of
Figure 549888DEST_PATH_IMAGE162
Updating the correction factor for the window sliding prediction of the next time to be
Figure 397758DEST_PATH_IMAGE166
In which
Figure 791830DEST_PATH_IMAGE164
Then, returning to execute the step S4;
and S8, performing inverse normalization on a series of flow prediction results obtained by passing the window sliding window through the flow sequence to obtain a predicted flow sequence.
The network topology under the multilink convergence scene is modeled and simulated based on NS3, and the simulation effect of the PyViz visualized network flow obtained through simulation is shown in FIG. 5. The constructed network topology structure has 9 nodes, wherein D1-D8 nodes represent flow sending nodes, R nodes represent flow receiving nodes, P2P communication links are adopted among the nodes, flow at 500 continuous moments is captured through wireshark packet capturing software, and the obtained simulated flow sequence is shown in figure 6. As can be seen from fig. 6, the network traffic obtained by simulation has strong burstiness, and therefore, a high requirement is put on the prediction accuracy of the traffic prediction method.
Setting an initial value of a feedback factor
Figure 999826DEST_PATH_IMAGE167
Error threshold value
Figure 61323DEST_PATH_IMAGE168
Feedback regulation of gradient
Figure 80095DEST_PATH_IMAGE169
Based on the parameter setting condition, the network traffic prediction is performed on the simulation traffic sequence, and the obtained traffic prediction effect is shown in fig. 7. As can be seen from FIG. 7, the predicted flow rate curve substantially coincides with the true flow rate curve, and the correlation coefficient of the flow rate prediction accuracy is calculated
Figure 695884DEST_PATH_IMAGE170
Approaching to 1, which shows that the flow prediction data of the invention is very close to the real data and the prediction effect is good. To verify the adaptive feedback gradient adjustment mechanism, the adaptive feedback gradient adjustment mechanism is applied to the flow prediction process
Figure 458304DEST_PATH_IMAGE171
The adjustment process of (2) is tracked and the data tracking curve is shown in fig. 8. In the context of figure 8 of the drawings,
Figure 108728DEST_PATH_IMAGE171
the value of (b) is in a fluctuating state, indicating that the adaptive feedback gradient regulation mechanism participates in the whole flow predictionAnd (6) measuring.
To verify the universality of the method, choose
Figure 547668DEST_PATH_IMAGE172
And
Figure 181912DEST_PATH_IMAGE173
respectively predicting the network flow and normalizing the absolute value error sum of the two predictions
Figure 482443DEST_PATH_IMAGE174
The predicted normalized absolute value errors are compared as shown in fig. 9. As can be seen in fig. 9, at the beginning of the flow prediction,
Figure 987374DEST_PATH_IMAGE175
and
Figure 347948DEST_PATH_IMAGE176
the normalized absolute error of the three is larger, and the normalized absolute errors of the three are gradually consistent with the prediction, which shows that the normalized absolute errors of the three are gradually consistent with each other
Figure 938330DEST_PATH_IMAGE177
In the flow prediction process, the adaptive adjustment of the network flow prediction precision can be realized, so that the adaptive adjustment of the network flow prediction precision can be realized
Figure 291819DEST_PATH_IMAGE178
The value of the method does not influence the flow prediction precision, so the method has universality.
In order to verify the advantages of the method, the conventional GM (1, 1) gray prediction model and the ewboGM (1, 1) gray prediction model with the addition of the exponential weakening buffer operator are respectively adopted to predict the simulated network traffic, and the prediction result is compared with the network traffic prediction result of the method, and the prediction effect is shown in fig. 10, fig. 11 and fig. 12.
Wherein, FIG. 10 is a comparison graph of the predicted value and the true value based on the conventional GM (1, 1) model, from which it can be seen that there is a large prediction error at the extreme value of the flow curveCalculating the flow prediction accuracy
Figure 182415DEST_PATH_IMAGE179
(ii) a FIG. 11 is a comparison graph of predicted values and actual values based on the ewboGM (1, 1) model, from which it can be seen that the prediction error is alleviated, but a large prediction error still exists at the extreme value of the flow curve, and the flow prediction accuracy is calculated
Figure 448311DEST_PATH_IMAGE180
The prediction precision is improved; FIG. 12 is a graph of normalized absolute error comparison of predicted and true values for the flow of three models, from which it can be seen that the GM (1, 1) gray predictive model has the largest normalized absolute error; although the normalized absolute value error of the ewboGM (1, 1) model is reduced, the normalized absolute value error is still obvious; compared with the former two models, the flow prediction method has the minimum normalized absolute error.
Through calculation, the flow prediction precision of the method of the invention
Figure 791568DEST_PATH_IMAGE181
Prediction compared to the GM (1, 1) model
Figure 433902DEST_PATH_IMAGE182
Improved by 6.49 percent compared with the prediction of the ewbogM (1, 1) model
Figure 913425DEST_PATH_IMAGE183
1.55% is improved, the normalized absolute error of flow prediction is minimum, and the average prediction time length of the algorithm is about 0.65 second under an Intel (R) Core (TM) i7-9700 CPU @3GHz processor, a 32GB memory and a 64-bit Win10 operating system.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims (8)

1. A kind ofThe grey prediction method of the network flow based on the self-adaptive feedback regulation mechanism is characterized in that the method is based on the constructed network topology and the target continuous flow sequence to be predicted of the network topology
Figure 579846DEST_PATH_IMAGE002
Executing the following steps S1 to S8 to obtain the target continuous flow rate sequence to be predicted
Figure 28144DEST_PATH_IMAGE003
The predicted flow sequence of (1);
step S1, flow sequence
Figure 339040DEST_PATH_IMAGE004
Carrying out normalization processing to obtain a normalized flow sequence
Figure 281719DEST_PATH_IMAGE005
；
Step S2, constructing a correction factor
Figure 627250DEST_PATH_IMAGE006
Feedback correction function of
Figure 184133DEST_PATH_IMAGE008
；
Step S3, initializing a feedback correction function
Figure 982325DEST_PATH_IMAGE009
Of the feedback factor
Figure 977963DEST_PATH_IMAGE006
Is provided with
Figure 430197DEST_PATH_IMAGE006
Is at an initial value of
Figure 954720DEST_PATH_IMAGE010
；
S4, according to the normalized flow sequence
Figure 443470DEST_PATH_IMAGE012
Feedback correction function
Figure 977219DEST_PATH_IMAGE013
And a feedback correction function
Figure 31763DEST_PATH_IMAGE014
Feedback factor of
Figure 196028DEST_PATH_IMAGE006
For normalized flow sequence
Figure 719544DEST_PATH_IMAGE015
Applying a feedback modification function to each element in (1)
Figure 525826DEST_PATH_IMAGE016
Obtaining a sequence of pre-processing flows
Figure 966035DEST_PATH_IMAGE018
；
S5, adopting a window sliding one-step prediction mode pair
Figure 832360DEST_PATH_IMAGE019
Performing GM (1, 1) grey prediction, and sequentially retaining the window sliding prediction value at each moment;
step S6, according to the window sliding predicted value at each moment in the step S5, respectively corresponding the window sliding predicted value at each moment to the window sliding predicted value at each moment
Figure 295702DEST_PATH_IMAGE020
Subtracting and comparing the flow values at the next moment to obtain the error of each flow sequence, and predicting the threshold error value based on the preset flow
Figure 154943DEST_PATH_IMAGE021
And obtaining the magnitude relation between three predicted values and the true value: the difference value between the predicted value and the true value is approximately equal to the preset range, the predicted value is greater than the true value, and the predicted value is smaller than the true value;
s7, when the difference value between the predicted value and the true value is approximately equal to the preset range, keeping the feedback factor of the window sliding prediction at the next time
Figure 449658DEST_PATH_IMAGE006
The size is not changed, i.e.
Figure 752463DEST_PATH_IMAGE022
Then, returning to execute the step S4;
when the predicted value is larger than the true value, the preset feedback factor is used for judging whether the predicted value is larger than the true value
Figure 234260DEST_PATH_IMAGE006
Corrected gradient of
Figure 851186DEST_PATH_IMAGE023
Updating the correction factor for the window sliding prediction of the next time to be
Figure 751140DEST_PATH_IMAGE024
Then, returning to execute the step S4;
when the predicted value is smaller than the true value, the preset feedback factor is used
Figure 224847DEST_PATH_IMAGE006
Modified gradient of
Figure 397202DEST_PATH_IMAGE023
Updating the correction factor for the window sliding prediction of the next time to be
Figure 348978DEST_PATH_IMAGE025
Then, returning to execute the step S4;
and S8, performing inverse normalization on a series of flow prediction results obtained by traversing the flow sequence through the sliding window to obtain a predicted flow sequence.
2. The method for predicting the gray of the network traffic based on the adaptive feedback regulation mechanism according to claim 1, wherein the step S1 specifically comprises: to the flow sequence
Figure 618285DEST_PATH_IMAGE026
Normalization is carried out to obtain a normalized flow sequence
Figure 526809DEST_PATH_IMAGE005
，
Figure 983198DEST_PATH_IMAGE027
。
3. The grey prediction method for network traffic based on adaptive feedback adjustment mechanism according to claim 2, characterized in that in step S2, the correction factor-based grey prediction method is constructed according to the following formula
Figure 941927DEST_PATH_IMAGE006
Feedback correction function of
Figure 800162DEST_PATH_IMAGE028
：
Figure 615671DEST_PATH_IMAGE029
Wherein,
Figure 762618DEST_PATH_IMAGE006
is a feedback factor.
4. According to the rightThe grey prediction method for network traffic based on the adaptive feedback adjustment mechanism according to claim 3, wherein step S4 specifically comprises: generating a pretreatment flow sequence according to the following formula
Figure 72508DEST_PATH_IMAGE030
:
Figure 785249DEST_PATH_IMAGE031
。
5. The grey prediction method for network traffic based on adaptive feedback adjustment mechanism according to claim 4, characterized in that step S5 comprises the following sub-steps:
s5.1, flow sequence
Figure 771660DEST_PATH_IMAGE032
Initializing to obtain initialized flow sequence
Figure 202641DEST_PATH_IMAGE033
；
S5.2, presetting the sliding prediction window size asmWill be
Figure 549178DEST_PATH_IMAGE034
In a continuous neighborhood ofmOne flow sample as initialization sequence of each prediction
Figure 585267DEST_PATH_IMAGE033
：
Figure 742579DEST_PATH_IMAGE036
Wherein
Figure 660856DEST_PATH_IMAGE037
；
S5.3, calculating
Figure 765078DEST_PATH_IMAGE039
Generating a sequence by a single accumulation
Figure 186832DEST_PATH_IMAGE041
：
Figure 531357DEST_PATH_IMAGE043
；
Wherein,
Figure 874614DEST_PATH_IMAGE044
；
s5.4, calculating according to the following formula
Figure 579265DEST_PATH_IMAGE045
To generate a sequence of closely adjacent means
Figure 386684DEST_PATH_IMAGE046
：
Figure 89060DEST_PATH_IMAGE047
，
Wherein,
Figure DEST_PATH_IMAGE048
s5.5, constructing according to the following formula
Figure 765286DEST_PATH_IMAGE049
And
Figure 8048DEST_PATH_IMAGE050
gray differential equation of (a):
Figure 607657DEST_PATH_IMAGE051
wherein,
Figure 543252DEST_PATH_IMAGE052
the coefficient of development of the grey colour,
Figure 657838DEST_PATH_IMAGE053
the amount of gray effect;
s5.6, constructing the parameter vector of the gray differential equation of the step S5.5
Figure 641975DEST_PATH_IMAGE054
And solving the solution by using a least square method according to the following formula:
Figure 909139DEST_PATH_IMAGE056
，
wherein,
Figure 15636DEST_PATH_IMAGE058
；
s5.7, obtained by solving according to step S5.6
Figure 617518DEST_PATH_IMAGE060
、
Figure DEST_PATH_IMAGE061
The time response function of the gray differential equation is calculated as follows
Figure 202083DEST_PATH_IMAGE062
：
Figure DEST_PATH_IMAGE063
S5.8, solved according to step S5.7
Figure 556710DEST_PATH_IMAGE064
The predicted result is calculated as follows
Figure 834108DEST_PATH_IMAGE066
：
Figure DEST_PATH_IMAGE067
。
6. The network flow gray prediction method based on the adaptive feedback regulation mechanism according to claim 5, wherein the magnitude relationship between three predicted values and the true value obtained in step S6 is specifically:
s6.1, judgment
Figure 392128DEST_PATH_IMAGE068
If the difference value is equal to the preset range, determining that the difference value between the predicted value and the true value is approximately equal to the preset range; otherwise, executing step S6.2;
s6.2, judgment
Figure DEST_PATH_IMAGE069
And if the judgment result is true, determining that the predicted value is greater than the true value, otherwise, determining that the predicted value is less than the true value.
7. The method of claim 6, wherein the network topology includes at least one traffic sending node and one traffic receiving node, and the traffic sending node and the traffic receiving node are in P2P communication connection.
8. The method according to claim 7, wherein wireshark packet capturing software is used to continuously capture traffic to be predicted in the network topology.
CN202211199477.8A 2022-09-29 2022-09-29 Network flow gray prediction method based on self-adaptive feedback regulation mechanism Pending CN115567405A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211199477.8A CN115567405A (en) 2022-09-29 2022-09-29 Network flow gray prediction method based on self-adaptive feedback regulation mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211199477.8A CN115567405A (en) 2022-09-29 2022-09-29 Network flow gray prediction method based on self-adaptive feedback regulation mechanism

Publications (1)

Publication Number Publication Date
CN115567405A true CN115567405A (en) 2023-01-03

Family

ID=84742114

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211199477.8A Pending CN115567405A (en) 2022-09-29 2022-09-29 Network flow gray prediction method based on self-adaptive feedback regulation mechanism

Country Status (1)

Country Link
CN (1) CN115567405A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116566841A (en) * 2023-05-09 2023-08-08 北京有元科技有限公司 Flow trend prediction method and device based on network flow query
CN116599860A (en) * 2023-07-11 2023-08-15 南京信息工程大学 Network traffic gray prediction method based on reinforcement learning

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116566841A (en) * 2023-05-09 2023-08-08 北京有元科技有限公司 Flow trend prediction method and device based on network flow query
CN116566841B (en) * 2023-05-09 2023-12-01 北京有元科技有限公司 Flow trend prediction method based on network flow query
CN116599860A (en) * 2023-07-11 2023-08-15 南京信息工程大学 Network traffic gray prediction method based on reinforcement learning
CN116599860B (en) * 2023-07-11 2023-10-13 南京信息工程大学 Network traffic gray prediction method based on reinforcement learning

Similar Documents

Publication Publication Date Title
US11409347B2 (en) Method, system and storage medium for predicting power load probability density based on deep learning
CN115567405A (en) Network flow gray prediction method based on self-adaptive feedback regulation mechanism
CN110245801A (en) A kind of Methods of electric load forecasting and system based on combination mining model
US20160357169A1 (en) Model Predictive Control with Uncertainties
CN106249599A (en) A kind of network control system fault detection method based on neural network prediction
CN113641722A (en) Long-term time series data prediction method based on variant LSTM
CN107506865A (en) A kind of load forecasting method and system based on LSSVM optimizations
CN110471768B (en) FastPCA-ARIMA-based load prediction method
CN113852432A (en) RCS-GRU model-based spectrum prediction sensing method
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
Chen et al. Novel adaptive neural networks control with event-triggered for uncertain nonlinear system
CN115983438A (en) Method and device for determining operation strategy of data center terminal air conditioning system
CN110880044A (en) Markov chain-based load prediction method
CN109242190A (en) Mid-long term load forecasting method and system based on BFGS-FA optimization fractional order gray model
CN109034497A (en) Prediction technique, system, medium and the equipment of polycrystalline reduction process energy consumption value
CN114972429B (en) Target tracking method and system for cloud edge cooperative self-adaptive reasoning path planning
CN102055694B (en) Non-linear system identification method based on particle swarm
Liu et al. Wavelet neural network based link quality prediction for fluctuating low power wireless links
CN115423149A (en) Incremental iterative clustering method for energy internet load prediction and noise level estimation
Zhang et al. Adaptive multiple model control for a class of nonlinear discrete time systems: second-level adaption design approach
CN114826948A (en) SDN network flow prediction method based on graph convolution network
Zhang et al. A hybrid load forecasting method based on neural network in smart grid
Yao et al. Preconditioned temporal difference learning
Li et al. Quality Prediction of Strip in Finishing Rolling Process Based on GBDBN‐ELM
CN110472191A (en) A kind of the service evaluation calculation method and device of dynamic self-adapting

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