A training method, device and computer equipment for channel gain prediction model

Abstract

The present application relates to a training method, device and computer equipment for a channel gain prediction model. The method includes: obtaining a first sample set of a target area, clustering the sample points based on the geographical location information and channel gain of the sample points to obtain a first subset of sub-areas; training to obtain a corresponding initial channel gain prediction model based on the correspondence between the geographical location information and the channel gain of the sample points in the first subset of each sub-area; obtaining and determining the sample sampling rate of each sub-area based on the accuracy of each initial channel gain prediction model; sampling the sample points of the sub-area according to the sampling number of the sub-area to obtain a second subset of the sub-area; training to obtain a corresponding channel gain prediction model based on the second subset of the sub-area and the first subset. The disclosed embodiment improves the generation accuracy and generation efficiency of the channel gain.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication intelligent processing, and in particular to a method for training a channel gain prediction model, a method for generating a channel gain map, a device, a computer device, a storage medium and a computer program product. Background Art

[0002] With the rapid development of 5G, the future 6G will involve wireless channels of larger dimensions. However, common channel modeling methods such as random modeling are mathematically easy to handle, but they are only related to environmental parameters in a statistical sense, but have nothing to do with the actual environment. Therefore, in the field of wireless communications, achieving environmental perception has become a research focus. Among them, channel knowledge map (CKM) has attracted much attention as an emerging technology. CKM is a database that marks the precise locations of transmitters and receivers, and provides channel knowledge related to environmental objects, such as location, shape, and distance. CKM contains a variety of special instances, such as channel path map (CPM), channel gain map (CGM), and channel angle map (CAM).

[0003] CGM can be used to predict the channel gain at a specific location. Among them, a feasible method to establish CGM is offline numerical simulation technology, such as ray tracing algorithm. However, the ray tracing algorithm requires the reconstruction of the entire scene, including the material and size of the building, and the reconstructed scene will still be different from the reality, resulting in poor results. In addition, the ray tracing algorithm involves a large number of calculations and simulation operations, which is time-consuming when updating channel information. Therefore, a fast and direct method is to use interpolation technology, whose goal is to reconstruct the entire CGM based on the measurement results obtained at a limited number of locations. However, the interpolation-based method cannot capture the intrinsic structure of the channel information and requires considerable storage space.

[0004] How to reconstruct the entire channel gain map based on channel information at limited locations urgently requires a new method that can achieve high-precision reconstruction and make full use of limited training information. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention provides a method for training a channel gain prediction model, a method for generating a channel gain map, an apparatus, a computer device, a storage medium and a computer program product, which overcome the challenges brought about by the huge differences in training information caused by complex occlusion relationships in the scene, improve the efficiency of channel gain map reconstruction, and greatly reduce the required training samples.

[0006] In a first aspect, the present application provides a method for training a channel gain prediction model, the method comprising:

[0007] Acquire a first sample set of a target area, wherein the first sample set includes geographic location information of sample points and annotated channel gains;

[0008] Based on the geographical location information and channel gain of the sample points, clustering the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of the clusters;

[0009] Based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-regions and the channel gain, a corresponding initial channel gain prediction model is trained;

[0010] Obtaining and determining a sample sampling rate for each sub-region based on the accuracy of each initial channel gain prediction model;

[0011] Determine the number of samples of the sub-region based on the total number of samples of the second sample set of the target region and the sample sampling rate of the sub-region;

[0012] Sampling sample points of the sub-region according to the sampling quantity of the sub-region to obtain a second subset of the sub-region;

[0013] Based on the second subset of the sub-regions and the first subset, a corresponding channel gain prediction model is trained.

[0014] In one embodiment, the obtaining and determining the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model includes:

[0015] Obtaining an average difference between a predicted amount and an annotated amount of channel gain of an initial channel gain prediction model, a sampling quantity weight of a sub-region corresponding to the initial channel gain prediction model, and a cumulative deviation of an annotated amount of channel gain of the initial channel gain prediction model;

[0016] The sample sampling rate of each sub-region is determined based on the average difference of each initial channel gain prediction model, the accumulated deviation and the sampling quantity weight.

[0017] In one embodiment, the training based on the second subset of the sub-regions and the first subset to obtain a corresponding channel gain prediction model includes:

[0018] performing fusion processing on the first subset and the second subset to obtain a third subset;

[0019] Obtaining the maximum distance between each sample point in the third subset and the corresponding cluster center;

[0020] Acquire reference sample points from a sub-region adjacent to the sub-region;

[0021] When the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, adding the reference sample point to the third subset to obtain a fourth subset;

[0022] Based on the fourth subset of the sub-regions, a corresponding channel gain prediction model is trained.

[0023] In one of the embodiments, clustering the sample points based on the geographic location information and channel gains of the sample points to obtain a first subset of sub-regions includes:

[0024] Based on the number of intermediate clusters and the geographical location information of the sample points in the first sample set, clustering the sample points to obtain a first subset of the number of intermediate clusters;

[0025] Based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate cluster quantity and the channel gain, a corresponding intermediate channel gain prediction model is trained;

[0026] Obtaining a first average difference between a predicted amount and a marked amount of the channel gain of each intermediate channel model;

[0027] When the first average difference is smaller than the reference average difference, updating the reference average difference to the first average difference;

[0028] The number of intermediate clusters is updated, and based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain, a corresponding updated channel gain prediction model is trained;

[0029] Obtaining a second average difference between the predicted value and the labeled value of each updated channel gain prediction model;

[0030] When the second average difference is less than the reference average difference, updating the reference average difference to the second average difference, and continuing to update the updated number of intermediate clusters until the updated number of intermediate clusters reaches a threshold;

[0031] Based on the intermediate cluster number corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, clustering processing is performed on the sample points to obtain a first subset of sub-regions.

[0032] In one embodiment, the geographic location information includes location information of the sample point and distance information between the base station antenna and the sample point, and the step of acquiring a first sample set of the target area includes:

[0033] Acquire location information of sample points in the target area and distance information between the sample points and the base station antenna;

[0034] The position information and the distance information are respectively normalized to obtain a first sample set of the target area.

[0035] In a second aspect, the present application also provides a method for generating a channel gain map, the method comprising:

[0036] Acquire a corresponding third set in the target area; wherein the third set includes a plurality of data points and corresponding geographic location information;

[0037] Inputting the geographical location information of the location points in the third set into the steps of the method described in any one of the embodiments of the present disclosure to generate a channel gain prediction model, and outputting the channel gain of the location points;

[0038] A channel gain map of the target area is generated based on the first sample set, the second sample set, and the third sample set.

[0039] In a third aspect, the present application also provides a training device for a channel gain prediction model, the device comprising:

[0040] A first acquisition module, used to acquire a first sample set of a target area, wherein the first sample set includes geographic location information of sample points and annotated channel gains;

[0041] A clustering module, configured to perform clustering processing on the sample points based on the geographical location information and channel gains of the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of clustering;

[0042] A first training module, configured to train a corresponding initial channel gain prediction model based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-areas and the channel gain;

[0043] A first determination module, used to obtain and determine a sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model;

[0044] A second determination module, configured to determine the number of samples of the sub-region based on the total number of samples of the second sample set of the target region and the sample sampling rate of the sub-region;

[0045] A sampling module, configured to sample sample points of the sub-region according to the sampling quantity of the sub-region to obtain a second subset of the sub-region;

[0046] The second training module is used to train and obtain a corresponding channel gain prediction model based on the second subset of the sub-regions and the first subset.

[0047] In one embodiment, the first determining module is further configured to:

[0048] Obtaining an average difference between a predicted amount and an annotated amount of channel gain of an initial channel gain prediction model, a sampling quantity weight of a sub-region corresponding to the initial channel gain prediction model, and a cumulative deviation of an annotated amount of channel gain of the initial channel gain prediction model;

[0049] The sample sampling rate of each sub-region is determined based on the average difference of each initial channel gain prediction model, the accumulated deviation and the sampling quantity weight.

[0050] In one embodiment, the second training module is further used for:

[0051] performing fusion processing on the first subset and the second subset to obtain a third subset;

[0052] Obtaining the maximum distance between each sample point in the third subset and the corresponding cluster center;

[0053] Acquire reference sample points from a sub-region adjacent to the sub-region;

[0054] When the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, adding the reference sample point to the third subset to obtain a fourth subset;

[0055] Based on the fourth subset of the sub-regions, a corresponding channel gain prediction model is trained.

[0056] In one embodiment, the clustering module is further used to:

[0057] Based on the number of intermediate clusters and the geographical location information of the sample points in the first sample set, clustering the sample points to obtain a first subset of the number of intermediate clusters;

[0058] Based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate cluster quantity and the channel gain, a corresponding intermediate channel gain prediction model is trained;

[0059] Obtaining a first average difference between a predicted amount and a marked amount of the channel gain of each intermediate channel model;

[0060] When the first average difference is smaller than the reference average difference, updating the reference average difference to the first average difference;

[0061] The number of intermediate clusters is updated, and based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain, a corresponding updated channel gain prediction model is trained;

[0062] Obtaining a second average difference between the predicted value and the labeled value of each updated channel gain prediction model;

[0063] When the second average difference is less than the reference average difference, updating the reference average difference to the second average difference, and continuing to update the updated number of intermediate clusters until the updated number of intermediate clusters reaches a threshold;

[0064] Based on the intermediate cluster number corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, clustering processing is performed on the sample points to obtain a first subset of sub-regions.

[0065] In one embodiment, the first acquisition module is further used for:

[0066] Acquire location information of sample points in the target area and distance information between the sample points and the base station antenna;

[0067] The position information and the distance information are respectively normalized to obtain a first sample set of the target area.

[0068] In a fourth aspect, the present application also provides a device for generating a channel gain map, the device comprising:

[0069] A second acquisition module is used to acquire a corresponding third set in the target area; wherein the third set includes a plurality of data points and corresponding geographic location information;

[0070] A prediction module, configured to input the geographical location information of the location points in the third set into the steps of the method according to any one of the embodiments of the present disclosure to generate a channel gain prediction model, and output the channel gain of the location points;

[0071] A generating module is used to generate a channel gain map of the target area based on the first sample set, the second sample set, and the third set.

[0072] In a fifth aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method described in any one of the embodiments of the present disclosure are implemented.

[0073] In a sixth aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of the embodiments of the present disclosure are implemented.

[0074] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of the embodiments of the present disclosure are implemented.

[0075] The training method of the channel gain prediction model, the generation method of the channel gain map, the device, the computer equipment, the storage medium and the computer program product, by clustering the first sample set of the target area, obtain the first subsets corresponding to each sub-area, after the clustering process, the correlation of each sample point in the first subset is stronger. Therefore, the initial channel gain prediction model trained based on the sample points in the first subset predicts the position points in the corresponding sub-area, and the accuracy of the obtained channel gain is higher. Further, according to the accuracy of the initial channel gain prediction model, the sample sampling rate of each sub-area is determined, wherein the accuracy of the initial channel gain prediction model can reflect the complexity of the communication environment of the corresponding sub-area, the accuracy is low, the corresponding complexity is high; the accuracy is high, the corresponding complexity is low. Based on this, a larger number of new sample points are set for areas with higher complexity of the communication environment. When the total number of sample points is limited, the embodiment of the present disclosure further improves the generation accuracy and generation efficiency of the channel gain. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0077] Figure 1 A schematic diagram of a flow chart of a method for training a channel gain prediction model in one embodiment;

[0078] Figure 2 A schematic diagram of a structure in which a target area is clustered into sub-areas in one embodiment;

[0079] Figure 3 A schematic diagram of a flow chart of a method for training a channel gain prediction model in one embodiment;

[0080] Figure 4 A schematic diagram of a flow chart of a method for training a channel gain prediction model in one embodiment;

[0081] Figure 5 A schematic diagram of a flow chart of a method for training a channel gain prediction model in one embodiment;

[0082] Figure 6 A channel gain map obtained by using a ray tracing algorithm in one embodiment;

[0083] Figure 7 A channel gain map obtained by using the channel gain map generation method of this case in one embodiment;

[0084] Figure 8 A system architecture diagram of a method for generating a channel gain map in accordance with an embodiment of the present invention;

[0085] Fig. 9 It is a structural schematic diagram of a training device for a channel gain prediction model in one embodiment;

[0086] Fig.10 is a schematic structural diagram of a device for generating a channel gain map in one embodiment;

[0087] Fig.11 is an internal structure diagram of a computer device in one embodiment;

[0088] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0089] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0090] In one embodiment, Figure 1 As shown, a training method for a channel gain prediction model is provided. This embodiment uses the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0091] Step S101: Acquire a first sample set of a target area, where the first sample set includes geographic location information of sample points and annotated channel gains.

[0092] The geographical location information may include location information and distance information between the base station and the antenna. The channel gain of the sample point can be expressed as in, Indicates the antenna transmission power, represents the average received power, and i represents the i-th sample point in the first sample set. In an exemplary embodiment, the geographical location information and channel gain of sample point i can be expressed as: in and represents the horizontal and vertical coordinates of sampling point i, It is expressed as the distance between the antenna and the sample point. i Calculated as d i =||c bs -c sp，i ||, where ||·|| represents the calculation formula of Euclidean distance. Therefore, the channel gain map dataset for sparse representation is constructed as follows: Among them, M scgm Represented as the total number of data points in the sparse channel gain map.

[0093] The constructed sparse channel gain map contains 1000 records, namely M scgm =1000, data set It is expressed as follows:

[0094]

[0095] In addition, the constructed test set contains 200 records, namely M te =200, the data point format is similar to the data set The same as in.

[0096] Step S103: clustering the sample points based on the geographical location information and channel gains of the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of the clusters.

[0097] In an exemplary embodiment, a variety of clustering algorithms can be used to cluster the sample points, such as K-means clustering, hierarchical clustering, density clustering, and spectral clustering. The geographical location information of the sample points in the target area is clustered to obtain a first subset corresponding to the multiple sub-areas. Figure 2 As shown, the target area 201 is clustered into four sub-areas, such as sub-area 203. The sub-areas correspond to the categories of clustering, that is, each sub-area corresponds to an information cluster.

[0098] Step S105 : Based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-regions and the channel gain, a corresponding initial channel gain prediction model is trained.

[0099] In one embodiment, for any sub-region, the geographic location information and channel gain of the sample points in the first set of the sub-region are obtained. The geographic location information is input into the convolutional neural network model, the predicted channel gain is output, and the convolutional neural network model is iteratively adjusted based on the difference between the predicted channel gain and the actual channel gain to obtain an initial channel gain prediction model.

[0100] In a specific embodiment, the first set k can be expressed as: The input data can be represented as The corresponding training output is expressed as in M k Indicates the number of sample points contained in the first set k. Transformed into the corresponding input feature map, expressed as Among them, G and D represent the number and length of feature maps respectively, and they are input into the corresponding sub-neural network. Each sub-neural network has L layers, each layer contains a convolution layer and a pooling layer, then the output of the lth layer of the sub-neural network can be expressed as Then, the output of layer l+1 is The t-th output feature map of can be calculated as:

[0101]

[0102] Among them, f down (·) represents the maximum pooling function, represents the output after the convolution operation, which can be calculated as follows:

[0103]

[0104] in, represents the jth input feature map of the i-th entity in layer l, represents the convolution operation, and Represent the corresponding weight matrix and bias term, f relu (·) represents the ReLU activation function.

[0105] After the convolution and pooling operations are completed, the output of the training sample is obtained through a fully connected layer. All predicted outputs within an information cluster are expressed as Then, the model parameter Λ of the sub-neural network k It can be optimized and calculated as follows:

[0106]

[0107] in, represents the optimized model parameters, f mse (·) represents the mean square error loss function.

[0108] In one embodiment, a test set is selected Data points within v z The information of the first three dimensions constitutes the test sample x z At the same time, select the cluster center o k The first three dimensions of information constitute o'k , according to the Euclidean distance d' z，k =||x z -o' k The size of || classifies each test sample into the corresponding sub-area, where ||·|| represents the calculation formula of Euclidean distance. The test subset can be expressed as in Indicates the number of test samples in the test subset. Input into the kth sub-neural network corresponding to the sub-region to obtain the predicted channel gain, which can be calculated as follows:

[0109]

[0110] in, represents the predicted channel gain, represents the prediction function of the k-th neural network, Represents the corresponding input feature map.

[0111] Step S107 , obtaining and determining the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model.

[0112] The accuracy of the initial channel gain prediction model can be measured by the difference between the channel gain prediction amount and the labeled amount. In one embodiment, step S107 includes:

[0113] Step S301, obtaining the average difference between the predicted amount and the labeled amount of the channel gain of the initial channel gain prediction model, the sampling quantity weight of the sub-region corresponding to the initial channel gain prediction model, and the cumulative deviation of the channel gain labeled amount of the initial channel gain prediction model.

[0114] Among them, the average difference between the predicted value and the labeled value can be expressed by the root mean square error, for example:

[0115]

[0116] Among them, M te represents the total number of test samples, e z Represents the sample point x z The corresponding real channel gain (annotated amount), Test sample x z The predicted amount.

[0117] The accumulated deviation can be calculated by the variance of the channel gain of the sub-region sample points, for example: Among them, e iand e represent the annotated value (true value) and average channel gain (the average value of all true channel gains) of the first subset channel gain, respectively. k represents the number of sample points in the first subset. Wherein, the sampling weight θ of the sub-region corresponding to the initial channel gain prediction model k This can be done by: Calculated, M k represents the number of sample points in the first subset, M scgm Indicates the total number of sample points in the first sample set.

[0118] Step S303: determining the sample sampling rate of each sub-region based on the average difference of each initial channel gain prediction model, the accumulated deviation and the sampling quantity weight.

[0119] Specifically, the sample sampling rate λ of the initial channel gain prediction model corresponding to the sub-region k It can be expressed as follows:

[0120]

[0121] In the disclosed embodiment, the cumulative deviation is the difference in the actual channel gain within a sub-region. A large cumulative deviation indicates that the channel gain of each sampling point in the sub-region varies greatly, which means that the radio propagation environment is complex. We need to collect more comprehensive information to allow the neural network to learn the channel gain in various situations to improve the prediction efficiency. Therefore, the sampling rate uses: the average difference, which reflects the performance of the model, from the perspective of the model; and the cumulative deviation, which reflects the complexity of the internal environment of the sub-region, which are all effective, so that the obtained sampling rate is more scientific and effective.

[0122] Step S109: determining the number of samples of the sub-region based on the total number of samples of the second sample set of the target region and the sample sampling rate of the sub-region.

[0123] In the embodiment of the present disclosure, the total number of samples of the second sample set can be expressed as N, that is, if a total of N sample points are sampled, the number of samples allocated to each sub-region can be calculated as N k =N×λ k .

[0124] Step S111 : sampling sample points of the sub-region according to the sampling quantity of the sub-region to obtain a second subset of the sub-region.

[0125] The method for obtaining the sample points in the second subset may be the same as the method for obtaining the first sample set in the above embodiment, and the embodiments of the present disclosure will not be described in detail herein.

[0126] Step S113: Based on the second subset of the sub-regions and the first subset, a corresponding channel gain prediction model is trained.

[0127] In an exemplary embodiment, N in the second subset may be k New sample points are added to the first subset In the example, the first subsets of all sub-regions are updated, and the corresponding channel gain model is obtained by training using the updated first subsets.

[0128] In the above embodiment, by clustering the first sample set of the target area, the first subsets corresponding to each sub-area are obtained. After the clustering process, the correlation between each sample point in the first subset is stronger. Therefore, the initial channel gain prediction model trained based on the sample points in the first subset predicts the position points in the corresponding sub-area, and the channel gain obtained is more accurate. Further, according to the accuracy of the initial channel gain prediction model, the sample sampling rate of each sub-area is determined, wherein the accuracy of the initial channel gain prediction model can reflect the complexity of the communication environment of the corresponding sub-area, and the lower the accuracy, the higher the corresponding complexity; the higher the accuracy, the lower the corresponding complexity. Based on this, a larger number of new sample points are set for areas with higher complexity of the communication environment. When the total number of sample points is limited, the embodiment of the present disclosure further improves the accuracy of the channel gain prediction model.

[0129] In one embodiment, reference Figure 4 As shown, sampling of sample points is performed on the sub-region according to the sampling quantity of the sub-region to obtain a second subset of the sub-region, including:

[0130] Step S401: merge the first subset and the second subset to obtain a third subset.

[0131] Specifically, the first subset and the second subset may be combined to obtain a third subset.

[0132] Step S403: Obtain the maximum distance between each sample point in the third subset and the corresponding cluster center.

[0133] In an exemplary embodiment, the distance may include each sample point To cluster center o k The Euclidean distance is calculated as follows:

[0134] d i，k =||v i -o k || (7)

[0135] Furthermore, the largest Euclidean distance is selected from the Euclidean distances corresponding to the sample points as the maximum distance.

[0136] In an exemplary embodiment, the kth third subset The maximum distance is calculated as follows:

[0137]

[0138] Among them, D k Represents the third subset The maximum distance, max(·) means calculating the maximum value of each data, d a,k Represents the calculated data point v a With cluster center o k The distance Indicates the third subset after adding new sample points The total number of data points in .

[0139] Step S405: Acquire reference sample points from a sub-region adjacent to the sub-region.

[0140] Step S407 , when the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, the reference sample point is added to the third subset to obtain a fourth subset.

[0141] In an exemplary embodiment, the information cluster is calculated Data points outside To cluster center o k The Euclidean distance d β,k , then the distance The difference in coverage distance D β,k It can be calculated as D β,k =d β,k -D k . Set the distance threshold σ, if D β，k ≤σ, then the data point v β Add to the corresponding third subset middle.

[0142] In an exemplary embodiment, the sample point information of each third subset is updated in the above manner, and the optimized fourth subset can be expressed as

[0143] Step S409: Based on the fourth subset of the sub-regions, a corresponding channel gain prediction model is trained.

[0144] Specifically, the embodiment of the present disclosure is based on the fourth subset of the sub-regions, and the corresponding channel gain prediction model trained is the same as the above embodiment, and the embodiment of the present disclosure will not be repeated here.

[0145] In the above embodiment, on the basis of the third subset, reference sample points that meet the preset conditions are selected from adjacent sub-regions. These sample points at the edge of the sub-region corresponding to the third subset will also contain certain information of other sub-regions. The addition of the above reference sample points is equivalent to expanding the data points of the third subset, thereby improving the prediction accuracy of the model. This method effectively divides the limited sampling information and fully explores the uniqueness and correlation of different partitions, which has a strong application value for channel gain prediction.

[0146] In one embodiment, reference Figure 5 As shown, the clustering process is performed on the sample points based on the geographical location information and the channel gain of the sample points to obtain a first subset of sub-areas, including:

[0147] Step S501: Based on the number of intermediate clusters and the geographical location information of the sample points in the first sample set, clustering processing is performed on the sample points to obtain a first subset of the number of intermediate clusters.

[0148] In an exemplary embodiment, the number of intermediate clusters may be represented as K, and the initial value may be set to 1. At this time, the first sample set and the first subset are the same. Of course, the initial value of K may also be set to 2, 3, 4, etc. This embodiment of the present disclosure does not limit this.

[0149] Step S503: Based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate cluster quantity and the channel gain, a corresponding intermediate channel gain prediction model is trained.

[0150] Specifically, when K is 1, All sample points in are used as training samples, that is, geographic environment information As input, the channel gain e i As output, a neural network model is trained, and a convolutional neural network is used here to obtain an intermediate channel gain model. When K is not equal to 1, K intermediate channel gain models are respectively trained based on the sample points of the K first subsets. The process of clustering the first sample set to obtain the K first subsets can be obtained in the following manner:

[0151] Step (1): Randomly initialize K cluster centers, then the set of cluster centers can be expressed as Calculate each data point To cluster center o k The Euclidean distance ofi Update to the closest cluster In which, the calculation formula of Euclidean distance is as follows:

[0152] d i，k =||v i -o k || (9)

[0153] Then, the corresponding information cluster number k can be calculated as

[0154] Step (2): After all data points are classified into their respective information clusters, the cluster center is updated by the mean of each dimension of all data points. The updated cluster center can be calculated as follows:

[0155]

[0156] Among them, M k Represents the number of data points contained in information cluster k.

[0157] Step (3): After obtaining the new cluster center, repeat steps (1) and (2) until the calculated new cluster center does not change or steps (1) and (2) are repeated for more than the maximum number of iterations I. max , the classification is now complete.

[0158] Step S505: Obtain a first average difference between the predicted amount and the marked amount of the channel gain of each intermediate channel model.

[0159] The first average difference can be realized by a root mean square error value. Specifically, the first average difference R K It can be expressed as follows:

[0160]

[0161] Among them, M te represents the total number of test samples for each intermediate channel model, e z Represents the test sample x z The corresponding real channel gain (annotated amount), Represents the predicted quantity of channel gain for each intermediate channel model.

[0162] Step S507: When the first average difference is smaller than the reference average difference, update the reference average difference to the first average difference.

[0163] In an exemplary embodiment, if the first average difference is less than the reference average difference, for example, R K <R com , the reference mean difference is expressed as R com , then RK Replace R with the value com .

[0164] Step S509: updating the number of intermediate clusters, and training a corresponding updated channel gain prediction model based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain.

[0165] In an exemplary embodiment, for example, the number of intermediate clusters is updated to K+1. The sample points in the first sample set are clustered to obtain K+1 first subsets. Based on the correspondence between the geographical location information of the sample points in the K+1 first subsets and the channel gain, K+1 updated channel gain prediction models are trained. It should be noted that the number of updated intermediate clusters is not necessarily K+1, but may also be K+2, and the embodiments of the present disclosure do not limit this.

[0166] Step S511, obtaining a second average difference between the predicted value and the labeled value of each updated channel gain prediction model.

[0167] The second average difference can be realized by a root mean square error value. Specifically, the second average difference R K+1 This can be achieved by formula (11).

[0168] Step S513: when the second average difference is smaller than the reference average difference, the reference average difference is updated to the second average difference, and the updated number of intermediate clusters is continuously updated until the minimum reference average difference is determined.

[0169] In an exemplary embodiment, if the second average difference is less than the reference average difference, for example, R K+1 <R com , the reference mean difference is expressed as R com , then R K+1 Replace R with the value com In another exemplary embodiment, if the second average difference is greater than or equal to the reference average difference, the reference average difference is kept unchanged. Until the updated number of intermediate clusters reaches the threshold value K max .

[0170] Step S515 , based on the intermediate cluster number corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, clustering processing is performed on the sample points to obtain a first subset of sub-regions.

[0171] Among them, the minimum reference average difference can also be the last updated reference average difference, and the number of intermediate clusters corresponding to the minimum reference average difference can be expressed as K *In an exemplary embodiment, clustering is performed on the sample points in the first sample set to obtain K * The first subset.

[0172] In an exemplary embodiment, for example, in 9 classification cases R K The values ​​of are {2.799, 2.528, 2.446, 2.377, 2.491, 2.596, 2.596, 2.697, 2.601}, so when K = 4, R K The value of is the smallest and is finally determined to be K * =4.

[0173] In one embodiment, the geographic location information includes location information of the sample point and distance information between the base station antenna and the sample point, and the acquiring of the first sample set of the target area includes:

[0174] The location information of the sample points in the target area and the distance information between the sample points and the base station antenna are obtained.

[0175] The position information and the distance information are respectively normalized to obtain a first sample set of the target area.

[0176] Specifically, in an exemplary embodiment, the geographical location information and channel gain of the sample point i can be expressed as: in and represents the horizontal and vertical coordinates of sampling point i, It is expressed as the distance between the antenna and the sample point. i Calculated as d i =||c bs -c sp，i ||.

[0177] In another exemplary embodiment, the position information and the distance information are normalized respectively, and the normalization ensures that each feature contributes equally to the Euclidean distance calculation, thereby preventing certain features from dominating the clustering. The calculation formula can be expressed as:

[0178]

[0179] Among them, y represents the value of a certain dimension information in the data point, y' represents the normalized value of y, min(y) represents the minimum value of a certain dimension information in all data points, and max(y) represents the maximum value of a certain dimension information in all data points.

[0180] Based on the same inventive concept, an embodiment of the present application further provides a method for generating a channel gain map, the method comprising:

[0181] Acquire a corresponding third set in the target area; wherein the third set includes a plurality of location points and corresponding geographic location information;

[0182] Inputting the geographical location information of the location points in the third set into the steps of the method described in any one of the embodiments of the present disclosure to generate a channel gain prediction model, and outputting the channel gain of the data point;

[0183] A channel gain map of the target area is generated based on the first sample set, the second sample set, and the third sample set.

[0184] refer to Figure 8 As shown, in the embodiment of the present disclosure, the third set may include data points containing only geographic location information, and the geographic location information of these data points is input into the channel gain prediction model obtained by the steps of any of the methods in the above embodiments, and the channel gain of the data points is output. The first sample set and the second sample set both contain the geographic location information of the sample points and the corresponding channel gain. Therefore, by merging the first sample set and the second sample set with the third set, a more comprehensive channel gain of the target area can be obtained. Based on the geographic location information of the data points, the corresponding channel gain is marked, and a channel gain map can be obtained, such as Figure 7 shown.

[0185] Finally, in order to verify the effectiveness of the proposed channel gain map reconstruction method, the ray tracing algorithm is used to obtain the channel gain distribution in the scene and compare it with the predicted channel gain distribution map obtained based on the prediction model. The results are shown in the figure. Figure 6 and Figure 7 As shown (the white area in the figure is a building).

[0186] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0187] Based on the same inventive concept, the embodiment of the present application also provides a channel gain prediction model training device for implementing the channel gain prediction model training method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the training device for one or more channel gain prediction models provided below can refer to the limitations of the channel gain prediction model training method above, and will not be repeated here.

[0188] In one embodiment, the present application further provides a training device for a channel gain prediction model, the device 900 comprising:

[0189] A first acquisition module 901 is used to acquire a first sample set of a target area, wherein the first sample set includes geographic location information of sample points and annotated channel gains;

[0190] A clustering module 903, configured to perform clustering processing on the sample points based on the geographical location information and channel gains of the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of clustering;

[0191] A first training module 905 is used to train and obtain a corresponding initial channel gain prediction model based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-areas and the channel gain;

[0192] A first determination module 907, configured to obtain and determine a sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model;

[0193] A second determination module 909, configured to determine the number of samples of the sub-region based on the total number of samples of the second sample set of the target region and the sample sampling rate of the sub-region;

[0194] A sampling module 911 is used to sample sample points of the sub-region according to the sampling quantity of the sub-region to obtain a second subset of the sub-region;

[0195] The second training module 913 is used to train and obtain a corresponding channel gain prediction model based on the second subset of the sub-regions and the first subset.

[0196] In one embodiment, the first determining module is further configured to:

[0197] Obtaining an average difference between a predicted amount and an annotated amount of channel gain of an initial channel gain prediction model, a sampling quantity weight of a sub-region corresponding to the initial channel gain prediction model, and a cumulative deviation of an annotated amount of channel gain of the initial channel gain prediction model;

[0198] The sample sampling rate of each sub-region is determined based on the average difference of each initial channel gain prediction model, the accumulated deviation and the sampling quantity weight.

[0199] In one embodiment, the second training module is further used for:

[0200] performing fusion processing on the first subset and the second subset to obtain a third subset;

[0201] Obtaining the maximum distance between each sample point in the third subset and the corresponding cluster center;

[0202] Acquire reference sample points from a sub-region adjacent to the sub-region;

[0203] When the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, adding the reference sample point to the third subset to obtain a fourth subset;

[0204] Based on the fourth subset of the sub-regions, a corresponding channel gain prediction model is trained.

[0205] In one embodiment, the clustering module is further used to:

[0206] Based on the number of intermediate clusters and the geographical location information of the sample points in the first sample set, clustering the sample points to obtain a first subset of the number of intermediate clusters;

[0207] Based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate cluster quantity and the channel gain, a corresponding intermediate channel gain prediction model is trained;

[0208] Obtaining a first average difference between a predicted amount and a marked amount of the channel gain of each intermediate channel model;

[0209] When the first average difference is smaller than the reference average difference, updating the reference average difference to the first average difference;

[0210] The number of intermediate clusters is updated, and based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain, a corresponding updated channel gain prediction model is trained;

[0211] Obtaining a second average difference between the predicted value and the labeled value of each updated channel gain prediction model;

[0212] When the second average difference is less than the reference average difference, updating the reference average difference to the second average difference, and continuing to update the updated number of intermediate clusters until the updated number of intermediate clusters reaches a threshold;

[0213] Based on the intermediate cluster number corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, clustering processing is performed on the sample points to obtain a first subset of sub-regions.

[0214] In one embodiment, the first acquisition module is further used for:

[0215] Acquire location information of sample points in the target area and distance information between the sample points and the base station antenna;

[0216] The position information and the distance information are respectively normalized to obtain a first sample set of the target area.

[0217] In a fourth aspect, the present application further provides a device for generating a channel gain map, the device 100 comprising:

[0218] The second acquisition module 1001 is used to acquire a corresponding third set in the target area; wherein the third set includes a plurality of data points and corresponding geographic location information;

[0219] Prediction module 1003, configured to input geographic location information of the location point in the third set into the steps of generating a channel gain prediction model according to any one of the methods of the embodiments of the present disclosure, and output the channel gain of the location point;

[0220] The generating module 1005 is configured to generate a channel gain map of the target area based on the first sample set, the second sample set, and the third sample set.

[0221] In the above device embodiments, in addition to the device-specific embodiments, it is also necessary to write out the device item embodiments in which all method items correspond one-to-one with the claims.

[0222] Each module in the above-mentioned channel gain map generation device and channel gain prediction model training device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0223] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.11As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store training data of a channel gain prediction model. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a training method for a channel gain prediction model is implemented.

[0224] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.12 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a training method for a channel gain prediction model is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0225] Those skilled in the art will understand that Fig.12The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0226] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0227] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processors involved in each embodiment provided in this application may be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, data processing logic devices based on quantum computing, etc., but are not limited thereto.

[0228] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0229] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for training a channel gain prediction model, characterized in that: The method comprises: Acquire a first sample set of a target area, wherein the first sample set includes geographic location information of sample points and annotated channel gains; Based on the geographical location information and channel gain of the sample points, clustering the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of the clusters; Based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-areas and the channel gain, a corresponding initial channel gain prediction model is trained; based on the geographical location information and the channel gain of the sample points, the sample points are clustered to obtain the first subset of the sub-areas, including: based on the intermediate cluster number and the geographical location information of the sample points in the first sample set, the sample points are clustered to obtain the first subset of the intermediate cluster number; based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate cluster number and the channel gain, a corresponding intermediate channel gain prediction model is trained; a first average difference between the predicted amount and the marked amount of the channel gain of each intermediate channel model is obtained; when the first average difference is less than the reference average difference In the case where the reference average difference is less than the reference average difference, the reference average difference is updated to the first average difference; the number of intermediate clusters is updated, and based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain, a corresponding updated channel gain prediction model is trained; a second average difference between the predicted amount and the labeled amount of each updated channel gain prediction model is obtained; in the case where the second average difference is less than the reference average difference, the reference average difference is updated to the second average difference, and the updated number of intermediate clusters is continuously updated until the updated number of intermediate clusters reaches a threshold; based on the number of intermediate clusters corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, the sample points are clustered to obtain a first subset of the sub-region; Obtaining and determining the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model; the obtaining and determining the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model includes: obtaining the average difference between the predicted amount and the labeled amount of the channel gain of the initial channel gain prediction model, the sampling quantity weight of the sub-region corresponding to the initial channel gain prediction model, and the cumulative deviation of the labeled amount of the channel gain of the initial channel gain prediction model; determining the sample sampling rate of each sub-region based on the average difference between the predicted amount and the labeled amount of the channel gain of each initial channel gain prediction model, the cumulative deviation, and the sampling quantity weight; Determine the number of samples of the sub-area based on the total number of samples of the second sample set of the target area and the sample sampling rate of the sub-area; wherein the second sample set includes geographical location information of the sample points and the marked channel gain; and the total number of samples of the second sample set is the same as the total number of samples of the first sample set; Sampling sample points of the sub-area according to the sampling quantity of the sub-area to obtain a second subset of the sub-area; the second subset includes geographical location information of the sample points and marked channel gains; Based on the second subset and the first subset of the sub-region, a corresponding channel gain prediction model is trained; the training of the corresponding channel gain prediction model based on the second subset and the first subset of the sub-region includes: fusing the first subset and the second subset to obtain a third subset; obtaining the maximum distance from each sample point in the third subset to the corresponding cluster center; obtaining a reference sample point from a sub-region adjacent to the sub-region; when the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, adding the reference sample point to the third subset to obtain a fourth subset; based on the fourth subset of the sub-region, a corresponding channel gain prediction model is trained.

2. The method according to claim 1, characterized in that The geographical location information includes location information of the sample point and distance information between the base station antenna and the sample point. The acquiring a first sample set of the target area includes: Acquire location information of sample points in the target area and distance information between the sample points and the base station antenna; The position information and the distance information are respectively normalized to obtain a first sample set of the target area.

3. A method for generating a channel gain map, characterized in that: The method comprises: Acquire a corresponding third set in the target area; wherein the third set includes a plurality of data points and corresponding geographic location information; Inputting the geographical location information of the location points in the third set into the steps of the method according to any one of claims 1 or 2 to generate a channel gain prediction model, and outputting the channel gain of the location points; A channel gain map of the target area is generated based on the first sample set, the second sample set, and the third sample set.

4. A training device for a channel gain prediction model, characterized in that: The device comprises: A first acquisition module, used to acquire a first sample set of a target area, wherein the first sample set includes geographic location information of sample points and annotated channel gains; A clustering module, configured to perform clustering processing on the sample points based on the geographical location information and channel gains of the sample points to obtain a first subset of sub-regions; wherein the sub-regions correspond to the categories of clustering; The first training module is used to train a corresponding initial channel gain prediction model based on the correspondence between the geographical location information of the sample points in the first subset of each of the sub-areas and the channel gain; the first training module is also used to: cluster the sample points based on the number of intermediate clusters and the geographical location information of the sample points in the first sample set to obtain the first subset of the intermediate number of clusters; train a corresponding intermediate channel gain prediction model based on the correspondence between the geographical location information of the sample points in the first subset of the intermediate number of clusters and the channel gain; obtain a first average difference between the predicted amount and the marked amount of the channel gain of each intermediate channel model; and update the reference average difference when the first average difference is less than the reference average difference. is the first average difference; updating the number of intermediate clusters, and based on the correspondence between the geographical location information of the sample points in the first subset of the updated number of intermediate clusters and the channel gain, training to obtain the corresponding updated channel gain prediction model; obtaining the second average difference between the predicted amount and the labeled amount of each updated channel gain prediction model; when the second average difference is less than the reference average difference, updating the reference average difference to the second average difference, and continuing to update the updated number of intermediate clusters until the updated number of intermediate clusters reaches a threshold; based on the number of intermediate clusters corresponding to the minimum reference average difference and the geographical location information of the sample points in the first sample set, clustering the sample points to obtain a first subset of the sub-region; A first determination module is used to obtain and determine the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model; the obtaining and determining the sample sampling rate of each sub-region based on the accuracy of each initial channel gain prediction model includes: obtaining the average difference between the predicted amount and the labeled amount of the channel gain of the initial channel gain prediction model, the sampling quantity weight of the sub-region corresponding to the initial channel gain prediction model, and the cumulative deviation of the channel gain labeled amount of the initial channel gain prediction model; based on the average difference between the predicted amount and the labeled amount of the channel gain of each of the initial channel gain prediction models, the cumulative deviation and the sampling quantity weight, determine the sample sampling rate of each sub-region; A second determination module is used to determine the number of samples in the sub-area based on the total number of samples of the second sample set of the target area and the sample sampling rate of the sub-area; wherein the second sample set includes the geographical location information of the sample points and the marked channel gain; and the total number of samples in the second sample set is the same as the total number of samples in the first sample set; A sampling module, configured to sample sample points in the sub-area according to the sampling quantity of the sub-area to obtain a second subset of the sub-area; the second subset includes geographic location information of the sample points and annotated channel gains; A second training module is used to train a corresponding channel gain prediction model based on the second subset of the sub-region and the first subset; the second training module is also used to train a corresponding channel gain prediction model based on the second subset of the sub-region and the first subset, including: fusing the first subset and the second subset to obtain a third subset; obtaining the maximum distance from each sample point in the third subset to the corresponding cluster center; obtaining a reference sample point from a sub-region adjacent to the sub-region; when the difference between the distance between the reference sample point and the cluster center and the maximum distance is less than or equal to a preset threshold, adding the reference sample point to the third subset to obtain a fourth subset; and training a corresponding channel gain prediction model based on the fourth subset of the sub-region.

5. A device for generating a channel gain map, characterized in that: The device comprises: A second acquisition module is used to acquire a corresponding third set in the target area; wherein the third set includes a plurality of data points and corresponding geographic location information; A prediction module, configured to input the geographical location information of the location points in the third set into the channel gain prediction model generated according to the steps of the method of any one of claims 1 or 2, and output the channel gain of the location points; A generating module is used to generate a channel gain map of the target area based on the first sample set, the second sample set, and the third set.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 or 2 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 or 2 are implemented.

Images