Base station traffic prediction method and device, nonvolatile storage medium and electronic equipment

Abstract

The invention discloses a base station traffic prediction method and device, a nonvolatile storage medium and electronic equipment. The method comprises the steps that base station historical flow data are acquired, a graph structure is constructed according to the base station historical flow data, nodes in the graph structure are base stations, and edges in the graph structure are used for reflecting the incidence relation between the base stations; according to the graph structure, a feature matrix is determined, and elements in the feature matrix are traffic values of the nodes in the preset abnormal region; processing the feature matrix through a graph attention network model to obtain a target traffic feature, the target traffic feature being time series data; and analyzing the target traffic characteristics through the long short-term memory network model to obtain a traffic prediction result. According to the method and the device, the technical problem that various resources in a communication system cannot be efficiently scheduled due to the fact that the base station flow cannot be accurately predicted in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular, to a method, apparatus, non-volatile storage medium, and electronic device for predicting base station traffic. Background Art

[0002] In the related art, when predicting base station traffic, in the face of the sharp increase in data volume, the increase in dimensions, the enhancement of non-linearity, and the intensification of dynamic changes, it is impossible to accurately predict the base station traffic, resulting in the inability to efficiently schedule various resources in the communication system.

[0003] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] Embodiments of the present application provide a method, apparatus, non-volatile storage medium, and electronic device for predicting base station traffic, so as to at least solve the technical problem that various resources in the communication system cannot be efficiently scheduled due to the inability to accurately predict base station traffic in the related art.

[0005] According to one aspect of the embodiments of the present application, a method for predicting base station traffic is provided, including: obtaining historical base station traffic data, and constructing a graph structure based on the historical base station traffic data, where nodes in the graph structure are base stations, and edges in the graph structure are used to reflect the association relationship between base stations; determining a feature matrix according to the graph structure, where elements in the feature matrix are traffic values of nodes at a preset moment; processing the feature matrix through a graph attention network model to obtain a target traffic feature, where the target traffic feature is time series data; and analyzing the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

[0006] Optionally, constructing a graph structure based on the historical base station traffic data includes: determining that there is an edge connection between nodes corresponding to two base stations when there is an overlapping area between the coverage areas of the two base stations; or determining that there is an edge connection between nodes corresponding to the two base stations when the distance between the two base stations is less than a preset distance threshold; or determining whether there is an edge connection between nodes corresponding to base stations according to user movement information.

[0007] Optionally, the method further includes: when determining whether there is an edge connection between nodes corresponding to two base stations according to the overlapping area, the higher the overlapping degree of the coverage areas of the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between nodes corresponding to two base stations according to the distance, the shorter the distance between the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between nodes corresponding to base stations according to user mobility information, the higher the frequency of user mobility information indicating migration from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station, or the greater the number of user mobility information indicating migration from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station.

[0008] Optionally, processing the feature matrix through the graph attention network model includes: performing normalization processing on the feature matrix; processing the normalized feature matrix through the graph attention network model to determine the similarity coefficient between the neighboring base stations of the target base station and the target base station, where the similarity coefficient is used to reflect the influence degree of the neighboring base stations on the target base station; processing the traffic characteristics of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic characteristics.

[0009] Optionally, the graph attention network model further includes a multi-head attention mechanism; processing the traffic characteristics of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic characteristics includes: determining multiple groups of similarity coefficients through the multi-head attention mechanism, where each attention head of the multi-head attention mechanism independently determines a group of similarity coefficients; processing the traffic characteristics of the target base station and the neighboring base stations according to each group of similarity coefficients to obtain intermediate traffic characteristics; concatenating the intermediate traffic characteristics corresponding to each group of similarity coefficients to obtain the target traffic characteristics; or averaging the intermediate traffic characteristics corresponding to each group of similarity coefficients to obtain the target traffic characteristics.

[0010] Optionally, the base station historical traffic data includes at least one of the following: traffic periodic change information of the base station, location information of the base station, coverage area information of the base station, user distribution information of the base station, weather information, and holiday information.

[0011] Optionally, the attributes of the node include at least one of the following: location information of the base station corresponding to the node, coverage area information of the base station corresponding to the node, user distribution information of the base station corresponding to the node.

[0012] According to another aspect of the embodiments of the present application, there is also provided a base station traffic prediction device, including: a first processing module, configured to obtain base station historical traffic data and construct a graph structure based on the base station historical traffic data, wherein nodes in the graph structure are base stations, and edges in the graph structure are used to reflect the association relationship between base stations; a second processing module, configured to determine a feature matrix based on the graph structure, wherein elements in the feature matrix are traffic values of nodes at a preset moment; a third processing module, configured to process the feature matrix through a graph attention network model to obtain a target traffic feature, wherein the target traffic feature is time series data; and a fourth processing module, configured to analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

[0013] According to another aspect of the embodiments of the present application, there is also provided a non-volatile storage medium storing a program, wherein when the program runs, it controls the device where the non-volatile storage medium is located to execute the base station traffic prediction method.

[0014] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including a memory and a processor, where the processor is configured to run the program stored in the memory, and when the program runs, it executes the base station traffic prediction method.

[0015] According to another aspect of the embodiments of the present application, there is also provided a computer program product, including a computer program, where the computer program, when executed by a processor, implements the base station traffic prediction method.

[0016] In the embodiments of the present application, by obtaining base station historical traffic data and constructing a graph structure based on the base station historical traffic data, wherein nodes in the graph structure are base stations, and edges in the graph structure are used to reflect the association relationship between base stations; determining a feature matrix based on the graph structure, wherein elements in the feature matrix are traffic values of nodes at a preset moment; processing the feature matrix through a graph attention network model to obtain a target traffic feature, wherein the target traffic feature is time series data; and analyzing the target traffic feature through a long short-term memory network model to obtain a traffic prediction result, the base station traffic is predicted by combining the graph attention network model and the long short-term memory network model, achieving the purpose of accurately predicting the base station traffic, thereby realizing the technical effect of improving the resource scheduling efficiency of the communication system, and further solving the technical problem that various resources in the communication system cannot be efficiently scheduled due to the inability to accurately predict the base station traffic in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0018] Figure 1 It is a schematic structural diagram of a computer terminal (mobile terminal) provided according to an embodiment of the present application;

[0019] Figure 2 It is a schematic flowchart of a base station traffic prediction method provided according to an embodiment of the present application;

[0020] Figure 3 It is a schematic diagram of normalizing the attention coefficient provided according to an embodiment of the present application;

[0021] Figure 4 It is a schematic diagram of the multi-head attention mechanism provided according to the implementation of the present application;

[0022] Figure 5 It is a schematic architecture diagram of a long short-term memory network model provided according to an embodiment of the present application;

[0023] Figure 6 It is a schematic flowchart of the base station traffic prediction process provided according to an embodiment of the present application;

[0024] Figure 7 It is a schematic structural diagram of a base station traffic prediction device provided according to an embodiment of the present application. Detailed implementation manners

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily need to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0027] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0028] Graph Attention Networks: Graph Attention Networks (GAT) are advanced neural network models designed for graph structured data. The core of Graph Attention Networks is to integrate the attention mechanism to achieve adaptive weight allocation for neighbor nodes. Compared with traditional graph convolutional networks, GAT significantly enhances the model's ability to capture complex node relationships and features by dynamically calculating the correlation scores between nodes as the weight for aggregating neighbor information. This innovative mechanism enables GAT to understand the information flow in the graph structure more delicately, effectively extract local features, and demonstrate excellent performance in tasks such as graph classification, node classification, and link prediction. The introduction of the attention mechanism not only improves the expressiveness and generalization capabilities of the model, but also provides strong support for in-depth exploration of the deep-level features of graph data. As a cutting-edge technology in the field of graph machine learning, GAT, through its unique aggregation strategy and dynamic weight allocation mechanism, has brought new ideas and methods to the processing of complex graph structured data, and promoted the further development of graph neural network technology.

[0029] Long Short-Term Memory: Long Short-Term Memory (LSTM) is an advanced variant or upgraded version of Recurrent Neural Network (RNN), which is particularly good at processing long sequence data. With its unique gating mechanism, it effectively solves the gradient problem in long sequence training. The significant advantage of LSTM lies in its excellent time feature processing ability, which can accurately capture and retain long-term dependencies in the sequence, even when facing information with long time intervals. This feature enables LSTM to show excellent performance far exceeding traditional RNN when dealing with tasks such as time series analysis and language modeling that require a deep understanding of time dynamics, making it the preferred model for processing time series data in the field of deep learning.

[0030] Multi-head Attention Mechanism: Multi-head Attention Mechanism, a technology favored by the fields of computer science and machine learning, is inspired by a deep insight into human vision and cognitive processes. Drawing on the research results of psychology and neuroscience, when faced with massive amounts of information, humans will instinctively focus on certain key parts and temporarily ignore other non-core information. This is the biological basis of the attention mechanism. The core goal of this mechanism is to significantly improve the accuracy and efficiency of sequence data processing by accurately locating and focusing on specific input subsets.

[0031] Softmax: As a form of sigmoid in multi-classification scenarios, it plays a key role in converting classification results into probability distribution. It cleverly compresses the multi-channel output of the neural network into the (0,1) interval and ensures that the sum of these output values is 1. Each value can be regarded as the predicted probability of the corresponding category.

[0032] In today's era of rapid development of informatization, communication networks are the cornerstone of information transmission. Accurate prediction of traffic is of vital importance for network planning, efficient resource scheduling, and fault prevention. With the accelerated commercial deployment of 5G technology and the widespread access of Internet of Things (IoT) devices, the traffic characteristics of communication networks have shown unprecedented complexity and uncertainty. This is mainly reflected in the rapid changes in traffic patterns, the frequent occurrence of burst traffic, and the diversification of traffic composition, which brings huge challenges to network management and optimization.

[0033] Among the related technologies, time series analysis and machine learning technologies have played an important role in the field of network traffic prediction. However, in the era of 5G and the Internet of Things, facing the dramatic increase in data volume, the increase in dimensionality, the enhancement of nonlinearity, and the intensification of dynamic changes, these traditional methods have gradually shown inherent limitations such as limited generalization ability and poor prediction accuracy. Therefore, there is an urgent need for a prediction technology that can meet the characteristics of the new network environment and has excellent learning and generalization performance.

[0034] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.

[0035] According to an embodiment of the present application, a method embodiment of a base station traffic prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0036] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a base station traffic prediction method. Figure 1As shown, the computer terminal 10 (or mobile device 10) may include one or more processors 102 (shown as 102a, 102b, ……, 102n in the figure) (the processor 102 may include, but is not limited to, processing devices such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above-mentioned electronic device. For example, the computer terminal 10 may further include more or fewer components than Figure 1 shown in Figure 1 or have a different configuration from

[0037] It should be noted that the above one or more processors 102 and / or other data processing circuits are generally referred to as "data processing circuits" in this article. The data processing circuit may be embodied in whole or in part as software, hardware, firmware, or any combination thereof. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the computer terminal 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit is a processor control (such as the selection of a variable resistor terminal path connected to an interface).

[0038] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the base station traffic prediction method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implements the above-mentioned base station traffic prediction method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely located with respect to the processor 102, and these remote memories may be connected to the computer terminal 10 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0039] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0040] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0041] In the above operating environment, the embodiment of the present application provides a base station traffic prediction method, such as Figure 2 As shown, the method comprises the following steps:

[0042] Step S202, obtaining historical traffic data of the base station, and constructing a graph structure based on the historical traffic data of the base station, wherein the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationship between the base stations;

[0043] In the technical solution provided in step S202, the base station historical traffic data includes at least one of the following: base station traffic periodic change information, base station location information, base station coverage area information, base station user distribution information, weather information and holiday information.

[0044] In some embodiments of the present application, when acquiring the historical traffic data of the base station, statistical analysis methods can be used to eliminate obviously illogical traffic data. For the missing traffic data, interpolation estimation, regression analysis and other strategies are adopted to reasonably fill in the missing traffic data to ensure the integrity of the data set. Duplicate traffic records are completely removed to ensure the uniqueness and accuracy of the data set.

[0045] In addition, the flow data can be normalized to limit its value range to [0,1]. Further standardization is performed to make the flow data have zero mean and unit variance characteristics.

[0046] In some embodiments of the present application, when constructing a graph structure based on the historical traffic data of base stations, it is also necessary to further mine the data characteristics in the historical traffic data of base stations. For example, key time series characteristics can be extracted from the historical data traffic records first. This characteristic set synthesizes the detailed fluctuation patterns during the day and the periodic weekly change rules, aiming to accurately depict the dynamic evolution trajectory of the data traffic over time. Further, the present invention deepens the understanding of the time series data by calculating and analyzing a series of statistical metrics. These metrics specifically include the average traffic level to characterize the traffic baseline state, the traffic volatility to quantify the severity of traffic changes, and the long-term change trend to reveal the macroscopic development path of the traffic. The above-mentioned deep statistical characteristics, as the key elements to improve the performance of the prediction model, are innovatively incorporated into the prediction model, significantly enhancing the accuracy and stability of the model for predicting data traffic.

[0047] In addition, the spatial characteristics of the base stations and external factors can be integrated and analyzed. For the spatial characteristics, a spatial feature matrix can be constructed by combining spatial elements such as the geographical location of the base station, the coverage area, and the user distribution, so as to comprehensively reflect the geographical relevance of the traffic. Using Geographic Information System (GIS) technology, the spatial information is converted into numerical features recognizable by the model, enhancing the model's understanding ability of spatial relationships.

[0048] Regarding external factors, severe weather and the like may cause a surge in traffic, while holidays may change the network usage behavior of users. Through data integration technology, the external factor data and the network traffic data are organically integrated to form a more comprehensive and accurate input feature set, providing richer information support for the model. In this way, the potential impact of external factors such as weather conditions and holiday arrangements on network traffic can be fully considered.

[0049] As an alternative implementation manner, the attributes of the node include at least one of the following: the location information of the base station corresponding to the node, the coverage area information of the base station corresponding to the node, and the user distribution information of the base station corresponding to the node.

[0050] Optionally, each base station serves as a node in the graph structure, and it can also be considered that there is a one-to-one correspondence between the base station and the node in the graph structure. The attributes of the node can include basic information such as the geographical location (such as longitude and latitude), the coverage area, and the user distribution of the base station corresponding to the node.

[0051] In some embodiments of the present application, the steps of constructing a graph structure based on the historical traffic data of base stations include: when there is an overlapping area between the coverage areas of two base stations, determining that there is an edge connection between the nodes corresponding to these two base stations; or, when the distance between two base stations is less than a preset distance threshold, determining that there is an edge connection between the nodes corresponding to these two base stations; or, determining whether there is an edge connection between the nodes corresponding to the base stations according to the user mobility information.

[0052] Optionally, based on preset rules, it can be determined which base stations (nodes) are connected (edges). This is determined according to factors such as the physical distance between base stations, the degree of overlap of coverage areas, and user mobility patterns. For example, if the coverage areas of two base stations overlap, or the distance between them is less than a certain threshold, then it can be considered that there is an edge between them. Or the edge can be defined according to the frequency of users moving from one base station to another, and the higher the frequency, the more likely the edge exists.

[0053] As an optional implementation manner, when determining whether there is an edge connection between the nodes corresponding to two base stations based on the overlapping area, the higher the degree of overlap of the coverage areas of the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between the nodes corresponding to two base stations based on the distance, the shorter the distance between the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between the nodes corresponding to base stations based on user mobility information, the higher the frequency indicated by the user mobility information of migrating from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station, or the more the number indicated by the user mobility information of migrating from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station.

[0054] In some embodiments of the present application, the weight of the edge can reflect the association strength or similarity between base stations. The calculation of the weight can be comprehensively considered according to various factors, including but not limited to:

[0055] Physical distance: The closer the distance between two base stations, the greater the possible weight.

[0056] Coverage overlap: The higher the degree of overlap of the coverage areas of two base stations, the greater the possible weight.

[0057] User mobility: The frequency or number of users migrating from one base station to another, the higher the frequency or the more the number, the greater the possible weight.

[0058] Traffic correlation: The correlation between the traffic data of two base stations, the higher the correlation, the greater the possible weight.

[0059] As an optional implementation manner, the weight can also be normalized to limit its value within a specific range (such as [0,1]).

[0060] Step S204, determining a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment;

[0061] In the technical solution provided in step S204, the graph structure data can be constructed into a feature matrix X, and in the matrix Denote the traffic value of node i at time t - m, and then normalize the matrix data. After that, use the Graph Attention Network (GAT) model to process the normalized feature matrix.

[0062] As an alternative implementation, by combining the traffic value of node i at time t - m with the traffic data of other relevant nodes, the GAT model can gain insights into the spatial patterns and correlations of traffic distribution. This processing involves multiple layers of attention mechanisms in deep neural networks. Each layer of the attention mechanism focuses on extracting spatial dependencies at different scales, weights the traffic values of neighbor nodes according to their association strength, and finally generates the feature representation h_i of node i. It should be noted that the introduction of the multi - head attention mechanism in the GAT model further enhances this process. By independently calculating attention weights through multiple parallel attention heads, analyzing the relationships between nodes from different perspectives, and then integrating the results, it significantly improves the model's ability to capture the complex structure of graph data and generalization performance.

[0063] In addition, by combining the attention mechanism with the characteristics of graph - structured data, the GAT model can more subtly understand the traffic interaction patterns between different base stations and the trends of how these patterns evolve over time. This not only considers the physical characteristics of direct node connections, such as distance and coverage overlap, but also takes into account dynamic user flow patterns and statistical correlations between traffic data, enabling the feature representation of each node to comprehensively reflect its role and influence in the graph structure.

[0064] Step S206: Process the feature matrix through the Graph Attention Network model to obtain the target traffic feature, where the target traffic feature is time - series data;

[0065] In the technical solution provided in step S206, the steps of processing the feature matrix through the Graph Attention Network model include: normalizing the feature matrix; processing the normalized feature matrix through the Graph Attention Network model to determine the similarity coefficient between the adjacent base stations of the target base station and the target base station, where the similarity coefficient is used to reflect the degree of influence of the adjacent base station on the target base station; and processing the traffic features of the target base station and the adjacent base stations according to the similarity coefficient to obtain the target traffic feature.

[0066] In some embodiments of the present application, the above - mentioned similarity coefficient is also referred to as the attention coefficient. In traffic prediction, the similarity coefficient e ij between the adjacent sites and the target toll station can be calculated one by one through the formula, that is, the influence coefficient of node i on node j.

[0067] e ij = a([Wh i ||Wh j), j ∈ N i

[0068] where: N i is the time feature of the node; W is a shared parameter used to increase the dimension of the node features; || denotes concatenating the features of the transformed nodes. h i represents the feature vector of base station i, and h j represents the feature vector of base station j. a() represents an activation function. Then, the inner product operation can be performed on the concatenated vector and the new feature vector, and finally, the weights are activated through a function. Additionally, for convenient weight assignment, the correlation degrees calculated between the target node and all adjacent nodes need to be uniformly normalized. As Figure 3 shown, the Softmax normalization method can be adopted. The normalization formula is as follows:

[0069]

[0070] In the above formula, LeakyReLU() represents the activation function, k represents the node corresponding to any adjacent base station of the target base station, and exp() represents the ()-th power of the natural exponential e. α ij represents the normalized correlation coefficient of node j relative to node i. e ij represents the correlation coefficient of node j relative to node i before normalization, and e ik represents the correlation coefficient of node k relative to node i before normalization.

[0071] After that, the linear combination of the features corresponding to the normalized attention coefficients can be calculated. After passing through a non-linear activation function, the final output feature vector corresponding to each node is:

[0072]

[0073] In the above formula, σ() represents the non-linear activation function.

[0074] In some embodiments of the present application, the graph attention network model further includes a multi-head attention mechanism; processing the traffic features of the target base station and adjacent base stations according to the similarity coefficients to obtain the target traffic features includes: determining multiple groups of similarity coefficients through the multi-head attention mechanism, where each attention head of the multi-head attention mechanism independently determines a group of similarity coefficients; processing the traffic features of the target base station and adjacent base stations according to each group of similarity coefficients to obtain intermediate traffic features; concatenating the intermediate traffic features corresponding to each group of similarity coefficients to obtain the target traffic features; or averaging the intermediate traffic features corresponding to each group of similarity coefficients to obtain the target traffic features.

[0075] Optionally, the effect of the graph attention layer also needs to be stacked in multiple layers. Therefore, it can be adopted as Figure 4The multi-head attention mechanism shown above improves efficiency, and the formula is as follows:

[0076]

[0077] In the above formula, represents the concatenation of the output features of the graph attention layers from the first layer to the Kth layer. σ() represents the activation function. represents the correlation coefficient between nodes i and j corresponding to the kth layer, and W k represents the weight matrix of the kth layer. h′ i (K) represents the feature vector of node i obtained by the final concatenation.

[0078] The multi-head attention mechanism is an extension of the basic attention mechanism. By introducing multiple attention heads, each head independently calculates the attention, and then the results are concatenated together for linear transformation. Its core idea is to capture more fine-grained information by independently calculating the attention through multiple attention heads, thereby improving the performance of the model. As can be seen from Figure 4 , each head of the multi-head attention mechanism will generate a set of weights and calculate the corresponding feature vectors. Then, the feature vectors corresponding to each head can be fused by concatenation or averaging to obtain the final feature vector corresponding to each node.

[0079] Step S208: Analyze the target traffic characteristics through a long short-term memory network model to obtain the traffic prediction result.

[0080] In the technical solution provided in step S208, the long short-term memory network model is a special recurrent neural network model, which has both long-term and short-term memory, and improves the gradient dispersion and gradient explosion problems of traditional recurrent neural network models. The LSTM model has three threshold units: a forget gate, an input gate, and an output gate, and transmits information along the time series through linear operations. Since each operation will recycle the information and input it again, it has good memory, and the model realizes the screening of input information through the "gate" structure.

[0081] The structure of the long short-term memory network model provided in the embodiments of the present application is as shown in Figure 5 , including a forget gate, an input gate, and an output gate. Among them, the forget gate is used to distinguish the data that needs to be retained or discarded, and the formula is:

[0082] f t =σ g (W f x t +U f h t-1 +b f )

[0083] The input gate is used to determine new information that can be stored in the cell and is divided into two steps: 1) Confirm the value to be updated in the sigmoid layer to obtain i t , and then determine the data to be updated through the tanh(·) layer to generate the data vector S t : 2) Generate the state c of the new cell. The specific calculation formula is:

[0084] i t =σ g (Wix t +Uih t-1 +bi)

[0085] S t =tanh{W c ·(W c x t +U c h t-1 )+b c}

[0086] c t =f t ·c t-1 +i t ·s t

[0087] The output gate is used to determine how much of the information (hidden state) h at the previous moment, the input x at the current moment t-1 , and the previously obtained cell state c are respectively passed to the next moment and output. The formula is t o

[0088] o t =σ g (W o x t +U o h t-1 +b o )

[0089] h t =o t ·tanh(C t )

[0090] In the above formulas, W f , b f , W i , b i , W o , b o are the weights and bias values of each gate respectively; σ(·) and tanh(·) are both activation functions. The subscript t indicates that the parameter or variable corresponds to the t-th moment, and the subscript t-1 indicates the corresponding (t-1)-th moment, that is, the previous moment of the t-th moment.

[0091] According to an embodiment of the present application, there is also provided a traffic prediction process as Figure 6 shown, including the following steps:

[0092] Step S602: Collect the historical traffic data of the base station and perform preprocessing;

[0093] Step S604: Extract the spatial features and temporal features from the historical traffic data of the base station, and construct a graph structure according to the extracted features;

[0094] Step S606: Process the graph structure through the GAT model, so as to map the spatial features into the time series data group to obtain a feature vector group;

[0095] Step S608: Input the feature vector group output by the GAT model into the LSTM model, so that the model learns the temporal features of the traffic data;

[0096] Step S610: Test the prediction model on the test set, and use the prediction model to predict the base station traffic after passing the test, where the prediction model is composed of the GAT model and the LSTM model.

[0097] By obtaining the historical traffic data of the base station and constructing a graph structure based on the historical traffic data of the base station, where the nodes in the graph structure are base stations and the edges in the graph structure are used to reflect the association relationship between base stations; determining a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; processing the feature matrix through a graph attention network model to obtain target traffic features, where the target traffic features are time series data; analyzing the target traffic features through a long short-term memory network model to obtain a traffic prediction result, the method of predicting the base station traffic by combining the graph attention network model and the long short-term memory network model achieves the purpose of accurately predicting the base station traffic, thus realizing the technical effect of improving the resource scheduling efficiency of the communication system, and further solving the technical problem that various resources in the communication system cannot be efficiently scheduled due to the inability to accurately predict the base station traffic in the related art.

[0098] The method provided by the embodiments of this application combines two deep learning models, GAT and LSTM, to form a unique network traffic prediction framework. GAT is responsible for capturing the spatial relationships and features between base stations, while LSTM focuses on processing the temporal features of traffic data. The combination of the two enables efficient prediction of base station traffic data. Moreover, in the embodiments of this application, the base stations are defined in detail as nodes in the graph structure, and the edges and their weights determined based on factors such as physical distance, coverage overlap, user mobility, and traffic correlation are also defined. This way of constructing the graph structure comprehensively reflects the complex relationships between base stations and provides rich input features for the GAT model. Additionally, a multi-head attention mechanism is introduced in this application. By independently calculating attention through multiple attention heads, more fine-grained information can be captured. This mechanism significantly enhances the expressive power and generalization ability of the model and improves the prediction accuracy. Furthermore, the data preprocessing steps provided by the embodiments of this application also include data purification, standardization, temporal feature mining, spatial feature integration, and external factor fusion analysis. These steps together ensure the quality and integrity of the input data and provide a basis for model training.

[0099] The embodiments of this application provide a base station traffic prediction device, Figure 7 which is a schematic structural diagram of the device. As can be seen from Figure 7 it, the device includes: a first processing module 70, configured to obtain historical traffic data of base stations and construct a graph structure based on the historical traffic data of base stations, where the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationships between base stations; a second processing module 72, configured to determine a feature matrix based on the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; a third processing module 74, configured to process the feature matrix through a graph attention network model to obtain target traffic features, where the target traffic features are temporal data; and a fourth processing module 76, configured to analyze the target traffic features through a long short-term memory network model to obtain a traffic prediction result.

[0100] In some embodiments of this application, the historical traffic data of base stations includes at least one of the following: traffic periodic change information of base stations, location information of base stations, coverage area information of base stations, user distribution information of base stations, weather information, and holiday information.

[0101] In some embodiments of this application, the attributes of the nodes include at least one of the following: location information of the base stations corresponding to the nodes, coverage area information of the base stations corresponding to the nodes, and user distribution information of the base stations corresponding to the nodes.

[0102] In some embodiments of the present application, the steps for the first processing module 70 to construct a graph structure based on the historical traffic data of the base stations include: when there is an overlapping area between the coverage areas of two base stations, determining that there is an edge connection between the nodes corresponding to the two base stations; or, when the distance between two base stations is less than a preset distance threshold, determining that there is an edge connection between the nodes corresponding to the two base stations; or, determining whether there is an edge connection between the nodes corresponding to the base stations according to the user mobility information.

[0103] In some embodiments of the present application, when determining whether there is an edge connection between the nodes corresponding to two base stations according to the overlapping area, the higher the overlapping degree of the coverage areas of the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between the nodes corresponding to two base stations according to the distance, the shorter the distance between the two base stations, the greater the weight of the corresponding edge; when determining whether there is an edge connection between the nodes corresponding to the base stations according to the user mobility information, the higher the frequency of the user mobility information indicating migration from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station, or the more the number of migrations from the first base station to the second base station indicated by the user mobility information, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station.

[0104] In some embodiments of the present application, the steps for the third processing module 74 to process the feature matrix through the graph attention network model include: performing normalization processing on the feature matrix; processing the normalized feature matrix through the graph attention network model to determine the similarity coefficient between the neighboring base stations of the target base station and the target base station, where the similarity coefficient is used to reflect the influence degree of the neighboring base stations on the target base station; processing the traffic characteristics of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic characteristics.

[0105] In some embodiments of the present application, the graph attention network model further includes a multi-head attention mechanism; the steps for the third processing module 74 to process the traffic characteristics of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic characteristics include: determining multiple groups of similarity coefficients through the multi-head attention mechanism, where each attention head of the multi-head attention mechanism independently determines a group of similarity coefficients; processing the traffic characteristics of the target base station and the neighboring base stations according to each group of similarity coefficients to obtain intermediate traffic characteristics; splicing the intermediate traffic characteristics corresponding to each group of similarity coefficients to obtain the target traffic characteristics; or taking the average of the intermediate traffic characteristics corresponding to each group of similarity coefficients to obtain the target traffic characteristics.

[0106] It should be noted that each module in the above base station traffic prediction device can be a program module (for example, a set of program instructions for implementing a specific function), or a hardware module. For the latter, it can be presented in the following forms, but not limited to: the manifestation form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0107] According to an embodiment of the present application, a non-volatile storage medium is provided. A program is stored in the non-volatile storage medium. When the program runs, it controls the device where the non-volatile storage medium is located to execute the following base station traffic prediction method: obtain base station historical traffic data, and construct a graph structure based on the base station historical traffic data, where the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationship between base stations; determine a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; process the feature matrix through a graph attention network model to obtain a target traffic feature, where the target traffic feature is time series data; analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

[0108] According to an embodiment of the present application, an electronic device is further provided, including a memory and a processor. The processor is used to run the program stored in the memory. When the program runs, it executes the following base station traffic prediction method: obtain base station historical traffic data, and construct a graph structure based on the base station historical traffic data, where the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationship between base stations; determine a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; process the feature matrix through a graph attention network model to obtain a target traffic feature, where the target traffic feature is time series data; analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

[0109] According to an embodiment of the present application, a computer program product is further provided, including a computer program. When the computer program is executed by a processor, it implements the following base station traffic prediction method: obtain base station historical traffic data, and construct a graph structure based on the base station historical traffic data, where the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationship between base stations; determine a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; process the feature matrix through a graph attention network model to obtain a target traffic feature, where the target traffic feature is time series data; analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

[0110] In the above embodiments of the present application, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0111] In several embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0112] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0113] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0114] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the related technology, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0115] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A base station traffic prediction method, characterized in that, Including: Obtain the historical traffic data of base stations, and construct a graph structure based on the historical traffic data of base stations. Wherein, the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the correlation relationship between base stations; Determine a feature matrix based on the graph structure. Wherein, the elements in the feature matrix are the traffic values of nodes at a preset moment; Process the feature matrix through a graph attention network model to obtain a target traffic feature. Wherein, the target traffic feature is time-series data; Analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

2. The base station traffic prediction method according to claim 1, wherein Constructing a graph structure based on the historical traffic data of base stations includes: In the case where there is an overlapping area between the coverage areas of two base stations, determine that there is an edge connection between the nodes corresponding to the two base stations; or, In the case where the distance between two base stations is less than a preset distance threshold, determine that there is an edge connection between the nodes corresponding to the two base stations; or, Determine whether there is an edge connection between the nodes corresponding to the base stations according to the user mobility information.

3. The base station traffic prediction method according to claim 2, wherein The method further includes: In the case of determining whether there is an edge connection between the nodes corresponding to two base stations according to the overlapping area, the higher the overlapping degree of the coverage areas of the two base stations, the greater the weight of the corresponding edge; In the case of determining whether there is an edge connection between the nodes corresponding to two base stations according to the distance, the shorter the distance between the two base stations, the greater the weight of the corresponding edge; In the case of determining whether there is an edge connection between the nodes corresponding to the base stations according to the user mobility information, the higher the frequency indicated by the user mobility information for migrating from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station, or the more the number indicated by the user mobility information for migrating from the first base station to the second base station, the greater the weight of the edge between the nodes corresponding to the first base station and the second base station.

4. The base station traffic prediction method according to claim 1, characterized in that Processing the feature matrix through a graph attention network model includes: Perform normalization processing on the feature matrix; Process the normalized feature matrix through the graph attention network model to determine the similarity coefficient between the neighboring base stations of the target base station and the target base station. Wherein, the similarity coefficient is used to reflect the influence degree of the neighboring base stations on the target base station; Process the traffic features of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic feature.

5. The base station traffic prediction method according to claim 4, wherein The graph attention network model also includes a multi-head attention mechanism; processing the traffic features of the target base station and the neighboring base stations according to the similarity coefficient to obtain the target traffic feature includes: Determine multiple groups of the similarity coefficients through the multi-head attention mechanism. Wherein, each attention head of the multi-head attention mechanism independently determines a group of the similarity coefficients; Process the traffic features of the target base station and the neighboring base stations according to each group of the similarity coefficients respectively to obtain intermediate traffic features; Concatenate the intermediate traffic features corresponding to each group of the similarity coefficients to obtain the target traffic feature; or take the average of the intermediate traffic features corresponding to each group of the similarity coefficients to obtain the target traffic feature.

6. The base station traffic prediction method according to claim 1, characterized in that The base station historical traffic data includes at least one of the following: the traffic periodic change information of the base station, the location information of the base station, the coverage area information of the base station, the user distribution information of the base station, weather information, and holiday information.

7. The base station traffic prediction method according to claim 1, wherein The attributes of the node include at least one of the following: the location information of the base station corresponding to the node, the coverage area information of the base station corresponding to the node, and the user distribution information of the base station corresponding to the node.

8. A base station traffic prediction device, characterized in that, Comprising: A first processing module, configured to obtain base station historical traffic data and construct a graph structure based on the base station historical traffic data, where the nodes in the graph structure are base stations, and the edges in the graph structure are used to reflect the association relationship between base stations; A second processing module, configured to determine a feature matrix according to the graph structure, where the elements in the feature matrix are the traffic values of the nodes at a preset moment; A third processing module, configured to process the feature matrix through a graph attention network model to obtain a target traffic feature, where the target traffic feature is time series data; A fourth processing module, configured to analyze the target traffic feature through a long short-term memory network model to obtain a traffic prediction result.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores a program, where when the program runs, it controls the device where the non-volatile storage medium is located to execute the base station traffic prediction method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Comprising: A memory and a processor, where the processor is configured to run the program stored in the memory, and when the program runs, it executes the base station traffic prediction method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, Including a computer program, where when the computer program is executed by a processor, it implements the base station traffic prediction method according to any one of claims 1 to 7.

Images