Feature generation method, apparatus, equipment and storage medium for network field strength prediction

Abstract

This invention belongs to the field of artificial intelligence technology and discloses a method, apparatus, device, and storage medium for generating features for network field strength prediction. The method includes: acquiring network field strength prediction samples; performing label interpolation on the network field strength prediction samples to generate rasterized label feature values; determining basic feature samples based on the network field strength prediction samples and generating basic feature values ​​based on the basic feature samples; generating network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, wherein the network field strength prediction features are used to train a network field strength prediction model. Through the above method, multiple feature values ​​are applied to feature vector extraction to generate features for mobile network field strength prediction, which are then input into the neural network for model training. Compared to a simple feature extraction process, the prediction accuracy is significantly improved, and the model training effect is better.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a feature generation method, apparatus, device, and storage medium for predicting network field strength. Background Technology

[0002] Wireless network planning utilizes methods such as service scenario positioning and user density analysis to seek network deployment solutions that meet requirements and are cost-effective. Link budget is a key foundation for coverage planning, and it requires selecting a propagation model suitable for the actual environment. However, existing propagation models do not mention the method for generating raster labels during the rasterization process of sample data. They typically use a simple mean method to calculate raster label values, which results in a large error between the obtained raster label values ​​and the samples, significantly impacting model training and leading to poor training performance.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this invention is to provide a feature generation method, apparatus, device, and storage medium for network field strength prediction, aiming to solve the technical problem of poor model training performance in existing technologies.

[0005] To achieve the above objectives, this invention provides a feature generation method for network field strength prediction, the method comprising the following steps:

[0006] Obtain network field strength prediction samples;

[0007] The network field strength prediction samples are subjected to label interpolation processing to generate rasterized label feature values;

[0008] Based on the network field strength prediction samples, basic feature samples are determined, and basic feature values ​​are generated based on the basic feature samples.

[0009] Network field strength prediction features are generated based on the rasterized label feature values ​​and basic feature values, and these features are used to train the network field strength prediction model.

[0010] Optionally, before generating the network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, the method further includes:

[0011] The network field strength prediction samples are processed by representation learning to obtain preset ground feature characteristics;

[0012] The step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​includes:

[0013] Network field strength prediction features are generated based on the preset land feature relationship features, rasterized label feature values, and basic feature values.

[0014] Optionally, the step of performing representation learning processing on the network field strength prediction samples to obtain preset land cover relationship features includes:

[0015] Based on the network field strength prediction samples, determine the visibility parameters and category parameters of ground features;

[0016] Preset feature relationship characteristics are determined based on the feature visibility parameters and feature category parameters.

[0017] Optionally, before generating the network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, the following steps are included:

[0018] The terrain type characteristics of the community are determined based on the network field strength prediction samples.

[0019] The step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​also includes:

[0020] Based on the terrain type characteristics of the community, the rasterized label feature values, and the basic feature values, a network field strength prediction feature is generated;

[0021] or,

[0022] Network field strength prediction features are generated based on the preset land feature relationship features, community terrain type features, rasterized label feature values, and basic feature values.

[0023] Optionally, determining the cell terrain type characteristics based on the network field strength prediction samples includes:

[0024] Retrieve preset type information;

[0025] Determine the parameters of the current sample cell based on the network field strength prediction samples;

[0026] The parameters of the current sample cell are matched with preset type information to obtain the cell terrain type features.

[0027] Optionally, the step of performing label interpolation processing on the network field strength prediction samples to generate rasterized label feature values ​​includes:

[0028] The network field strength prediction samples are rasterized to obtain the transmitter coordinates and the raster information of each signal grid.

[0029] Based on the transmitter coordinates and grid information, tag interpolation is performed on each signal grid to obtain rasterized tag feature values.

[0030] Optionally, the step of determining basic feature samples based on the network field strength prediction samples and generating basic feature values ​​based on the basic feature samples includes:

[0031] The network field strength prediction samples are filtered to obtain basic feature samples. The basic feature samples include at least the height difference data between the transmitter and receiver, the distance data between the transmitter and receiver, and the angle data between the transmitter and the signal direction.

[0032] Basic feature values ​​are generated based on the basic feature samples.

[0033] Furthermore, to achieve the above objectives, the present invention also proposes a feature generation device for network field strength prediction, the feature generation device for network field strength prediction comprising:

[0034] The acquisition module is used to acquire network field strength prediction samples;

[0035] The processing module is used to perform label interpolation processing on the network field strength prediction samples to generate rasterized label feature values;

[0036] The processing module is further configured to determine basic feature samples based on the network field strength prediction samples, and generate basic feature values ​​based on the basic feature samples.

[0037] The processing module is further configured to generate network field strength prediction features based on the rasterized label feature values ​​and basic feature values, and the network field strength prediction features are used to train the network field strength prediction model.

[0038] Furthermore, to achieve the above objectives, the present invention also proposes a feature generation device for network field strength prediction. The feature generation device for network field strength prediction includes: a memory, a processor, and a feature generation program for network field strength prediction stored in the memory and executable on the processor. The feature generation program for network field strength prediction is configured to implement the steps of the feature generation method for network field strength prediction as described above.

[0039] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a feature generation program for network field strength prediction, wherein when the feature generation program for network field strength prediction is executed by a processor, it implements the steps of the feature generation method for network field strength prediction as described above.

[0040] This invention obtains network field strength prediction samples; performs label interpolation on the network field strength prediction samples to generate rasterized label feature values; determines basic feature samples based on the network field strength prediction samples, and generates basic feature values ​​based on the basic feature samples; generates network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, and uses these network field strength prediction features to train the network field strength prediction model. Through this method, multiple feature values ​​are applied to feature vector extraction to generate features for mobile network field strength prediction, which are then input into the neural network for model training. Compared to a simple feature extraction process, this significantly improves prediction accuracy and enhances model training performance. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the feature generation device for network field strength prediction in the hardware operating environment involved in the embodiments of the present invention;

[0042] Figure 2 This is a flowchart illustrating the first embodiment of the feature generation method for network field strength prediction of the present invention.

[0043] Figure 3 This is a schematic diagram of a module of an embodiment of the feature generation method for network field strength prediction of the present invention;

[0044] Figure 4 This is a schematic diagram of the sampling point distribution in an embodiment of the feature generation method for network field strength prediction of the present invention;

[0045] Figure 5 This is a schematic diagram of engineering parameters of an embodiment of the feature generation method for network field strength prediction of the present invention;

[0046] Figure 6 This is a schematic diagram of the orientation angle of an embodiment of the feature generation method for network field strength prediction of the present invention;

[0047] Figure 7 This is a flowchart illustrating the second embodiment of the feature generation method for network field strength prediction of the present invention.

[0048] Figure 8 This is a schematic diagram of EMD histogram calculation in an embodiment of the feature generation method for network field strength prediction of the present invention.

[0049] Figure 9 This is a structural block diagram of the first embodiment of the feature generation device for network field strength prediction of the present invention.

[0050] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0052] Reference Figure 1 , Figure 1 This is a schematic diagram of the feature generation device structure for network field strength prediction in the hardware operating environment involved in the embodiments of the present invention.

[0053] like Figure 1 As shown, the feature generation device for network field strength prediction may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0054] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the feature generation device for network field strength prediction, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0055] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a feature generation program for network field strength prediction.

[0056] exist Figure 1In the feature generation device for network field strength prediction shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the feature generation device for network field strength prediction of the present invention can be set in the feature generation device for network field strength prediction. The feature generation device for network field strength prediction calls the feature generation program for network field strength prediction stored in the memory 1005 through the processor 1001 and executes the feature generation method for network field strength prediction provided in the embodiment of the present invention.

[0057] This invention provides a feature generation method for network field strength prediction, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of a feature generation method for predicting network field strength according to the present invention.

[0058] In this embodiment, the feature generation method for network field strength prediction includes the following steps:

[0059] Step S10: Obtain network field strength prediction samples.

[0060] It should be noted that the execution subject of this embodiment is a smart terminal with feature extraction function. The smart terminal can be a server, a computer, or other smart devices with the same or similar functions as a server.

[0061] It should be noted that this embodiment applies to the feature extraction process of sample data before the network field strength prediction model. By optimizing the feature extraction process of sample data, it ensures that the feature data input to the model to be trained is more comprehensive and reliable. In order to overcome the defects in the prior art, this embodiment provides a feature generation method for mobile network field strength prediction, realizing feature extraction of the original sample data and generating feature engineering that can be applied to deep learning networks and has a strong correlation with mobile network field strength prediction. This is because feature engineering occupies an important position in deep learning neural networks, and its purpose is to extract features from the original data to the maximum extent for use by the algorithm and model. Previous feature selection methods mostly adopted a single method, and the correlation between feature engineering and label values ​​was low, resulting in poor model training effect. However, the feature engineering proposed in this embodiment comprehensively analyzes the dispersion of features, their correlation with the predictor variable, and their importance to the predictor variable, making the feature selection process more reasonable and the training results more accurately close to the predicted values, achieving good training effect.

[0062] In specific implementations, such as Figure 3As shown, this embodiment uses a deep learning neural network model to feed feature vectors into the network's input layer. For example, in a deep learning neural network, the layers are fully connected. Assuming the model has three hidden layers, each neuron in the next layer is connected to all neurons in the current input layer. Features are synthesized in this way, passing through three hidden layers until the output layer to obtain the field strength prediction value applied to the mobile network. Based on the training process of the above-described example network model, an optimized sample data feature extraction process is proposed.

[0063] It should be noted that the network field strength prediction sample contains a large amount of data, such as: the spatial location of the base station on the map, the spatial location of each signal detection point, the terrain in the detection area, the type of each block in the detection area (divided into grids in the sample), and the signal status of each signal cell, etc.

[0064] Step S20: Perform label interpolation processing on the network field strength prediction samples to generate rasterized label feature values.

[0065] It should be noted that the label interpolation process consists of two parts. First, it's necessary to set the grid number and grid center coordinates. Based on the base station's location, a sample area is defined centered on the base station. The signal region in the network field strength prediction sample centered on the base station is divided. If a signal grid already exists, it needs to be integrated and adjusted to fit the grid size of the model to be trained. Then, the center coordinates of each grid are used as the grid coordinates. Second, it's also necessary to calculate the grid label value and associate it with the corresponding grid number. The grid label value is the influence factor of the N sample points within the grid on the field strength value at the grid center point. Therefore, it can be seen that the rasterized label feature value is used to reflect the signal characteristics of each area surrounding the base station. Figure 3 The label interpolation generation module performs the processing.

[0066] In this embodiment, the network field strength prediction samples are rasterized to obtain the transmitter coordinates and raster information of each signal raster; based on the transmitter coordinates and raster information, label interpolation is performed on each signal raster to obtain rasterized label feature values. Here, the transmitter is the signal base station.

[0067] In its specific implementation, this embodiment proposes a preferred scheme for generating rasterized label feature values ​​and uses it as an example for illustration. For instance: First, the map is rasterized with the base station as the center, with a resolution of 5m × 5m. Based on the geographic coordinates of the sample values, the label of the raster where each sample value is located is determined. Let the coordinates of the cell center point be (cell_x, cell_y), r be the preset cell coverage radius, and x be the preset raster size (default 5m), then the coordinates of the upper left corner of the cell coverage area are:

[0068] x_min=cell_x-r

[0069] y_min = cell_y - r

[0070] The raster cells are labeled 0, 1, ..., (p×p-1) from left to right and top to bottom (assuming p = 2r / x, rounded up). Based on the geographic coordinates of the sample values ​​and the preset raster size, the raster cell number for each sample value is determined as follows: Assuming the coordinates of the sample value are (point_x...). j ,point_y j If the sample value is located in a grid cell, then the grid cell number i is:

[0071] i = p * (floor((point_y) j -y_min) / x))+floor((point_x j -x_min) / x)

[0072] The coordinates of the i-th grid cell are the center coordinates of the i-th grid cell (grid_x). i grid_y i The calculation formula is:

[0073] grid_x i = x_min + x*(i%p+1 / 2)

[0074] grid_y i = y_min + x * (floor(i / p) + 1 / 2)

[0075] Here, % represents the remainder operation, and floor represents the floor operation. The process of rasterizing the samples is completed in the above way, and each grid is numbered.

[0076] Furthermore, based on this, it is necessary to further calculate the raster label value. The influence factor of N sample points within the raster on the field strength value at the raster center point is calculated using a label interpolation algorithm. The field strength label value at the raster center point is then obtained through interpolation based on this influence factor. The interpolation calculation uses the BiCubic basis function.

[0077]

[0078] Where: a is -0.5.

[0079] When using the label interpolation method, the x and y coordinates of the sample points are calculated separately, and the field strength label value A(point_x) at the center point of the sample is calculated. i ,point_y i The corresponding x-coordinate coefficient is k_x i The coefficient of the ordinate is k_yi The calculation formula is as follows:

[0080]

[0081] Similarly, the coefficients corresponding to all sample points are obtained. The weight of each sample point can be obtained by multiplying the weight coefficients of the horizontal and vertical axes. Finally, the field strength label value of the grid center point is obtained by the summation formula, as follows:

[0082]

[0083] in, Let B(X,Y) be the field strength label value of the sample point, and B(X,Y) be the field strength label value of the grid center point, where k_x is the field strength label value. i The weights on the x-coordinate of the sample points are represented by k_y. i This represents the weight on the y-axis of the sample point. For example... Figure 4 As shown, the large point B represents the desired center point of the grid, and the randomly distributed points A1 to A5 around it represent sample points within that grid point.

[0084] It should be noted that this embodiment proposes to calculate label features through a label interpolation algorithm. The calculated target value has a smaller error, and the prediction result when sent to the neural network for model training is closer to the true value, which improves the model training effect and makes the predicted value more realistic when the model is used.

[0085] Step S30: Determine basic feature samples based on the network field strength prediction samples, and generate basic feature values ​​based on the basic feature samples.

[0086] It should be noted that, Figure 3 The basic feature module processes fundamental feature values. These features are constructed by combining engineering parameters and map data from the network field strength prediction samples to create grid height, angle, and distance features, which serve as the data foundation for basic feature processing. The spatial relationship between each sampling point and the base station in the sample data is also considered. This is because the relative spatial relationship between the transmitter (base station) and receiver (signal sampling point) affects signal strength. For example, the relative height difference (elevation angle) affects signal propagation because ground reflection can amplify or cancel the signal. Such effects can be accurately predicted using basic feature values.

[0087] In this embodiment, the network field strength prediction samples are filtered to obtain basic feature samples. The basic feature samples include at least the height difference data between the transmitter and receiver, the distance data between the transmitter and receiver, and the angle data between the transmitter and the signal direction. Basic feature values ​​are generated based on the basic feature samples.

[0088] It should be noted that the necessary data needs to be filtered out based on the network field strength prediction samples. For example, engineering parameters and geographical data should be used to extract profile features, height features, distance features, and angle features. Figure 5 As shown. Further feature extraction is performed. The basic feature samples at least include the height difference data between the transmitter and receiver, the distance data between the transmitter and receiver, and the angle data between the transmitter and the signal direction. Other parameters may also be included; this embodiment does not limit this. The angle data between the transmitter and the signal line direction is the angle between the grid center point and the signal strength, such as... Figure 6 The horizontal direction angle MsAzimuth.

[0089] Among them, the signal direction angle data is Figure 6 In this context, the antenna horizontal azimuth angle, the user horizontal azimuth angle, and the angle between the grid center point and true north are all considered.

[0090] In its specific implementation, this embodiment combines Figure 5 An optimal scheme for obtaining basic feature values ​​is proposed and illustrated using this as an example, as follows: First, feature extraction is performed based on the transmitter, receiver, and relevant profile height: Figure 5 As shown, this mainly includes the height characteristics of the base station location and the grid center point, as well as the height characteristics of the line connecting the two. The specific formula is as follows:

[0091] TRRelHeight=TxHeight+TxAltitude-TRHeight-TRAltitude

[0092] TxHeightProfile1=TxHeight+TxAltitude-TxHeight1-TxAltitude1

[0093] TxHeightProfile2=TxHeight+TxAltitude–TxHeight2–TxAltitude2

[0094] RxHeightProfile1=TxHeight+TxAltitude-RxHeight1-RxAltitude1

[0095] RxHeightProfile2=TxHeight+TxAltitude–RxHeight2–RxAltitude2

[0096] TRMaxEdgeHeight=TxHeight+TxAltitude–MaxHeight–MaxAltitude

[0097] Where: TRRelHeight is the relative height difference between the base station and the grid center point, TxHeight is the height of the base station, because the base station itself is generally a base station tower, and its own height needs to be considered, TxAltitude is the altitude of the base station, TRHeight is the actual height of the grid center point, and TRAltitude is the altitude of the grid center point.

[0098] TxHeightProfile1 is the height of the sampling point 5m away from the base station on the line connecting the base station and the center point of the grid; TxHeight1 is the height of the building where the sampling point 5m away from the base station is located; TxAltitude1 is the altitude of the sampling point 5m away from the base station.

[0099] TxHeightProfile2 is the height of the sampling point 10m away from the base station on the line connecting the base station and the grid center point; TxHeight2 is the height of the building where the sampling point 10m away from the base station is located; TxAltitude2 is the altitude of the sampling point 10m away from the base station.

[0100] RxHeightProfile1 is the height of the sampling point 5m away from the grid center point on the line connecting the base station and the grid center point; RxHeight1 is the height of the building where the sampling point 5m away from the grid center point is located; RxAltitude1 is the altitude of the sampling point 5m away from the grid center point.

[0101] RxHeightProfile2 is the height of the sampling point 10m away from the grid center point on the line connecting the base station and the grid center point; RxHeight2 is the height of the building where the sampling point 10m away from the grid center point is located; RxAltitude2 is the altitude of the sampling point 10m away from the grid center point.

[0102] TRMaxEdgeHeight is the relative height of the highest peak on the line connecting the base station and the grid center point; MaxHeight is the building height at the highest peak on the line connecting the base station and the grid center point; and MaxAltitude is the altitude of the highest peak on the line connecting the base station and the grid center point.

[0103] Furthermore, feature extraction is required for the distances between the transmitter, receiver, and related profiles: Figure 5 As shown, this specifically includes constructing the 2D planar distance TRDistance between the grid location of the receiving point and the transmitting point, and the horizontal 2D distance EdgeDistanceToRx between the highest peak of the receiving connection and the receiving point. Where (CellX, CellY) represents the geographical location of the base station, (X, Y) represents the coordinates of the grid center point, and (X_max, Y_max) represents the grid location of the highest peak of the transmitting and receiving connection. The specific calculation formula is as follows:

[0104]

[0105]

[0106] Finally, feature extraction is required between the grid center point and the signal angle: such as Figure 6 As shown, this mainly includes the angle MsAzimuth between the grid center point and the direction of signal strength, and the elevation angle MsElevation between the grid center point and the top of the base station. The deflection angle between the direction of signal strength and true north is introduced, i.e., the antenna horizontal angle Azimuth, and the angle between the grid center point and true north is θ. The specific formula is as follows:

[0107]

[0108] MsAzimuth=θ-Azimuth

[0109] MsElevation=arctan(TRRelHeight / TRDistance)

[0110] Normalizing the above linear features (height, distance, and angle features) using a normal distribution can make the feature distribution more uniform, thus accelerating the convergence speed of the model. The specific formula for normalization is as follows:

[0111]

[0112] Where: x represents the sample value, i.e., the linear feature vector mentioned above; μ is the sample mean; σ is the sample standard deviation; and z is the result of normalization. The basic feature values ​​are obtained by normalizing the above linear features (height feature, distance feature, and angle feature). This is not intended to limit this embodiment; the basic feature values ​​can also be obtained directly without normalization, but the results will be relatively worse.

[0113] Step S40: Generate network field strength prediction features based on the rasterized label feature values ​​and basic feature values. The network field strength prediction features are used to train the network field strength prediction model.

[0114] It should be noted that the network field strength prediction features generated based on the rasterized label feature values ​​and basic feature values ​​are sufficient to train the network field strength prediction model. By introducing spatial location relationships and the label values ​​of each grid, the trained network field strength prediction model can accurately predict the network field strength of base stations under terrain conditions.

[0115] It should be noted that performing raster digitization on the original sample data based on the label interpolation algorithm and generating the corresponding mobile network field strength values ​​for each raster can reduce the error of field strength value averaging, resulting in more accurate training results. Furthermore, extracting profile features, height features, distance features, and angle features from sample data, engineering parameters, and geographical data improves their correlation with the field strength sample values, accelerates model convergence, and yields a more general and accurate intelligent propagation model.

[0116] This embodiment obtains network field strength prediction samples; performs label interpolation on the network field strength prediction samples to generate rasterized label feature values; determines basic feature samples based on the network field strength prediction samples, and generates basic feature values ​​based on the basic feature samples; generates network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, and uses these network field strength prediction features to train the network field strength prediction model. By applying multiple feature values ​​to feature vector extraction using the above method, features for mobile network field strength prediction are generated and input into the neural network for model training. Compared to a simple feature extraction process, the prediction accuracy is significantly improved, and the model training effect is better.

[0117] refer to Figure 7 , Figure 7 This is a flowchart illustrating the second embodiment of a feature generation method for predicting network field strength according to the present invention.

[0118] Based on the first embodiment described above, the feature generation method for network field strength prediction in this embodiment further includes, before and after step S40:

[0119] Step S401: Determine the terrain type characteristics of the cell based on the network field strength prediction samples.

[0120] It should be noted that different terrains can also affect signal reception and propagation. Therefore, extracting the terrain type features of each cell and adding them to the features used for training can help overcome this and improve the accuracy of model predictions.

[0121] In this embodiment, preset type information is obtained; the parameters of the current sample cell are determined based on the network field strength prediction sample; the parameters of the current sample cell are matched with the preset type information to obtain the cell terrain type features.

[0122] It should be noted that, due to the uncertainty of the terrain type of a residential area, different preset residential area types need to be set in advance to help extract the terrain type features. For example... Figure 3 The small-area classification module is used to extract the terrain type features of the small-area.

[0123] In this specific implementation, this embodiment proposes an optimal scheme for extracting cell terrain type features, for example: determining cell terrain category features based on the EMD algorithm. In mobile network field strength prediction, there are some features that neural networks cannot learn but have a significant impact on the prediction results. These features need to be extracted and processed uniformly. Therefore, based on a large amount of sample data and prior knowledge, cells are divided into 20 preset types, and the bulldozer distance algorithm is used to determine the cell terrain type. The specific algorithm is as follows: Figure 8 As shown.

[0124] Where, P = <p1,p2,…,p N >,Q= <q1,q2,…,q N >These represent two different terrain distributions, u i Let X represent the weights of each item, and v represent the weights of each item. j Let Y be the weights of each term. Calculate the similarity between the two terrain distributions using EMD.

[0125]

[0126] Where, p i and q j The flow between them is f ij p i and q j The distance between them is d ij ,and

[0127] ∑ i f ij =v j , ∑ j f ij =u i d ij =abs(p i -q j )

[0128] Based on the above algorithm, the similarity between the community and 20 preset community terrain distribution types can be obtained, and then the distribution type of the community can be determined.

[0129] Applying all the above methods to feature vector extraction generates feature engineering for mobile network field strength prediction. This feature is then input into the neural network for model training. Compared to simple feature engineering, the prediction accuracy is significantly improved, and the model training effect is better. Extracting the cell terrain category feature based on the EMD algorithm, which has a strong correlation with mobile network field strength prediction, results in higher model training accuracy and better training effect.

[0130] Step S41: Generate network field strength prediction features based on the terrain type features of the community, rasterized label feature values, and basic feature values; or, generate network field strength prediction features based on the preset land feature relationship features, terrain type features of the community, rasterized label feature values, and basic feature values.

[0131] It should be noted that the terrain type features of the residential area can be combined with the above-mentioned rasterized label feature values ​​and basic feature value generation network field strength prediction features as feature inputs for training the network model. Alternatively, the model can be trained based on the preset land feature relationship features, residential area terrain type features, rasterized label feature values, and basic feature value generation network field strength prediction features.

[0132] Among them, the preset ground feature relationship features are the pre-acquired spatial relationships between the transmitting and receiving ends. These can be directly input or based on... Figure 3 The learning module extracts features from the network field strength prediction samples. Similarly, preset land cover relationship features can be combined with rasterized label feature values ​​and basic feature values ​​as features for training the network model.

[0133] In this embodiment, the network field strength prediction samples are processed by representation learning to obtain preset land feature relationship features; the step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​includes: generating network field strength prediction features based on the preset land feature relationship features, rasterized label feature values ​​and basic feature values.

[0134] Understandably, the network field strength prediction samples undergo representation learning processing, primarily to characterize ground feature relationships using representation learning methods. Ground feature relationships refer to the distribution of ground features between the base station and the actual signal receiving grid, such as oceans, wetlands, lakes, and forests, which affect signal propagation efficiency.

[0135] In this embodiment, the visibility parameters and category parameters of ground features are determined based on the network field strength prediction samples; and the preset ground feature relationship features are determined based on the visibility parameters and category parameters of ground features.

[0136] It should be noted that this embodiment proposes a preferred scheme for extracting ground feature relationship features. For example: First, considering the three-dimensional terrain structure and combining it with a ray tracing algorithm, it determines whether there is visibility between the transmitting point and the receiving grid, obtaining the visibility parameter of the ground features, and representing this determination result using a local representation method. The local representation can usually be represented in the form of a One-Hot vector. Second, based on the distribution of ground features between the base station and the actual signal receiving grid, as shown in Table 1, ground feature vectors are extracted and represented using a distributed representation learning method. The distributed representation can usually be represented as a low-dimensional dense vector, that is, using a neural network to map the high-dimensional local representation space to a very low-dimensional distributed representation space. In this low-dimensional space, each feature is no longer a point on the coordinate axis, but is dispersed throughout the low-dimensional space, maintaining the topological relationship between different feature values.

[0137] This embodiment provides a classification method for land feature indexes, with a total of 20 categories. Each index value represents a different land feature category, as shown in Table 1.

[0138] It should be noted that, based on the aforementioned distance, the distributed representation learning method in representation learning records the land feature index features along the line connecting the transmitter and receiver grid points, and then uses the distributed representation learning method to map the land feature indexes into 6-dimensional dense vectors. This distributed representation method enhances the representational power of land feature index features, reduces the dimensionality of the vectors, solves the high-dimensional sparsity problem, and improves the generalization ability of neural networks under sparse input data.

[0139] Table 1: Meaning of Geographic Feature Index

[0140]

[0141] It should be noted that the occlusion judgment result is represented by a local representation learning method. This discrete representation has good interpretability and is very computationally efficient when used in linear models.

[0142] This embodiment determines the cell terrain type features based on the network field strength prediction samples; it generates network field strength prediction features based on the cell terrain type features, rasterized label feature values, and basic feature values; or, it generates network field strength prediction features based on the preset land feature relationship features, cell terrain type features, rasterized label feature values, and basic feature values. By extracting multiple features and using them to train the network field strength prediction model, the adaptability of the network field strength prediction model is improved, making the predictions no longer affected by the regional environment, and maintaining high accuracy even in complex terrain or obstacle situations.

[0143] Furthermore, this embodiment of the invention also proposes a storage medium storing a feature generation program for network field strength prediction. When the feature generation program for network field strength prediction is executed by a processor, it implements the steps of the feature generation method for network field strength prediction as described above.

[0144] Reference Figure 9 , Figure 9 This is a structural block diagram of the first embodiment of the feature generation device for network field strength prediction of the present invention.

[0145] like Figure 9 As shown, the feature generation device for network field strength prediction proposed in this embodiment of the invention includes an acquisition module 10, which is used to acquire network field strength prediction samples.

[0146] The processing module 20 is used to perform label interpolation processing on the network field strength prediction samples to generate rasterized label feature values.

[0147] The processing module 20 is further configured to determine basic feature samples based on the network field strength prediction samples, and generate basic feature values ​​based on the basic feature samples.

[0148] The processing module 20 is further configured to generate network field strength prediction features based on the rasterized label feature values ​​and basic feature values, and the network field strength prediction features are used to train the network field strength prediction model.

[0149] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0150] In this embodiment, the acquisition module 10 acquires network field strength prediction samples; the processing module 20 performs label interpolation on the network field strength prediction samples to generate rasterized label feature values; the processing module 20 determines basic feature samples based on the network field strength prediction samples and generates basic feature values ​​based on the basic feature samples; the processing module 20 generates network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, and the network field strength prediction features are used to train the network field strength prediction model. Through the above method, multiple feature values ​​are applied to feature vector extraction to generate features for mobile network field strength prediction, which are then input into the neural network for model training. Compared to a simple feature extraction process, the prediction accuracy is significantly improved, and the model training effect is better.

[0151] In one embodiment, the processing module 20 is further configured to perform representation learning processing on the network field strength prediction samples to obtain preset ground feature relationship characteristics;

[0152] The step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​includes:

[0153] Network field strength prediction features are generated based on the preset land feature relationship features, rasterized label feature values, and basic feature values.

[0154] In one embodiment, the processing module 20 is further configured to determine the ground feature visibility parameters and ground feature category parameters based on the network field strength prediction samples;

[0155] Preset feature relationship characteristics are determined based on the feature visibility parameters and feature category parameters.

[0156] In one embodiment, the processing module 20 is further configured to determine the cell terrain type characteristics based on the network field strength prediction samples;

[0157] Based on the terrain type characteristics of the community, the rasterized label feature values, and the basic feature values, a network field strength prediction feature is generated;

[0158] or,

[0159] Network field strength prediction features are generated based on the preset land feature relationship features, community terrain type features, rasterized label feature values, and basic feature values.

[0160] In one embodiment, the processing module 20 is further configured to acquire preset type information;

[0161] Determine the parameters of the current sample cell based on the network field strength prediction samples;

[0162] The parameters of the current sample cell are matched with preset type information to obtain the cell terrain type features.

[0163] In one embodiment, the processing module 20 is further configured to perform rasterization processing on the network field strength prediction sample to obtain the transmitter coordinates and raster information of each signal raster.

[0164] Based on the transmitter coordinates and grid information, tag interpolation is performed on each signal grid to obtain rasterized tag feature values.

[0165] In one embodiment, the processing module 20 is further configured to filter the network field strength prediction samples to obtain basic feature samples, wherein the basic feature samples include at least the height difference data between the transmitting end and the receiving end, the distance data between the transmitting end and the receiving end, and the angle data between the transmitting end and the signal direction.

[0166] Basic feature values ​​are generated based on the basic feature samples.

[0167] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0168] In addition, for technical details not described in detail in this embodiment, please refer to the feature generation method for network field strength prediction provided in any embodiment of the present invention, which will not be repeated here.

[0169] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0170] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0172] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A feature generation method for predicting network field strength, characterized in that, The feature generation method for network field strength prediction includes: Obtain network field strength prediction samples; The network field strength prediction samples are subjected to label interpolation to generate rasterized label feature values. The rasterized label feature values ​​are calculated by using a label interpolation algorithm to calculate the influence factors of N sample points in the raster on the field strength value of the raster center point. The field strength label value of the raster center point is obtained by interpolation based on the influence factors. Based on the network field strength prediction samples, basic feature samples are determined, and basic feature values ​​are generated based on the basic feature samples. Network field strength prediction features are generated based on the rasterized label feature values ​​and basic feature values, and these features are used to train the network field strength prediction model. The step of determining basic feature samples based on the network field strength prediction samples and generating basic feature values ​​based on the basic feature samples includes: The network field strength prediction samples are filtered to obtain basic feature samples. The basic feature samples include at least the height difference data between the transmitter and receiver, the distance data between the transmitter and receiver, and the angle data between the transmitter and the signal direction. Basic feature values ​​are generated based on the basic feature samples.

2. The method as described in claim 1, characterized in that, Before generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values, the process also includes: The network field strength prediction samples are processed by representation learning to obtain preset ground feature characteristics; The step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​includes: Network field strength prediction features are generated based on the preset land feature relationship features, rasterized label feature values, and basic feature values.

3. The method as described in claim 2, characterized in that, The process of performing representation learning on the network field strength prediction samples to obtain preset land cover relationship features includes: Based on the network field strength prediction samples, determine the visibility parameters and category parameters of ground features; Preset feature relationship characteristics are determined based on the feature visibility parameters and feature category parameters.

4. The method according to any one of claims 1 to 3, characterized in that, Before generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values, the process includes: The terrain type characteristics of the community are determined based on the network field strength prediction samples. The step of generating network field strength prediction features based on the rasterized label feature values ​​and basic feature values ​​also includes: Based on the terrain type characteristics of the community, the rasterized label feature values, and the basic feature values, a network field strength prediction feature is generated; or, Network field strength prediction features are generated based on preset land feature relationship features, community terrain type features, rasterized label feature values, and basic feature values.

5. The method as described in claim 4, characterized in that, The step of determining the cell terrain type characteristics based on the network field strength prediction samples includes: Retrieve preset type information; Determine the parameters of the current sample cell based on the network field strength prediction samples; The parameters of the current sample cell are matched with preset type information to obtain the cell terrain type features.

6. The method according to any one of claims 1 to 3, characterized in that, The step of generating rasterized label feature values ​​by performing label interpolation on the network field strength prediction samples includes: The network field strength prediction samples are rasterized to obtain the transmitter coordinates and the raster information of each signal grid. Based on the transmitter coordinates and grid information, tag interpolation is performed on each signal grid to obtain rasterized tag feature values.

7. A feature generation device for predicting network field strength, characterized in that, The feature generation device for network field strength prediction includes: The acquisition module is used to acquire network field strength prediction samples; The processing module is used to perform label interpolation processing on the network field strength prediction samples to generate rasterized label feature values. The rasterized label feature values ​​are calculated by using a label interpolation algorithm to calculate the influence factors of N sample points in the raster on the field strength value of the raster center point. The field strength label value of the raster center point is obtained by interpolation based on the influence factors. The processing module is further configured to determine basic feature samples based on the network field strength prediction samples, and generate basic feature values ​​based on the basic feature samples. The processing module is further configured to generate network field strength prediction features based on the rasterized label feature values ​​and basic feature values, and the network field strength prediction features are used to train the network field strength prediction model. The processing module is also used to filter the network field strength prediction samples to obtain basic feature samples. The basic feature samples include at least the height difference data between the transmitter and receiver, the distance data between the transmitter and receiver, and the angle data between the transmitter and the signal direction. Basic feature values ​​are generated based on the basic feature samples.

8. A feature generation device for predicting network field strength, characterized in that, The device includes: a memory, a processor, and a feature generation program for network field strength prediction stored in the memory and executable on the processor, the feature generation program for network field strength prediction being configured to implement the steps of the feature generation method for network field strength prediction as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a feature generation program for network field strength prediction, which, when executed by a processor, implements the steps of the feature generation method for network field strength prediction as described in any one of claims 1 to 6.

Images