Spectrum coverage map construction method and device based on heterogeneous graph neural network
Through the spectrum coverage map construction method based on heterogeneous graph neural network, network coverage prediction is performed using base station communication parameters and grid position data, the problems of low prediction accuracy and limited application range in the prior art are solved, and a higher Computational accuracy and a wider range of applications.
Patent Information
- Application Number
- CN202411124907.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-16
AI Technical Summary
The prior art has limited accuracy in network coverage prediction, and it is impossible to predict network coverage without road test areas. It relies on road test data, which limits the scope of the prediction application.
Using the spectrum coverage map construction method based on heterogeneous graph neural network, the spectrum coverage map is constructed by obtaining base station communication parameters and raster position data, generating an adjacency matrix, inputting a pre-trained graph neural network model for numerical prediction, and constructing a spectrum coverage map without relying on road measurement. data.
It improves the calculation accuracy and application range of network coverage prediction, can accurately predict in untracked areas, and enhances the intelligence level of network optimization.
Smart Images

Figure CN118678377B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of communication technology, and in particular to a method and device for constructing a spectrum coverage map based on a heterogeneous graph neural network. Background Art
[0002] Intelligent network optimization refers to the simultaneous adjustment and optimization of large-scale base station antennas to ensure that the optimized network has high performance. The core of intelligent network optimization is to intelligently adjust the parameters of large-scale base stations and predict the network coverage after parameter adjustment. The network coverage prediction is an important feedback on the performance of parameter adjustment.
[0003] In related technologies, the process of network coverage prediction first involves road testing of the grids in part of the target area to obtain the measurement data corresponding to the Reference Signal Receiving Power (RSRP) in the beam space, so as to obtain the spectrum coverage based on the measurement data. Then, the localized statistical channel modeling technology is used to construct the statistical parameters of the Angular Power Spectrum (APS) of each transmission path in three-dimensional space. When the engineering parameters such as the downtilt angle and azimuth angle of the base station antenna change, the RSRP after the antenna parameters are adjusted can be predicted by multiplying the inferred APS with the modified antenna coefficient matrix. However, this method has limited prediction accuracy on the one hand, and cannot predict network coverage in areas that have not been road tested on the other. Summary of the invention
[0004] The main purpose of the embodiments of the present application is to propose a method and device for constructing a spectrum coverage map based on a heterogeneous graph neural network, to improve the calculation accuracy of network coverage prediction, and to increase the scope of network coverage prediction.
[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a spectrum coverage map construction method based on a heterogeneous graph neural network, comprising: obtaining base station communication parameters of a target base station and grid position data of a plurality of target grids;
[0006] Connecting the target base station and the target grid, connecting different target grids located within a preset distance based on the grid position data, and generating an adjacency matrix according to the connection relationship;
[0007] Obtaining base station node characteristics according to the base station communication parameters, obtaining grid node characteristics according to the grid position data, and obtaining communication node characteristics according to the base station node characteristics and the grid node characteristics;
[0008] The communication node features and the adjacency matrix are input into a pre-trained graph neural network model for numerical prediction to obtain the reference signal predicted received power of the beam space of the target base station on each of the target grids, and a spectrum coverage map is constructed based on the reference signal predicted received power.
[0009] In some embodiments, obtaining grid node features according to the grid position data includes:
[0010] If the angular power spectrum of the target grid is obtained, the grid node features are obtained according to the grid position data and the angular power spectrum;
[0011] Otherwise, grid node features are obtained according to the grid position data.
[0012] In some embodiments, the training process of the graph neural network model includes the following steps:
[0013] Acquire test data, grid position parameters, and base station communication sample data of each test grid in the test area, and acquire the maximum beam reference signal received power corresponding to the main serving base station of the test grid from the test data as a sample label;
[0014] Determine the relevant grid of each of the test base stations according to the base station number in the base station communication sample data and the base station identifier in the test data, determine the adjacent grid of the relevant grid according to the grid position parameter, and generate a connection relationship diagram of the test base station, the relevant grid and the adjacent grid;
[0015] Obtaining node feature data according to the base station communication sample data and the grid position parameters;
[0016] Inputting the node feature data and the connection relationship graph into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid;
[0017] The loss value is calculated according to the sample prediction data and the sample label, and the graph neural network model is adjusted based on the loss value until the iteration termination condition is reached to obtain the trained graph neural network model.
[0018] In some embodiments, obtaining the maximum beam reference signal received power corresponding to the primary serving base station of the test grid from the test data as a sample label includes:
[0019] Acquire the signal strength of at least one demodulation reference signal from the test data, and select the base station with the largest signal strength as the primary serving base station;
[0020] All beam reference signal received powers of the primary serving base station are selected from the test data, and a maximum value of the beam reference signal received powers is selected as the maximum beam reference signal received power.
[0021] In some embodiments, obtaining node feature data according to the base station communication sample data and the grid position parameters includes:
[0022] Obtaining base station sample characteristics according to the base station communication sample data;
[0023] Calculate the angular power spectrum estimation value of the test grid according to the test data and the base station communication sample data of the corresponding primary serving base station;
[0024] If the angle power spectrum estimation value is valid, obtaining the grid sample feature according to the angle power spectrum estimation value and the grid position parameter; otherwise, obtaining the grid sample feature according to the grid position parameter;
[0025] The node feature data is obtained according to the base station sample features and the grid sample features.
[0026] In some embodiments, the calculating the angular power spectrum estimation value of the test grid according to the test data and the base station communication sample data of the corresponding primary serving base station includes:
[0027] Obtaining an antenna coefficient matrix of the primary serving base station according to the base station communication sample data;
[0028] Acquire all beam reference signal received powers of the primary serving base station from the test data;
[0029] Compressed sensing calculation is performed according to the beam reference signal received power and the antenna coefficient matrix to obtain the angle power spectrum estimation value.
[0030] In some embodiments, the step of inputting the node feature data and the connection relationship graph into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid includes:
[0031] Acquire adjacent nodes with which each node has a connection relationship based on the connection relationship graph;
[0032] Calculating, using a graph attention layer, a feature weight of the node feature data of each node relative to the node feature data of the adjacent nodes;
[0033] The node feature data of the adjacent nodes are weighted based on the feature weights to obtain aggregate feature data corresponding to each node, and sample prediction data corresponding to the aggregate feature data is calculated using a fully connected layer.
[0034] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a spectrum coverage map construction device based on a heterogeneous graph neural network, comprising:
[0035] Data acquisition module: used to acquire base station communication parameters of a target base station and grid position data of multiple target grids;
[0036] A connection relationship generating module: used to connect the target base station and the target grid, connect different target grids located within a preset distance based on the grid position data, and generate an adjacency matrix according to the connection relationship;
[0037] Feature extraction module: used to obtain base station node features according to the base station communication parameters, obtain grid node features according to the grid position data, and obtain communication node features according to the base station node features and the grid node features;
[0038] Prediction module: used to input the communication node features and the adjacency matrix into a pre-trained graph neural network model for numerical prediction, obtain the reference signal predicted received power of the beam space of the target base station on each of the target grids, and construct a spectrum coverage map based on the reference signal predicted received power.
[0039] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.
[0040] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a storage medium, which is a storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.
[0041] The spectrum coverage map construction method and device based on heterogeneous graph neural network proposed in the embodiment of the present application obtains the base station communication parameters of the target base station and the grid position data of multiple target grids, connects the target base station and the target grid, connects different target grids located within a preset distance based on the grid position data, generates an adjacency matrix according to the connection relationship, obtains the base station node characteristics according to the base station communication parameters, obtains the grid node characteristics according to the grid position data, and obtains the communication node characteristics according to the base station node characteristics and the grid node characteristics. Finally, the communication node characteristics and the adjacency matrix are input into the pre-trained graph neural network model for numerical prediction, and the reference signal predicted received power of the beam space of the target base station on each target grid is obtained, and the spectrum coverage map is constructed based on the reference signal predicted received power. The embodiment of the present application trains the graph neural network model for data prediction, and obtains the corresponding node characteristics in combination with the communication parameters of the base station and the position data of the grid. Then, based on the connection relationship between the base station and the grid, the adjacency matrix is used to represent the correlation between the nodes. The node characteristics and the correlation are considered simultaneously in the prediction process. Compared with the method of directly multiplying the APS calculated in the road test process and the changed antenna coefficient matrix in the related art, this method improves the calculation accuracy. In addition, during the application process, there is no need to rely on road test data, and predictions can be made based on the location data of the target grid, thereby expanding the application scope of network coverage prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a flowchart of a method for constructing a spectrum coverage map based on a heterogeneous graph neural network provided in an embodiment of the present application.
[0043] Figure 2 It is a schematic diagram of the training process of the graph neural network model provided in an embodiment of the present application.
[0044] Figure 3 It is a schematic diagram of obtaining the maximum beam reference signal received power corresponding to the main serving base station of the test grid as a sample label from the test data provided by an embodiment of the present application.
[0045] Figure 4 It is a schematic diagram of obtaining node feature data based on base station communication sample data and grid position parameters provided in an embodiment of the present application.
[0046] Figure 5 It is a flowchart for calculating the angular power spectrum estimate of the test grid provided in an embodiment of the present application.
[0047] Figure 6 It is a schematic diagram of the connection relationship diagram provided in the embodiment of the present application.
[0048] Figure 7 It is a schematic diagram of numerical prediction using the graph neural network model provided in an embodiment of the present application.
[0049] Figure 8 It is a flowchart provided by an embodiment of the present application for inputting node feature data and a connection relationship diagram into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid.
[0050] Fig. 9 This is a schematic diagram of the performance of the graph neural network model provided in an embodiment of the present application.
[0051] Fig.10 It is a schematic diagram of the influence of the connection relationship between the test grids provided in the embodiment of the present application on the prediction result.
[0052] Fig.11 It is a schematic diagram of the accuracy of prediction results obtained by training using angular power spectrum estimation values provided in an embodiment of the present application.
[0053] Fig.12 It is a schematic diagram of the mean absolute percentage error in the test set provided by an embodiment of the present application.
[0054] Fig.13 This is a structural block diagram of a spectrum coverage map construction device based on a heterogeneous graph neural network provided in another embodiment of the present application.
[0055] Fig.14 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION
[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0057] It should be noted that although the functional modules are divided in the device schematic and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart.
[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0059] First, some nouns involved in this application are analyzed:
[0060] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0061] Graph Neural Networks (GNNs) is a deep learning method specifically designed for processing graph data. It learns the representation of nodes, edges, and the entire graph by transferring and aggregating information between the nodes of the graph. GNNs can effectively capture the complex patterns and relationships of the graph. The core of GNNs is to use the structural information of the graph and iteratively update the node representation to ultimately achieve in-depth understanding and prediction of graph data.
[0062] Angular Power Spectrum (APS) is a statistic that describes the energy distribution of a signal or data in the spatial angle domain. The angular power spectrum reflects the power distribution of the signal in different directions. By calculating the angular power spectrum, the spatial features of the signal can be effectively extracted, and it is widely used in tasks such as target detection, signal separation, and direction estimation.
[0063] Reference Signal Received Power (RSRP): is a key indicator for measuring downlink signal strength in communication networks. It indicates the received power level of the reference signal sent by the base station at the user equipment (UE). This value is measured in dBm and usually ranges from -140dBm to -44dBm, where -140dBm indicates a very weak signal and -44dBm indicates a very strong signal. Accurate measurement of RSRP is of great significance for optimizing network coverage, improving signal quality and ensuring communication quality. It is one of the important parameters for network performance evaluation and troubleshooting.
[0064] In recent years, due to the rapid development of mobile communication technologies such as 5G and 6G and the increase in the scale of base stations, intelligent network optimization of large-scale base stations has become urgent. Traditional network optimization relies on manual adjustment of base station antennas and subsequent coverage test iterations, which consumes a lot of manpower costs. Intelligent network optimization refers to the simultaneous adjustment and optimization of large-scale base station antennas to ensure that the optimized network has high performance. The core of intelligent network optimization is to intelligently adjust the parameters of large-scale base stations and predict the network coverage after parameter adjustment. The network coverage prediction is an important feedback on the performance of parameter adjustment.
[0065] In the related art, the process of network coverage prediction first involves road testing of the grid in part of the target area to obtain the measurement data corresponding to the reference signal received power RSRP in the beam space. Then, the statistical parameters of the angular power spectrum APS of each transmission path are constructed in three-dimensional space using localized statistical channel modeling technology. When the engineering parameters such as the downtilt angle and azimuth angle of the base station antenna change, the RSRP after the antenna parameters are adjusted can be predicted by multiplying the inferred APS with the modified antenna coefficient matrix. However, on the one hand, when the beam space information is limited and the channel structure is relatively complex, the prediction results of the prediction method based on the localized statistical channel model are limited in accuracy, and on the other hand, it is impossible to predict the network coverage of areas that have not been road tested.
[0066] Based on this, an embodiment of the present application provides a method and device for constructing a spectrum coverage map based on a heterogeneous graph neural network, trains a graph neural network model to perform data prediction, and combines the communication parameters of the base station and the position data of the grid to obtain the corresponding node features. Then, based on the connection relationship between the base station and the grid, an adjacency matrix is used to represent the correlation between the nodes. Both node characteristics and correlation are considered in the prediction process. Compared with the method in the related art of directly multiplying the angular power spectrum calculated during the road test and the changed antenna coefficient matrix, this method improves the calculation accuracy. In addition, during the application process, there is no need to rely on road test data, and predictions can be made based on the location data of the target grid, thereby expanding the application scope of network coverage prediction.
[0067] The embodiments of the present application provide a method and device for constructing a spectrum coverage map based on a heterogeneous graph neural network, which is specifically illustrated by the following embodiments. First, the method for constructing a spectrum coverage map based on a heterogeneous graph neural network in the embodiments of the present application is described.
[0068] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.
[0069] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0070] The spectrum coverage map construction method based on heterogeneous graph neural network provided in the embodiment of the present application relates to the field of communication technology. The spectrum coverage map construction method based on heterogeneous graph neural network provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be a computer program running in the terminal or the server side. For example, the computer program can be a native program or software module in the operating system; it can be a local (Native) application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as a client that supports communication data prediction, or it can be a small program, that is, a program that can be run only after being downloaded to the browser environment; it can also be a small program that can be embedded in any APP. In short, the above-mentioned computer program can be any form of application, module or plug-in. Among them, the terminal communicates with the server through the network. The spectrum coverage map construction method based on heterogeneous graph neural network can be executed by the terminal or the server, or by the terminal and the server in collaboration.
[0071] In some embodiments, the terminal may be a smart phone, tablet computer, laptop computer, desktop computer or smart watch, etc. The server may be an independent server, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms; or a service node in a blockchain system, where each service node in the blockchain system forms a peer-to-peer (P2P, PeerTo Peer, P2P) network, and the P2P protocol is an application layer protocol running on the Transmission Control Protocol (TCP) protocol. The terminal and the server may be connected via Bluetooth, Universal Serial Bus (USB) or network communication connection methods, which are not limited in this embodiment.
[0072] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0073] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with the relevant laws, regulations, and standards of the relevant countries and regions. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.
[0074] The following describes a method for constructing a spectrum coverage map based on a heterogeneous graph neural network in an embodiment of the present application.
[0075] Figure 1 is an optional flowchart of a spectrum coverage map construction method based on a heterogeneous graph neural network provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps 110 to 130. It can also be understood that this embodiment Figure 1 The order of step 110 to step 130 is not specifically limited, and the order of steps can be adjusted or some steps can be reduced or added according to actual needs.
[0076] Step 110: Acquire base station communication parameters of the target base station and grid position data of multiple target grids.
[0077] In one embodiment, the target area to be tested is divided into multiple target grids of the same or different sizes, and the target grids are used to represent the target area. For example, the target area is divided into target grids of 10 meters by 10 meters. It is understandable that the size of the target grid can be set according to actual needs.
[0078] In one embodiment, a point can be randomly selected within the position range of the target grid to represent the position of the target grid. The selected point can be a center point or a vertex of the target grid, etc. This embodiment does not limit the selection of the point. Thus, the grid position data of each target grid can be obtained.
[0079] Next, in one embodiment, one is selected from the base stations in the target area as the target base station, and the base station communication parameters include the base station location of the target base station and the engineering parameter data of the base station. For example, the engineering parameter data may be: the beamforming change matrix of the base station, the active antenna unit (AAU) category of the base station, the departure downtilt angle, the departure azimuth angle, etc. It can be understood that the embodiment of the present application does not limit the engineering parameter data, and may also include more data than the example. The more data, the greater the amount of information contained, and the more accurate the subsequent prediction. However, more data will also bring higher computational complexity and affect the prediction efficiency. Therefore, the required engineering parameter data can be selected according to actual needs.
[0080] Among them, the beamforming change matrix of the base station is a technology used to adjust the direction of the antenna beam, and controls the direction of the beam by changing the phase of the signal. The active antenna unit of the base station is the core component of the base station, responsible for sending and receiving signals. According to the application scenarios and requirements, the active antenna unit can be divided into many types, including active antenna units that support Massive MIMO technology, active antenna units for deep coverage, and active antenna units that support multiple frequency bands. The departure downtilt angle refers to the angle between the main lobe direction of the antenna and the vertical line, which is used to control the vertical coverage range of the signal. The departure azimuth angle refers to the horizontal angle between the main lobe direction of the antenna and the due north direction, which is used to control the horizontal coverage range of the signal.
[0081] Step 120: Connect the target base station and the target grid, connect different target grids within a preset distance based on the grid position data, and generate an adjacency matrix according to the connection relationship.
[0082] In one embodiment, both the target base station and the target grid are taken as nodes, and then the target base station and the target grid are connected. For example, when connecting, the target base station can be taken as the central node, and the target grid and the target base station can be connected in a star-shaped manner. Then, according to the grid position data of the target grid, the distance between the target grids is calculated, and the target grids whose distance is less than the preset distance are connected. Finally, the adjacency matrix is obtained based on the connection relationship between the target base station and the target grid and the connection relationship between the target grids. The dimension of the adjacency matrix is related to the total number of all nodes. For example, if the total number of nodes of the target base station and the target grid is n, the dimension of the adjacency matrix is n×n. In the adjacency matrix, if there is a connection relationship, the matrix value of the corresponding position is 1, otherwise it is 0. Therefore, according to the 0 or 1 of the element value in the adjacency matrix, the connection relationship between all nodes can be obtained.
[0083] Step 130: obtain base station node characteristics according to base station communication parameters, obtain grid node characteristics according to grid position data, and obtain communication node characteristics according to base station node characteristics and grid node characteristics.
[0084] Among them, the base station node characteristics corresponding to the target base station are obtained by representing the base station location and engineering parameter data.
[0085] In one embodiment, since not all areas in the target area may not have been road tested, that is, some target grids may have some road test data in the previous road test process. Therefore, in order to improve data utilization, the embodiment of the present application can also query whether the target grid has an angle power spectrum (APS) from the historical road test data. If the APS of the target grid is obtained, the corresponding target grid node features can be obtained based on the grid position data and the angle power spectrum APS. Otherwise, if the angle power spectrum cannot be obtained, the node features of the target grid are determined only based on the grid position data. In this way, the above process divides the target grid into two categories: one is the grid with an angle power spectrum obtained by historical measurement, and the other is the grid without an angle power spectrum. The reason may be that it has not been road tested, or although it has been road tested, the angle power spectrum cannot be obtained in the historical data.
[0086] In one embodiment, the base station node features and the grid node features are obtained by embedding, and after the base station node features and the grid node features are obtained, they are spliced to obtain the communication node features.
[0087] Step 140: Input the communication node features and adjacency matrix into a pre-trained graph neural network model for numerical prediction, obtain the reference signal predicted received power of the target base station's beam space on each target grid, and construct a spectrum coverage map based on the reference signal predicted received power.
[0088] In one embodiment, after the communication node features and the adjacency matrix are obtained, they are simultaneously input into a pre-trained graph neural network model for numerical prediction to obtain the reference signal predicted received power of the beam space of the target base station on each target grid. The reference signal predicted received power of the target base station on each target grid is obtained through the above process, and the target base station can also be replaced and the above prediction process can be repeated. The main service base station of each target grid can be determined based on the base station corresponding to the maximum value of the reference signal predicted received power, and the spectrum coverage map can be constructed based on the information such as the main service base station.
[0089] The following describes the training process of the neural network model in the embodiment of the present application.
[0090] In one embodiment, referring to Figure 2 , Figure 2 This is a schematic diagram of the training process of the graph neural network model provided in the embodiment of the present application, which specifically includes:
[0091] Step 210: Obtain test data, grid position parameters and base station communication sample data of each test grid in the test area, and obtain the maximum beam reference signal receiving power corresponding to the main service base station of the test grid from the test data as a sample label.
[0092] In one embodiment, a region of interest is selected as a test region, and the test region is divided according to the division method of the target grid to obtain a corresponding plurality of test grids. For example, the number of test grids is N, which is expressed as: Among them, the test area includes M test base stations, which are expressed as: Each test base station has pre-acquired base station communication sample data and its own base station number, for example, the base station number is represented as enodeBID, and the base station communication sample data corresponds to the above-mentioned base station communication parameters.
[0093] Next, a road test is performed in the test area by driving a car or operating a drone carrying a dedicated road test device, and the beam reference signal receiving power is discretely collected for each test grid to obtain test data including the beam reference signal receiving power.
[0094] In one embodiment, in order to reduce the test error during the drive test, multiple sampling is performed at each sampling point, and the expected beam reference signal received power obtained at the sampling point is calculated as the beam reference signal received power at the sampling point, thereby obtaining a test data set of the test grid. , where S represents the sampled and Ns represents the number of sampled test grids.
[0095] According to the above-mentioned road test process, multiple beam reference signal receiving powers in each test grid can be obtained, but each test grid has a main service base station and a neighboring base station, and the test data includes multiple beam reference signal receiving powers of the current main service base station of the test grid and multiple beam reference signal receiving powers of possible neighboring base stations, and the test measurements of all test grids must include the maximum beam reference signal receiving power of the main service base station. Therefore, the embodiment of the present application also needs to distinguish the obtained beam reference signal receiving power, determine which test base station corresponds to the data, and thus determine the maximum beam reference signal receiving power of the main service base station.
[0096] In one embodiment, referring to Figure 3 , Figure 3 This is a schematic diagram of obtaining the maximum beam reference signal received power corresponding to the primary serving base station of the test grid from the test data as a sample label provided by an embodiment of the present application, which specifically includes the following steps:
[0097] Step 310: Obtain the signal strength of at least one demodulation reference signal from the test data, and select the base station with the largest signal strength as the primary serving base station.
[0098] Among them, after the user equipment (UE) obtains the physical cell corresponding to the test grid, it performs beam scanning here to obtain the main synchronization signal (SSB) sent by different test base stations. The main synchronization signal SSB corresponds to a demodulation reference signal. At this time, the maximum signal strength of the demodulation reference signal is selected, and the test base station corresponding to the main synchronization signal SSB corresponding to the maximum signal strength is used as the main service base station, and the others are used as neighboring base stations.
[0099] Step 320: Select all beam reference signal received powers of the primary serving base station from the test data, and select the maximum value of the beam reference signal received power as the maximum beam reference signal received power.
[0100] Among them, after determining the main service base station corresponding to each test grid, all beam reference signal receiving powers of the main service base station are selected from the test data, and the maximum value of the beam reference signal receiving power is selected to obtain the maximum beam reference signal receiving power of the main service base station, and the maximum beam reference signal receiving power is used as a sample label to instruct the graph neural network model to predict the maximum beam reference signal receiving power based on the input information.
[0101] Step 220: Determine the relevant grid of each test base station according to the base station number in the base station communication sample data and the base station identifier in the test data, determine the adjacent grid of the relevant grid according to the grid position parameter, and generate a connection relationship diagram of the test base station, the relevant grid and the adjacent grid.
[0102] In one embodiment, the test data also includes a base station identifier (EUTRAN Cell Identifier, ECI). At this time, the base station identifier is matched with the base station number enodeB ID in the base station communication sample data. If the two are consistent, it is considered that the target grid matches the test base station. Then, all test grids matching the test base station are obtained in this way as related grids of the test base station. Then, the test grids located within a preset distance are taken as neighboring grids. It can be understood that each test base station is the main service base station of its related grid.
[0103] Next, a connection relationship diagram between the test base station, the relevant grid, and the adjacent grid is generated. In one embodiment, a connection relationship diagram between each test base station and its corresponding relevant grid is first generated, and the connection relationship diagram can be a star. Then, based on the star relationship diagram, different adjacent grids are connected to generate a connection relationship diagram between the test base station, the relevant grid, and the adjacent grid.
[0104] Step 230: Obtain node feature data based on base station communication sample data and grid position parameters.
[0105] In one embodiment, referring to Figure 4 , Figure 4This is a schematic diagram of obtaining node feature data according to base station communication sample data and grid position parameters provided by an embodiment of the present application, which specifically includes the following steps:
[0106] Step 410: Obtain base station sample characteristics based on base station communication sample data.
[0107] Among them, the base station communication sample data includes the base station location and base station working parameter data, so the base station sample features can be obtained by embedding the base station location and base station working parameter data.
[0108] Step 420: Calculate the angular power spectrum estimation value of the test grid based on the test data and the base station communication sample data of the corresponding primary serving base station.
[0109] In one embodiment, referring to Figure 5 , Figure 5 : is a flowchart of calculating the angular power spectrum estimation value of the test grid provided by an embodiment of the present application, which specifically includes the following steps:
[0110] Step 510: Obtain the antenna coefficient matrix of the primary serving base station according to the base station communication sample data.
[0111] Among them, the antenna coefficient matrix obtained according to the base station communication sample data is expressed as:
[0112]
[0113]
[0114] in, represents the beam m at an angle ( ), where P represents the transmit power, represents the departure downtilt angle of the primary serving base station, represents the departure azimuth, represents the number of base stations that are evenly dispersed in the direction away from the downtilt angle, represents the number of base stations that are uniformly discrete in the direction away from the azimuth, represents the number of antennas of the base station in the x direction, represents the number of antennas of the base station in the y direction, Indicates that at an angle ( ) is used to indicate the ability of an antenna to send or receive signals in a specific direction. The covariance of the antenna phase error is used to measure the phase inconsistency of each element in the antenna array. represents the beam m at the antenna position ( ), Indicated by the antenna position ( ) and angle ( ) caused by the phase difference of beam m, and represents the interval between adjacent antennas, Represents the cosine value of the phase difference corresponding to different antenna positions, ( ) indicate different antenna positions.
[0115] Step 520: Obtain all beam reference signal received powers of the primary serving base station from the test data.
[0116] Among them, the beam reference signal received power of beam m is expressed as .
[0117] Step 530: Perform compressed sensing calculation according to the beam reference signal received power and the antenna coefficient matrix to obtain an angle power spectrum estimation value.
[0118] In one embodiment, the angular power spectrum estimate is expressed as:
[0119]
[0120] in, Indicates the angle ( ) on the angular power spectrum estimation value, and the calculation is performed for each beam to obtain all the angular power spectrum estimation values of the test grid. In one embodiment, the angular power spectrum estimation value can be calculated using a compressed sensing algorithm such as a non-negative orthogonal matching pursuit algorithm, a weighted non-negative orthogonal matching pursuit algorithm, or a weighted non-negative orthogonal matching algorithm based on second-order statistics, and this embodiment does not limit the algorithm.
[0121] It can be understood that each test grid can obtain the angle power spectrum estimation value from the primary service base station to the corresponding grid through the above process, and the angle power spectrum estimation value is represented by a vector. The set of angle power spectrum estimation values obtained by each test grid is expressed as: .
[0122] Step 430: If the angle power spectrum estimate is valid, obtain the grid sample features according to the angle power spectrum estimate and the grid position parameters; otherwise, obtain the grid sample features according to the grid position parameters.
[0123] In one embodiment, since there may be errors in the calculation process of the angle power spectrum estimate, the angle power spectrum estimate of some test grids is actually unavailable. The embodiment of the present application can determine the validity of the angle power spectrum estimate. If the angle power spectrum estimate is valid, the grid sample feature is obtained according to the angle power spectrum estimate and the grid position parameter, otherwise the grid sample feature is obtained according to the grid position parameter. It can be understood that the grid position parameter is expressed in longitude and latitude.
[0124] In one embodiment, the calculated angle power spectrum estimate and the antenna coefficient matrix can be used to estimate the beam reference signal received power in reverse, and then the error between the estimated value and the measured value of the beam reference signal received power is determined. If the error is less than a certain range, the angle power spectrum estimate is considered valid. For example, the estimated beam reference signal received power is , the measured beam reference signal received power is By comparing the errors between the two, we can determine whether the angle power spectrum estimation is valid.
[0125] Step 440: Obtain node feature data according to the base station sample features and the grid sample features.
[0126] In one embodiment, the base station sample features and the grid sample features are obtained by embedding, and after the base station sample features and the grid sample features are obtained, they are spliced to obtain the node feature data.
[0127] In one embodiment, referring to Figure 6 , Figure 6 It is a schematic diagram of the connection relationship diagram provided in the embodiment of the present application. Figure 6 The data set includes: test data and base station communication sample data, where the test data includes: test grid number, grid location parameters (expressed in longitude and latitude), base station identifier of the primary service base station, maximum beam reference signal received power and angle power spectrum estimation. The base station communication sample data includes: base station number, base station location (expressed in longitude and latitude), departure downtilt angle, departure azimuth angle, etc.
[0128] Next, the test base station and test grid are represented as nodes, and the base station sample characteristics of the base station node are obtained according to the base station position and base station engineering parameter data in the base station communication sample data. For nodes with valid angular power spectrum estimation values, the grid sample characteristics of the grid nodes with valid APS are obtained according to the angular power spectrum estimation values and grid position parameters. For nodes with unavailable angular power spectrum estimation values, the grid sample characteristics of the grid nodes with invalid APS are obtained according to the grid position parameters.
[0129] Then, the relevant grid of each test base station is determined according to the base station number in the base station communication sample data and the base station identifier in the test data, and a relationship diagram between the test base station and the relevant grid is drawn. On the basis of the relationship diagram between the test base station and the relevant grid, a connection is established between the adjacent grids of the relevant grid according to the grid position parameters. Finally, all the relationship diagrams are summarized to obtain the final connection relationship diagram.
[0130] Step 240: Input the node feature data and the connection relationship diagram into the graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid.
[0131] In one embodiment, for example Figure 6 The obtained connection relationship graph calculates the adjacency matrix, and the adjacency matrix and the node feature data corresponding to each node are input into the graph neural network model for numerical prediction. When performing numerical prediction, since the dimensions of the node feature data of different nodes in the embodiment of the present application may be different, it is necessary to perform unified feature extraction of the node feature data.
[0132] Reference Figure 7 , Figure 7 Schematic diagram of numerical prediction using the graph neural network model provided in the embodiment of the present application. Figure 7 The process of unified feature extraction of node feature data is as follows. Since the node feature data includes position features obtained based on position information and other features obtained based on other information, the other features in the node feature data are mapped to a unified dimension by a fully connected layer. For node feature data that does not have an angular power spectrum estimation value, the default value is filled and mapped. Then the mapping result and the position feature are concatenated to obtain node feature data of unified dimension.
[0133] Next, refer to Figure 8 , Figure 8 The present invention provides a flowchart of inputting node feature data and a connection relationship graph into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid, which specifically includes the following steps:
[0134] Step 810: Obtain adjacent nodes with which each node has a connection relationship based on the connection relationship graph.
[0135] Among them, if the element value of the adjacency matrix in the connection relationship graph is 1, it can be considered that the nodes at these two positions have a connection relationship, so the adjacent nodes of each node can be obtained according to the adjacency matrix.
[0136] Step 820: Use the graph attention layer to calculate the feature weight of the node feature data of each node relative to the node feature data of adjacent nodes.
[0137] Among them, combined Figure 7 ,In the graph attention layer, each node calculates the feature weight of its node feature data compared with the node feature data of adjacent nodes through the softmax activation function.
[0138] Step 830: Weight the node feature data of adjacent nodes based on the feature weights to obtain the aggregate feature data corresponding to each node, and use the fully connected layer to calculate the sample prediction data corresponding to the aggregate feature data.
[0139] Among them, after obtaining the feature weights, the weighted features are aggregated through the message propagation mechanism, that is, the node feature data of adjacent nodes are weighted based on the feature weights, that is, the aggregate feature data corresponding to each node is obtained according to the sum of the product of the feature weights and the node feature data, thereby completing the mutual attention calculation between nodes. Finally, the aggregated feature data after attention aggregation is input into a shared fully connected layer, and finally the sample prediction data of the corresponding node position is output.
[0140] It can be understood that since the nodes in the input graph neural network model also include the nodes corresponding to the base stations, theoretically, the base station nodes will also have a prediction result related to the beam reference signal receiving power. The embodiment of the present application masks the prediction results of the base station nodes when outputting, that is, the prediction results of the base station nodes are not output, and only the prediction results of the nodes corresponding to the grid are output.
[0141] Step 250: Calculate the loss value according to the sample prediction data and the sample label, and adjust the graph neural network model based on the loss value until the iteration termination condition is reached to obtain a trained graph neural network model.
[0142] In one embodiment, the sample prediction data is compared with the sample label as the true value, the loss value is calculated using the mean square error as the loss function, and the graph neural network model is adjusted based on the loss value through the Adam gradient back propagation algorithm until the iteration termination condition is reached to obtain a trained graph neural network model. The iteration termination condition here can be a preset number of iterations, or the performance of the trained graph neural network model reaches a preset standard, which is not limited in this embodiment.
[0143] In one embodiment, the embodiment of the present application sets the initial learning rate of the model to 0.01 during the training process, sets the maximum number of training rounds to 2000, and includes test grids with sample labels and test grids without sample labels in the training data set, wherein the proportion of test grids with sample labels in the training data set is p, and 50% of the test grids in the training data set have valid angular power spectrum estimates, and the other 50% of the test grids do not have valid angular power spectrum estimates. The test grids without sample labels are used as the test set to perform error evaluation of the mean absolute error. The experiment is repeated ten times with a random seed of 1-10, and the average value is taken as the evaluation curve result.
[0144] Reference Fig. 9 , Fig. 9 This is a schematic diagram of the performance of the graph neural network model provided in an embodiment of the present application. Fig. 9It is used to evaluate the effect of unified feature extraction, so only the main service base station and the test grid are connected to avoid mutual influence between the test grids. The results of three different cases are compared: unified feature extraction, no unified feature extraction, no unified feature extraction and the input node feature data is adjusted to have the same vector dimension as after unified feature extraction. It can be seen that as the ratio p increases, the mean absolute error MAE in the three cases becomes smaller and smaller, and the mean absolute error is the smallest when unified feature extraction is performed. Therefore, the unified feature extraction in the embodiment of the present application can improve the accuracy of the prediction results.
[0145] Reference Fig.10 , Fig.10 It is a schematic diagram of the influence of the connection relationship between the test grids provided in the embodiment of the present application on the prediction result. Fig.10 In order to test the influence of the connection relationship between the test grids on the prediction results, the embodiment of the present application compares the average absolute error performance of the test grids with available angular power spectrum estimation values and the test grids without available angular power spectrum estimation values at a ratio p of 20% under different preset distance settings (when the preset distance is 0, it means that no connection between the test grids is performed). It can be seen that the connection between the test grids can significantly improve the accuracy of the prediction results. At the same time, when the preset distance is in the range of 30 meters to 90 meters, the average absolute error MAE performance of the prediction results is lower than 4.5dB. It can be seen that the spectrum coverage map construction method based on heterogeneous graph neural network in the embodiment of the present application has strong robustness to the preset examples. At the same time, when the preset distance is in the range of 30 meters to 130 meters, the accuracy of the prediction results trained using the angular power spectrum estimation value is also higher than the result without using the angular power spectrum estimation value. It can be seen that the use of the angular power spectrum estimation value improves the performance.
[0146] Reference Fig.11 , Fig.11 It is a schematic diagram of the accuracy of prediction results obtained by training using angular power spectrum estimation values provided in an embodiment of the present application. Fig.11 exist Fig.10 On this basis, the preset distance is set to 50m. The prediction results trained with the angular power spectrum estimation are more accurate than those without the angular power spectrum estimation. It can be seen that the performance is improved by using the angular power spectrum estimation.
[0147] Reference Fig.12 , Fig.12 It is a schematic diagram of the mean absolute percentage error in the test set provided by an embodiment of the present application. Fig.12In order to intuitively show the effectiveness of the spectrum coverage map construction method based on heterogeneous graph neural network provided in the embodiment of the present application, the result of the mean absolute percentage error MAPE of the test set is visually displayed: when the proportion p is 20%, the preset distance is 50 meters. From the results, it can be seen that the mean absolute percentage error MAPE of about 75% of the test grids is less than 10%, and the mean absolute percentage error MAPE of all test grids is 6.65%, indicating the high reliability of the spectrum coverage map construction method based on heterogeneous graph neural network provided in the embodiment of the present application.
[0148] The technical solution provided by the embodiment of the present application is to obtain the base station communication parameters of the target base station and the grid position data of multiple target grids, connect the target base station and the target grid, connect different target grids located within a preset distance based on the grid position data, generate an adjacency matrix according to the connection relationship, obtain the base station node characteristics according to the base station communication parameters, obtain the grid node characteristics according to the grid position data, and obtain the communication node characteristics according to the base station node characteristics and the grid node characteristics. Finally, the communication node characteristics and the adjacency matrix are input into a pre-trained graph neural network model for numerical prediction to obtain the reference signal predicted received power of the beam space of the target base station on each target grid. The embodiment of the present application trains a graph neural network model for data prediction, and obtains the corresponding node characteristics in combination with the communication parameters of the base station and the position data of the grid. Then, based on the connection relationship between the base station and the grid, the adjacency matrix is used to represent the correlation between the nodes. Node characteristics and correlation are considered simultaneously in the prediction process. Compared with the method of directly multiplying the APS calculated during the road test and the changed antenna coefficient matrix in the related art, this method improves the calculation accuracy. In addition, during the application process, there is no need to rely on road test data, and predictions can be made based on the location data of the target grid, thereby expanding the application scope of network coverage prediction.
[0149] The embodiment of the present application also provides a spectrum coverage map construction device based on a heterogeneous graph neural network, which can implement the above-mentioned spectrum coverage map construction method based on a heterogeneous graph neural network, referring to Fig.13 , the device comprises:
[0150] The data acquisition module 1310 is used to acquire the base station communication parameters of the target base station and the grid position data of multiple target grids.
[0151] The connection relationship generating module 1320 is used to connect the target base station and the target grid, connect different target grids located within a preset distance based on the grid position data, and generate an adjacency matrix according to the connection relationship.
[0152] Feature extraction module 1330: used to obtain base station node features according to base station communication parameters, obtain grid node features according to grid position data, and obtain communication node features according to base station node features and grid node features.
[0153] Prediction module 1340: used to input communication node features and adjacency matrix into a pre-trained graph neural network model for numerical prediction, and obtain the reference signal predicted received power of the beam space of the target base station on each target grid.
[0154] The specific implementation of the spectrum coverage map construction device based on heterogeneous graph neural network in this embodiment is basically consistent with the specific implementation of the spectrum coverage map construction method based on heterogeneous graph neural network mentioned above, and will not be repeated here.
[0155] The present application also provides an electronic device, including:
[0156] at least one memory;
[0157] at least one processor;
[0158] at least one program;
[0159] The program is stored in the memory, and the processor executes the at least one program to implement the above-mentioned spectrum coverage map construction method based on heterogeneous graph neural network implemented in this application. The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a car computer, etc.
[0160] See also Fig.14 , Fig.14 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:
[0161] The processor 1401 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;
[0162] The memory 1402 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1402 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1402, and the processor 1401 calls and executes the spectrum coverage map construction method based on heterogeneous graph neural network in the embodiment of this application;
[0163] Input / output interface 1403, used to implement information input and output;
[0164] Communication interface 1404, used to realize communication interaction between the device and other devices, which can be realized by wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.); and
[0165] A bus 1405 that transmits information between various components of the device (e.g., the processor 1401, the memory 1402, the input / output interface 1403, and the communication interface 1404);
[0166] The processor 1401 , the memory 1402 , the input / output interface 1403 and the communication interface 1404 are connected to each other in communication within the device via a bus 1405 .
[0167] An embodiment of the present application also provides a storage medium, which is a storage medium that stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned spectrum coverage map construction method based on heterogeneous graph neural network.
[0168] As a non-transitory storage medium, the memory can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0169] The spectrum coverage map construction method and device based on heterogeneous graph neural network proposed in the embodiment of the present application, by obtaining the base station communication parameters of the target base station and the grid position data of multiple target grids, connecting the target base station and the target grid, connecting different target grids located within a preset distance based on the grid position data, generating an adjacency matrix according to the connection relationship, obtaining the base station node characteristics according to the base station communication parameters, obtaining the grid node characteristics according to the grid position data, and obtaining the communication node characteristics according to the base station node characteristics and the grid node characteristics. Finally, the communication node characteristics and the adjacency matrix are input into the pre-trained graph neural network model for numerical prediction, and the reference signal predicted received power of the beam space of the target base station on each target grid is obtained. The embodiment of the present application trains the graph neural network model for data prediction, and obtains the corresponding node characteristics in combination with the communication parameters of the base station and the position data of the grid. Then, based on the connection relationship between the base station and the grid, the adjacency matrix is used to represent the correlation between the nodes. The node characteristics and the correlation are considered simultaneously in the prediction process. Compared with the method of directly multiplying the APS calculated in the road test process and the changed antenna coefficient matrix in the related art, this method improves the calculation accuracy. In addition, during the application process, there is no need to rely on road test data, and predictions can be made based on the location data of the target grid, thereby expanding the application scope of network coverage prediction.
[0170] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.
[0171] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.
[0172] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0173] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.
[0174] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, 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 "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0175] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.
[0176] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0177] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0178] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0179] If the 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 this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store programs.
[0180] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.
Claims
1. A spectrum coverage map construction method based on heterogeneous graph neural network, characterized in that: include: Acquiring base station communication parameters of a target base station and grid position data of a plurality of target grids; Connecting the target base station and the target grid, connecting different target grids located within a preset distance based on the grid position data, and generating an adjacency matrix according to the connection relationship; Obtaining base station node characteristics according to the base station communication parameters, obtaining grid node characteristics according to the grid position data, and obtaining communication node characteristics according to the base station node characteristics and the grid node characteristics; The communication node features and the adjacency matrix are input into a pre-trained graph neural network model for numerical prediction to obtain the reference signal predicted received power of the beam space of the target base station on each of the target grids, and a spectrum coverage map is constructed based on the reference signal predicted received power.
2. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 1, characterized in that: The obtaining grid node features according to the grid position data comprises: If the angular power spectrum of the target grid is obtained, the grid node features are obtained according to the grid position data and the angular power spectrum; Otherwise, grid node features are obtained according to the grid position data.
3. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 1, characterized in that: The training process of the graph neural network model includes the following steps: Acquire test data, grid position parameters, and base station communication sample data of each test grid in the test area, and acquire the maximum beam reference signal received power corresponding to the main serving base station of the test grid from the test data as a sample label; Determine the relevant grid of each of the test base stations according to the base station number in the base station communication sample data and the base station identifier in the test data, determine the adjacent grid of the relevant grid according to the grid position parameter, and generate a connection relationship diagram of the test base station, the relevant grid and the adjacent grid; Obtaining node feature data according to the base station communication sample data and the grid position parameters; Inputting the node feature data and the connection relationship graph into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid; The loss value is calculated according to the sample prediction data and the sample label, and the graph neural network model is adjusted based on the loss value until the iteration termination condition is reached to obtain the trained graph neural network model.
4. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 3, characterized in that: The obtaining, from the test data, the maximum beam reference signal received power corresponding to the primary serving base station of the test grid as a sample label includes: Acquire the signal strength of at least one demodulation reference signal from the test data, and select the base station with the largest signal strength as the primary serving base station; All beam reference signal received powers of the primary serving base station are selected from the test data, and a maximum value of the beam reference signal received powers is selected as the maximum beam reference signal received power.
5. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 3, characterized in that: The obtaining of node feature data according to the base station communication sample data and the grid position parameters includes: Obtaining base station sample characteristics according to the base station communication sample data; Calculate the angular power spectrum estimation value of the test grid according to the test data and the base station communication sample data of the corresponding primary serving base station; If the angle power spectrum estimation value is valid, obtaining the grid sample feature according to the angle power spectrum estimation value and the grid position parameter; otherwise, obtaining the grid sample feature according to the grid position parameter; The node feature data is obtained according to the base station sample features and the grid sample features.
6. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 5, characterized in that: The calculating the angle power spectrum estimation value of the test grid according to the test data and the base station communication sample data of the corresponding primary serving base station comprises: Obtaining an antenna coefficient matrix of the primary serving base station according to the base station communication sample data; Acquire all beam reference signal received powers of the primary serving base station from the test data; Compressed sensing calculation is performed according to the beam reference signal received power and the antenna coefficient matrix to obtain the angle power spectrum estimation value.
7. The method for constructing a spectrum coverage map based on a heterogeneous graph neural network according to claim 3, characterized in that: The step of inputting the node feature data and the connection relationship graph into a graph neural network model for numerical prediction to obtain sample prediction data corresponding to each test grid includes: Acquire adjacent nodes with which each node has a connection relationship based on the connection relationship graph; Calculating, using a graph attention layer, a feature weight of the node feature data of each node relative to the node feature data of the adjacent nodes; The node feature data of the adjacent nodes are weighted based on the feature weights to obtain aggregate feature data corresponding to each node, and sample prediction data corresponding to the aggregate feature data is calculated using a fully connected layer.
8. A spectrum coverage map construction device based on heterogeneous graph neural network, characterized in that: include: Data acquisition module: used to acquire base station communication parameters of a target base station and grid position data of multiple target grids; A connection relationship generating module: used to connect the target base station and the target grid, connect different target grids located within a preset distance based on the grid position data, and generate an adjacency matrix according to the connection relationship; Feature extraction module: used to obtain base station node features according to the base station communication parameters, obtain grid node features according to the grid position data, and obtain communication node features according to the base station node features and the grid node features; Prediction module: used to input the communication node features and the adjacency matrix into a pre-trained graph neural network model for numerical prediction, obtain the reference signal predicted received power of the beam space of the target base station on each of the target grids, and construct a spectrum coverage map based on the reference signal predicted received power.
9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the spectrum coverage map construction method based on heterogeneous graph neural network as described in any one of claims 1 to 7 when executing the computer program.
10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing a spectrum coverage map based on a heterogeneous graph neural network as described in any one of claims 1 to 7 is implemented.
