Method and apparatus for air base station deployment based on traffic prediction

Abstract

The application provides an air base station deployment method and device based on traffic prediction, which comprises the following steps: preprocessing historical traffic data obtained in advance to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and the periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on the proximity traffic data and time features extracted based on the proximity traffic data and the periodic traffic data to perform traffic prediction; the traffic prediction model is obtained by training training sample data and corresponding traffic labels; and the air base station is deployed according to the traffic prediction result. The spatial and time sequence features of the proximity traffic data and the periodic traffic data after preprocessing are extracted to predict potential traffic hotspots in advance, so that the air base station can be deployed on demand on the basis of a sudden communication request.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and apparatus for deploying airborne base stations based on traffic prediction. Background Technology

[0002] With the rapid increase in user demand for communication services, traditional ground base stations are unable to meet the needs of cellular communication, potentially leading to bottlenecks. Unmanned aerial vehicles (UAVs), which are aircraft capable of autonomous flight through remote control, offer numerous potential applications in wireless communication systems due to their mobility and affordability. UAVs carrying mobile base stations can be deployed as airborne base stations to provide wireless network services to areas experiencing temporary hotspots. Applying airborne base stations in traditional commercial cellular networks offers several advantages, such as eliminating the need for pre-installed equipment and site rental costs, reducing base station deployment costs, and increasing the flexibility of the cellular network, enabling it to quickly provide flexible and dynamic services based on specific service scenarios. Deploying UAVs as airborne base stations in areas with high service volume can meet the communication needs of hotspots experiencing temporary surges in traffic, thereby providing better mobile communication services.

[0003] To avoid wasting network resources during periods of low traffic, network operators need to predict potential traffic hotspots in advance. Furthermore, to meet sudden surges in communication requests, the on-demand deployment of airborne base stations requires them to constantly change locations. However, most current traffic prediction methods abstract the cellular traffic prediction problem into a time series analysis and solve it, lacking the ability to grasp the spatial relationships and dynamic characteristics of traffic. Summary of the Invention

[0004] This invention provides a method and apparatus for deploying airborne base stations based on traffic prediction, which addresses the shortcomings of existing terrestrial base stations in meeting the communication needs of hotspot areas with temporary traffic surges. It predicts spatial and temporal traffic within a target time period based on the spatial and temporal characteristics of data, improving the accuracy of traffic prediction and enabling rapid on-demand planning of airborne base station deployment locations to provide timely mobile communication services.

[0005] This invention provides a method for deploying airborne base stations based on traffic prediction, comprising: preprocessing pre-acquired historical traffic data to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein, the traffic prediction model uses spatial features extracted based on the proximity traffic data and temporal features extracted based on the proximity traffic data and the periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and its corresponding traffic labels; and deploying airborne base stations according to the traffic prediction result.

[0006] According to the present invention, a method for deploying an airborne base station based on traffic prediction is provided. The traffic prediction model includes: a spatial prediction block, which extracts spatial features from input proximity traffic data and predicts them to obtain spatially predicted traffic; a temporal prediction block, which extracts temporal features from the proximity traffic data and the periodic traffic data, respectively, predicts them, and fuses them to obtain temporally predicted traffic; and a traffic fusion layer, which fuses the spatially predicted traffic and the temporally predicted traffic to obtain a traffic prediction result.

[0007] According to a method for deploying an airborne base station based on traffic prediction provided by the present invention, the step of extracting spatial features from input proximity traffic data and predicting spatially predicted traffic includes: sorting all grids related to the current grid in descending order based on pre-acquired grid relevance; selecting grids from all grids in descending order based on a preset number, and concatenating the proximity traffic data corresponding to the selected grids into a sequence; extracting spatial features from the sequence, and predicting traffic based on the extracted spatial features and pre-acquired attention.

[0008] According to a traffic prediction-based airborne base station deployment method provided by the present invention, the time prediction block includes: a proximity converter, which extracts time features from the proximity traffic data and performs traffic prediction on the extracted time features to obtain proximity time-predicted traffic; a periodic converter, which extracts time features from the periodic traffic data and performs traffic prediction on the extracted time features to obtain periodic time-predicted traffic; and a time fusion converter, which mixes the proximity time-predicted traffic and the periodic time-predicted traffic to generate time-predicted traffic.

[0009] According to the present invention, a method for deploying airborne base stations based on traffic prediction further includes, before extracting time features from the proximity traffic data, performing data augmentation on the proximity traffic data.

[0010] According to a traffic prediction-based airborne base station deployment method provided by the present invention, after inputting the proximity traffic data and periodic traffic data into a traffic prediction model, the method includes: generating corresponding embedded elements for each element in the input proximity traffic data and periodic traffic data; and mapping the embedded elements corresponding to each element to a preset tuple to obtain corresponding attention.

[0011] According to a method for deploying an airborne base station based on traffic prediction provided by the present invention, the preprocessing of pre-acquired historical traffic data includes: acquiring historical traffic data, the historical traffic data including a target area and historical traffic within the target area; dividing the target area into grids based on a preset size, and merging the historical traffic within each grid based on a first preset time to obtain merged traffic for each grid; summarizing the merged traffic for each grid based on a second preset time to obtain summarized data for the corresponding grid; acquiring summarized data of other grids within a preset range of the current grid to obtain proximity traffic data; and acquiring summarized data of each grid within a preset period to obtain periodic traffic data.

[0012] The present invention also provides an airborne base station deployment device based on traffic prediction, comprising: a data acquisition module for preprocessing pre-acquired historical traffic data to obtain proximity traffic data and periodic traffic data; a traffic prediction module for inputting the proximity traffic data and periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein, the traffic prediction model uses spatial features extracted based on the proximity traffic data and temporal features extracted based on the proximity traffic data and the periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and its corresponding traffic labels; and a base station deployment module for deploying airborne base stations according to the traffic prediction result.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described traffic prediction-based airborne base station deployment method.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described traffic prediction-based airborne base station deployment method.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described traffic prediction-based airborne base station deployment method.

[0016] The present invention provides a method and apparatus for deploying airborne base stations based on traffic prediction. By extracting the spatial and temporal series features of preprocessed proximity traffic data and periodic traffic data, it can predict potential traffic hotspots in advance from both temporal and spatial perspectives. This ensures that the location of airborne base stations can be continuously adjusted based on sudden communication requests, enabling on-demand deployment, capacity enhancement, and providing on-demand and timely wireless services to ground users. By deploying airborne base stations based on traffic prediction, the initiative in deployment is enhanced, facilitating timely perception of system changes and adjustment of the deployment location of airborne base stations based on predicted spatial and temporal traffic over a future period, thereby achieving on-demand capacity enhancement. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the airborne base station deployment method based on traffic prediction provided by the present invention.

[0019] Figure 2 This is a schematic diagram comparing the predicted and actual values ​​of spatiotemporal flow data using the historical average (HA) method and the spatiotemporal-Transformer method provided by this invention.

[0020] Figure 3 This is a schematic diagram showing the average deployment error of the prediction-based method and the historical data-based method at different times t in the last 100 periods provided by the present invention;

[0021] Figure 4 This is a schematic diagram of the structure of the airborne base station deployment device based on traffic prediction provided by the present invention;

[0022] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Figure 1 This invention illustrates a flowchart of an airborne base station deployment method based on traffic prediction, the method comprising:

[0025] S11, preprocess the previously acquired historical traffic data to obtain proximity traffic data and periodic traffic data;

[0026] S12, input the proximity traffic data and periodic traffic data into the traffic prediction model to obtain the traffic prediction result output by the traffic prediction model; wherein, the traffic prediction model uses the spatial features extracted based on the proximity traffic data and the temporal features extracted based on the proximity traffic data and the periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on the training sample data and its corresponding traffic labels.

[0027] S13, deploy the airborne base station based on the traffic prediction results.

[0028] It should be noted that S1N in this specification does not represent the order of deployment methods for airborne base stations based on traffic prediction. The following details will explain this in conjunction with... Figures 2-3 The present invention describes a method for deploying airborne base stations based on traffic prediction.

[0029] Step S11: Preprocess the previously acquired historical traffic data to obtain proximity traffic data and periodic traffic data.

[0030] In this embodiment, the pre-acquired historical traffic data is preprocessed, including: acquiring historical traffic data, which includes a target area and historical traffic within the target area; dividing the target area into grids based on a preset size, and merging the historical traffic within each grid based on a first preset time to obtain the merged traffic for each grid; summarizing the merged traffic for each grid based on a second preset time to obtain summary data; acquiring the summary data of other grids within a preset range of the current grid to obtain proximity traffic data; and acquiring the summary data of each grid within a preset period to obtain periodic traffic data.

[0031] It's worth noting that the acquired historical traffic data can serve as a metric for the level of interaction between users and the mobile network. For example, assuming a preset size of 100×100, a first preset time of 10 minutes, and a second preset time of 1 hour, the target area is divided into a 100×100 grid. Historical traffic within each grid is merged every 10 minutes, and then the merged traffic is aggregated hourly to obtain summary data. This summary data includes the grid index and time, represented as follows: t represents time, and i represents the grid index.

[0032] Step S12: Input the proximity traffic data and periodic traffic data into the traffic prediction model to obtain the traffic prediction result output by the traffic prediction model; wherein, the traffic prediction model uses spatial features extracted based on proximity traffic data and temporal features extracted based on proximity traffic data and periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and its corresponding traffic labels.

[0033] In this embodiment, the traffic prediction model includes:

[0034] SA1, the spatial prediction block, extracts spatial features from the input proximity flow data and makes predictions to obtain spatially predicted flow.

[0035] Specifically, spatial feature extraction and prediction are performed on the input proximity flow data to obtain spatially predicted flow. This includes: sorting all grids related to the current grid in descending order based on pre-acquired grid correlations; selecting grids from all grids in descending order based on a preset number, and concatenating the proximity flow data corresponding to the selected grids into a sequence; extracting spatial features from the sequence, and predicting flow based on the extracted spatial features and pre-acquired attention. It should be added that the acquired grid correlations include: calculating coefficients based on the Pearson correlation coefficient, and determining the correlation between each grid and other grids based on the calculated coefficients.

[0036] In an optional embodiment, the flow prediction model employs a Transformer model, and accordingly, the spatial prediction block is a spatial Transformer block (STB). It should be noted that for each grid, considering its relationship with all other grids would result in a fairly long sequence of grids; however, not all grids have a significant impact on the flow state of other grids. Therefore, including the grid itself, the model selects the K grids most relevant to the current grid, that is, ranking each grid in descending order of relevance to all other grids and selecting the top K most important grids. Subsequently, from X... c The selected traffic data in (t) are concatenated into a new sequence X. s (t) serves as the input to the space transformer.

[0037] SA2, the time prediction block, extracts, predicts, and fuses the time features of the proximity traffic data and the periodic traffic data to obtain the time-predicted traffic.

[0038] In this embodiment, the time prediction block includes: a proximity transformer, which extracts time features from proximity traffic data and predicts traffic based on the extracted time features to obtain proximity time prediction traffic; a periodic transformer, which extracts time features from periodic traffic data and predicts traffic based on the extracted time features to obtain periodic time prediction traffic; and a time fusion unit, which mixes proximity time prediction traffic and periodic time prediction traffic to generate time prediction traffic.

[0039] It should be noted that the traffic prediction model adopts the Transformer model, and correspondingly, the time prediction block is the Time Transformer Block (TTB). The space-time-Transformer method is used to predict space-time traffic to obtain the traffic prediction results. The time prediction block extracts the proximity time trend and periodic features of wireless traffic from proximity traffic data and periodic traffic data respectively, and makes predictions. With the help of the time fusion unit, the prediction information of different sequences is fused, thereby improving the accuracy of time traffic prediction.

[0040] In an optional embodiment, prior to extracting temporal features from the proximity traffic data, the method further includes data augmentation of the proximity traffic data. It should be added that, during the data augmentation process, the calculated correlation matrix, combined with pre-acquired grid correlations, selects the top Q relevant grids for each grid, and concatenates these data with the data from the grid itself to generate a source sequence.

[0041] SA3, the traffic fusion layer, fuses spatial and temporal predicted traffic to obtain traffic prediction results.

[0042] In an optional embodiment, before inputting the proximity traffic data and periodic traffic data into the traffic prediction model, the method further includes training the traffic prediction model, specifically including: obtaining training sample data and the traffic labels corresponding to the training sample data; using the training sample data as input data for training and the traffic labels corresponding to the training sample data as labels for training, training the model to be trained to obtain a traffic prediction model for predicting traffic.

[0043] It should be noted that obtaining training sample data and the corresponding traffic labels includes: obtaining historical training data and the corresponding traffic labels; preprocessing the historical training data to obtain training sample data, which includes proximity training traffic data and periodic training traffic data.

[0044] It should be noted that the network to be trained can be an existing network built into the training device. This existing network typically includes a network structure, or it can be other networks specified by the user, such as a neural network like the Transformer. The network to be trained typically includes a spatial prediction block for spatial flow prediction of the input proximity flow data, a temporal prediction block for temporal flow prediction of the input proximity flow data and periodic flow data, a flow fusion layer for fusing the predicted spatial and temporal flow, and a loss function. According to a preset iteration rule, the above training sample data is input into the model to be trained for training, resulting in the trained flow prediction model.

[0045] Specifically, training the model to be trained includes: inputting training sample data into the model to be trained to obtain the training traffic prediction results output by the model to be trained; constructing a loss function based on the training traffic prediction results and the traffic labels corresponding to the training sample data; and ending the training based on the convergence of the loss function.

[0046] Step S13: Deploy the airborne base station based on the traffic prediction results.

[0047] In one alternative embodiment, deploying an airborne base station based on traffic prediction results includes: adjusting the location of the airborne base station using a DQN reinforcement learning model based on the traffic prediction results to achieve on-demand capacity enhancement.

[0048] Specifically, each airborne base station can be viewed as a single agent. At the start of the task, the airborne base station selects an action according to the ∈-greedy policy. Then, the environment sends the next state and returns the reward to the agent. The agent uses the reward returned by the environment to update its knowledge and evaluate the previous action. The above steps are repeated until the airborne base station task ends.

[0049] It should be added that the state S = {traffic prediction value of the selected grid, location of the air base station}, where the location of each air base station can be represented by (x j ,y j The action is the movable direction of the airborne base station, i.e., action A = {forward, backward, left, right, hover}; the reward is the total traffic on the grid where the airborne base station is located.

[0050] In one optional embodiment, the main process of the DQN algorithm includes:

[0051] Step 1: First, initialize the experience replay pool D with a capacity of N;

[0052] Step 2: Initialize the Q-network and its neural network parameters ω; Initialize the target Q-network and its neural network parameters ω. -;

[0053] Step 3: Iterate through the rounds episode = 1, 2, ..., M:

[0054] Step 3.1 Initialize the state set S;

[0055] Step 3.2 Iterate through step = 1, 2, ..., T:

[0056] Step 3.2.1 Use the ∈-greedy strategy to take action strategy A, A = {forward, backward, left, right, hover};

[0057] Step 3.2.2 Execute action A, calculate the cost reward R of the system taking action A in state S, and the system will reach the new base station planning state S' in the next time step to obtain the reward and the new state S';

[0058] Step 3.2.3 Store the sample (S,A,R,S') into the experience replay pool D;

[0059] Step 3.2.4 Calculate the target Q value using the uniformly randomly sampled Minibatch from the empirical replay pool. i =R + γ·max A Q(S′,A;ω - Update the Q network parameters ω to reduce the loss function [y]. i -Q(S,A;ω)] 2 ;

[0060] Step 3.2.5 Update the parameters of the base station's planned target Q network every C steps, i.e., ω. - =ω.

[0061] Step 4 algorithm ends.

[0062] In one optional embodiment, the system model and DQN simulation parameters are set as shown in the table below:

[0063] parameter value Proximity data length 3 Periodic data length 3 Traffic prediction model learning rate 0.001 Batch size of traffic prediction model 4 Training set to validation set ratio 9:1 DQN model learning rate 0.1 Discount factor 0.9 Greed factor 0.9 Batch size of DQN model 128 Experience pool capacity 2000 Maximum step size of random walk 200

[0064] In one alternative embodiment, reference Figure 2 The diagram comparing the predicted and actual values ​​of space-time traffic data using the historical average (HA) method and the space-time-Transformer method shows that the space-time traffic prediction based on the space-time-Transformer model performs well. Therefore, the infinite traffic prediction of this model will be effective for the deployment of airborne base stations.

[0065] Additionally, refer to Figure 3The diagram illustrates the average deployment error of the prediction-based method and the historical data-based method over the last 100 periods at different times t. It can be seen that the prediction-based aerial base station deployment method is more effective.

[0066] In summary, the embodiments of the present invention extract the spatial and temporal series features of preprocessed proximity traffic data and periodic traffic data to predict potential traffic hotspots in advance from both temporal and spatial perspectives. This ensures that the location of airborne base stations can be continuously adjusted based on sudden communication requests for on-demand deployment, thereby enhancing capacity and providing on-demand and timely wireless services to ground users. By deploying airborne base stations based on traffic prediction, the initiative in deployment is enhanced, facilitating timely perception of system changes and adjustment of airborne base station deployment locations based on predicted spatial and temporal traffic over a future period, thus achieving on-demand capacity enhancement.

[0067] The air base station deployment apparatus based on traffic prediction provided by the present invention will be described below. The air base station deployment apparatus based on traffic prediction described below can be referred to in correspondence with the air base station deployment method based on traffic prediction described above.

[0068] Figure 4 A schematic diagram of a traffic prediction-based airborne base station deployment device is shown. The device includes:

[0069] The data acquisition module 41 preprocesses the previously acquired historical traffic data to obtain proximity traffic data and periodic traffic data;

[0070] The traffic prediction module 42 inputs proximity traffic data and periodic traffic data into the traffic prediction model to obtain the traffic prediction result output by the traffic prediction model. The traffic prediction model uses spatial features extracted based on proximity traffic data and temporal features extracted based on proximity traffic data and periodic traffic data to perform traffic prediction. The traffic prediction model is trained based on training sample data and its corresponding traffic labels.

[0071] The base station deployment module 43 deploys the airborne base station based on the traffic prediction results.

[0072] In this embodiment, the data acquisition module 41 includes: a historical data acquisition unit for acquiring historical traffic data, which includes a target area and historical traffic within the target area; a region division unit for dividing the target area into grids based on a preset size, and merging the historical traffic within each grid based on a first preset time to obtain the merged traffic for each corresponding grid; a data aggregation unit for aggregating the merged traffic for each corresponding grid based on a second preset time to obtain aggregated data; and a traffic data acquisition unit for acquiring aggregated data of other grids within a preset range of the current grid to obtain proximity traffic data, and acquiring aggregated data of each of the grids within a preset period to obtain periodic traffic data.

[0073] The traffic prediction module 42 includes: a data input unit that inputs adjacent traffic data and periodic traffic data into the traffic prediction model; a traffic prediction unit that uses the traffic prediction model to predict traffic based on spatial features extracted from adjacent traffic data and temporal features extracted from adjacent traffic data and periodic traffic data, and obtains traffic prediction results; and a data output model that outputs the traffic prediction results obtained by the traffic prediction model.

[0074] The traffic flow prediction model includes: a spatial prediction block, which extracts spatial features from the input proximity traffic data and predicts the spatial predicted traffic flow; and a temporal prediction block, which extracts temporal features from the proximity traffic data and the periodic traffic data, predicts the temporal predicted traffic flow, and fuses them.

[0075] Furthermore, the spatial prediction block includes: a sorting unit, which sorts all grids related to the current grid from largest to smallest based on pre-acquired grid relevance; a data processing unit, which selects grids from all grids from largest to smallest based on a preset quantity, and concatenates the proximity flow data corresponding to the selected grids into a sequence; and a spatial flow prediction unit, which extracts spatial features from the sequence and performs flow prediction based on the extracted spatial features and pre-acquired attention.

[0076] The time prediction block includes: a proximity transformer, which extracts time features from proximity traffic data and predicts traffic based on the extracted time features to obtain proximity time prediction traffic; a periodic transformer, which extracts time features from periodic traffic data and predicts traffic based on the extracted time features to obtain periodic time prediction traffic; and a time fusion unit, which mixes proximity time prediction traffic and periodic time prediction traffic to generate time prediction traffic.

[0077] In one alternative embodiment, the time prediction block includes: a data augmentation unit that augments the proximity traffic data before extracting time features from the proximity traffic data.

[0078] In an optional embodiment, the apparatus further includes a training module for training the traffic prediction model before inputting the proximity traffic data and periodic traffic data into the traffic prediction model. Specifically, the training module includes: a training data acquisition unit for acquiring training sample data and the traffic labels corresponding to the training sample data; and a training unit for using the training sample data as input data for training and the traffic labels corresponding to the training sample data as labels for training, to train the model to be trained and obtain a traffic prediction model for predicting traffic.

[0079] Furthermore, the training data acquisition unit includes: a training data acquisition subunit for acquiring historical training data and the traffic labels corresponding to the historical training data; and a preprocessing subunit for preprocessing the historical training data to obtain training sample data, which includes proximity training traffic data and periodic training traffic data.

[0080] Furthermore, the training unit includes: a traffic prediction subunit, which inputs training sample data into the model to be trained to obtain the training traffic prediction result output by the model to be trained; and a training subunit, which constructs a loss function based on the training traffic prediction result and the traffic labels corresponding to the training sample data, and terminates training based on the convergence of the loss function.

[0081] The base station deployment module 43 includes: a base station deployment unit that, based on traffic prediction results, uses a DQN reinforcement learning model to adjust the location of the airborne base station in order to achieve the task of on-demand capacity enhancement.

[0082] Specifically, each airborne base station can be viewed as a single agent. At the start of the task, the airborne base station selects an action according to the ∈-greedy policy. Then, the environment sends the next state and returns the reward to the agent. The agent uses the reward returned by the environment to update its knowledge and evaluate the previous action. The above steps are repeated until the airborne base station task ends.

[0083] In summary, this embodiment of the invention extracts the spatial and temporal series characteristics of proximity traffic data and periodic traffic data preprocessed by the data acquisition module through the traffic prediction module. This allows for the prediction of potential traffic hotspots from both temporal and spatial perspectives. This ensures that the location of airborne base stations can be continuously adjusted based on sudden communication requests, enabling on-demand deployment, capacity enhancement, and the provision of on-demand and timely wireless services to ground users. Furthermore, by deploying airborne base stations based on traffic prediction, the initiative in deployment is enhanced. This facilitates timely detection of system changes and adjustment of airborne base station deployment locations based on predicted spatial and temporal traffic over a future period, achieving on-demand capacity enhancement.

[0084] Figure 5An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 51, a communications interface 52, a memory 53, and a communication bus 54, wherein the processor 51, communications interface 52, and memory 53 communicate with each other via the communication bus 54. The processor 51 can call logical instructions in the memory 53 to execute a method for deploying airborne base stations based on traffic prediction. This method includes: preprocessing pre-acquired historical traffic data to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and periodic traffic data into a traffic prediction model to obtain the traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on proximity traffic data and temporal features extracted based on proximity traffic data and periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and their corresponding traffic labels; and deploying airborne base stations based on the traffic prediction result.

[0085] Furthermore, the logical instructions in the aforementioned memory 53 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the 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 cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the airborne base station deployment method based on traffic prediction provided by the above methods. The method includes: preprocessing pre-acquired historical traffic data to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on proximity traffic data and temporal features extracted based on proximity traffic data and periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and its corresponding traffic labels; and deploying airborne base stations based on the traffic prediction results.

[0087] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, performs the above-described method for deploying an airborne base station based on traffic prediction. This method includes: preprocessing previously acquired historical traffic data to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on proximity traffic data and temporal features extracted based on proximity traffic data and periodic traffic data to perform traffic prediction; the traffic prediction model is trained based on training sample data and its corresponding traffic labels; and deploying an airborne base station based on the traffic prediction result.

[0088] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for air base station deployment based on traffic prediction, the method comprising: The method comprises: ​ preprocessing the pre-acquired historical traffic data to obtain proximity traffic data and periodic traffic data; inputting the proximity traffic data and the periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on the proximity traffic data, time features extracted based on the proximity traffic data and the periodic traffic data to perform traffic prediction; the traffic prediction model is trained according to training sample data and corresponding traffic labels; deploying aerial base stations according to the traffic prediction result; wherein the traffic prediction model comprises: a spatial prediction block that extracts spatial features from input proximity traffic data and performs prediction to obtain spatial prediction traffic; a time prediction block that extracts time features from the proximity traffic data and the periodic traffic data respectively, performs prediction and fusion to obtain time prediction traffic; a traffic fusion layer that fuses the spatial prediction traffic and the time prediction traffic to obtain a traffic prediction result; wherein the spatial feature extraction and prediction from the input proximity traffic data to obtain the spatial prediction traffic comprises: based on pre-acquired grid correlation, sorting all grids related to the current grid from large to small; based on a preset number, selecting grids from all grids from large to small, and concatenating the proximity traffic data corresponding to the selected grids into a sequence; extracting spatial features from the sequence, and performing traffic prediction based on the extracted spatial features and pre-acquired attention; wherein the acquired grid correlation comprises: calculating a coefficient based on the Pearson correlation coefficient, and determining the correlation of each grid with other grids according to the calculated coefficient; after inputting the proximity traffic data and the periodic traffic data into the traffic prediction model, comprising: for each element in the input proximity traffic data and periodic traffic data, generating a corresponding embedded element; mapping the embedded element corresponding to each element to a preset tuple to obtain the corresponding attention. 2.The method of claim 1, wherein, The time prediction block comprises: a proximity transformer that extracts time features from the proximity traffic data and performs traffic prediction on the extracted time features to obtain proximity time prediction traffic; a periodic transformer that extracts time features from the periodic traffic data and performs traffic prediction on the extracted time features to obtain periodic time prediction traffic; a time fusion device that mixes the proximity time prediction traffic and the periodic time prediction traffic to generate time prediction traffic. 3.The method of claim 2, wherein, Before extracting time features from the proximity traffic data, it further comprises data enhancement on the proximity traffic data. 4.The method of claim 1, wherein, The preprocessing of the pre-acquired historical traffic data comprises: acquiring historical traffic data, the historical traffic data comprising a target area and historical traffic in the target area; dividing the target area into grids based on a preset size, and merging the historical traffic in each grid based on a first preset time to obtain merged traffic corresponding to each grid; based on the second preset time, data of the merged traffic corresponding to each of the grids is aggregated to obtain aggregated data of the corresponding grid; obtaining aggregated data of other grids within a preset range of the current grid to obtain proximity traffic data, and obtaining aggregated data of each of the grids within a preset period to obtain periodic traffic data.

5. An apparatus for air base station deployment based on traffic prediction, the apparatus comprising: a processor configured to: determine a traffic prediction for a geographic area; and determine a location for a base station based on the traffic prediction. comprising: a data acquisition module, which pre-processes historical traffic data obtained in advance to obtain proximity traffic data and periodic traffic data; a traffic prediction module, which inputs the proximity traffic data and the periodic traffic data into a traffic prediction model to obtain a traffic prediction result output by the traffic prediction model; wherein the traffic prediction model uses spatial features extracted based on the proximity traffic data, and time features extracted based on the proximity traffic data and the periodic traffic data, to perform traffic prediction; the traffic prediction model is trained according to training sample data and corresponding traffic labels; a base station deployment module, which deploys aerial base stations according to the traffic prediction result; wherein the traffic prediction model comprises: a spatial prediction block, which extracts spatial features from the input proximity traffic data and performs prediction to obtain spatial prediction traffic; a time prediction block, which extracts time features from the proximity traffic data and the periodic traffic data respectively, performs prediction and fusion, and obtains time prediction traffic; a traffic fusion layer, which fuses the spatial prediction traffic and the time prediction traffic to obtain a traffic prediction result; wherein the extraction of spatial features from the input proximity traffic data and the prediction to obtain spatial prediction traffic comprises: based on pre-acquired grid correlation, all grids related to the current grid are sorted from large to small; based on a preset number, grids are selected from all grids from large to small, and the proximity traffic data corresponding to the selected grids are concatenated into a sequence; spatial features are extracted from the sequence, and traffic prediction is performed based on the extracted spatial features and pre-acquired attention; wherein the acquired grid correlation comprises: calculating a coefficient based on a Pearson correlation coefficient, and determining the correlation of each grid with other grids according to the calculated coefficient; after the proximity traffic data and the periodic traffic data are input into the traffic prediction model, comprising: for each element in the input proximity traffic data and periodic traffic data, generating a corresponding embedded element; mapping the embedded element corresponding to each of the elements to a preset tuple to obtain corresponding attention.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the method for deploying aerial base stations based on traffic prediction according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method for deploying aerial base stations based on traffic prediction according to any one of claims 1 to 4.

Images