CN104506378A - Data flow prediction device and method - Google Patents

Abstract

The present application provides a device and method for predicting data traffic. The device includes: a signal acquisition module; a determination module, which is used to determine that the waveform corresponding to the data flow signal has self-similarity according to the data flow signal; a signal processing module, which is used to decompose and reconstruct the waveform by wavelet analysis technology; the determination module, It is also used to determine at least one approximation signal and a signal whose stationarity is less than a first preset threshold among the plurality of detail signals as the first type of data traffic signal; and determine a signal whose stationarity is greater than or equal to the first preset threshold It is the second type of data flow signal; the calculation module is used to predict the first type of data flow signal by using the compressed sensing model; use the linear model to predict the second type of data flow signal; synthesize the first type of prediction results and the second type forecast result. By adopting the device or method of the present application, the prediction accuracy for data traffic can be improved.

Description

A device and method for predicting data traffic

technical field

The present application relates to the communication field, in particular to a device and method for predicting data traffic.

Background technique

With the continuous development of communication technology, the data traffic generated by communication between electronic devices such as mobile terminals is also increasing.

In the prior art, in order to reasonably allocate network resources, during the process of transmitting data traffic, the amount of data to be transmitted in a future period of time is predicted based on actual data traffic information. According to the prediction results, the network resources are allocated to transmit data more efficiently.

However, in the prior art, for data traffic prediction, usually a certain method is mainly used for prediction. However, this method usually only has high prediction accuracy for the data traffic generated by one or several specific types of services, while for the data traffic generated by other services, the prediction accuracy is often low.

Contents of the invention

The purpose of this application is to provide a device and method for predicting data traffic, which can predict data traffic by using a prediction method that matches the waveform characteristics of data traffic according to the waveform characteristics of business data traffic, thereby improving the data traffic. prediction accuracy.

In order to achieve the above object, the application provides the following scheme:

According to a first possible implementation manner of the first aspect of the present application, the present application provides a device for predicting data traffic, including:

A signal acquisition module, configured to acquire a data flow signal within a preset time length;

A determining module, configured to determine, according to the data flow signal, that the waveform corresponding to the data flow signal has self-similarity;

A signal processing module, configured to decompose and reconstruct the waveform using wavelet analysis technology to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and multiple detail signals;

The determination module is further configured to determine the at least one approximation signal and a signal whose stationarity is less than a first preset threshold among the plurality of detail signals as the first type of data traffic signal;

determining a signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and the plurality of detail signals as a second type of data traffic signal;

A calculation module, configured to predict the first type of data traffic signal using a compressed sensing model to obtain a first type of prediction result;

Predicting the second type of data traffic signal using a linear model to obtain a second type of prediction result;

Synthesizing the first type of prediction result and the second type of prediction result.

In combination with the second possible implementation of the first aspect, the signal acquisition module is specifically used for:

Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order;

From the data traffic corresponding to each sampling point arranged in chronological order, the data traffic corresponding to the sampling point within a preset time length is intercepted.

In combination with the third possible implementation manner of the first aspect, the determining module is specifically used for:

Calculate the Hurst exponent of the waveform corresponding to the data flow signal by using the rescaled range analysis method;

It is determined that the value of the Hurst exponent is greater than a second preset threshold.

With reference to a specific implementation manner of the second possible implementation manner of the first aspect, the determining module is specifically configured to:

According to the formula calculating a sample autocorrelation function of the at least one approximation signal and the plurality of detail signals;

Will The signal is determined as the first type of data flow signal;

Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1,2, 3,...K; θ is the first preset threshold.

With reference to the fourth possible implementation manner of the first aspect, the determining module is further configured to:

determining that the waveform corresponding to the data flow signal does not have self-similarity;

The calculation module is also used for:

When the determining module determines that the waveform corresponding to the data flow signal does not have self-similarity, a linear model is used to predict the data flow signal.

According to the first possible implementation manner of the second aspect of the present application, the present application provides a method for predicting data traffic, including:

Obtaining data flow signals within a preset time length;

According to the data flow signal, determine that the waveform corresponding to the data flow signal has self-similarity;

Decomposing and reconstructing the waveform by using wavelet analysis technology to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and a plurality of detail signals;

Determining a signal whose stationarity is less than a first preset threshold among the at least one approximation signal and the plurality of detail signals as a first-type data traffic signal;

determining a signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and the plurality of detail signals as a second-type data traffic signal,

Predicting the first type of data flow signal using a compressed sensing model to obtain a first type of prediction result;

Predicting the second type of data traffic signal using a linear model to obtain a second type of prediction result;

Synthesizing the first type of prediction result and the second type of prediction result.

With reference to the second possible implementation of the second aspect, the acquisition of the data flow signal within a preset time length specifically includes:

Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order;

From the data traffic corresponding to each sampling point arranged in chronological order, the data traffic corresponding to the sampling point within a preset time length is intercepted.

With reference to the third possible implementation manner of the second aspect, the determining that the waveform corresponding to the data traffic signal has self-similarity specifically includes:

Calculate the Hurst exponent of the waveform corresponding to the data flow signal by using the rescaled range analysis method;

It is determined that the value of the Hurst exponent is greater than a second preset threshold.

With reference to a specific implementation manner of the second possible implementation manner of the second aspect, the determination of the at least one approximation signal and a signal whose stationarity is smaller than a first preset threshold among the at least one approximation signal and multiple detail signals as the first A class of data traffic signals, specifically including:

According to the formula calculating a sample autocorrelation function of the at least one approximation signal and the plurality of detail signals;

Will The signal is determined as the first type of data flow signal;

Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1,2, 3,...K; θ is the first preset threshold.

In combination with the fourth possible implementation of the second aspect, it also includes:

determining that the waveform corresponding to the data flow signal does not have self-similarity;

A linear model is used to predict the data traffic signal.

According to the specific embodiments provided by the application, the application discloses the following technical effects:

The device or method for predicting data traffic disclosed in this application determines that the waveform corresponding to the data traffic signal has self-similarity, and uses wavelet analysis technology to decompose and reconstruct the waveform; according to the stationarity, the reconstructed Classify the data flow signal; for the signal with low stationarity, use the compressed sensing model to predict; for the signal with high stationarity, use the linear model to predict; synthesize each prediction result; according to the waveform characteristics of the service data flow The data flow is predicted by using a prediction method matching the waveform characteristic of the data flow, thereby improving the prediction accuracy of the data flow.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some of the present application. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without paying creative labor.

FIG. 1 is a schematic diagram of an application scenario of a method for predicting data traffic in the present application;

Fig. 2 is a structural diagram of a device for predicting data traffic of the present application;

FIG. 3 is a flow chart of a method for predicting data traffic of the present application;

FIG. 4 is a flow chart of another method for predicting data traffic of the present application;

FIG. 5 is a structural diagram of a computing node of the present application.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the application with reference to the drawings in the embodiments of the application. Apparently, the described embodiments are only some of the embodiments of the application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

In order to make the above objects, features and advantages of the present application more obvious and comprehensible, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

FIG. 1 is a schematic diagram of an application scenario of the method for predicting data traffic of the present application. The application scenarios in FIG. 1 are just examples, and do not include all application scenarios. As shown in Figure 1, the network architecture in this scenario includes:

A gateway 10, a first cell 101, a second cell 102 and a third cell 103. Each cell includes at least one base station and a user terminal. The base stations of each cell, and each base station and the gateway 10 can communicate with each other. The network architecture may be a cellular network. The gateway 10 may be located in the core network, and the base station may be located in the access network.

The method and device for predicting data traffic of the present invention can be applied at the gateway 10 or at the base station.

FIG. 2 is a structural diagram of a device for predicting data traffic according to the present application. As shown in Figure 2, the device may include:

A signal acquisition module 201, configured to acquire a data traffic signal within a preset time length;

The data flow signal may represent the amount of data transmitted through the network within a certain period of time. A plurality of data traffic signals can be acquired within the preset time length. The multiple data traffic signals have a sequence in time.

A determination module 202, configured to determine, according to the data flow signal, that the waveform corresponding to the data flow signal has self-similarity;

In layman's terms, a self-similar object is nearly or indeed similar to a part of it. If a curve is self-similar, it means that every part of the curve has a small part similar to it. Self-similarity is an important property of fractals.

Generally, the waveform with self-similarity has poor stability, and the waveform without self-similarity (that is, non-self-similar) has high stability, and a linear model can be used for prediction.

The signal processing module 203 is configured to decompose and reconstruct the waveform using wavelet analysis technology to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and multiple detail signals;

The waveform can be decomposed into at least one approximation signal and multiple detail signals by using Haar wavelet analysis technique and mallat algorithm. In practical applications, the number of decomposed detail signals can be selected and set according to actual requirements. For example, the waveform can be decomposed into at least one approximation signal and three detail signals.

The determination module 202 is further configured to determine the at least one approximation signal and a signal whose stationarity is less than a first preset threshold among the at least one approximation signal and the plurality of detail signals as a first-type data traffic signal;

A signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and the plurality of detail signals is determined as a second-type data traffic signal.

Regarding stationarity, it can be understood as follows: Assume that each element in the time series X={X _t }, t=1,2,... is randomly generated by a random process that obeys a certain probability distribution, if X satisfies The following conditions:

1. The mean value E(X)=μ is a constant independent of time t;

2. Variance var(X)=σ ² is a constant independent of time t;

3. The covariance cov(X _t ,X _t+k )=γ _k is a constant that is only related to the time interval k but not to the time t.

Then, the time series X is stationary.

When testing the stationarity of time series by definition, the sequence is required to be infinitely long, which is difficult to achieve in reality. Usually, in practical applications, sample sequences of finite length are collected. In the embodiment of the present invention, the stationarity of the sample sequence is judged by the sample autocorrelation function. A detailed description of the autocorrelation function will be given later.

The approximation signal and each detail signal may be judged separately to judge whether the stationarity of each signal is less than a first preset threshold. When the stationarity of the signal is less than a preset threshold, it can be determined that the signal does not have stationarity; when the stationarity of the signal is greater than or equal to the first preset threshold, it can be judged that the signal has stationarity.

A calculation module 204, configured to predict the first type of data traffic signal using a compressed sensing model to obtain a first type of prediction result;

Predicting the second type of data traffic signal using a linear model to obtain a second type of prediction result;

Synthesizing the first type of prediction result and the second type of prediction result.

Compressed sensing (Compressed sensing), also known as compressed sampling (Compressive sampling), sparse sampling (Sparse sampling), compressed sensing. Compressed sensing is a new sampling theory. By exploiting the sparse characteristics of the signal, random sampling can be used to obtain discrete samples of the signal under conditions much smaller than the Nyquist sampling rate, and then the signal can be reconstructed through a nonlinear reconstruction algorithm.

Usually, a part of the reconstructed data traffic signals belongs to the first type of data traffic signals, and another part belongs to the second type of data traffic signals. For each of the reconstructed data traffic signals, a prediction result can be obtained, and the prediction result can represent a part of the data volume to be transmitted in the network at a future moment. The combination may be adding up the prediction results. After synthesis, a total forecast result can be obtained, and the total forecast result can represent the total amount of data that needs to be transmitted in the network in the future. According to the predicted total amount of data that needs to be transmitted in the network in the future, relevant devices in the network can allocate network resources, so as to perform data transmission more efficiently.

In summary, in the embodiment shown in Figure 2, by determining that the waveform corresponding to the data flow signal has self-similarity, wavelet analysis technology is used to decompose and reconstruct the waveform; Classify the structured data traffic signals; for signals with low stationarity, use the compressed sensing model to predict; for signals with high stationarity, use a linear model for prediction; synthesize each prediction result; Waveform characteristics, using a prediction method that matches the waveform characteristics of the data flow to predict the data flow, thereby improving the prediction accuracy of the data flow.

In practical applications, the signal acquisition module 201 can specifically be used for:

Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order;

From the data traffic corresponding to each sampling point arranged in chronological order, the data traffic corresponding to the sampling point within a preset time length is intercepted.

In practical applications, the determining module 202 may specifically be used for:

Calculate the Hurst exponent of the waveform corresponding to the data flow signal by using the rescaled range analysis method;

It is determined that the value of the Hurst exponent is greater than a second preset threshold.

In practical applications, the determination module 202 can also be specifically used for:

According to the formula calculating a sample autocorrelation function of the at least one approximation signal and the plurality of detail signals;

Will The signal is determined as the first type of data flow signal;

Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1,2, 3,...K; θ is the first preset threshold.

In practical applications, the determining module 202 may also be used for:

determining that the waveform corresponding to the data flow signal does not have self-similarity;

The calculation module 204 can also be used for:

When the determining module determines that the waveform corresponding to the data flow signal does not have self-similarity, a linear model is used to predict the data flow signal.

The present application also provides a method for predicting data traffic.

FIG. 3 is a flow chart of a method for predicting data traffic in the present application. As shown in Figure 3, the method may include:

Step 301: Acquiring data traffic signals within a preset time length;

The data flow signal may represent the amount of data transmitted through the network within a certain period of time. A plurality of data traffic signals can be acquired within the preset time length. The multiple data traffic signals have a sequence in time.

Step 302: According to the data flow signal, determine that the waveform corresponding to the data flow signal has self-similarity;

In layman's terms, a self-similar object is nearly or indeed similar to a part of it. If a curve is self-similar, it means that every part of the curve has a small part similar to it. Self-similarity is an important property of fractals.

Generally, the waveform with self-similarity has poor stability, and the waveform without self-similarity (that is, non-self-similar) has high stability, and a linear model can be used for prediction.

Step 303: Using wavelet analysis technology to decompose and reconstruct the waveform to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and multiple detail signals;

The waveform can be decomposed into at least one approximation signal and multiple detail signals by using Haar wavelet analysis technique and mallat algorithm. In practical applications, the number of decomposed detail signals can be selected and set according to actual requirements. For example, the waveform can be decomposed into at least one approximation signal and three detail signals.

Step 304: Determining the signal whose stationarity is smaller than a first preset threshold among the at least one approximation signal and the plurality of detail signals as the first type of data traffic signal;

Regarding stationarity, it can be understood as follows: Assume that each element in the time series X={X _t }, t=1,2,... is randomly generated by a random process that obeys a certain probability distribution, if X satisfies The following conditions:

1. The mean value E(X)=μ is a constant independent of time t;

2. Variance var(X)=σ ² is a constant independent of time t;

3. The covariance cov(X _t ,X _t+k )=γ _k is a constant that is only related to the time interval k but not to the time t.

Then, the time series X is stationary.

When testing the stationarity of time series by definition, the sequence is required to be infinitely long, which is difficult to achieve in reality. Usually, in practical applications, sample sequences of finite length are collected. In the embodiment of the present invention, the stationarity of the sample sequence is judged by the sample autocorrelation function. A detailed description of the autocorrelation function will be given later.

The approximation signal and each detail signal may be judged separately to judge whether the stationarity of each signal is less than a first preset threshold. When the stationarity of the signal is less than a preset threshold, it can be determined that the signal does not have stationarity; when the stationarity of the signal is greater than or equal to the first preset threshold, it can be judged that the signal has stationarity.

Step 305: Determining a signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and multiple detail signals as the second type of data traffic signal;

Step 306: Predict the first type of data traffic signal using a compressed sensing model to obtain a first type of prediction result;

Compressed sensing (Compressed sensing), also known as compressed sampling (Compressive sampling), sparse sampling (Sparse sampling), compressed sensing. Compressed sensing is a new sampling theory. By exploiting the sparse characteristics of the signal, random sampling can be used to obtain discrete samples of the signal under conditions much smaller than the Nyquist sampling rate, and then the signal can be reconstructed through a nonlinear reconstruction algorithm.

Step 307: Predict the second type of data traffic signal using a linear model to obtain a second type of prediction result;

Step 308: Synthesize the first type of prediction result and the second type of prediction result.

Usually, a part of the reconstructed data traffic signals belongs to the first type of data traffic signals, and another part belongs to the second type of data traffic signals. For each of the reconstructed data traffic signals, a prediction result can be obtained, and the prediction result can represent a part of the data volume to be transmitted in the network at a future moment. The combination may be adding up the prediction results. After synthesis, a total forecast result can be obtained, and the total forecast result can represent the total amount of data to be transmitted in the network at a future moment. According to the predicted total amount of data that needs to be transmitted in the network in the future, relevant devices in the network can allocate network resources, so as to perform data transmission more efficiently.

To sum up, in the embodiment shown in Figure 3, by determining that the waveform corresponding to the data flow signal has self-similarity, wavelet analysis technology is used to decompose and reconstruct the waveform; Classify the structured data traffic signals; for signals with low stationarity, use the compressed sensing model to predict; for signals with high stationarity, use a linear model for prediction; synthesize each prediction result; Waveform characteristics, using a prediction method that matches the waveform characteristics of the data flow to predict the data flow, thereby improving the prediction accuracy of the data flow.

It should be noted that, in the above embodiment, when it is determined that the waveform corresponding to the data traffic signal has no self-similarity, the following step may be further included: using a linear model to predict the data traffic signal. Because non-self-similar waveforms usually have high stationarity, a linear model can be used to directly predict the data flow signal.

It should also be noted that, in the above steps, the acquisition of the data flow signal within the preset time length can specifically be carried out in the following manner:

Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order; from the data flow corresponding to each sampling point arranged in chronological order, intercept within the preset time length The data flow corresponding to the sampling point.

The determining that the waveform corresponding to the data traffic signal has self-similarity may specifically adopt the following methods:

calculating the Hurst exponent of the waveform corresponding to the data flow signal by using a rescaled range analysis method; judging whether the value of the Hurst exponent is greater than a second preset threshold.

Rescaled Range Analysis, also known as R/S analysis, is a method proposed by hydrologist Hurst on the basis of a large number of empirical studies.

Usually, the second preset threshold can be set to 0.5.

Using the rescaled range analysis method to calculate the Hurst exponent of the waveform corresponding to the data flow signal, the following methods can be adopted:

The data traffic corresponding to the sampling points within the preset time length may constitute a service time series. For business time series X={X _i ,i>1}, assume that the sequence length is N, that is, X={X ₁ ,X ₂ ,...,X _N }. divide the sequence into subsequences, where each subsequence has the same length, which is n.

For the kth subsequence Calculate its mean and standard deviation $S_k=\sqrt{\frac{1}{\mathrm{no}}\underset{i=(k-1)\mathrm{no}+1}{\overset{\mathrm{k\; n}}{\mathrm{\Σ}}}{(x_i-{\mathrm{E.}}_k)}^2}.$

For the kth subsequence, calculate the deviation between each sample point Z _i,k and E _k , that is, Xi _,k =Z _i,k -E _k , the subscript i,k represents the ith in the kth subsequence element. Calculate the cumulative deviation of the kth subsequence i=1,2,...,n, calculate the cumulative deviation limit difference R _k =max{Y _1,k ,...,Y _n,k }-min{Y _{1, k} ,...,Y _n,k }.

Calculate the S _k and R _k of all subsequences, and finally calculate the R/S statistics of the original sequence, namely Obviously, is a function of n. Statistics show that The relationship with n is approximately expressed as (R/S) _n ～cn ^H , and logarithms are taken on both sides to obtain log(R/S) _n = logc+Hlogn, where logc is a constant. When n takes different values in the logarithmic coordinate system, all (logn,log(R/S) _n ) points can be observed, and these points can be observed to be approximately on a straight line, and the straight line can be obtained by linear fitting through the least square method The slope is the H parameter.

The H parameter can express the self-similarity of the business. The value range of H is 0<H<1. If 0.5<H<1, it indicates self-similarity; the larger the H value, the stronger the self-similarity.

The determining that the stationarity of the reconstructed data traffic signal is less than a first preset threshold may be specifically performed in the following manner:

According to the formula Calculating a sample autocorrelation function of the data flow corresponding to the sampling point within the preset time length;

judge whether it is established;

Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1, 2,3,...K; K is the upper limit of k, which can be set according to requirements. θ is the first preset threshold.

The values of K and θ are related to N. When N is greater than 100, K is 15 and θ is 1.5. In the actual simulation, it will be adjusted according to the specific prediction object.

The first type of data flow signal is predicted using a compressed sensing model, specifically in the following manner:

Step A: Construct the first detail signal matrix;

Specifically, for the first type of data traffic signal, a business time series is extracted.

Sparse representation of business time series. That is, each element in the business time series is compared against a sparsification threshold. If it is lower than the thinning threshold, set the element to 0, otherwise it remains unchanged. The second threshold may be determined according to simulation results.

The detailed signal time series after sparse representation is used as training data, combined with the data to be predicted represented by 0, to construct the first detailed signal matrix X (m, n represent the number of users and the number of time slots, respectively). The matrix must be a sparse matrix, that is, the number of 0 elements in the matrix is much greater than the number of non-zero elements.

$\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}& <span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}\mathrm{<span\; class="notranslate">\; no</span>}}& \mathrm{<span\; class="notranslate">\; 0</span>}& <span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>& \mathrm{<span\; class="notranslate">\; 0</span>}\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&CenterDot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>& <span\; class="notranslate">\; \&Center\; Dot;</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>& {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; mn</span>}}& \mathrm{<span\; class="notranslate">\; 0</span>}& <span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span>& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]$

Step B: Determining the temporal correlation of the first detail signal matrix

The present invention is by analyzing the characteristics of the D1 detail signal matrix X after sparse, the time correlation matrix is designed as:

Where ω _m,n (i),i=1,...,N _T is the weight coefficient obtained by linear regression of the known N _T data in the first detail signal matrix X, that is, obtained by solving the following linear equations.

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ;</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>\\ <span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&Center\; Dot;</span>\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ;</span>\end{array}\right.$

Step C: Considering the temporal correlation, an optimization model about the approximate decomposition of the first detail signal matrix is established.

Considering the first detail signal matrix and its corresponding time correlation, the optimization model is established as:

$\mathrm{<span\; class="notranslate">\; min</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; LR</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; (</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; L</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; LR</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; T</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

in Both β and α are weight parameters, and the time correlation matrix T describes the stationarity characteristics of the data in the first detail signal matrix in the time dimension, and includes the correlation degree information between matrix elements in the time dimension. Among them, the values of α and β are related to the low-rank property of matrix decomposition and time correlation. That is, if α is too large, it means that the matrix decomposition results are more low-rank, and if β is too large, it means that the matrix decomposition results are more time-dependent. In general, the recommended value of α is 0.1, and the recommended value of β is 0.001, which can be adjusted according to the actual situation.

Step D: Solve the optimization model and determine the factorization matrices L and R.

The optimization problem is solved using the alternating least squares algorithm, and the low-rank decomposition matrices L and R are obtained.

Step E: Reconstruct the first detail signal matrix to obtain the data to be predicted.

According to the obtained low-rank decomposition matrices L and R, the original D1 detail signal matrix is reconstructed: That is, the predicted value of the data to be predicted in the first detail signal matrix is obtained. Specifically, in step A, the data to be predicted is substituted into the original matrix with a value of 0, and after reconstruction, the data at the position corresponding to the original 0 value is the predicted value (usually a non-zero value) for the data to be predicted. The chronological forecasts in the matrix correspond to different time points.

In the above embodiment, the linear model is used to predict the second type of data traffic signal, and the following methods can be used specifically:

Express the traffic at the current moment as the linear weighted sum of the traffic at the previous n moments, that is

Where {a _t }, t=1,2,...,N is a white noise sequence coefficient Using least squares estimation, the model order n can be obtained from the final prediction error criterion function (FPE). where NID stands for independent normal distribution.

Least squares estimation parameters The detailed process can be:

From the expression of the linear weighted sum of the business volume at the first n moments, the following linear equations can be obtained

Expressed in matrix form as in

According to multiple regression theory, The least squares estimate of is

The calculation formula of the FPE function can be Where N is the original sequence length, n is the model order, has been given above, namely

The FPE criterion takes the n value when the variance of one-step forecast error approaches the minimum value as the model order, that is L is the upper bound of the preset model order, and the value is related to the number of training samples.

FIG. 4 is a flow chart of another method for predicting data traffic according to the present application. As shown in Figure 4, the method may include:

Step 401: Acquiring data flow signals within a preset time length;

Step 402: According to the data flow signal, judge whether the waveform corresponding to the data flow signal has self-similarity, and obtain a judgment result; when the judgment result indicates that the waveform corresponding to the data flow signal does not have self-similarity, Execute step 410 .

Step 403: When the judgment result indicates that the waveform corresponding to the data traffic signal has self-similarity, use wavelet analysis technology to decompose and reconstruct the waveform to obtain the reconstructed data traffic signal; the reconstruction The subsequent data traffic signal includes at least one approximation signal and a plurality of detail signals;

Step 404: Determine whether the stationarity of the reconstructed data traffic signal is less than a first preset threshold; if it is less than the first preset threshold, perform step 405; otherwise, perform step 406;

Step 405: Determine the reconstructed data traffic signal as the first type of data traffic signal;

Step 406: Determine the reconstructed data traffic signal as the second type of data traffic signal;

Step 407: Using the compressed sensing model to predict the first type of data traffic signal, and obtain the first type of prediction result;

Step 408: Predict the second type of data traffic signal using a linear model to obtain a second type of prediction result;

Step 409: Synthesize the first type of prediction result and the second type of prediction result.

Step 410: Use a linear model to predict the data flow signal.

In the embodiment shown in Fig. 4, by judging whether the waveform corresponding to the data traffic signal has self-similarity, when there is similarity, the wavelet analysis technology is used to decompose and reconstruct the waveform; Classify the reconstructed data flow signal; for the signal with low stationarity, use the compressed sensing model to predict; for the signal with high stationarity, use the linear model to predict; synthesize each prediction result; when the data flow When the waveform corresponding to the signal has no similarity, the linear model is used to predict the data flow signal; for the data flow signal with similarity and the data flow signal without similarity, the matching prediction method can be used to predict the data flow prediction, so as to improve the prediction accuracy of the data flow, and expand the scope of application of the method for predicting the data flow of the present application.

In addition, the embodiment of the present application also provides a computing node. The computing node may be a host server with computing capabilities, or a personal computer PC, or a portable portable computer or terminal, etc. The specific embodiments of the present application do not The specific implementation of computing nodes is limited.

FIG. 5 is a structural diagram of a computing node of the present application. As shown in FIG. 5, computing node 500 includes:

A processor (processor) 510 , a communication interface (Communications Interface) 520 , a memory (memory) 530 , and a bus 540 .

The processor 510 , the communication interface 520 , and the memory 530 communicate with each other through the bus 540 .

The processor 510 is configured to execute the program 532 .

Specifically, the program 532 may include program codes including computer operation instructions.

The processor 510 may be a central processing unit CPU, or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

The memory 530 is used to store a program 532 . The memory 530 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one magnetic disk memory.

The memory 530 is used to store computer-executable instructions. The processor 510 is connected to the memory 530 through the bus. When the device for predicting data flow is running, the processor 510 executes the instructions stored in the memory 530. The computer executes the instruction 532, so that the processor executes the method for predicting data traffic provided in the embodiment of the present application.

Finally, it should also be noted that in this text, relational terms such as first and second etc. are only used to distinguish one entity or operation from another, and do not necessarily require or imply that these entities or operations, any such actual relationship or order exists. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Through the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware platforms, and of course all can be implemented by hardware, but in many cases the former is better implementation. Based on this understanding, all or part of the contribution made by the technical solution of the present application to the background technology can be embodied in the form of software products, and the computer software products can be stored in storage media, such as ROM/RAM, magnetic disks, optical disks, etc. , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods described in various embodiments or some parts of the embodiments of the present application.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part.

In this paper, specific examples are used to illustrate the principles and implementation methods of the application. The descriptions of the above embodiments are only used to help understand the method and core idea of the application; meanwhile, for those of ordinary skill in the art, according to the application Thoughts, there will be changes in specific implementation methods and application ranges. To sum up, the contents of this specification should not be understood as limiting the application.

Claims (10)

1. A device for predicting data traffic, comprising: A signal acquisition module, configured to acquire a data flow signal within a preset time length; A determining module, configured to determine, according to the data flow signal, that the waveform corresponding to the data flow signal has self-similarity; A signal processing module, configured to decompose and reconstruct the waveform using wavelet analysis technology to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and multiple detail signals; The determination module is further configured to determine the at least one approximation signal and a signal whose stationarity is less than a first preset threshold among the plurality of detail signals as the first type of data traffic signal; determining a signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and the plurality of detail signals as a second type of data traffic signal; A calculation module, configured to predict the first type of data traffic signal using a compressed sensing model to obtain a first type of prediction result; Predicting the second type of data traffic signal using a linear model to obtain a second type of prediction result; Synthesizing the first type of prediction result and the second type of prediction result. 2. The device according to claim 1, wherein the signal acquisition module is specifically used for: Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order; From the data traffic corresponding to each sampling point arranged in chronological order, the data traffic corresponding to the sampling point within a preset time length is intercepted. 3. The device according to claim 1, wherein the determining module is specifically used for: Calculate the Hurst exponent of the waveform corresponding to the data flow signal by using the rescaled range analysis method; It is determined that the value of the Hurst exponent is greater than a second preset threshold. 4. The device according to claim 2, wherein the determining module is specifically used for: According to the formula calculating a sample autocorrelation function of the at least one approximation signal and the plurality of detail signals; Will The signal is determined as the first type of data traffic signal; Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1,2, 3,...K; θ is the first preset threshold. 5. The device according to claim 1, wherein the determining module is also used for: determining that the waveform corresponding to the data flow signal does not have self-similarity; The calculation module is also used for: When the determining module determines that the waveform corresponding to the data flow signal does not have self-similarity, a linear model is used to predict the data flow signal. 6. A method for predicting data traffic, comprising: Obtaining data flow signals within a preset time length; According to the data flow signal, determine that the waveform corresponding to the data flow signal has self-similarity; Decomposing and reconstructing the waveform by using wavelet analysis technology to obtain a reconstructed data flow signal; the reconstructed data flow signal includes at least one approximation signal and a plurality of detail signals; Determining a signal whose stationarity is less than a first preset threshold among the at least one approximation signal and the plurality of detail signals as a first-type data traffic signal; determining a signal whose stationarity is greater than or equal to the first preset threshold among the at least one approximation signal and the plurality of detail signals as a second-type data traffic signal, Predicting the first type of data flow signal using a compressed sensing model to obtain a first type of prediction result; Predicting the second type of data traffic signal using a linear model to obtain a second type of prediction result; Synthesizing the first type of prediction result and the second type of prediction result. 7. The method according to claim 6, wherein the acquiring the data flow signal within a preset time length specifically comprises: Sampling the generated data flow according to the preset time interval to obtain the data flow corresponding to each sampling point arranged in chronological order; From the data traffic corresponding to each sampling point arranged in chronological order, the data traffic corresponding to the sampling point within a preset time length is intercepted. 8. The method according to claim 6, wherein the determining that the waveform corresponding to the data flow signal has self-similarity specifically comprises: Calculate the Hurst exponent of the waveform corresponding to the data flow signal by using the rescaled range analysis method; It is determined that the value of the Hurst exponent is greater than a second preset threshold. 9. The method according to claim 7, wherein the at least one approximation signal and a signal whose stationarity is smaller than a first preset threshold among the at least one approximation signal and a plurality of detail signals is determined as a first-type data flow signal , including: According to the formula calculating a sample autocorrelation function of the at least one approximation signal and the plurality of detail signals; Will The signal is determined as the first type of data flow signal; Wherein, _Xi is the data flow corresponding to the i-th sampling point within the preset time length; _EX is the mean value of X; N is the number of sampling points within the preset time length; k=1,2, 3,...K; θ is the first preset threshold. 10. The method of claim 6, further comprising: determining that the waveform corresponding to the data flow signal does not have self-similarity; A linear model is used to predict the data traffic signal.