Network traffic prediction method and apparatus, storage medium

Abstract

The present disclosure provides a network traffic prediction method and device, and a storage medium. The network traffic prediction method comprises: clustering data objects in a traffic data set to obtain a weekday data set and a non-weekday data set; predicting the weekday data set by using a long short-term memory network model to obtain a first prediction result; determining a residual value by using the first prediction result and historical data; adding the residual value to the weekday data set to obtain an updated data set; predicting the updated data set by using a gated recurrent network model to obtain a second prediction result; and obtaining a network traffic prediction result according to the first prediction result and the second prediction result.

Description

Technical Field

[0001] The present disclosure relates to the field of communications, and in particular to a network traffic prediction method and device, and a storage medium. Background Art

[0002] As telecom operators continue to expand their business, the amount of traffic on their networks is growing exponentially. Improper allocation of network traffic resources can lead to network paralysis. Predicting future traffic trends and properly planning network resources in advance can minimize network congestion. Therefore, efficiently allocating network resources is crucial.

[0003] Currently, traffic scheduling methods are mainly based on the following three methods: statistical model-based, empirical pattern analysis-based, and machine learning model-based. Statistical model-based methods use differential methods to smooth and preprocess traffic data, then capture the characteristics of network traffic. The smoothed model is fitted using a statistical model. Empirical pattern analysis-based methods decompose traffic data into countless IMF (Intrinsic Mode Function) components and residual components, and then predict them separately. Machine learning model-based methods use the learning capabilities of machine learning models to find the optimal solution. Summary of the Invention

[0004] The inventors note that among the aforementioned related technologies, methods based on statistical models require stationary data and are unable to extract nonlinear feature relationships. Methods based on empirical mode analysis increase uncertainty for components with large amplitude variations. Methods based on machine learning models typically use a single model, which cannot produce highly accurate predictions of network traffic data.

[0005] Based on this, the present disclosure provides a network traffic prediction solution. By utilizing a hybrid model composed of a long short-term memory network model (Long Short Term Memory, abbreviated as: LSTM) and a gated recurrent network model (Gated Recurrent Unit, abbreviated as: GRU), it can extract features from traffic data from different dimensions, thereby effectively improving the prediction results of network traffic data.

[0006] In a first aspect of the present disclosure, a network traffic prediction method is provided, which is performed by a network traffic prediction device. The method includes: clustering data objects in a traffic data set to obtain a working day data set and a non-working day data set; predicting the working day data set using a long short-term memory network model to obtain a first prediction result; determining a residual value using the first prediction result and historical data; adding the residual value to the working day data set to obtain an updated data set; predicting the updated data set using a gated recurrent network model to obtain a second prediction result; and obtaining a network traffic prediction result based on the first prediction result and the second prediction result.

[0007] In some embodiments, obtaining a network traffic prediction result based on the first prediction result and the second prediction result includes: fusing the first prediction result and the second prediction result to obtain the network traffic prediction result.

[0008] In some embodiments, clustering a traffic data set includes: determining the number of clusters K; calculating the density value of each data object in the traffic data set; selecting K data objects in descending order of the density values ​​as K initial cluster centers; and clustering the data objects in the traffic data set based on the K initial cluster centers.

[0009] In some embodiments, the density value of the i-th data object in the traffic dataset is determined by the distance between the i-th data object and the j-th data object in the traffic dataset, 1≤j≤N and j≠i, N is the total number of data objects in the traffic dataset.

[0010] In some embodiments, the density value ρ of the i-th data object i for

[0011]

[0012] where dist(x i , x j ) is the i-th data object x i With the jth data object x j The distance between c is the preset cutoff distance.

[0013] In some embodiments, determining the number of clusters K includes: setting a cluster number set, wherein the cluster number set includes a plurality of initial cluster numbers; clustering the traffic data set using each initial cluster number in the cluster number set in ascending order of the initial cluster numbers to obtain a plurality of clustering results; calculating a Davidson-Borg-Ding index for each of the plurality of clustering results; if the Davidson-Borg-Ding index of the mth cluster result is less than the Davidson-Borg-Ding index of the m+1th cluster result, the m+2th cluster result, and the m+3th cluster result, then using the mth initial cluster number as the cluster number K, where 1≤m≤M-3. M is the total number of initial clusters.

[0014] In some embodiments, the Davidson-Borgin index of the mth clustering result is determined by the average distance of the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between each two cluster centers in the mth clustering result.

[0015] In some embodiments, the Davidson-Borgin Index DBI(m) of the m-th clustering result is:

[0016]

[0017] Wherein, k is the number of clusters in the m-th clustering result, avg(C i ) is the average distance from the data objects in the ith cluster to the cluster center of the ith cluster, avg(C j ) is the average distance from the data objects in the jth cluster to the cluster center of the jth cluster, d cen (u i ,u j ) is the distance from the center of the i-th cluster to the center of the j-th cluster, 1≤i, j≤k and j≠i.

[0018] In some embodiments, the minimum number of the plurality of initial clusters is 2, and the maximum number is no more than N is a natural number, and N is the total number of data objects in the traffic data set.

[0019] In a second aspect of the present disclosure, a network traffic prediction device is provided, comprising: a first processing module, configured to cluster data objects in a traffic data set to obtain a working day data set and a non-working day data set; a second processing module, configured to predict the working day data set using a long short-term memory network model to obtain a first prediction result; a third processing module, configured to determine a residual value using the first prediction result and historical data, add the residual value to the working day data set to obtain an updated data set, and predict the updated data set using a gated recurrent network model to obtain a second prediction result; and a fourth processing module, configured to obtain a network traffic prediction result based on the first prediction result and the second prediction result.

[0020] In some embodiments, the fourth processing module is configured to fuse the first prediction result and the second prediction result to obtain the network traffic prediction result.

[0021] In some embodiments, the first processing module is configured to determine the number of clusters K, calculate the density value of each data object in the traffic data set, select K data objects in descending order of the density values ​​as K initial cluster centers, and cluster the data objects in the traffic data set according to the K initial cluster centers.

[0022] In some embodiments, the density value of the i-th data object in the traffic dataset is determined by the distance between the i-th data object and the j-th data object in the traffic dataset, 1≤j≤N and j≠i, N is the total number of data objects in the traffic dataset.

[0023] In some embodiments, the density value ρ of the i-th data object i for

[0024]

[0025] where dist(x i , x j ) is the i-th data object x i With the jth data object x j The distance between c is the preset cutoff distance.

[0026] In some embodiments, the first processing module is configured to set a cluster number set, wherein the cluster number set includes multiple initial cluster numbers, cluster the traffic data set using each initial cluster number in the cluster number set in ascending order of the initial cluster numbers to obtain multiple clustering results, calculate the Davidson-Borgin index of each clustering result in the multiple clustering results, and if the Davidson-Borgin index of the mth clustering result is simultaneously less than the Davidson-Borgin index of the m+1th clustering result, the m+2th clustering result, and the m+3th clustering result, then the mth initial cluster number is used as the cluster number K, where 1≤m≤M-3. M is the total number of initial clusters.

[0027] In some embodiments, the Davidson-Borgin index of the mth clustering result is determined by the average distance of the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between each two cluster centers in the mth clustering result.

[0028] In some embodiments, the Davidson-Borgin Index DBI(m) of the m-th clustering result is:

[0029]

[0030] Wherein, k is the number of clusters in the m-th clustering result, avg(C i ) is the average distance from the data objects in the ith cluster to the cluster center of the ith cluster, avg(C j ) is the average distance from the data objects in the jth cluster to the cluster center of the jth cluster, d cen (u i ,u j ) is the distance from the center of the i-th cluster to the center of the j-th cluster, 1≤i, j≤k and j≠i.

[0031] In some embodiments, the minimum number of the plurality of initial clusters is 2, and the maximum number is no more than N is a natural number, and N is the total number of data objects in the traffic data set.

[0032] In a third aspect of the present disclosure, a network traffic prediction device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the method described in any of the above embodiments based on instructions stored in the memory.

[0033] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method described in any of the above embodiments is implemented.

[0034] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0036] Figure 1 A flowchart of a network traffic prediction method according to an embodiment of the present disclosure is provided;

[0037] Figure 2 A flow chart of a traffic data set clustering method according to an embodiment of the present disclosure;

[0038] Figure 3 A schematic flow chart of a method for determining the number of clusters according to an embodiment of the present disclosure;

[0039] Figure 4 This is a schematic diagram of the structure of a network traffic prediction device according to an embodiment of the present disclosure;

[0040] Figure 5 This is a schematic structural diagram of a network traffic prediction device according to another embodiment of the present disclosure;

[0041] Figure 6 Schematic diagram for time comparison of different models;

[0042] Figure 7A and Figure 7B Schematic diagram of RMSE performance of different models.

[0043] Figure 8 Schematic diagram of the accuracy of different models;

[0044] Figure 9 Schematic diagram for comparing predicted values ​​with true values. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is in no way intended to limit the present disclosure and its application or use. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0046] Unless specifically stated otherwise, the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure.

[0047] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0048] Technologies, methods and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, such technologies, methods and equipment should be considered part of the authorization specification.

[0049] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0050] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0051] Figure 1 FIG2 is a flow chart of a network traffic prediction method according to an embodiment of the present disclosure. In some embodiments, the following network traffic prediction method is performed by a network traffic prediction device.

[0052] In step 101 , data objects in a traffic data set are clustered to obtain a working day data set and a non-working day data set.

[0053] In some embodiments, the step of clustering the traffic data set is as follows: Figure 2 shown.

[0054] Figure 2 FIG2 is a flow chart of a method for clustering traffic data sets according to an embodiment of the present disclosure. In some embodiments, the following method for clustering traffic data sets is performed by a network traffic prediction device.

[0055] In step 201, the number of clusters K is determined.

[0056] In some embodiments, the step of determining the number of clusters K is as follows: Figure 3 shown.

[0057] Figure 3 FIG2 is a flow chart of a method for determining the number of clusters according to an embodiment of the present disclosure. In some embodiments, the following method for determining the number of clusters is performed by a network traffic prediction device.

[0058] In step 301 , a cluster number set is set, wherein the cluster number set includes a plurality of initial cluster numbers.

[0059] In some embodiments, the minimum value of the multiple initial cluster numbers is 2 and the maximum value is no more than N is a natural number, and N is the total number of data objects in the traffic dataset.

[0060] In step 302 , the traffic data set is clustered using each initial cluster number in the cluster number set in ascending order of the initial cluster number to obtain a plurality of clustering results.

[0061] For example, if the first initial number of clusters in the cluster number set is 2, then the traffic dataset is clustered with the number of clusters being 2 to obtain the first clustering result. Next, if the second initial number of clusters in the cluster number set is 3, then the traffic dataset is clustered with the number of clusters being 3 to obtain the second clustering result. And so on.

[0062] In step 303 , the Davies-Bouldin index of each clustering result in the plurality of clustering results is calculated.

[0063] In some embodiments, the Davidson-Botting index of the mth clustering result is determined by the average distance of the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between each two cluster centers in the mth clustering result.

[0064] For example, the Davidson-Borgin index DBI(m) of the m-th clustering result is shown in formula (1).

[0065]

[0066] Among them, k is the number of clusters in the m-th clustering result, avg(C i ) is the average distance from the data objects in the ith cluster to the cluster center of the ith cluster, avg(C j ) is the average distance from the data objects in the jth cluster to the cluster center of the jth cluster, d cen (u i ,u j ) is the distance from the center of the i-th cluster to the center of the j-th cluster, 1≤i, j≤k and j≠i.

[0067] In step 304, if the Davidson-Borg index of the mth clustering result is less than the Davidson-Borg index of the m+1th clustering result, the m+2th clustering result, and the m+3th clustering result, the mth initial cluster number is used as the cluster number K, where 1≤m≤M-3. M is the total number of initial clusters.

[0068] It should be noted that the use of the Davidson-Boulding index can better balance the independence and cohesion of the data set, thereby selecting the optimal cluster value.

[0069] return Figure 2 In step 202, the density value of each data object in the traffic data set is calculated.

[0070] In some embodiments, the density value of the i-th data object in the traffic dataset is determined by the distance between the i-th data object and the j-th data object in the traffic dataset, 1≤j≤N and j≠i, where N is the total number of data objects in the traffic dataset.

[0071] For example, the density value ρ of the i-th data object i As shown in formula (2).

[0072]

[0073] where dist(x i , x j ) is the i-th data object x i With the jth data object x j The distance between c The preset cutoff distance.

[0074] In step 203, K data objects are selected in descending order of density values ​​to serve as K initial cluster centers.

[0075] In step 204, data objects in the traffic data set are clustered according to the K initial cluster centers.

[0076] It should be noted here that by using the density value of the data object to select K initial cluster centers, it is closer to the actual cluster center, thereby effectively reducing the number of iterations of the clustering process.

[0077] In some embodiments, after determining the K initial cluster centers, the data objects in the traffic data set are clustered using the k-means clustering algorithm. Since the k-means clustering algorithm itself is not the inventive point of the present disclosure, it will not be described in detail here.

[0078] return Figure 1 In step 102, a long short-term memory network model (LSTM) is used to predict the workday data set to obtain a first prediction result.

[0079] LSTM is a time-recurrent neural network designed to solve the long-term dependency problem of general RNNs (Recurrent Neural Networks). Since LSTM itself is not the inventive point of this disclosure, it will not be described in detail here.

[0080] In some embodiments, a stationarity test is performed on the data objects in the working day data set, and differential processing is performed when the data objects are not stationary, so that the data objects in the working day data set are stationary time series, and then the data objects in the working day data set are input into LSTM for prediction.

[0081] When training the LSTM, training parameters are determined according to the Akaike Information Criterion (AIC), and then the LSTM is trained using the training parameters.

[0082] It should be noted that during model training, both the weekday and non-weekday datasets are used for LSTM training. During prediction, only the weekday dataset is fed into the LSTM.

[0083] In step 103, a residual value is determined using the first prediction result and historical data.

[0084] For example, the residual value is determined according to the difference between the first prediction result and the historical data.

[0085] At step 104 , the residual values ​​are added to the workday dataset to obtain an updated dataset.

[0086] In step 105, a gated recurrent network (GRU) model is used to predict the updated data set to obtain a second prediction result.

[0087] The GRU is a variant of the LSTM. The LSTM uses three gate functions: an input gate, a forget gate, and an output gate to control input, memory, and output values. In contrast, the GRU has only two gate functions: an update gate and a reset gate. Since the GRU itself is not the core of this disclosure, it will not be described in detail here.

[0088] In step 106, a network traffic prediction result is obtained according to the first prediction result and the second prediction result.

[0089] In some embodiments, the first prediction result and the second prediction result are fused to obtain a network traffic prediction result.

[0090] For example, if the first prediction result is l t , the second prediction result is n t , then the network traffic prediction result x t As shown in formula (3).

[0091] x t =l t +n t (3)

[0092] In the network traffic prediction method provided in the above embodiments of the present disclosure, by utilizing a hybrid model composed of a long short-term memory network model and a gated recurrent network model, features of traffic data can be extracted from different dimensions, thereby effectively improving the prediction results of network traffic data.

[0093] Figure 4 FIG. 1 is a schematic diagram of the structure of a network traffic prediction device according to an embodiment of the present disclosure. Figure 4 As shown, the network traffic prediction device includes a first processing module 41 , a second processing module 42 , a third processing module 43 and a fourth processing module 44 .

[0094] The first processing module 41 is configured to cluster the data objects in the traffic data set to obtain a working day data set and a non-working day data set.

[0095] In some embodiments, the first processing module 41 determines the number of clusters K, calculates the density value of each data object in the traffic data set, selects K data objects in descending order of density value as K initial cluster centers, and clusters the data objects in the traffic data set according to the K initial cluster centers.

[0096] In some embodiments, the density value of the i-th data object in the traffic dataset is determined by the distance between the i-th data object and the j-th data object in the traffic dataset, 1≤j≤N and j≠i, where N is the total number of data objects in the traffic dataset.

[0097] For example, the density value ρ of the i-th data object i As shown in formula (2).

[0098] It should be noted here that by using the density value of the data object to select K initial cluster centers, it is closer to the actual cluster center, thereby effectively reducing the number of iterations of the clustering process.

[0099] In some embodiments, the first processing module 41 sets a cluster number set, where the cluster number set includes multiple initial cluster numbers. The traffic data set is clustered using each initial cluster number in the cluster number set in ascending order of the initial cluster numbers to obtain multiple clustering results. The Davidson-Borgin index of each clustering result in the multiple clustering results is calculated. If the Davidson-Borgin index of the mth clustering result is less than the Davidson-Borgin index of the m+1th clustering result, the m+2th clustering result, and the m+3th clustering result, the mth initial cluster number is used as the cluster number K, where 1≤m≤M-3. M is the total number of initial clusters.

[0100] In some embodiments, the Davidson-Botting index of the mth clustering result is determined by the average distance of the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between each two cluster centers in the mth clustering result.

[0101] For example, the Davidson-Borgin index DBI(m) of the m-th clustering result is shown in formula (1).

[0102] It should be noted that the use of the Davidson-Boulding index can better balance the independence and cohesion of the data set, thereby selecting the optimal cluster value.

[0103] In some embodiments, the minimum value of the multiple initial cluster numbers is 2 and the maximum value is no more than N is a natural number, and N is the total number of data objects in the traffic dataset.

[0104] The second processing module 42 is configured to use the long short-term memory network model to predict the weekday data set to obtain a first prediction result.

[0105] The third processing module 43 is configured to determine a residual value using the first prediction result and historical data, add the residual value to the workday dataset to obtain an updated dataset, and predict the updated dataset using the gated recurrent network model to obtain a second prediction result.

[0106] For example, the residual value is determined according to the difference between the first prediction result and the historical data.

[0107] The fourth processing module 44 is configured to obtain a network traffic prediction result according to the first prediction result and the second prediction result.

[0108] In some embodiments, the fourth processing module 44 fuses the first prediction result and the second prediction result to obtain a network traffic prediction result.

[0109] For example, if the first prediction result is l t , the second prediction result is n t , then the network traffic prediction result x t As shown in formula (3).

[0110] In the network traffic prediction device provided in the above embodiments of the present disclosure, by utilizing a hybrid model composed of a long short-term memory network model and a gated recurrent network model, features of traffic data can be extracted from different dimensions, thereby effectively improving the prediction results of network traffic data.

[0111] Figure 5 FIG. 1 is a schematic diagram of the structure of a network traffic prediction device according to another embodiment of the present disclosure. Figure 5 As shown, the network traffic prediction device includes a memory 51 and a processor 52.

[0112] The memory 51 is used to store instructions. The processor 52 is coupled to the memory 51. The processor 52 is configured to execute the instructions stored in the memory. Figures 1 to 3 The method according to any one of the embodiments.

[0113] like Figure 5 As shown, the network traffic prediction device further includes a communication interface 53 for exchanging information with other devices. At the same time, the network traffic prediction device further includes a bus 54 through which the processor 52, the communication interface 53, and the memory 51 communicate with each other.

[0114] The memory 51 may include high-speed RAM memory or non-volatile memory, such as at least one disk storage device. The memory 51 may also be a memory array. The memory 51 may also be divided into blocks, and the blocks may be combined into virtual volumes according to certain rules.

[0115] Furthermore, the processor 52 may be a central processing unit (CPU), or may be an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the present disclosure.

[0116] The present disclosure also relates to a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the following Figures 1 to 3 The method according to any one of the embodiments.

[0117] The present disclosure is described below through specific examples.

[0118] Traffic data is a typical example of time series data and serves as a valuable reference for system performance management, risk prevention, and anomaly detection. For example, traffic data from a particular operator from September 1 to October 15, 2022, was collected for experimental training, testing, and validation. The data was sampled every five minutes, yielding a total of 1,200 data points. The data was first preprocessed, cleaned, and filtered to identify 880 valid data sets. 600 of these were used for model training, and the remaining 280 were used to test the network model's performance.

[0119] In order to evaluate the effectiveness of the solution provided by the present disclosure, the root mean square error (RMSE) is used as an evaluation indicator, as shown in formula (4).

[0120]

[0121] In formula (4), n is the total number of data, x i is the original data, x′ iFor forecast data.

[0122] In order to verify the time efficiency of the solution provided by this disclosure, ARIMA (Autoregressive Integrated Moving Average), GRU and LSTM algorithms are used for comparative analysis. The experimental results are shown in the figure. Figure 6 shown.

[0123] Depend on Figure 6 It can be seen that when the data scale increases, the network convergence time of the ARIMA algorithm and the GRU algorithm shows a quadratic growth trend, and the clustering time of the LSTM algorithm shows a linear growth trend, but its convergence speed is not as good as the convergence speed of the solution provided by the present disclosure. It can be seen that the solution provided by the present disclosure has greatly improved the time efficiency, and the time complexity has been reduced from the original O(n 2 ) becomes O(np).

[0124] In addition, the performance test and evaluation results of the scheme and other algorithms provided by this disclosure in real data sets S1 and S2 are as follows: Figure 7A and Figure 7B shown. Figure 7A Corresponding test results of the S1 dataset, Figure 7B Corresponding test results of S2 dataset. Figure 7A and Figure 7B In the figure, identifier 1 is used to represent the RMSE of ARIMA, identifier 2 is used to represent the RMSE of GRU, identifier 3 is used to represent the RMSE of LSTM, and identifier 4 is used to represent the RMSE of the solution provided in the present disclosure.

[0125] like Figure 7A and Figure 7B As shown, since ARIMA can only predict the linear part of the data, its prediction effect is the worst and the RMSE value is the largest, while the prediction effects of GRU and LSTM models are in the middle. The prediction mean square error of the solution provided by the present disclosure is the smallest, indicating that the solution provided by the present disclosure has high prediction accuracy.

[0126] The average of the three experimental results of 20 minutes, 40 minutes and 60 minutes is taken, and the obtained accuracy data is shown in Table 1. Figure 8 shown.

[0127]

[0128] Table 1 Comparison of model prediction performance

[0129] As shown in Table 1, the accuracy of the solution provided by this disclosure is higher than that of other existing technologies, regardless of the training time, whether it is 20 minutes, 40 minutes, or 60 minutes. Taking the 60-minute data as an example, the solution provided by this disclosure improves the accuracy of ARIMA, GRU, and LSTM by 11.2%, 14.5%, and 6.7%, respectively. This is due to the hybrid model's strong ability to learn time series, which allows the solution provided by this disclosure to achieve better accuracy.

[0130] Figure 9 A comparison diagram of the predicted value and the true value is given. Figure 9 As shown, label 1 is used to represent the original data, label 2 is used to represent the prediction result of ARIMA, label 3 is used to represent the prediction result of GRU, label 4 is used to represent the prediction result of LSTM, and label 5 is used to represent the prediction result of the solution provided by the present disclosure.

[0131] like Figure 9 As shown in the figure, the prediction error of the prediction results of the solution provided by the present disclosure is the smallest, and it is basically consistent with the actual flow data. By comparison, it can be seen that the prediction results of the solution provided by the present disclosure have significantly improved the prediction effect at the point of sudden flow changes, and are more stable than the results predicted by the baseline model.

[0132] In some embodiments, the functional units described above may be implemented as general-purpose processors, programmable logic controllers (PLC), digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any appropriate combination thereof, for performing the functions described in the present disclosure.

[0133] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0134] The description of the present disclosure is provided for purposes of illustration and description and is not intended to be exhaustive or to limit the disclosure to the disclosed form. Many modifications and variations will be apparent to those skilled in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present disclosure and to enable those skilled in the art to understand the present disclosure and design various embodiments with various modifications suitable for specific applications.

Claims

1. A network traffic prediction method, performed by a network traffic prediction device, comprising: Cluster the data objects in the traffic dataset to obtain the working day dataset and the non-working day dataset; Predicting the workday dataset using a long short-term memory network model to obtain a first prediction result; Determining a residual value using the first prediction result and historical data; Adding the residual value to the workday dataset to obtain an updated dataset; Using a gated recurrent network model to predict the updated data set to obtain a second prediction result; Obtaining a network traffic prediction result according to the first prediction result and the second prediction result, wherein the first prediction result and the second prediction result are fused to obtain the network traffic prediction result; Among them, clustering the traffic data set includes: Determine the number of clusters K; Calculating a density value of each data object in the traffic data set; Selecting K data objects in descending order of density values ​​as K initial cluster centers; Clustering the data objects in the traffic data set according to the K initial cluster centers; The density value of the i-th data object in the traffic data set is determined by the distance between the i-th data object and the j-th data object in the traffic data set. and , N is the total number of data objects in the traffic data set, The density value of the i-th data object for in For the i-th data object With the jth data object The distance between is the preset cutoff distance; Determining the number of clusters K includes: Setting a cluster number set, wherein the cluster number set includes a plurality of initial cluster numbers; Clustering the traffic data set using each initial cluster number in the cluster number set in ascending order of the initial cluster number to obtain a plurality of clustering results; Calculating the Davidson-Botting Index of each clustering result in the plurality of clustering results; If the Davidson-Borg index of the mth clustering result is less than the Davidson-Borg index of the m+1th clustering result, the m+2th clustering result, and the m+3th clustering result, the mth initial cluster number is used as the cluster number K, where , M is the total number of initial clusters.

2. The method according to claim 1, wherein The Davidson-Borgin index of the mth clustering result is determined by the average distance from the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between every two cluster centers in the mth clustering result.

3. The method according to claim 2, wherein: Davidson-Botting index of the mth clustering result for: Wherein, k is the number of clusters in the m-th clustering result, is the average distance from the data objects in the i-th cluster to the cluster center of the i-th cluster, is the average distance from the data objects in the jth cluster to the cluster center of the jth cluster, is the distance from the center of the i-th cluster to the center of the j-th cluster, and .

4. The method according to claim 1, wherein The minimum number of the multiple initial clusters is 2, and the maximum number is no more than N is a natural number, and N is the total number of data objects in the traffic data set.

5. A network traffic prediction device, comprising: The first processing module is configured to cluster the data objects in the traffic data set to obtain a workday data set and a non-workday data set, wherein the number of clusters K is determined, the density value of each data object in the traffic data set is calculated, K data objects are selected in descending order of the density values ​​as K initial cluster centers, and the data objects in the traffic data set are clustered based on the K initial cluster centers, wherein the density value of the i-th data object in the traffic data set is determined by the distance between the i-th data object and the j-th data object in the traffic data set. and , N is the total number of data objects in the traffic data set, the density value of the i-th data object for in For the i-th data object With the jth data object The distance between is a preset cutoff distance, wherein the first processing module is configured to set a cluster number set, wherein the cluster number set includes multiple initial cluster numbers, and cluster the traffic data set using each initial cluster number in the cluster number set in order of the initial cluster numbers from small to large to obtain multiple clustering results, and calculate the Davidson-Borg index of each clustering result in the multiple clustering results. If the Davidson-Borg index of the mth clustering result is less than the Davidson-Borg index of the m+1th clustering result, the m+2th clustering result, and the m+3th clustering result, then the mth initial cluster number is used as the cluster number K, wherein ,M is the total number of initial clusters; a second processing module configured to predict the workday dataset using a long short-term memory network model to obtain a first prediction result; a third processing module configured to determine a residual value using the first prediction result and historical data, add the residual value to the workday dataset to obtain an updated dataset, and perform prediction on the updated dataset using a gated recurrent network model to obtain a second prediction result; The fourth processing module is configured to obtain a network traffic prediction result based on the first prediction result and the second prediction result, wherein the first prediction result and the second prediction result are fused to obtain the network traffic prediction result.

6. The device according to claim 5, wherein The Davidson-Borgin index of the mth clustering result is determined by the average distance from the data objects in each cluster in the mth clustering result to the cluster center of each cluster, and the distance between every two cluster centers in the mth clustering result.

7. The device according to claim 6, wherein Davidson-Botting index of the mth clustering result for: Wherein, k is the number of clusters in the m-th clustering result, is the average distance from the data objects in the i-th cluster to the cluster center of the i-th cluster, is the average distance from the data objects in the jth cluster to the cluster center of the jth cluster, is the distance from the center of the i-th cluster to the center of the j-th cluster, and .

8. The device according to claim 5, wherein The minimum number of the multiple initial clusters is 2, and the maximum number is no more than N is a natural number, and N is the total number of data objects in the traffic data set.

9. A network traffic prediction device, comprising: Memory; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 4 based on instructions stored in the memory.

10. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images