Cellular communication flow prediction method, system and program product

Abstract

The invention belongs to the technical field of cellular communication, and particularly discloses a cellular communication flow prediction method, system and program product, and the method comprises the steps: building a historical cellular communication flow sequence through employing the cellular communication flow measurement data of each network node in a target network area at each continuous historical time point, and inputting the historical cellular communication flow sequence into the dynamic space-time diagram convolutional network for analysis to obtain a corresponding cellular communication flow prediction result. According to the method, the time-space characteristic analysis of the historical cellular communication flow is carried out on the target area through the corresponding dynamic time-space diagram convolutional network, so that the accurate prediction of the cellular communication flow in the subsequent time period of the area can be realized, and the method is proved to have an advanced effect through the ground real cellular flow information evaluation; the problems that an existing cellular communication flow method is low in efficiency and low in precision can be solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cellular communications, and in particular relates to a cellular communication traffic prediction method, system and program product. Background Art

[0002] With the rapid growth of cellular network devices and the fifth generation of communication networks, predicting cellular traffic has become increasingly complex and has become a key module in communication network management. In this context, the main task of cellular traffic prediction is to model the dynamic spatiotemporal characteristics of large-scale cellular traffic information. In recent years, many targeted prediction methods have been studied, such as the linear method historical average (HA), the autoregressive integrated moving average (ARIMA) method based on the clustering model, and the support vector regression (SVR) method with excellent abstraction ability and small error. However, from the current perspective, these methods generally have the problems of low efficiency and low accuracy, and their performance needs to be improved. Summary of the invention

[0003] The purpose of the present invention is to provide a cellular communication traffic prediction method, system and program product to solve the above problems existing in the prior art.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] In a first aspect, a cellular communication traffic prediction method is provided, comprising:

[0006] Obtaining cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point;

[0007] Based on the cellular communication traffic measurement data of each network node at each continuous historical time point, construct a cellular communication traffic measurement sequence of each network node in the historical time period, and construct a historical cellular communication traffic sequence of the target network area based on the cellular communication traffic measurement sequence of each network node in the historical time period;

[0008] The historical cellular communication traffic sequence of the target network area is input into a preset dynamic spatiotemporal graph convolutional network for traffic prediction to obtain the cellular communication traffic prediction result of the target network area in the next time period. The dynamic spatiotemporal graph convolutional network includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence. The input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence. The stacked spatiotemporal layer includes a plurality of stacked spatiotemporal modules. The spatiotemporal module is used to extract spatiotemporal features from the output of the input layer. The spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules. The output layer is used to output the cellular communication traffic prediction result.

[0009] In one possible design, the input layer includes a first linear transformation module, which is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence and output the processing result to each space-time module. The output layer includes a Leaky ReLU activation function module and a second linear transformation module. The Leaky ReLU activation function module is used to perform Leaky ReLU activation function processing on the summary output of each space-time module to obtain the activation function processing result. The second linear transformation module is used to perform feature dimension conversion processing on the activation function processing result and output the cellular communication traffic prediction result.

[0010] In one possible design, the spatiotemporal module includes a gated temporal convolutional network unit for performing spatiotemporal data modeling, a dynamic graph convolutional network unit for performing dynamic graph structure processing of spatiotemporal data, and a hypergraph convolutional network unit for capturing multi-level features of spatiotemporal data.

[0011] In one possible design, the expression of the gated temporal convolutional network unit is

[0012]

[0013]

[0014] Among them, GTCN(-) represents the gated temporal convolutional network model, S represents the input cellular communication traffic sequence, TCN ω (·) Characterize the temporal convolutional network model, TCN ω1 (·) and TCN ω2 (·) respectively represent the temporal convolutional network models with different weight parameters ω1 and ω2, Conv h (·) represents the convolution kernel expansion convolution in the time dimension, th(·) represents the tanh activation function, sigmoid(·) represents the sigmoid activation function, and ⊙ represents the XOR operation.

[0015] In one possible design, the expression of the hypergraph convolutional network unit is

[0016]

[0017] H f =diag(R f )

[0018] in, Characterize the hypergraph convolutional network model, Represents the input feature matrix with power index n, D k Characterize the input feature matrix, Υ i is the discovery parameter, and i∈γ=[γ0,...,γ I-1 ],i∈[0,I-1],M ks and M kt They represent the diagonal matrix of hyperedge degree and the diagonal matrix of hypernode degree respectively, E represents the correlation matrix, R f represents the adaptive weight vector for discovering hyperedges in a hypergraph, and diag(·) represents the diagonal matrix function.

[0019] In one possible design, the expression of the dynamic graph convolutional network unit is

[0020]

[0021] Among them, DynamicGCN(·) represents the dynamic graph convolutional network model, R j and R k Respectively represent different adaptive adjacency matrices, γ j , γ b , Υ f , γ e and γ c are different discovery parameters, and all of them are ∈γ=[γ0,...,γ I-1 ], Reshape(·) represents the reshape function.

[0022] In one possible design, the dynamic spatiotemporal graph convolutional network is obtained through training and testing, and the loss function used in the training and testing process is

[0023]

[0024] Among them, depl represents the loss function, MAE(·) represents the mean absolute error operation, represents the set of cellular communication traffic prediction results from time point x+1 to time point x+X, and x represents the reference time point, represents the set of measured results of cellular communication traffic from time point x+1 to time point x+X, and

[0025] In a second aspect, a cellular communication traffic prediction system is provided, comprising a data acquisition unit, a sequence construction unit and a traffic prediction unit, wherein:

[0026] A data acquisition unit, used to acquire cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point;

[0027] A sequence construction unit, configured to construct a cellular communication traffic measurement sequence of each network node within a historical time period based on the cellular communication traffic measurement data of each network node at each continuous historical time point, and to construct a historical cellular communication traffic sequence of a target network area based on the cellular communication traffic measurement sequence of each network node within the historical time period;

[0028] A traffic prediction unit is used to input the historical cellular communication traffic sequence of the target network area into a preset dynamic spatiotemporal graph convolutional network for traffic prediction, and obtain the cellular communication traffic prediction result of the target network area in the next time period. The dynamic spatiotemporal graph convolutional network includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence. The input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence. The stacked spatiotemporal layer includes a plurality of stacked spatiotemporal modules. The spatiotemporal module is used to extract spatiotemporal features from the output of the input layer, and the spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules. The output layer is used to output the cellular communication traffic prediction result.

[0029] In a third aspect, a cellular communication traffic prediction system is provided, comprising:

[0030] A memory for storing instructions;

[0031] A processor is used to read the instructions stored in the memory and execute any one of the methods described in the first aspect according to the instructions.

[0032] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored on the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes any one of the methods described in the first aspect. In addition, a computer program product is provided, and when the computer program product is executed on a computer, the computer executes any one of the methods described in the first aspect.

[0033] Beneficial effects: The present invention analyzes the spatiotemporal characteristics of historical cellular communication traffic in the target area through the corresponding dynamic spatiotemporal graph convolutional network, and then can achieve accurate prediction of the cellular communication traffic in the subsequent time period of the area. It is evaluated by the real cellular traffic information on the ground and has been proven to have advanced effects. It can solve the problems of low efficiency and low precision of existing cellular communication traffic methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 This is a schematic diagram of the steps of the method in Example 1 of the present invention;

[0036] Figure 2 This is a schematic diagram of the structure of the system in Example 2 of the present invention;

[0037] Figure 3 A schematic diagram of iterative training loss values ​​of a dynamic spatiotemporal graph convolutional network of the present invention;

[0038] Figure 4 A schematic diagram of prediction results of the dynamic spatiotemporal graph convolutional network of the present invention in a specific cellular service field;

[0039] Figure 5 Schematic diagram of the system structure in Example 3 of the present invention. DETAILED DESCRIPTION

[0040] It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. The specific structures and functional details disclosed herein are only used to describe the exemplary embodiments of the present invention. However, the present invention can be embodied in many alternative forms, and it should not be understood that the present invention is limited to the embodiments set forth herein.

[0041] It should be understood that, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the embodiments can be understood according to specific circumstances.

[0042] In the following description, certain details are provided to facilitate a complete understanding of the example embodiments. However, it will be appreciated by those of ordinary skill in the art that the example embodiments may be implemented without these certain details. For example, devices may be shown in block diagrams to avoid obscuring the examples with unnecessary details. In other embodiments, well-known processes, structures, and techniques may not be shown in unnecessary detail to avoid obscuring the embodiments.

[0043] Embodiment 1:

[0044] This embodiment provides a cellular communication traffic prediction method, which can be applied to corresponding processing terminals, such as Figure 1 As shown, the method comprises the following steps:

[0045] S1. Obtain cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point.

[0046] During specific implementation, the processing terminal needs to first obtain the cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point.

[0047] S2. Based on the cellular communication traffic measurement data of each network node at each continuous historical time point, a cellular communication traffic measurement sequence of each network node within the historical time period is constructed, and based on the cellular communication traffic measurement sequence of each network node within the historical time period, a historical cellular communication traffic sequence of the target network area is constructed.

[0048] In specific implementation, the processing terminal constructs a cellular communication traffic measurement sequence of each network node within a historical time period based on the cellular communication traffic measurement data of each network node at each continuous historical time point. in, Represents the cellular communication traffic measurement data of network node M at time point x. Then, based on the cellular communication traffic measurement sequence of each network node in the historical time period, the historical cellular communication traffic sequence of the target network area is constructed. in, Represents a historical cellular communication traffic measurement sequence from the 1st time point to the xth time point.

[0049] S3. Input the historical cellular communication traffic sequence of the target network area into a preset dynamic spatiotemporal graph convolutional network for traffic prediction, and obtain the cellular communication traffic prediction result of the target network area in the next time period, the dynamic spatiotemporal graph convolutional network comprises an input layer, a stacked spatiotemporal layer and an output layer connected in sequence, the input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence, the stacked spatiotemporal layer comprises a plurality of stacked spatiotemporal modules, the spatiotemporal module is used to extract spatiotemporal features of the output of the input layer, and the spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules, and the output layer is used to output the cellular communication traffic prediction result.

[0050] In the specific implementation, the processing terminal inputs the historical cellular communication traffic sequence of the target network area into the preset dynamic spatiotemporal graph convolutional network for traffic prediction, and obtains the cellular communication traffic prediction result of the target network area in the next time period.

[0051] like Figure 2As shown in the figure, the dynamic spatiotemporal graph convolutional network (DSGCN) includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence, wherein the input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence, and the stacked spatiotemporal layer includes multiple stacked spatiotemporal modules, which are used to extract spatiotemporal features from the output of the input layer, and the spatiotemporal features are aggregated and output to the output layer through skip connection operations between the spatiotemporal modules, and the output layer is used to output the cellular communication traffic prediction results. The main concept of the dynamic spatiotemporal graph convolutional network is to construct a cellular traffic graph based on the historical cellular communication traffic sequence to reveal the complex spatiotemporal features at different time points.

[0052] The input layer includes a first linear transformation module Linear Transform, which is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence and output the processing result to each spatiotemporal module. The output layer includes a Leaky ReLU activation function module and a second linear transformation module Linear Transform, which is used to perform Leaky ReLU activation function processing on the summary output of each spatiotemporal module to obtain the activation function processing result, and the second linear transformation module is used to perform feature dimension conversion processing on the activation function processing result and output the cellular communication traffic prediction result. The linear transformation module Linear Transform changes the dimension of the data or performs feature conversion by multiplying the data by a weight matrix and adding a bias term. One of the main functions of linear transformation is to change the dimension of the data; by multiplying the weight matrix, the features of the input data are projected into a new feature space, which can help the model better capture the correlation in a specific dimension; adding a bias term helps the model's fitting ability and improves the flexibility of the model. The Leaky ReLU activation function is a variant of the ReLU activation function. Its function is to allow certain negative values ​​to pass when the input is negative, instead of directly truncating the negative value to zero like the ReLU activation function. It enhances the expressiveness of the network by avoiding the problem of dead nodes (i.e. neurons with zero activation value).

[0053] The stacked connection between each spatiotemporal module is a Remainder Link, which allows the output data of the previous module to be directly passed to the subsequent module. This is similar to the concept of residual connection. Its purpose is to ensure that information in the deep network is not lost, while alleviating the problem of gradient disappearance and helping the model retain the key information extracted in the previous layers. Each spatiotemporal module uses the Cascade mechanism to form a cascaded information flow to stack the feature representation of each module output layer by layer. This method generates richer and more abstract feature expressions by combining the features of different modules. Unlike traditional hierarchical processing, the Cascade mechanism allows the model to integrate features from different depths more flexibly. Each spatiotemporal module has a remainder link, and the results of the module are summarized and added to the result layer through jump connections, that is, the information of multi-hop neighbors is aggregated to the central node through Hop Joins. In the model, nodes interact with the information of neighboring nodes through Hop Joins, thereby aggregating the features of distant nodes at a specific level, which helps to capture more distant spatiotemporal dependencies in the graph structure.

[0054] The spatiotemporal module includes a gated temporal convolutional network unit GTCN for spatiotemporal data modeling, a dynamic graph convolutional network unit Dynamic-GCN for spatiotemporal data dynamic graph structure processing, and a hypergraph convolutional network unit HCN for multi-level feature capture of spatiotemporal data. The main function of the gated temporal convolutional network unit GTCN is to extract the spatiotemporal features of network nodes by combining graph convolution and temporal convolution; the dynamic graph convolutional network unit Dynamic-GCN can dynamically adjust the topological structure or weight of the graph according to time changes, so as to better capture spatiotemporal dependencies and improve the prediction ability of the model; the hypergraph convolutional network unit HCN can extract information at different spatiotemporal scales, so as to perform more detailed feature extraction and modeling on the input data, which models the local spatiotemporal patterns in the data through convolution operations, and gradually extracts high-level features through hierarchical structures. The spatiotemporal module also contains an abstract data extraction part for spatiotemporal data extraction. Abstraction can simplify and refine complex spatiotemporal data, extract the core information or features in the data, thereby reducing the complexity of the input and improving the generalization ability of the model. Through Abstraction, the model can capture important patterns and high-level features in the data, helping subsequent model layers to process and predict more effectively.

[0055] Among them, the expression of the gated time convolutional network unit of this model is:

[0056]

[0057] Among them, GTCN(-) represents the gated temporal convolutional network model, Characterizing the input cellular communication traffic sequence, TCN ω (·) Characterize the temporal convolutional network model, TCN ω1 (·) and TCN ω2 (·) respectively represent the temporal convolutional network models with different weight parameters ω1 and ω2, Conv h (·) represents the convolution kernel expansion convolution in the time dimension, th(·) represents the tanh activation function, sigmoid(·) represents the sigmoid activation function, and ⊙ represents the XOR operation.

[0058] In the current communication flow prediction framework based on graph convolutional networks (GCN), the adjacency matrix W is used to collect the spatial relationships between individual nodes at a specific time. Kipf et al. proposed the formula for GCN:

[0059]

[0060] in, is the normalized adjacency moment, D represents the input feature matrix, which can be generated by the pre-trained model, and Y represents the discovery parameter. The hypergraph convolutional network (HCN) proposed by Feng et al. is an upgrade of GCN. The expression of HCN is

[0061]

[0062] Based on this, the expression of the hypergraph convolutional network unit of this model is:

[0063]

[0064] H f =diag(R f )

[0065] in, Characterize the hypergraph convolutional network model, Represents the input feature matrix with power index n, D k Characterize the input feature matrix, Υ i is the discovery parameter, and i ∈γ=[γ0,...,γ I-1 ],i∈[0,I-1],M ks and M ktThe diagonal matrices representing the hyperedge degree and the diagonal matrix representing the hypernode degree respectively (the hyperedge degree refers to the number of supernodes connected to a specific hyperedge. In a hypergraph, a hyperedge can connect multiple supernodes, so the degree of a hyperedge is equal to the number of supernodes connected to the hyperedge. The supernode degree refers to the number of hyperedges connected to a specific supernode, and the degree of each supernode is equal to the number of hyperedges it participates in. Hyperedges and supernodes are two basic concepts in hypergraphs. Supernodes are similar to nodes in ordinary graphs, representing entities or objects; hyperedges are different from edges in ordinary graphs. Hyperedges can connect multiple supernodes at the same time, representing multi-dimensional relationships), E represents the association matrix, and R f represents the adaptive weight vector for discovering hyperedges in a hypergraph, and diag(·) represents the diagonal matrix function.

[0066] The expression of the dynamic graph convolutional network unit of this model is:

[0067]

[0068] Among them, DynamicGCN(·) represents the dynamic graph convolutional network model, R j and R k Respectively represent different adaptive adjacency matrices, γ j , γ b , Υ f , γ e and γ c are different discovery parameters, and all of them are ∈γ=[γ0,...,γ I-1 ], Reshape(·) represents the reshape function.

[0069] The dynamic spatiotemporal graph convolutional network of this method is obtained through training and testing, and the loss function used in the training and testing process is:

[0070]

[0071] Among them, depl represents the loss function, MAE(·) represents the mean absolute error operation, represents the set of cellular communication traffic prediction results from time point x+1 to time point x+X, and x represents the reference time point, represents the set of measured results of cellular communication traffic from time point x+1 to time point x+X, and

[0072] In order to evaluate the prediction effect of the dynamic spatiotemporal graph convolutional network of this method, a ground-truth mobile traffic dataset of a certain network area is collected, and the aggregated cellular traffic information of 20×20 areas within it is recorded every hour. The cellular traffic information is divided into training set, validation set and test set in a ratio of 7:1:2. Before making the prediction, the data of the previous 7 days and 3 hours were used as historical cellular traffic information. The DSGCN model of this method is then compared with several existing traffic prediction methods, and the comparison indicators include mean absolute error (MAE), root mean square error (RMSE) and accuracy. The final comparison results are shown in Table 1 below:

[0073] Table 1

[0074]

[0075] The results show that the DSGCN model of this method performs better than other methods in all indicators. According to the data comparison, the traffic prediction performance of DSGCN is 61% and 37% lower in MAE and RMSE respectively, and the accuracy is more than 13% higher.

[0076] In order to evaluate the training effectiveness of the DSGCN model and the stability of the optimization process, the training loss value of each time is recorded. As the number of iterations increases, convergence is an important indicator in the GCN training method. The training loss value of each epoch after the DSGCN model is iterated 100 times is as follows: Figure 3 As shown in the figure, it can be seen that as the number of iterations increases, the training loss of the DSGCN model gradually decreases and tends to stabilize, showing convergence. This shows that the model is gradually being optimized, and the difference between the model prediction results and the ground truth cellular traffic information is decreasing. In other words, the DSGCN model is discovering key features from the data and is able to fit the training data well. For comparison, the traffic prediction results of the DSGCN model of this method for 120 hours are visualized on the basis of the ground truth measurement data set, as shown in the figure. Figure 4 As shown in Figure 1, the prediction results of the proposed method DSGCN in a specific cellular service area are shown. The proposed method can predict the peak and valley values ​​of cellular services. That is, the proposed method can achieve good performance and excellent performance in capturing spatial correlation and dependence. In addition, based on Figure 4 The prediction curve of DSGCN shows that except for the difference in peak flow, the curves in other time periods tend to be consistent, which shows that the DSGCN of this method can effectively collect complex spatiotemporal features between different nodes.

[0077] Embodiment 2:

[0078] This embodiment provides a cellular communication traffic prediction system, including a data acquisition unit, a sequence construction unit and a traffic prediction unit, wherein:

[0079] A data acquisition unit, used to acquire cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point;

[0080] A sequence construction unit, configured to construct a cellular communication traffic measurement sequence of each network node within a historical time period based on the cellular communication traffic measurement data of each network node at each continuous historical time point, and to construct a historical cellular communication traffic sequence of a target network area based on the cellular communication traffic measurement sequence of each network node within the historical time period;

[0081] A traffic prediction unit is used to input the historical cellular communication traffic sequence of the target network area into a preset dynamic spatiotemporal graph convolutional network for traffic prediction, and obtain the cellular communication traffic prediction result of the target network area in the next time period. The dynamic spatiotemporal graph convolutional network includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence. The input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence. The stacked spatiotemporal layer includes a plurality of stacked spatiotemporal modules. The spatiotemporal module is used to extract spatiotemporal features from the output of the input layer, and the spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules. The output layer is used to output the cellular communication traffic prediction result.

[0082] Embodiment 3:

[0083] This embodiment provides a cellular communication traffic prediction system. Figure 5 As shown, at the hardware level, it includes:

[0084] Data interface, used to establish data connection between the processor and the external data terminal;

[0085] A memory for storing instructions;

[0086] The processor is used to read the instructions stored in the memory and execute the cellular communication traffic prediction method in Example 1 according to the instructions.

[0087] Optionally, the system further includes an internal bus, through which the processor, the memory and the data interface can be interconnected, and the internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0088] The memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO) and / or first in last out (FILO) memory, etc. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0089] Embodiment 4:

[0090] This embodiment provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed on a computer, the computer executes the cellular communication traffic prediction method in Embodiment 1. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick, etc., and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0091] This embodiment further provides a computer program product, which, when executed on a computer, executes the cellular communication traffic prediction method in Embodiment 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0092] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cellular communication traffic prediction method, characterized in that: include: Obtaining cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point; Based on the cellular communication traffic measurement data of each network node at each continuous historical time point, construct a cellular communication traffic measurement sequence of each network node in the historical time period, and construct a historical cellular communication traffic sequence of the target network area based on the cellular communication traffic measurement sequence of each network node in the historical time period; The historical cellular communication traffic sequence of the target network area is input into a preset dynamic spatiotemporal graph convolutional network for traffic prediction to obtain the cellular communication traffic prediction result of the target network area in the next time period. The dynamic spatiotemporal graph convolutional network includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence. The input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence. The stacked spatiotemporal layer includes a plurality of stacked spatiotemporal modules. The spatiotemporal module is used to extract spatiotemporal features from the output of the input layer. The spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules. The output layer is used to output the cellular communication traffic prediction result.

2. A cellular communication traffic prediction method according to claim 1, characterized in that: The input layer includes a first linear transformation module, which is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence and output the processing result to each spatiotemporal module. The output layer includes a Leaky ReLU activation function module and a second linear transformation module, which is used to perform Leaky ReLU activation function processing on the summary output of each spatiotemporal module to obtain the activation function processing result. The second linear transformation module is used to perform feature dimension conversion processing on the activation function processing result and output the cellular communication traffic prediction result.

3. A cellular communication traffic prediction method according to claim 1, characterized in that: The spatiotemporal module includes a gated time convolutional network unit for performing spatiotemporal data modeling processing, a dynamic graph convolutional network unit for performing spatiotemporal data dynamic graph structure processing, and a hypergraph convolutional network unit for capturing multi-level features of spatiotemporal data.

4. A cellular communication traffic prediction method according to claim 3, characterized in that: The expression of the gated temporal convolutional network unit is Among them, CTCN(·) represents the gated temporal convolutional network model, Characterizing the input cellular communication traffic sequence, TCN ω (·) Characterize the temporal convolutional network model, TCN ω1 (·) and TCN ω2 (·) respectively represent the temporal convolutional network models with different weight parameters ω1 and ω2, Conv h (·) represents the convolution kernel expansion convolution in the time dimension, th(·) represents the tanh activation function, sigmoid(·) represents the sigmoid activation function, and ⊙ represents the XOR operation.

5. A cellular communication traffic prediction method according to claim 4, characterized in that: The expression of the hypergraph convolutional network unit is H f =diag(R f ) in, Characterize the hypergraph convolutional network model, Represents the input feature matrix with power index n, D k Characterize the input feature matrix, Υ i is the discovery parameter, and i ∈Υ=[Υ0,…,Υ I-1 ],i∈[0,I-1],M ks and M kt They represent the diagonal matrix of hyperedge degree and the diagonal matrix of hypernode degree respectively, E represents the correlation matrix, R f represents the adaptive weight vector for discovering hyperedges in a hypergraph, and diag(·) represents the diagonal matrix function.

6. A cellular communication traffic prediction method according to claim 5, characterized in that: The expression of the dynamic graph convolutional network unit is Among them, DynamicGCN(·) represents the dynamic graph convolutional network model, R j and R k Respectively represent different adaptive adjacency matrices, Υ j , Υ b , Υ f , Υ e And Υ c are different discovery parameters, and all of them are ∈Υ=[Υ0,…,Υ I-1 ], Reshape(·) represents the reshape function.

7. A cellular communication traffic prediction method according to claim 1, characterized in that: The dynamic spatiotemporal graph convolutional network is obtained through training and testing, and the loss function used in the training and testing process is Among them, depl represents the loss function, MAE(·) represents the mean absolute error operation, represents the set of cellular communication traffic prediction results from time point x+1 to time point x+X, and x represents the reference time point, represents the set of measured results of cellular communication traffic from time point x+1 to time point x+X, and 8. A cellular communication traffic prediction system, characterized in that: It includes a data acquisition unit, a sequence construction unit and a traffic prediction unit, wherein: A data acquisition unit, used to acquire cellular communication traffic measurement data of each network node in the target network area at each continuous historical time point; A sequence construction unit, configured to construct a cellular communication traffic measurement sequence of each network node within a historical time period based on the cellular communication traffic measurement data of each network node at each continuous historical time point, and to construct a historical cellular communication traffic sequence of a target network area based on the cellular communication traffic measurement sequence of each network node within the historical time period; A traffic prediction unit is used to input the historical cellular communication traffic sequence of the target network area into a preset dynamic spatiotemporal graph convolutional network for traffic prediction, and obtain the cellular communication traffic prediction result of the target network area in the next time period. The dynamic spatiotemporal graph convolutional network includes an input layer, a stacked spatiotemporal layer and an output layer connected in sequence. The input layer is used to perform feature dimension conversion processing on the input historical cellular communication traffic sequence. The stacked spatiotemporal layer includes a plurality of stacked spatiotemporal modules. The spatiotemporal module is used to extract spatiotemporal features from the output of the input layer, and the spatiotemporal features are summarized and output to the output layer through jump connection operations between the spatiotemporal modules. The output layer is used to output the cellular communication traffic prediction result.

9. A cellular communication traffic prediction system, characterized in that: include: A memory for storing instructions; A processor is used to read the instructions stored in the memory and execute the cellular communication traffic prediction method according to any one of claims 1-7 according to the instructions.

10. A computer program product, characterized in that When the computer program product is run on a computer, the cellular communication traffic prediction method according to any one of claims 1 to 7 is executed.

Images