A network traffic prediction method, system, computer device and storage medium

Abstract

The present invention provides a network traffic prediction method, system, computer device and storage medium, wherein the method is to pre-process the network traffic time series data to be processed of the current prediction period obtained according to the preset sampling time interval based on the preset multi-scale time domain period to obtain the traffic time series data to be analyzed of multiple time scales and the graph Laplace matrix, and then generate a directed graph adjacency matrix based on the multi-scale fused graph Laplace matrix obtained by weighted fusion of the graph Laplace matrices of different time scales with the optimization goal of minimizing the total variation of the graph, and then obtain the network traffic prediction value of the current prediction period based on the directed graph adjacency matrix and the traffic time series data to be analyzed of the preset prediction time scale based on the preset traffic prediction model. The present invention can adaptively learn the real intrinsic correlation of traffic data in a complex dynamic environment based on the time domain periodicity of traffic, thereby effectively improving the reliability and accuracy of real-time prediction of network traffic.

Description

Technical Field

[0001] The present invention relates to the field of communication network technology, and in particular to a network traffic prediction method, system, computer equipment and storage medium. Background Art

[0002] With the continuous development of technologies such as the Internet of Things, cloud computing, and edge computing, communication systems have also put forward more stringent requirements on network stability and user experience quality. AI-enhanced network management needs to rely on accurate perception of network status to achieve self-optimization of communication networks and automated energy resource allocation. For example, operators can perceive network resource allocation requirements by predicting communication network traffic, and can also design sleep mechanisms based on traffic prediction results to effectively reduce the energy consumption of communication devices (such as wireless base stations) that work continuously in the network. It can also manage and improve user experience based on possible network congestion based on traffic prediction results.

[0003] However, communication network traffic has complex spatiotemporal characteristics: user mobility causes traffic data in different regions to contain different degrees of spatial correlation (local or global), and the difference in regional network functions leads to different traffic characteristics; in addition, while communication traffic presents different dynamic patterns in the time dimension, various external factors such as holidays, the number of base stations or routers, and group activities will also make the spatiotemporal characteristics more complex, making it a major challenge to accurately predict communication traffic. Although existing scholars have found that graph neural networks can capture the spatiotemporal characteristics of traffic data, these spatiotemporal characteristics are mainly reflected in the graph structure of traffic data and the graph structure directly affects the prediction performance of graph neural networks. However, existing related work does not pay attention to the time domain periodicity of traffic data and cannot truly and reasonably reliably model and analyze the spatiotemporal correlation of network traffic data and reflect it in the corresponding graph structure, resulting in the graph structure used for prediction containing errors, missing or topology that cannot truly reflect the data prior information, causing noise or even errors to propagate in the graph, reducing the accuracy of traffic prediction, and making it difficult to meet actual application needs. Summary of the invention

[0004] The purpose of the present invention is to provide a network traffic prediction method, which introduces a multi-scale fusion graph structure learning mechanism designed based on the time domain periodicity analysis of network traffic, adaptively learns the real intrinsic correlation of traffic data in a complex dynamic environment, facilitates the extraction of spatiotemporal features that are more in line with the actual communication network application scenarios, and effectively improves the reliability and accuracy of real-time prediction of network traffic.

[0005] In order to achieve the above-mentioned purpose, a network traffic prediction method, system, computer device and storage medium are provided.

[0006] In a first aspect, an embodiment of the present invention provides a network traffic prediction method, the method comprising the following steps:

[0007] According to the preset sampling time interval, the time series data of the network traffic to be processed in the current prediction period is obtained; the time series data of the network traffic to be processed includes the time series data of the traffic between each communication node pair;

[0008] Preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes traffic change cycles at multiple time scales;

[0009] Taking minimizing the total variation of the graph as the optimization goal, weighted fusion is performed on the graph Laplacian matrices of different time scales to obtain a multi-scale fused graph Laplacian matrix, and the corresponding directed graph adjacency matrix is ​​obtained according to the scale fused graph Laplacian matrix;

[0010] According to the directed graph adjacency matrix and the traffic time series data to be analyzed corresponding to the preset prediction time scale, traffic prediction is performed based on the preset traffic prediction model to obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling time interval of the network traffic time series data to be processed.

[0011] Furthermore, the step of obtaining the preset multi-scale time domain period includes:

[0012] The network traffic time series data to be processed is classified and counted by communication nodes to obtain the traffic time series of each communication node;

[0013] The traffic time series of all communication nodes are merged according to the sampling time to obtain the corresponding network traffic time series;

[0014] According to the network traffic time series, a corresponding network traffic power spectrum is obtained;

[0015] According to the network traffic power spectrum, traffic variation cycles at different time scales are obtained.

[0016] Furthermore, the step of preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain period to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplacian matrix includes:

[0017] Resampling the network traffic time series data to be processed based on a preset multi-scale time domain period to obtain traffic time series data at multiple time scales;

[0018] When the current prediction period is the first prediction period, data reconstruction and graph Laplace matrix joint optimization are performed on the flow time series data of each time scale respectively to obtain the flow time series data to be analyzed and the graph Laplace matrix of the corresponding time scale;

[0019] When the current prediction cycle is not the first prediction cycle, the corresponding traffic time series data are reconstructed based on the graph Laplace matrices of each time scale of the previous prediction cycle to obtain the traffic time series data to be analyzed at the corresponding time scale, and the graph Laplace matrix of the corresponding time scale is updated according to the traffic time series data to be analyzed.

[0020] Furthermore, the step of updating the graph Laplacian matrix of the corresponding time scale according to the traffic time series data to be analyzed includes:

[0021] Obtaining an optimized graph Laplace matrix corresponding to the traffic time series data to be analyzed;

[0022] The optimized graph Laplace matrix and the graph Laplace matrix of the time scale corresponding to the previous prediction period are weightedly fused to obtain the graph Laplace matrix of the time scale corresponding to the current prediction period.

[0023] Furthermore, the step of weighted fusion of graph Laplacian matrices of different time scales with minimizing the total variation of the graph as the optimization goal to obtain a multi-scale fused graph Laplacian matrix includes:

[0024] Taking the time domain period weights corresponding to the graph Laplace matrices of different time scales as optimization variables, a time domain period weight optimization model is constructed based on the optimization goal of minimizing the total variation of the graph, and the time domain period weight optimization model is solved according to the traffic time series data to be analyzed at different time scales and the corresponding graph Laplace matrices to obtain the time domain period weights of different time scales;

[0025] The corresponding graph Laplacian matrices are weightedly fused based on the time-domain periodic weights of different time scales to obtain the multi-scale fused graph Laplacian matrix.

[0026] Furthermore, the step of obtaining a corresponding directed graph adjacency matrix according to the scale fusion graph Laplacian matrix includes:

[0027] Obtaining a corresponding undirected graph adjacency matrix according to the scale fusion graph Laplacian matrix and the corresponding degree matrix;

[0028] According to the undirected graph adjacency matrix, the weight proportion corresponding to each edge of each communication node is obtained, and the connectivity relationship between the corresponding communication node and other communication nodes is obtained according to the comparison result of the weight proportion corresponding to each edge and the preset proportion threshold;

[0029] According to the connectivity relationship between each communication node and other communication nodes, the directed graph adjacency matrix is ​​generated based on the undirected graph adjacency matrix.

[0030] Furthermore, the preset traffic prediction model includes a graph convolutional neural network, a graph attention network and a fully connected network; the step of performing traffic prediction based on the preset traffic prediction model to obtain the network traffic prediction value of the current prediction period according to the directed graph adjacency matrix and the traffic time series data to be analyzed corresponding to the preset prediction time scale includes:

[0031] Inputting the to-be-analyzed traffic time series data corresponding to the preset prediction time scale and the directed graph adjacency matrix into the graph convolutional neural network for global feature extraction to obtain the global spatiotemporal features of the traffic;

[0032] Inputting the global spatiotemporal features of the traffic into the graph attention network to extract local features, and obtaining corresponding local spatial features of the traffic;

[0033] The global spatiotemporal features of the traffic and the local spatial features of the traffic are input into the fully connected network for feature fusion prediction to obtain the network traffic prediction value of the current prediction period.

[0034] In a second aspect, an embodiment of the present invention provides a network traffic prediction system, the system comprising:

[0035] A data acquisition module is used to obtain the time series data of the network traffic to be processed in the current prediction period according to a preset sampling time interval; the time series data of the network traffic to be processed includes the time series data of the traffic between each communication node pair;

[0036] A data processing module, used to pre-process the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes a traffic change cycle at multiple time scales;

[0037] A directed graph construction module is used to perform weighted fusion of graph Laplacian matrices of different time scales with the optimization goal of minimizing the total variation of the graph to obtain a multi-scale fused graph Laplacian matrix, and obtain a corresponding directed graph adjacency matrix based on the scale fused graph Laplacian matrix;

[0038] The traffic prediction module is used to perform traffic prediction based on the preset traffic prediction model according to the traffic time series data to be analyzed corresponding to the directed graph adjacency matrix and the preset prediction time scale, and obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling time interval of the network traffic time series data to be processed.

[0039] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0040] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0041] The present invention provides a network traffic prediction method, system, computer device and storage medium. The method is used to obtain the to-be-processed network traffic time series data including the traffic time series data between each communication node pair in the current prediction period according to the preset sampling time interval, pre-process the to-be-processed network traffic time series data based on the preset multi-scale time domain period including the traffic change period of multiple time scales to obtain the to-be-analyzed traffic time series data and graph Laplace matrix of multiple time scales, perform weighted fusion on the graph Laplace matrices of different time scales with minimizing the total variation of the graph as the optimization goal to obtain the multi-scale fused graph Laplace matrix, obtain the corresponding directed graph adjacency matrix according to the scale fused graph Laplace matrix, and then perform traffic prediction based on the directed graph adjacency matrix and the to-be-analyzed traffic time series data corresponding to the preset prediction time scale based on the preset traffic prediction model to obtain the network traffic prediction value of the current prediction period. Technical solution. Compared with the existing technology, this network traffic prediction method introduces a multi-scale fusion graph structure learning mechanism designed based on the time domain periodicity analysis of network traffic. It adaptively learns the true intrinsic correlation of traffic data in a complex dynamic environment, facilitates the extraction of spatiotemporal features that are more in line with real communication network application scenarios, and effectively improves the reliability and accuracy of real-time prediction of network traffic. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of a network traffic prediction method according to an embodiment of the present invention;

[0043] Figure 2 It is a schematic diagram of the structure of a preset flow prediction model in an embodiment of the present invention;

[0044] Figure 3 It is a schematic diagram of the prediction trend of Internet traffic in a scenario of urban cellular network traffic prediction and the comparison between the predicted value and the actual value in an embodiment of the present invention;

[0045] Figure 4 It is a schematic diagram of Internet traffic prediction error in a scenario based on urban cellular network traffic prediction in an embodiment of the present invention;

[0046] Figure 5It is a schematic diagram of the cumulative distribution of absolute errors in Internet traffic prediction in a scenario of urban cellular network traffic prediction in an embodiment of the present invention;

[0047] Figure 6 is a schematic diagram of the structure of a network traffic prediction system in an embodiment of the present invention;

[0048] Figure 7 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and beneficial effects of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described below are part of the embodiments of the present invention and are only used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The network traffic prediction method provided by the present invention can be understood as a communication network traffic prediction scheme that introduces a multi-scale fusion graph structure learning mechanism designed based on the time domain periodicity analysis of network traffic, and adaptively learns the real intrinsic correlation of traffic data in a complex dynamic environment. The following embodiments will explain the network traffic prediction method of the present invention in detail.

[0051] In one embodiment, Figure 1 As shown, a network traffic prediction method is provided, comprising the following steps:

[0052] S11. Obtain the time series data of the network traffic to be processed in the current prediction period according to the preset sampling time interval; the time series data of the network traffic to be processed includes the traffic time series data between each pair of communication nodes; wherein, the preset sampling time interval can be understood as the traffic collection time of the communication network in actual applications, which can be set in seconds, minutes, hours, days and weeks in principle, but in order to balance the performance consumption of traffic data processing and analysis while ensuring the accuracy of traffic perception, this embodiment preferably sets the sampling time interval at the minute level, and the time scale corresponding to the current prediction period only needs to be not less than the preset sampling time interval.

[0053] The time series data of network traffic to be processed in the current prediction period in this embodiment is understood as traffic data of a certain period of time (several days, weeks or months, etc.) obtained by monitoring and collecting voice, text messages, data and other related business traffic of each communication node such as routers and business terminals in the communication network based on the preset sampling time interval according to the actual traffic prediction data volume demand. The time series data of network traffic to be processed can be understood as a spatial matrix of network traffic corresponding to each sampling moment, and the value of each matrix element in the spatial matrix corresponds to the traffic data between two communication nodes. Assume that in a communication network containing communication nodes, and the traffic data is sampled every preset time interval Fixed collection once, a total of Second, in indivual The time series data of network traffic to be processed in the time period can be described as a space-time matrix set :

[0054] ,

[0055] In the formula,

[0056]

[0057] in, express The space-time matrix corresponding to the moment; is the set of communication node pairs in the communication network; express Time communication node pair The flow value between Represents the field of real numbers.

[0058] S12. Preprocess the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes traffic change cycles at multiple time scales, wherein the time of each time scale is not less than the collection time interval of the network traffic time series data to be processed.

[0059] In this embodiment, the traffic change cycle of each time scale can be understood as the duration of the periodic regular changes in the traffic data on the corresponding time scale. For example, if the traffic characteristics at 8:00 every day are relatively similar, it can be understood that the traffic change cycle of the daily time scale is one day. If the traffic characteristics on Mondays are relatively similar, it can be understood that the traffic variable cycle of the periodic time scale is one week, etc.; that is, network traffic has the characteristics of periodic changes. This periodic feature will cause the graph structure calculated based on the traffic data at different times to be a function of time, and will fluctuate regularly over time. This fluctuation has periodic components on different time scales, and multiple time scales contribute to the fluctuation. Based on this, if you want to ensure the accuracy of traffic prediction, when performing traffic prediction at a smaller time scale (such as minutes), you must also consider the impact of larger time scales (such as hours, days, and weeks) on the current graph topology.

[0060] In order to obtain the time domain fluctuation characteristics of the graph structure to improve the accuracy of the graph structure, this embodiment preferably obtains the traffic change cycles of different time scales (for example, minute level, hour level, day level and week level, etc.), and analyzes the graph structure of the actually collected network traffic time series data to be processed based on the traffic change cycles of multiple time scales obtained, so as to facilitate the subsequent traffic prediction. Specifically, the acquisition steps of the preset multi-scale time domain cycle include:

[0061] The network traffic time series data to be processed is classified and counted by communication nodes to obtain the traffic time series of each communication node; that is, the space-time matrix of each moment in the network traffic time series data to be processed is The elements of each row of the matrix are accumulated to obtain the flow data of the corresponding communication node, and then the flow data of the communication node at each moment are summarized to obtain the corresponding flow time series, which is expressed as:

[0062]

[0063] in, express Time communication node Traffic data; express Time communication node pair The flow value between The total number of moments representing the flow time series; Represents the total number of communication nodes in the communication network.

[0064] The traffic time series of all communication nodes are merged according to the sampling time to obtain the corresponding network traffic time series; wherein, the network traffic data at each sampling time in the network traffic time series can be immediately obtained by merging the traffic data of each communication node in the communication network; the specific network traffic time series is expressed as:

[0065]

[0066] in, express The network traffic of the communication network at all times.

[0067] According to the network traffic time series, the corresponding network traffic power spectrum is obtained; wherein, the network traffic power spectrum is obtained based on traffic analysis of the network traffic time series. In principle, any power spectrum estimation method is applicable. In order to ensure the high efficiency of data analysis and processing, this embodiment preferably uses the periodogram method to perform Fourier transform on the autocorrelation function of the traffic time series data, and then analyzes the energy distribution of different frequencies, and then obtains the required network traffic power spectrum. The specific implementation details can be referred to the relevant existing technology, which will not be repeated here.

[0068] According to the network traffic power spectrum, traffic change cycles at different time scales are obtained; that is, after obtaining the network traffic power spectrum, the peak cycles of the power spectrum at different time scales such as minute level, hour level, day level and week level are located according to the needs, and the corresponding peak cycles are recorded as the corresponding traffic change cycles, for example, the traffic change cycles at minute level, hour level, day level and week level are recorded as , and use it as the basis for subsequent graph structure acquisition and network traffic time series data processing. It should be noted that the time scale that can be used in practical applications is not limited to minutes, hours, days, weeks, etc., and can be selected according to actual application requirements.

[0069] As described above, the network traffic time series data to be processed in this embodiment is data collected based on a higher sampling frequency relative to each time scale. In order to accurately analyze the impact of different time scales on the graph structure, this embodiment preferably analyzes the data based on the network traffic time series data to be processed, obtains data at different time scales based on multi-time scale resampling, and then calculates the graph structure of the corresponding time scale based on the data at different time scales. Specifically, the steps of preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain period to obtain the traffic time series data to be analyzed at multiple time scales and the graph Laplacian matrix include:

[0070] Based on the preset multi-scale time domain period, the network traffic time series data to be processed is resampled to obtain traffic time series data of multiple time scales; specifically, the process of obtaining traffic time series data of each time scale through resampling includes: taking values ​​of the network traffic time series data to be processed at equal time intervals according to the traffic change period of each time scale in the preset multi-scale time domain period, and obtaining traffic data at different times on the corresponding time scale; or, filtering the network traffic time series data to be processed, after obtaining the filtered traffic time series data, downsampling the filtered traffic time series data according to the traffic change period of each time scale in the preset multi-scale time domain period, and obtaining traffic time series data of the corresponding time scale. It should be noted that the detailed implementation details of the above-mentioned equal time interval value taking method or the post-filtering downsampling processing method can be referred to the relevant technical implementation, and will not be described in detail here.

[0071] In actual traffic prediction, considering that traffic data often has a small amount of missing, abnormal or noise effects during the collection process, if it is directly discarded, it will waste precious data resources, and ultimately weaken the effect of graph structure learning and traffic prediction. In this embodiment, the graph Laplace matrix is ​​preferably used to reconstruct the collected data to suppress the noise of the original data and eliminate the influence of missing values ​​and abnormal values. At the same time, the learning of the graph structure is further improved based on the reconstructed data to ensure accurate prediction and analysis based on the learned graph structure. In practical applications, when the method of the present invention is used to perform network traffic prediction for the first time, there is no graph Laplace matrix that matches the prediction time scale available for data reconstruction and processing, and when the network traffic prediction is not performed for the first time, the graph Laplace matrix of the corresponding time scale generated in the previous prediction period can be used for data reconstruction, but directly using the graph Laplace matrix of the previous prediction period for traffic prediction may have the problem of difficulty in real-time perception of traffic feature changes, resulting in traffic prediction being difficult to fit the actual application scenario. Based on this, in this embodiment, when reconstructing the traffic time series data of each time scale and obtaining the graph Laplace matrix, the first prediction and the non-first prediction are distinguished and processed according to the following method:

[0072] When the current prediction cycle is the first prediction cycle, the traffic time series data of each time scale are reconstructed and jointly optimized and solved with the graph Laplace matrix to obtain the traffic time series data and graph Laplace matrix to be analyzed of the corresponding time scale; wherein, the first prediction cycle can be understood as the first time that the method of the present invention is used to perform network traffic prediction; the specific process of reconstructing the traffic time series data of each time scale and jointly optimizing and solving the graph Laplace matrix can be understood as first constructing a corresponding joint optimization model of data reconstruction and graph Laplace matrix based on the data reconstruction requirements and the graph Laplace matrix solution requirements, and then solving the joint optimization model of data reconstruction and graph Laplace matrix based on the traffic time series data of the time scale to obtain the traffic time series data and graph Laplace matrix to be analyzed of the required time scale.

[0073] The above data reconstruction and graph Laplacian matrix joint optimization model can be expressed as:

[0074]

[0075] In the formula, Indicates time scale The time series data (reconstructed data) of the traffic to be analyzed is The column straightening vector of the reconstructed space-time matrix; Indicates time scale The Laplacian matrix of the graph below; Representation and The corresponding reconstruction vector variable; Representation and The corresponding graph Laplacian matrix variable is positive semidefinite; and Respectively and The corresponding transpose; Indicates time scale The traffic time series data of There is no abnormal traffic data between communication node pairs in the column straightened vector of the space-time matrix (original data); Indicates the time in the original collected traffic time series data with noise, missing values ​​and outliers Column straightening vector of space-time matrix; and Respectively and The Normal traffic data, reserve The normal traffic data in the , and the abnormal data are the content that needs to be optimized and solved; represents the regularization coefficient; p represents the regularization norm determined by the prior information of the graph structure in the actual traffic scenario. For example, when the prior is sparse, the value of p can be 1, and when the prior is Gaussian distribution, the value of p can be 2. It should be noted that, depending on the different time scales of the data, The traffic change cycle can be minute, hour, day or week. Can be Used to represent the time scale The corresponding graph structure of .

[0076] In practical applications, the solution process of the joint optimization model of data reconstruction and graph Laplace matrix can be understood as converting the joint optimization model into the following two sub-problems, and then solving the convex optimization problem through alternating minimization, so as to simultaneously obtain the traffic time series data to be analyzed and the graph Laplace matrix at each time scale.

[0077] 1) Graph Laplacian matrix optimization problem model:

[0078]

[0079] Among them, the interpretation and description of relevant parameters in the model can be found in the detailed description of the above-mentioned joint optimization model.

[0080] 2) Data reconstruction optimization problem model:

[0081]

[0082] For the explanatory descriptions of all the variables in the formula, please refer to the descriptions of the relevant variables in the above-mentioned data reconstruction and graph Laplace matrix joint optimization model, which will not be repeated here.

[0083] The above method steps can be used to realize data reconstruction and graph Laplace matrix solution of traffic time series data at various time scales when executing traffic prediction for the first time; it should be noted that the process of alternating minimum optimization based on the graph Laplace matrix optimization problem model and data reconstruction optimization problem model given above can refer to the solution method of the existing related convex optimization model, which will not be described in detail here.

[0084] When the current prediction cycle is not the first prediction cycle, the corresponding traffic time series data are reconstructed based on the graph Laplace matrices of each time scale of the previous prediction cycle to obtain the traffic time series data to be analyzed at the corresponding time scale, and the graph Laplace matrix of the corresponding time scale is updated according to the traffic time series data to be analyzed; wherein, the process of reconstructing the corresponding traffic time series data based on the graph Laplace matrices of each time scale of the previous prediction cycle can be understood as solving the preset data reconstruction model based on the traffic time series data of the corresponding time scale, and the corresponding preset data reconstruction model can be expressed as:

[0085]

[0086] in, Indicates time scale The time series data (reconstructed data) of the traffic to be analyzed is The column straightening vector of the reconstructed space-time matrix; Indicates the time scale of the previous forecast period The Laplacian matrix of the graph below; Representation and The corresponding reconstruction vector variable; express The corresponding transpose; Indicates time scale The traffic time series data of There is no abnormal traffic data between communication node pairs in the column straightened vector of the space-time matrix (original data); Indicates the time in the original collected traffic time series data with noise, missing values ​​and outliers Column straightening vector of space-time matrix; and Respectively and The Normal traffic data, reserve The normal traffic data in the data, and the abnormal data are the content that needs to be optimized and solved; it should be noted that, according to the different time scales of the data, The traffic change cycle can be minute, hour, day or week. Can be The Laplacian matrix of the graph representing the minute, hour, day, and week time scales for the last forecast period.

[0087] In practical applications, according to the Laplace matrices of the graphs of different time scales in the previous forecast period, based on the preset data reconstruction model constructed in this example, the flow time series data of the corresponding time scales are reconstructed and optimized, and the spatiotemporal data without abnormal data of the corresponding time scale is restored. It should be noted that the specific method for solving the above preset data reconstruction model can adopt the solution algorithm of the relevant convex optimization model, which is not specifically limited here.

[0088] In addition, in order to ensure that the graph structures of different time scales can be dynamically updated to adapt to the dynamically changing complex communication traffic environment, thereby ensuring the reliability of real-time traffic prediction, in this embodiment, preferably, after the traffic time series data is reconstructed and optimized based on the graph Laplace matrices of different time scales of the previous prediction period, the graph Laplace matrices of each time scale of the previous prediction period are dynamically updated based on the reconstructed and optimized data; specifically, the step of updating the graph Laplace matrix of the corresponding time scale according to the traffic time series data to be analyzed includes:

[0089] Obtain the optimized graph Laplace matrix corresponding to the traffic time series data to be analyzed; wherein, the process of obtaining the optimized graph Laplace matrix can be understood as solving the following optimization model based on the traffic time series data to be analyzed to obtain a graph Laplace matrix matching the current traffic time series data to be analyzed. The specific solution algorithm can adopt the existing convex optimization model solution algorithm, which will not be described in detail here:

[0090]

[0091] in, Representation and time scale The optimized graph Laplacian matrix corresponding to the traffic time series data to be analyzed; Representation and The corresponding graph Laplacian matrix variable is positive semidefinite; Indicates time scale The time series data (reconstructed data) of the traffic to be analyzed is The column straightening vector of the reconstructed space-time matrix; and Respectively and Corresponding transposition; It should be noted that, depending on the time scale of the data, The traffic change cycle can be minute, hour, day or week. Can be Used to represent the time scale The optimized graph Laplacian matrix of ;

[0092] The optimized graph Laplace matrix and the graph Laplace matrix of the time scale corresponding to the previous prediction period are weightedly fused to obtain the graph Laplace matrix of the time scale corresponding to the current prediction period; wherein the graph Laplace matrix of the time scale corresponding to the current prediction period can be expressed as:

[0093]

[0094] in, Indicates the time scale corresponding to the current forecast period The graph Laplacian matrix of ; Representation and time scale The optimized graph Laplacian matrix corresponding to the traffic time series data to be analyzed; Indicates the time scale corresponding to the previous forecast period The graph Laplacian matrix of ; Represents the forgetting factor, which can be set according to the specific actual application scenario. Through this formula, the graph topology can be adaptively adjusted in real time, thereby dynamically optimizing the graph topology that reflects the spatiotemporal correlation characteristics, effectively improving the adaptability to dynamic traffic data, and thus providing a reliable guarantee for obtaining traffic prediction results that are more in line with actual application scenarios.

[0095] S13, weighted fusion of graph Laplacian matrices of different time scales with minimizing the total variation of the graph as the optimization goal, to obtain a multi-scale fusion graph Laplacian matrix, and according to the scale fusion graph Laplacian matrix, obtain the corresponding directed graph adjacency matrix; wherein, the multi-scale fusion graph Laplacian matrix can be understood as a graph Laplacian matrix that can fuse time domain prior information and reduce the time sensitivity of network topology by fusing graph structures of different time scales and eliminating dynamic influences such as time domain periodic changes. In the calculation process of the multi-scale fusion graph Laplacian matrix, the time domain periodic weights corresponding to the graph Laplacian matrices of each time scale are mainly determined. In order to ensure the rationality of the time domain periodic weight setting, this embodiment preferably establishes a corresponding optimization model with the time domain periodic weights of each time scale as the optimization variable with minimizing the total variation of the graph as the optimization goal, and obtains the required time domain periodic weights of different time scales by solving the model. Specifically, the step of weighted fusion of graph Laplacian matrices of different time scales with minimizing the total variation of the graph as the optimization goal to obtain a multi-scale fusion graph Laplacian matrix includes:

[0096] The time domain cycle weights corresponding to the graph Laplace matrices of different time scales are used as optimization variables. A time domain cycle weight optimization model is constructed based on the optimization objective of minimizing the total variation of the graph. The time domain cycle weight optimization model is solved according to the traffic time series data to be analyzed at different time scales and the corresponding graph Laplace matrix to obtain the time domain cycle weights of different time scales. The time domain cycle weight optimization model can be expressed as:

[0097]

[0098] In the formula, Indicates time scale The corresponding time domain period weight, and The flow change cycle can be at different time scales such as minute, hour, day, and week. For example, , min, hour, day and week represent minute-level scale, hour-level scale, day-level scale and week-level scale respectively, and are not specifically limited here; Indicates time scale The time series data (reconstructed data) of the traffic to be analyzed is The column straightening vector of the reconstructed space-time matrix; Indicates time scale The graph Laplacian matrix of ; express The transpose of .

[0099] The corresponding graph Laplace matrices are weightedly fused based on the time domain period weights of different time scales to obtain the multi-scale fused graph Laplace matrix; that is, the multi-scale fused graph Laplace matrix can be expressed as:

[0100]

[0101] in, and Respectively represent the time scale The corresponding graph Laplacian matrix and time domain periodic weights; Representing the multi-scale fusion graph Laplacian matrix, it characterizes the intrinsic graph structure of traffic data and can simultaneously reflect the dual characteristics of traffic data in time and space that are more in line with real communication network application scenarios.

[0102] The method of constructing a graph Laplacian matrix based on multi-time scale fusion provided by the time-domain periodic analysis results of network traffic data in this embodiment can not only reflect the correlation of traffic data between communication nodes or regions through the graph Laplacian matrix, but also make good use of the impact of time-domain dynamic traffic data on the graph structure, obtain network traffic characteristics that are more in line with actual application scenarios, and provide reliable guarantee for accurate prediction of network traffic based on graph structure.

[0103] The goal of the actual traffic prediction task may only be for a certain time scale, such as hourly prediction. However, since traffic data has different temporal correlations and spatial correlations at different time scales, there are actually differences in the spatiotemporal correlations at different time scales. In order to effectively capture this difference, it is very important to comprehensively consider the impact of the graph Laplacian matrix at different time scales on the prediction results. Based on this, this embodiment preferably uses the scale-fused graph Laplacian matrix obtained above to obtain a scene-adaptive graph adjacency matrix; specifically, the step of obtaining the corresponding directed graph adjacency matrix based on the scale-fused graph Laplacian matrix includes:

[0104] According to the scale fusion graph Laplacian matrix and the corresponding degree matrix, the corresponding undirected graph adjacency matrix is ​​obtained; wherein the undirected graph adjacency matrix is ​​expressed as:

[0105]

[0106] in, Represents the multi-scale fusion graph Laplacian matrix; The degree matrix representing the network traffic topology after scale fusion is a diagonal matrix, and the elements on the diagonal are the sum of the weights of the communication nodes, which are also the diagonal elements of the Laplace matrix after scale fusion; Represents an undirected graph adjacency matrix, whose corresponding weights are real numbers in the undirected graph topology.

[0107] In the undirected graph adjacency matrix Based on the corresponding graph structure, each communication node is processed to determine whether the edge connected to the communication node is connected or not connected, and the directionality of the edge between the communication nodes is determined accordingly. The specific method for determining connectivity is as follows:

[0108] According to the undirected graph adjacency matrix, the weight ratio corresponding to each edge of each communication node is obtained, and the connectivity relationship between the corresponding communication node and other communication nodes is obtained according to the comparison result of the weight ratio corresponding to each edge and the preset ratio threshold; wherein, the process of obtaining the weight ratio corresponding to each edge of a communication node can be understood as first calculating the sum of the weights of all edges of the communication node, and then comparing the weight corresponding to each edge with the sum of the weights of all edges, and obtaining the ratio as the weight ratio corresponding to each edge; the corresponding preset ratio threshold is between 0 and 1, and the specific value can be adjusted according to the actual network, and the default value can be 80%. In practical applications, for each communication node, the first several edges whose weight ratio is greater than the preset ratio threshold can be selected and marked as connected (the connectivity attribute value is set to 1, and a directed edge is obtained), and the remaining edges are marked as non-connected (the connectivity attribute value is set to 0), that is, the edge marked as connected indicates that the flow direction from the opposite communication node to the communication node is established, otherwise, it indicates that the flow direction from the opposite communication node to the communication node is not established.

[0109] According to the connectivity relationship between each communication node and other communication nodes, the directed graph adjacency matrix is ​​generated based on the undirected graph adjacency matrix; specifically, based on the connectivity relationship between each communication node and other communication nodes determined by the above method steps, the undirected edges in the topological graph corresponding to the undirected graph adjacency matrix can be converted into directed edges to generate the required directed graph adjacency matrix, and the directed graph neighbor matrix can be understood as corresponding to each communication node having both incoming and outgoing directions.

[0110] S14. According to the directed graph adjacency matrix and the traffic time series data to be analyzed corresponding to the preset prediction time scale, traffic prediction is performed based on the preset traffic prediction model to obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling duration interval of the network traffic time series data to be processed; the traffic time series data to be analyzed corresponding to the preset prediction time scale can be understood as the traffic time series data to be analyzed with a time scale that matches the preset prediction time scale obtained by the aforementioned steps.

[0111] In principle, the preset traffic prediction model in this embodiment can adopt the existing graph neural network. In order to ensure the efficiency and accuracy of traffic prediction in practical applications, this embodiment preferably sets the preset traffic prediction model as follows: Figure 2 The figure includes a graph convolutional neural network, a graph attention network and a fully connected network, wherein the graph convolutional neural network (GCN) is used to extract global spatiotemporal features, the graph attention network (GAT) is used to extract local spatial features, and the fully connected network (FCNN) is used to fuse global spatiotemporal features and local spatial features and predict traffic based on the fused features; in this embodiment, the graph convolutional neural network, the graph attention network and the fully connected network can directly adopt the existing general network structure, or can adopt the related improved network structure, which is not specifically limited here.

[0112] Specifically, the step of performing traffic prediction based on a preset traffic prediction model to obtain a network traffic prediction value for a current prediction period according to the directed graph adjacency matrix and the to-be-analyzed traffic time series data corresponding to the preset prediction time scale includes:

[0113] The traffic time series data to be analyzed corresponding to the preset prediction time scale and the directed graph adjacency matrix are input into the graph convolutional neural network for global feature extraction to obtain the global spatiotemporal features of the traffic; the graph convolution operation in the graph convolutional neural network in this embodiment is performed on the directed graph adjacency matrix (directed graph) obtained by the above method, and the remaining processing steps can refer to the relevant processing in the existing general GCN network, which is not specifically limited here.

[0114] The global spatiotemporal features of the traffic are input into the graph attention network for local feature extraction to obtain corresponding local spatial features of the traffic. In this embodiment, the local feature extraction of the graph attention network is also performed on the directed graph adjacency matrix (directed graph) obtained by the above method. The remaining processing steps can refer to the relevant processing in the existing general GAT network and are not specifically limited here.

[0115] The global spatiotemporal features of the traffic and the local spatial features of the traffic are input into the fully connected network for feature fusion prediction to obtain the network traffic prediction value of the current prediction period.

[0116] The above-mentioned method steps can be used to predict the network traffic conditions of different prediction periods in real time. This embodiment only illustrates the traffic prediction process of one prediction period. In actual applications, the network traffic prediction of each prediction period can be obtained by executing the above-mentioned method steps. It should be noted that the above-mentioned extraction of global spatiotemporal features of traffic based on graph convolutional neural networks and the detailed implementation process of extracting local spatial features of traffic based on graph attention networks can refer to the feature extraction processing of existing graph convolutional neural networks and graph attention networks, which will not be repeated here.

[0117] The embodiment of the present invention provides a method for obtaining the network traffic time series data to be processed including the traffic time series data between each communication node pair in the current prediction period according to a preset sampling time interval, preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain period including a traffic change period of multiple time scales to obtain the traffic time series data to be analyzed and a graph Laplace matrix of multiple time scales, weightedly fusion the graph Laplace matrices of different time scales with the optimization goal of minimizing the total variation of the graph to obtain a multi-scale fused graph Laplace matrix, obtaining the corresponding directed graph adjacency matrix according to the scale fused graph Laplace matrix, and then performing traffic prediction based on the directed graph adjacency matrix and the traffic time series data to be analyzed corresponding to the preset prediction time scale based on a preset traffic prediction model to obtain a solution for the network traffic prediction value of the current prediction period. By introducing a multi-scale fusion graph structure learning mechanism designed based on the time domain periodicity analysis of network traffic, the real intrinsic correlation of traffic data is adaptively learned in a complex dynamic environment, so as to facilitate the extraction of spatiotemporal features that are more in line with the application scenarios of real communication networks, and effectively improve the reliability and accuracy of real-time prediction of network traffic.

[0118] In order to verify the effectiveness of the method of the present invention, this embodiment also tests the traffic prediction scenario based on the urban cellular network (including SMS service, call service and Internet service), and obtains the following results: Figure 3-5 The verification results are as follows: Figure 3 The comparison results of the actual and predicted network traffic data in a test are shown. Figure 4 It shows the absolute value of the prediction error changing over time. Figure 5 The distribution of network traffic prediction errors is shown, and based on Figure 3-5The verification results show that the prediction results of the method of the present invention are in good agreement with the actual values, with low prediction errors, and can quickly achieve accurate predictions even in burst traffic conditions. At the same time, by comparing with the existing network traffic prediction methods, it is found that: in terms of RMSE evaluation indicators, compared with the existing methods, the SMS service performance of the method of the present invention is improved by 8.56%, the call service performance is improved by 11.76%, and the Internet service performance is improved by 7.01%; in terms of MAE indicators, the method of the present invention improves the SMS service performance by 11.56%, the call service performance by 6.98%, and the Internet service performance by 9.14%; at the same time, regarding R 2 Indicators, the method of the present invention shows similar improvements on various cellular traffic data sets as other types of traffic data. That is, the verification results of the urban cellular network traffic prediction scenario can strongly prove the effectiveness and reliability of the method of the present invention for communication traffic prediction.

[0119] In addition, in order to verify the effectiveness of the multi-scale fusion graph structure learning mechanism designed based on time domain periodicity analysis proposed in the method of the present invention, ablation experiments and comparative analysis of multiple groups of different graph structure learning mechanisms were carried out based on cellular traffic data, and the comparative results of directed graph structure modeling effects were obtained as shown in Table 1.

[0120] Table 1 Ablation experiment analysis results

[0121]

[0122] In Table 1, Indicates that no graph structure learning scheme is used. It means that the graph structure is introduced but it is only modeled as an undirected graph. Indicates that only the adaptive adjacency matrix is ​​used to adaptively learn the spatial structure. represents an adjacency matrix, which is a directed graph with two directions, It represents the graph structure finally used by the present invention, in which there are two-directional adjacency matrices and an adaptive adjacency matrix for correcting the spatial correlation fluctuations that occur over time. As shown in Table 1, in the traffic prediction task, The prediction error obtained is smaller than that obtained using The prediction error obtained shows that considering spatial dependence can reduce the prediction error; the directed graph structure designed by the present invention The error of modeling data is significantly smaller than that of undirected graphs Modeling data errors, the graph structure performs better than using only the adaptive adjacency matrix performance, and The graph structure of the pattern has the smallest prediction error in practical applications, which further proves the effectiveness and necessity of considering the time-domain periodicity of network traffic and modeling communication network traffic as a directed graph and designing a dynamic and adaptive graph topology learning mechanism.

[0123] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0124] In one embodiment, Figure 6 As shown, a network traffic prediction system is provided, the system comprising:

[0125] The data acquisition module 1 is used to obtain the time series data of the network traffic to be processed in the current prediction period according to the preset sampling time interval; the time series data of the network traffic to be processed includes the time series data of the traffic between each communication node pair;

[0126] Data processing module 2, used for preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes a traffic change cycle at multiple time scales;

[0127] The directed graph construction module 3 is used to perform weighted fusion of graph Laplacian matrices of different time scales with minimizing the total variation of the graph as the optimization goal to obtain a multi-scale fused graph Laplacian matrix, and obtain a corresponding directed graph adjacency matrix according to the scale fused graph Laplacian matrix;

[0128] The traffic prediction module 4 is used to perform traffic prediction based on the preset traffic prediction model according to the traffic time series data to be analyzed corresponding to the directed graph adjacency matrix and the preset prediction time scale, and obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling time interval of the network traffic time series data to be processed.

[0129] For the specific definition of the network traffic prediction system, please refer to the definition of the network traffic prediction method above, and the corresponding technical effects can also be obtained equivalently, which will not be repeated here. Each module in the above-mentioned network traffic prediction system can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0130] Figure 7FIG. 1 shows an internal structure diagram of a computer device in an embodiment, and the computer device may specifically be a terminal or a server. Figure 7 As shown, the computer device includes a processor, a memory, a network interface, a display, a camera and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a network traffic prediction method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0131] It can be understood by those skilled in the art that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computing device may include more or less components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0132] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0133] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0134] In summary, the embodiments of the present invention provide a network traffic prediction method and system, wherein the network traffic prediction method realizes obtaining the to-be-processed network traffic time series data including the traffic time series data between each communication node pair in the current prediction period according to the preset sampling time interval, preprocessing the to-be-processed network traffic time series data based on the preset multi-scale time domain period including the traffic change period of multiple time scales to obtain the to-be-analyzed traffic time series data and graph Laplace matrix of multiple time scales, and weighted fusion of graph Laplace matrices of different time scales with minimizing the total variation of the graph as the optimization goal to obtain the multi-scale fused graph Laplace matrix. The invention discloses a technical scheme for obtaining a network traffic prediction value of the current prediction period by performing traffic prediction based on a preset traffic prediction model according to the directed graph adjacency matrix and the time series data of the traffic to be analyzed corresponding to the preset prediction time scale. The method introduces a multi-scale fusion graph structure learning mechanism designed based on the time domain periodicity analysis of network traffic, and adaptively learns the true intrinsic correlation of traffic data in a complex dynamic environment, so as to extract spatiotemporal features that are more in line with the actual communication network application scenarios, and effectively improve the reliability and accuracy of real-time prediction of network traffic.

[0135] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0136] The above-mentioned embodiments only express several preferred implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principle of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the protection scope of the claims.

Claims

1. A network traffic prediction method, characterized in that: The method comprises the following steps: According to the preset sampling time interval, the time series data of the network traffic to be processed in the current prediction period is obtained; the time series data of the network traffic to be processed includes the time series data of the traffic between each communication node pair; Preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes traffic change cycles at multiple time scales; Taking minimizing the total variation of the graph as the optimization goal, weighted fusion is performed on the graph Laplacian matrices of different time scales to obtain a multi-scale fused graph Laplacian matrix, and according to the multi-scale fused graph Laplacian matrix, a corresponding directed graph adjacency matrix is ​​obtained; According to the directed graph adjacency matrix and the traffic time series data to be analyzed corresponding to the preset prediction time scale, traffic prediction is performed based on the preset traffic prediction model to obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling time interval of the network traffic time series data to be processed.

2. The network traffic prediction method according to claim 1, characterized in that: The step of obtaining the preset multi-scale time domain period includes: The network traffic time series data to be processed is classified and counted by communication nodes to obtain the traffic time series of each communication node; The traffic time series of all communication nodes are merged according to the sampling time to obtain the corresponding network traffic time series; According to the network traffic time series, a corresponding network traffic power spectrum is obtained; According to the network traffic power spectrum, traffic variation cycles at different time scales are obtained.

3. The network traffic prediction method according to claim 1, characterized in that: The step of preprocessing the network traffic time series data to be processed based on a preset multi-scale time domain period to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplacian matrix comprises: Resampling the network traffic time series data to be processed based on a preset multi-scale time domain period to obtain traffic time series data at multiple time scales; When the current prediction period is the first prediction period, data reconstruction and graph Laplace matrix joint optimization are performed on the flow time series data of each time scale respectively to obtain the flow time series data to be analyzed and the graph Laplace matrix of the corresponding time scale; When the current prediction cycle is not the first prediction cycle, the corresponding traffic time series data are reconstructed based on the graph Laplace matrices of each time scale of the previous prediction cycle to obtain the traffic time series data to be analyzed at the corresponding time scale, and the graph Laplace matrix of the corresponding time scale is updated according to the traffic time series data to be analyzed.

4. The network traffic prediction method according to claim 3, characterized in that: The step of updating the graph Laplacian matrix of the corresponding time scale according to the traffic time series data to be analyzed comprises: Obtaining an optimized graph Laplace matrix corresponding to the traffic time series data to be analyzed; The optimized graph Laplace matrix and the graph Laplace matrix of the time scale corresponding to the previous prediction period are weightedly fused to obtain the graph Laplace matrix of the time scale corresponding to the current prediction period.

5. The network traffic prediction method according to claim 1, characterized in that: The step of weighted fusion of graph Laplacian matrices of different time scales with minimization of the total variation of the graph as the optimization goal to obtain a multi-scale fused graph Laplacian matrix includes: Taking the time domain period weights corresponding to the graph Laplace matrices of different time scales as optimization variables, a time domain period weight optimization model is constructed based on the optimization goal of minimizing the total variation of the graph, and the time domain period weight optimization model is solved according to the traffic time series data to be analyzed at different time scales and the corresponding graph Laplace matrices to obtain the time domain period weights of different time scales; The corresponding graph Laplacian matrices are weightedly fused based on the time-domain periodic weights of different time scales to obtain the multi-scale fused graph Laplacian matrix.

6. The network traffic prediction method according to claim 1, characterized in that: The step of obtaining a corresponding directed graph adjacency matrix according to the multi-scale fusion graph Laplacian matrix comprises: According to the multi-scale fusion graph Laplacian matrix and the corresponding degree matrix, a corresponding undirected graph adjacency matrix is ​​obtained; According to the undirected graph adjacency matrix, the weight proportion corresponding to each edge of each communication node is obtained, and the connectivity relationship between the corresponding communication node and other communication nodes is obtained according to the comparison result of the weight proportion corresponding to each edge and the preset proportion threshold; According to the connectivity relationship between each communication node and other communication nodes, the directed graph adjacency matrix is ​​generated based on the undirected graph adjacency matrix.

7. The network traffic prediction method according to claim 1, characterized in that: The preset traffic prediction model includes a graph convolutional neural network, a graph attention network and a fully connected network; the step of performing traffic prediction based on the preset traffic prediction model according to the directed graph adjacency matrix and the to-be-analyzed traffic time series data corresponding to the preset prediction time scale to obtain the network traffic prediction value of the current prediction period includes: Inputting the to-be-analyzed traffic time series data corresponding to the preset prediction time scale and the directed graph adjacency matrix into the graph convolutional neural network for global feature extraction to obtain the global spatiotemporal features of the traffic; Inputting the global spatiotemporal features of the traffic into the graph attention network to extract local features, and obtaining corresponding local spatial features of the traffic; The global spatiotemporal features of the traffic and the local spatial features of the traffic are input into the fully connected network for feature fusion prediction to obtain the network traffic prediction value of the current prediction period.

8. A network traffic prediction system, characterized in that: The system comprises: A data acquisition module is used to obtain the time series data of the network traffic to be processed in the current prediction period according to a preset sampling time interval; the time series data of the network traffic to be processed includes the time series data of the traffic between each communication node pair; A data processing module, used to pre-process the network traffic time series data to be processed based on a preset multi-scale time domain cycle to obtain the traffic time series data to be analyzed at multiple time scales and a graph Laplace matrix; the preset multi-scale time domain cycle includes a traffic change cycle at multiple time scales; A directed graph construction module is used to perform weighted fusion of graph Laplacian matrices of different time scales with the optimization goal of minimizing the total variation of the graph to obtain a multi-scale fused graph Laplacian matrix, and obtain a corresponding directed graph adjacency matrix based on the multi-scale fused graph Laplacian matrix; The traffic prediction module is used to perform traffic prediction based on the preset traffic prediction model according to the traffic time series data to be analyzed corresponding to the directed graph adjacency matrix and the preset prediction time scale, and obtain the network traffic prediction value of the current prediction period; the duration of the preset prediction time scale is greater than or equal to the preset sampling time interval of the network traffic time series data to be processed.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images