Wireless parameter optimization method, device, electronic device and computer program product

Abstract

The present disclosure relates to wireless parameter optimization methods, devices, electronic devices and computer program products. The above-mentioned wireless parameter optimization method includes: obtaining multimodal data corresponding to multiple cells to be optimized; inputting the multiple multimodal data into the scene recognition model respectively, and determining the target scene categories corresponding to the multiple cells to be optimized; based on the target scene categories corresponding to each cell to be optimized, performing parameter optimization on the multiple cells to be optimized in parallel in the parameter optimization module, and obtaining the target parameter configurations corresponding to the multiple cells to be optimized. The present disclosure introduces multimodal data describing the wireless network communication conditions, two-dimensional signal distribution conditions and geographical environment of the cell to be optimized in the scene recognition process, so as to more comprehensively describe the actual operating environment of the wireless cell, improve the accuracy of scene recognition and the subsequent parameter optimization effect.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technologies, and in particular to a method, device, electronic device, and computer program product for wireless parameter optimization. Background Art

[0002] With the rapid development of wireless communication services, the number of users and traffic carried by wireless cells has increased significantly, resulting in high base station loads. Therefore, optimizing wireless parameters for wireless cells has become a key measure to improve communication network performance.

[0003] Related technologies often analyze the scenario characteristics of wireless cells based on wireless network communication data and optimize wireless parameters based on these characteristics. However, relying solely on wireless network communication data during scenario analysis cannot fully describe the actual operating environment of the wireless cell, resulting in low scenario recognition accuracy and, in turn, affecting the effectiveness of wireless parameter optimization. Summary of the Invention

[0004] In view of this, exemplary embodiments of the present disclosure provide a wireless parameter optimization method, apparatus, electronic device, and computer program product to solve the problems existing in the related art.

[0005] According to a first aspect of the exemplary embodiments of the present disclosure, a method for optimizing wireless parameters is provided, the method comprising:

[0006] Acquire multimodal data corresponding to a plurality of cells to be optimized; the multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions, and geographical environment of the cells to be optimized;

[0007] Inputting the plurality of multimodal data into a scene recognition model respectively to determine target scene categories corresponding to the plurality of cells to be optimized; the scene recognition model includes a scene coding submodel and a scene clustering module, the scene coding submodel is used to perform data coding processing on the multimodal data to obtain scene codes, and the scene clustering module is used to determine the target scene category of the cell to be optimized based on the scene codes;

[0008] Based on the target scenario categories respectively corresponding to the cells to be optimized, the parameter optimization module performs parameter optimization on the multiple cells to be optimized in parallel to obtain target parameter configurations respectively corresponding to the multiple cells to be optimized.

[0009] According to a second aspect of the exemplary embodiments of the present disclosure, a wireless parameter optimization device is provided, the device comprising:

[0010] A data acquisition module is used to acquire multimodal data corresponding to a plurality of cells to be optimized; the multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions and geographical environment of the cells to be optimized;

[0011] a data processing module, configured to input the plurality of multimodal data into a scene recognition model respectively, and determine the target scene categories corresponding to the plurality of cells to be optimized; the scene recognition model includes a scene coding submodel and a scene clustering module, the scene coding submodel is configured to perform data coding processing on the multimodal data to obtain a scene code, and the scene clustering module is configured to determine the target scene category of the cell to be optimized based on the scene code;

[0012] The data processing module is further used to perform parameter optimization on the multiple cells to be optimized in parallel in the parameter optimization module based on the target scenario categories corresponding to each cell to be optimized, so as to obtain the target parameter configurations corresponding to the multiple cells to be optimized.

[0013] According to a third aspect of the exemplary embodiments of the present disclosure, there is provided an electronic device, including:

[0014] at least one processor;

[0015] A memory for storing instructions executable by at least one processor; wherein the at least one processor is configured to execute the instructions to implement the method described in the exemplary embodiments of the present disclosure.

[0016] According to a fourth aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, the method described in the exemplary embodiments of the present disclosure is implemented.

[0017] At least one of the above technical solutions adopted by the exemplary embodiments of the present disclosure can achieve the following beneficial effects: obtaining multimodal data corresponding to multiple cells to be optimized; the multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions and geographical environment of the cells to be optimized; the multiple multimodal data are respectively input into the scene recognition model to determine the target scene categories corresponding to the multiple cells to be optimized; the scene recognition model includes a scene coding submodel and a scene clustering module, the scene coding submodel is used to perform data encoding processing on the multimodal data to obtain scene coding, and the scene clustering module is used to determine the target scene category of the cell to be optimized based on the scene coding; based on the target scene categories corresponding to each cell to be optimized, the parameter optimization module performs parameter optimization on the multiple cells to be optimized in parallel to obtain the target parameter configurations corresponding to the multiple cells to be optimized. The present disclosure introduces multimodal data describing the wireless network communication conditions, two-dimensional signal distribution conditions and geographical environment of the cell to be optimized in the scene recognition process, thereby more comprehensively describing the actual operating environment of the wireless cell, improving the accuracy of scene recognition and the subsequent parameter optimization effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the exemplary embodiments of the present disclosure, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A flowchart of a wireless parameter optimization method provided by an exemplary embodiment of the present disclosure;

[0020] Figure 2 A training flow chart of a wireless parameter optimization method provided by an exemplary embodiment of the present disclosure;

[0021] Figure 3 This is a schematic diagram of rasterization of drive test data provided by an exemplary embodiment of the present disclosure;

[0022] Figure 4 This is a schematic diagram of satellite map semantic segmentation provided as an example in the present disclosure;

[0023] Figure 5 This is a schematic diagram of the structure of multimodal scene recognition provided by an exemplary embodiment of the present disclosure;

[0024] Figure 6 This is a schematic diagram of the structure of an online parameter optimization module provided as an example in the present disclosure;

[0025] Figure 7 This is a flow chart of an exemplary parameter online optimization module provided in the present disclosure;

[0026] Figure 8 A flow chart showing a wireless parameter optimization method according to an exemplary embodiment of the present disclosure;

[0027] Figure 9 This is a schematic block diagram of the functional modules of a wireless parameter optimization device provided by an exemplary embodiment of the present disclosure;

[0028] Figure 10 A schematic block diagram of a chip provided as an example in the present disclosure;

[0029] Figure 11 This is a structural block diagram of an exemplary electronic device provided in the present disclosure. DETAILED DESCRIPTION

[0030] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0031] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0032] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0033] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0034] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0035] It is understandable that before using the technical solutions disclosed in the various embodiments of this disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0036] For example, in response to a user's active request, a prompt message is sent to the user to clearly inform the user that the operation requested will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the electronic device, application, server, storage medium, or other software or hardware that performs the operations of the disclosed technical solution based on the prompt message.

[0037] As an optional but non-limiting implementation method, in response to receiving the user's active request, the method of sending a prompt message to the user can be, for example, a pop-up window, and the prompt message can be presented in the form of text in the pop-up window. In addition, the pop-up window can also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device. It is understandable that the above notification and the process of obtaining user authorization are only illustrative and do not constitute a limitation on the implementation method of the present disclosure. Other methods that meet relevant laws and regulations can also be applied to the implementation method of the present disclosure.

[0038] Before introducing the embodiments of the present disclosure, the following definitions are given for the relevant terms involved in the embodiments of the present disclosure:

[0039] Pseudo-Mercator projection: Pseudo-Mercator projection is a projection used to map the three-dimensional earth surface onto a two-dimensional plane.

[0040] SegFormer (Semantic Segmentation): SegFormer is a semantic segmentation network model based on the Transformer architecture. It leverages the Transformer's multi-head attention mechanism to better capture global and long-range dependencies in images. Furthermore, SegFormer employs a shallow-to-deep feature pyramid structure to effectively fuse features at different scales. This improves semantic segmentation accuracy while capturing both detailed image information and global semantic information.

[0041] K-Means (K-Means Clustering) Algorithm: K-Means is a commonly used unsupervised clustering algorithm. Its core concept is to cluster sample points around K cluster centers, minimizing the distance between each sample point and its cluster center. The algorithm first randomly selects K points as initial cluster centers and then iteratively updates the cluster assignment of each sample point and the location of the cluster center until the cluster center stops changing or the maximum number of iterations is reached.

[0042] With the rapid development of wireless communication services, the number of users and traffic carried by wireless cells has increased significantly, resulting in high base station loads. Therefore, optimizing wireless parameters for wireless cells has become a key measure to improve communication network performance.

[0043] Related technologies often analyze the scenario characteristics of wireless cells based on wireless network communication data and optimize wireless parameters based on these characteristics. However, relying solely on wireless network communication data during scenario analysis cannot fully describe the actual operating environment of the wireless cell, resulting in low scenario recognition accuracy and, in turn, affecting the effectiveness of wireless parameter optimization.

[0044] Therefore, in order to solve the above problems, the exemplary embodiments of the present disclosure provide a wireless parameter optimization method, which obtains full multimodal data by preprocessing the multimodal data of the cell to be optimized, wherein the multimodal data may include wireless network communication data, road test data, and satellite map tile data. Furthermore, based on the full multimodal data, a neural network is trained to generate scene codes corresponding to the cell to be optimized. Then, a machine learning algorithm is used to cluster the scene codes to determine the scene category corresponding to the cell to be optimized. Finally, the parameter online optimization path is determined according to the scene category of the cell to be optimized, and the parameter optimization result of the cell to be optimized is obtained, completing the entire optimization process.

[0045] For example, Figure 1 This is a flow chart of a wireless parameter optimization method provided by an exemplary embodiment of the present disclosure, such as Figure 1 As shown, the method may include a multimodal data processing module 110 , a scene encoding module 120 , a scene clustering module 130 and a parameter online optimization module 140 .

[0046] The multimodal data processing module 110 is used to pre-process the multimodal data of the cell to be optimized to obtain full multimodal data. The multimodal data may include wireless network communication data, drive test data, and satellite map tile data.

[0047] Preprocessing can include: dimensionless or vectorizing wireless network communication data to obtain standardized wireless network communication data; rasterizing drive test data to obtain drive test raster data; and semantic segmenting satellite map tile data to obtain a semantic map. The full multimodal data includes standardized wireless network communication data, drive test raster data, and a semantic map.

[0048] The scene encoding module 120 generates a scene encoding for the cell to be optimized based on the full multimodal data. Specifically, this involves inputting standardized wireless network communication data into a multi-layer perceptron for feature fusion to obtain a northbound data encoding. Then, the drive test raster data is input into a convolutional network for feature extraction to obtain a rasterized drive test data encoding. Finally, the semantic map is input into a convolutional network for feature extraction to obtain a semantic map encoding. Finally, the northbound data encoding, rasterized drive test data encoding, and semantic map encoding are input into a scene encoding network for concatenation to obtain a scene encoding for the cell to be optimized.

[0049] The scene clustering module 130 uses a machine learning scene clustering method to determine the scene category of the cell to be optimized according to the scene coding of the cell to be optimized.

[0050] The online parameter optimization module 140 performs online parameter optimization based on the scenario category of the cell to be optimized, obtaining the parameter optimization results for the cell to be optimized. The online parameter optimization module includes n scenario categories. During the optimization process, the online parameter optimization path is determined based on the scenario category of the cell to be optimized, and the parameter optimization results for the cell to be optimized are obtained.

[0051] For example, Figure 2 This is a training flow chart of an exemplary wireless parameter optimization method provided by the present disclosure, such as Figure 2 As shown, the method may include the following steps:

[0052] The wireless parameter optimization method may include the following steps:

[0053] Step S210: Process the multi-source multi-modal data.

[0054] Illustratively, the multi-source multi-modal data may include wireless network communication data, drive test data, and satellite map tile data corresponding to multiple cells to be optimized.

[0055] Among them, wireless network communication data is usually numerical data or categorical data, which can be divided into four categories: engineering parameter configuration, service traffic, neighboring cell networking and measurement reports.

[0056] For example, for numerical fields, standardization can be used to perform dimensionless processing so that the data conforms to a data distribution with a mean of 0 and a standard deviation of 1. This step can be expressed as:

[0057]

[0058] Among them, μ represents the expectation of the data; σ represents the standard deviation of the data; X standard Represents the standardized data; X represents the numeric field value in the wireless network communication data.

[0059] For example, for categorical fields, the data can be vectorized using OneHot encoding. This step can be expressed as:

[0060]

[0061] Among them, A represents a specific set, if X i is a subset of set A, then the binary bit corresponding to the code is 1, otherwise, the binary bit corresponding to the code is 0; onehot Indicates the encoded vector data, X i Indicates the i-th value of the field value of the category type in the wireless network communication data.

[0062] For example, Table 1 shows the wireless network communication data fields provided by an example of the present disclosure:

[0063] Table 1 Schematic diagram of wireless network communication data fields

[0064]

[0065]

[0066]

[0067] Exemplarily, the drive test data may be subjected to rasterization processing to obtain drive test raster data.

[0068] Specifically, the latitude and longitude coordinates of the measurement points are first mapped onto a two-dimensional plane using a pseudo-Mercator projection to form a grid coordinate system. This step can be expressed as the following formula (3):

[0069]

[0070] Among them, lon represents the longitude of the measurement point; lat represents the latitude of the measurement point; z represents the grid zoom level; x represents the grid column number; y represents the grid row number; c i Indicates the signal value of the measurement point; n indicates the number of measurement points in the grid; C (x，y) Represents the signal measurement value of this raster.

[0071] In this embodiment, the zoom level z can be set to 20, indicating a grid size of 25*25 meters. A signal distribution feature map is constructed by selecting grids within a 600-meter range centered on the wireless cell location. Each pixel in the signal distribution feature map represents a grid, and the pixel value represents the signal measurement value of that grid.

[0072] Furthermore, for multiple measurement points falling within the same grid cell, the average value of their signal strength can be used as the representative measurement value of the grid to obtain the drive test grid data. The drive test grid data can be used to represent the signal strength corresponding to each grid. This step can be expressed as the following formula (4):

[0073]

[0074] Among them, c i Indicates the signal value of the measurement point; n indicates the number of measurement points in the grid; C (x，y) Represents the signal measurement value of this raster.

[0075] For example, Figure 3 This is a schematic diagram of the gridding of drive test data provided by an exemplary embodiment of the present disclosure, such as Figure 3 As shown, the resolution of the signal distribution feature map is 24*24 pixels, each pixel represents a grid, and the pixel value represents the signal measurement value of the grid.

[0076] Based on this, by performing rasterization processing on the drive test data, the drive test data can be converted into a two-dimensional signal distribution map, thereby better reflecting the signal strength distribution within the coverage area.

[0077] For example, satellite map tile data may be used to describe the geographical environment of a cell and its surroundings. Figure 4 This is a schematic diagram of satellite map semantic segmentation provided as an example in the present disclosure. Figure 4 As shown, the pre-processing process may include:

[0078] First, a high-resolution satellite image within a certain range (e.g., 500 meters) centered on the wireless cell location is selected to obtain a 512x512 resolution satellite map tile. The SegFormer semantic segmentation network is then used to perform semantic recognition and segmentation on the input satellite map tile image, yielding an initial semantic map of the same resolution.

[0079] In subsequent scene recognition, important geographic element information such as roads and buildings can be extracted from the semantic map, providing valuable geographic environment description for wireless planning.

[0080] Furthermore, by downsampling, the resolution of the initial semantic map can be reduced from 512*512 to 24*24 to obtain a semantic map. This semantic map maintains the same resolution as the rasterized drive test data mentioned above, facilitating subsequent feature fusion with other data sources (such as wireless network data and drive test data).

[0081] Based on this, wireless communication data can be dimensionlessly or vectorized to obtain standardized wireless network communication data reflecting network performance. Drive test data can be rasterized to obtain geographic features such as signal coverage. Satellite imagery can be semantically segmented to obtain semantic features of geographic environments such as roads and buildings. This multi-source and multi-modal data fusion provides richer input features for subsequent parameter optimization, thereby improving the overall effectiveness of wireless parameter planning and optimization.

[0082] Step S220: Perform multimodal scene recognition on all cells to obtain scene categories corresponding to all cells.

[0083] Multimodal scene recognition consists of two components: a scene encoding network and a machine learning clustering module. Data input is divided into three parts: standardized wireless network communication data, drive test grid data, and semantic maps. Three autoencoders are used to encode the standardized wireless network communication data, drive test grid data, and semantic maps, respectively, to generate corresponding codes. The machine learning clustering module then clusters the encoded data to determine the scene categories corresponding to all cells.

[0084] For example, Figure 5 This is a schematic diagram of the structure of multimodal scene recognition provided by an exemplary embodiment of the present disclosure.

[0085] Specifically, wireless network communication data undergoes dimensionless transformation and OneHot encoding, expanding the data dimension to 108. To reduce the data dimension and extract more effective feature representations, a five-layer symmetrical first autoencoder is used for encoding. The hidden layers of the first autoencoder can include FC-64, FC-32, FC-8, FC-32, and FC-64 neurons, with 64, 32, 8, 32, and 64 neurons, respectively. Reinforced Lu (ReLU) is used as the activation function, and the first autoencoder network structure is symmetrical.

[0086] Exemplarily, the training process of the first autoencoder for encoding wireless network communication data may include:

[0087] First, the standardized wireless network communication data is input into the input layer of the first autoencoder, and then passes through the FC-64, FC-32 and FC-8 encoder parts in sequence, gradually reducing the data dimension from high dimension to low dimension, and finally obtaining an 8-dimensional feature vector.

[0088] Then, the 8-dimensional feature vector is input into the decoder part of the first autoencoder, and passes through FC-32 and FC-64 in sequence, and finally restored to the original high-dimensional input data.

[0089] During the training process, the parameters of the autoencoder are optimized by minimizing the difference between the input data and the restored data (output data), so that it can effectively extract the key features of wireless communication data.

[0090] In order to reduce the amount of computation for subsequent processing, the output of the third hidden layer (FC-8) of the first autoencoder can be selected as the encoding result of the wireless communication data, with a dimension of 8, denoted as H 1 .

[0091] Compared with 64-dimensional features, 8-dimensional features H 1 This can better capture the potential characteristics of wireless communication data, allowing for subsequent fusion with grid drive test data and semantic map data to form a more comprehensive multimodal scene representation. This facilitates subsequent machine learning clustering, improving clustering efficiency and accuracy.

[0092] Based on this, through training, the first autoencoder can learn the ability to compress and encode high-dimensional wireless communication data into 8-dimensional feature vectors. This dimensionality reduction operation not only greatly reduces the computational complexity of subsequent processing, but also 1 This 8-dimensional feature vector also retains the core information of the original data, providing high-quality input features for multimodal scene recognition.

[0093] For example, for the road test grid data, it can be first structured into image data of size 24*24*1, and then the image data is encoded using a second autoencoder containing a 7-layer neural network. The second autoencoder is a convolutional autoencoder, and the specific network structure of the second autoencoder may include:

[0094] The first layer is the convolution layer Conv3-4, which contains four 3*3 convolution kernels with a step size of 2;

[0095] The second convolutional layer, Conv3-8, contains eight 3x3 convolution kernels with a stride of 2. After two layers of convolution, the image data size is reduced to 1 / 4 of its original size. If it is directly connected to a fully connected layer, the input dimension of the fully connected layer will be very high due to the small size of the feature map, resulting in a large number of parameters and prone to overfitting. Therefore, a Flatten layer is added after the second convolutional layer to flatten the two-dimensional feature map into a one-dimensional vector. This significantly reduces the input dimension of the fully connected layer, reduces the number of parameters, and improves the network's generalization ability.

[0096] The third layer is the fully connected layer FC-64, with 64 neurons;

[0097] The fourth layer is the fully connected layer FC-8, with 8 neurons;

[0098] The fifth layer is a fully connected layer FC-64 with 64 neurons. A Reshape layer is added after the fifth fully connected layer to restore the feature vector to the format of a two-dimensional feature map.

[0099] The sixth layer is the transposed convolution layer Deconv3-8, which contains 8 3*3 convolution kernels with a step size of 2;

[0100] The seventh layer is the transposed convolution layer Deconv3-4, which contains four 3*3 convolution kernels with a step size of 2.

[0101] Exemplarily, the training process of the second autoencoder encoding the drive test grid data may include:

[0102] First, the road test grid data is input into the encoder part of the convolutional autoencoder, and then through the convolution of the first and second layers and the pooling of the third and fourth layers, the 8-dimensional feature vector H is extracted. 2 .

[0103] Then, the 8-dimensional feature vector is input into the decoder part. After the fifth layer of pooling, the feature vector is restored to the format of a two-dimensional feature map. After the sixth and seventh layers of transposed convolution, it is restored to the original 24*24*1 image data.

[0104] During the training process, the parameters of the autoencoder are optimized by minimizing the difference between the input image and the restored image, enabling it to efficiently extract the key feature data of the road test raster data.

[0105] Since the H output of the fourth layer (FC-8) 2 It not only reduces the data dimension but also retains important features, which can better extract and characterize the key feature information of the drive test grid data, and can also be better compared with the encoding result H of the wireless communication data in the future. 1 And semantic map data are integrated to form a more comprehensive multimodal scene representation.

[0106] Therefore, the H output of the fourth layer (FC-8) of the second autoencoder can be 2 As the encoding result of the drive test grid data.

[0107] Based on this, by encoding the drive test data, not only the data dimension is greatly reduced, but also H 2 The feature vector also retains the core information of the original data and can provide high-quality feature input for subsequent multimodal scene recognition.

[0108] For example, for a semantic map, it can be first structured into 24*24*1 image data, and then encoded using a third autoencoder containing a 7-layer neural network. The third autoencoder is a convolutional autoencoder, and its specific network structure and training process are the same as those of the second autoencoder. Please refer to the above for details.

[0109] Since the output dimension of the fourth layer (FC-8) is 8, it not only reduces the data dimension but also retains important features, which can better extract and characterize the key feature information of the semantic map data, and can also be better compared with the encoding result H of the wireless communication data. 1 and H 2 Fusion is performed to form a more comprehensive multimodal scene representation.

[0110] Therefore, the H output of the fourth layer (FC-8) of the third autoencoder can be 3 As the encoding result of the drive test grid data.

[0111] For example, the training of the first, second, and third autoencoders can all use the mean square error loss function:

[0112]

[0113] in, Represents the autoencoder reconstruction vector; y i represents the real data vector; N represents the number of data set samples; L mse is the mean square error, L mse The smaller it is, the stronger the reconstruction performance of the autoencoder.

[0114] After obtaining the encoding results of the standardized wireless network communication data, drive test grid data, and semantic map, the three encoding results can be spliced ​​into a vector with a dimension of 24, and this vector is used as the scene code of the cell, denoted as H. This step can be expressed as:

[0115] H=[H 1 , H 2 , H 3 ] (6)

[0116] Among them, H represents the scene code of the cell; H 1 Represents the encoding result of standardized wireless network communication data; H 2 Indicates the encoding result of the drive test grid data; H 3 Represents the encoding result of the semantic map.

[0117] Exemplarily, the machine learning clustering module may use the K-Means algorithm to perform cluster analysis on the scene coding data of all cells. In this embodiment, the silhouette coefficient method is used to determine the optimal number of clusters k.

[0118] Specifically, the number of clusters, k, is given. A K-Means model is then trained using scene coding data from all cells to obtain clustering results corresponding to all cells. Next, z samples are randomly sampled from all cells, and the average silhouette coefficient of these z samples is calculated. This average silhouette coefficient is used as the silhouette coefficient corresponding to the number of clusters, k. Finally, the number of clusters, k, with the largest silhouette coefficient, is selected as the optimal number of clusters and used to train the K-Means model to obtain the scene categories corresponding to all cells.

[0119] For example, the silhouette coefficient calculation formula is as follows:

[0120]

[0121]

[0122] Among them, a i represents the cohesion of sample i; H i represents the scene code of sample i; H j represents the scene code of sample j in the same category as sample i; b i Indicates the separation degree of sample i, and the calculation formula is the same as a i Similarly, we need to traverse all other m clusters to obtain {b1(i), ..., b m (i)}, take the minimum value as b i The final result; n represents the number of data set samples. i Represents the silhouette coefficient of sample i, with a value range of [-1, 1]. The larger the value, the better the clustering effect.

[0123] The average value of the silhouette coefficient is used as the silhouette coefficient corresponding to the number of clusters k. The calculation formula is as follows:

[0124]

[0125] Among them, S i represents the silhouette coefficient of sample i; S represents the silhouette coefficient corresponding to the number of clusters k.

[0126] For example, the training process of the K-Means model may specifically include:

[0127] First, k cluster centers are randomly initialized. Then, for each sample point, its distance to each cluster center is calculated and assigned to the cluster center closest to it. Furthermore, the center of each cluster is recalculated, representing the average value of all sample points in that cluster. This "assign-update" process is repeated until the cluster center no longer changes or the maximum number of iterations is reached, ultimately yielding the scene categories corresponding to all cells.

[0128] Based on this, the silhouette coefficient method is used to determine the optimal number of clusters, and clustering is performed using the K-Means algorithm. This effectively divides the scene classifications in the entire cell scene data. In addition, evaluating the silhouette coefficient through random sampling can reduce the amount of calculation and improve efficiency.

[0129] Step S230: Parameter optimization is performed on multiple scene categories in parallel to obtain parameter optimization results corresponding to each scene category.

[0130] After obtaining the scenario categories corresponding to each cell through the above machine learning clustering algorithm, parameter optimization can be performed for each scenario category. During the parameter optimization process, it can be assumed that the optimal wireless parameters for cells in the same scenario category are consistent. Based on this assumption, the parameter optimization problem can be modeled as finding the optimal value of the corresponding parameter while maximizing the service quality indicator:

[0131] p M =αrgmax p∈P q (10)

[0132] Where P represents the parameter candidate value set; q represents the service quality indicator QoS; p M It represents the value of p when the service quality index is maximized, that is, the optimal value of the service parameter.

[0133] However, in practical applications, the relationship between the service quality indicator q and the parameter value P is unclear. Secondly, online evaluation of the service quality indicator q takes a long time, resulting in high trial-and-error costs. Finally, because the set of candidate parameter values ​​is often very large, comprehensive testing cannot be performed through exhaustive enumeration.

[0134] To address these issues, this embodiment proposes a parallel online parameter optimization module that uses a Bayesian parameter optimizer and a Bayesian parameter sampler to perform wireless parameter optimization. The Bayesian parameter optimizer finds optimal parameter values ​​by fitting the data distribution and performing iterative optimization. The Bayesian parameter sampler efficiently explores the parameter space with limited computing resources, quickly approaching the optimal parameter solution.

[0135] For example, Figure 6This is a schematic diagram of the structure of an online parameter optimization module provided by an exemplary embodiment of the present disclosure, such as Figure 6 As shown, the parameter online optimization module may include a Bayesian parameter sampler 610 , a Bayesian parameter optimizer 620 , a parameter sending module 630 and a data recovery module 640 .

[0136] The Bayesian parameter optimizer 620 is responsible for modeling the relationship function q=f(p) between the service indicator q and the wireless parameter p. The input data of the Bayesian parameter optimizer 620 is the wireless parameter value, and the output data is the estimated value of the service indicator.

[0137] The Bayesian parameter sampler 610 is used to determine the wireless parameter p that needs to be sampled and evaluated next time.

[0138] The parameter sending module 630 randomly selects parameter values ​​to be evaluated and sends them to existing cells, and performs random sampling among existing cells in the same scenario category.

[0139] The data recovery module 640 is used to record the time when the parameter value is issued, and the business indicators before and after the issuance, form a sample, add it to the optimized data set, and update the optimized data set.

[0140] The Bayesian parameter optimizer 620 fits the probability distribution model between the wireless parameters and the service indicators again according to the updated optimization data set, and updates the distribution function of the Gaussian process.

[0141] When the iteration output condition is met, the Bayesian parameter optimizer 620 outputs the target parameter configuration; when the iteration output condition is not met, the iteration cycle continues.

[0142] For example, a Gaussian process can be used to design an optimizer, and the relationship function f(p) between the service indicator q and the wireless parameter p is as follows:

[0143] f(p)~Normal(μ(p 1：n ),∑(p 1：n , p 1：n )) (11)

[0144] Among them, μ(p 1：n ) represents the expectation of business indicators in the sample set; ∑(p 1：n , p 1：n ) represents the covariance between the service indicators and the wireless parameters in the sample set.

[0145] When using the Bayesian parameter optimizer and sampler for online parameter optimization, we first collect the previous n historical parameter value samples and their corresponding business metrics. With this data, we can use it to estimate the posterior probability distribution of the n+1th sample. The posterior distribution can be understood as the probability or likelihood of a parameter value in the n+1th sample, given the previous n samples.

[0146] For example, the posterior distribution of the n+1th sample is as follows:

[0147] f(p)|f(p 1：n )~Normal(μ n (p),σ 2 (p)) (12)

[0148] Among them, μ n (p) represents the mean of the business indicators in the sampling set; σ(p) represents the standard deviation of the business indicators in the sampling set; f(p)|f(p 1：n ) represents the posterior distribution of the relationship function f(p) between the service indicator q and the wireless parameter p after sampling n samples.

[0149] Based on this, by constructing a Bayesian parameter optimizer, we can obtain the posterior distribution corresponding to any wireless parameter configuration, and from the posterior distribution we can obtain the expected service indicator μ under the wireless parameters. n+1 (p) and its standard deviation σ n+1 The expected value of a business indicator represents the estimated value of the business indicator, which is the objective function to be maximized. The standard deviation of the business indicator represents the uncertainty of this estimated value. The smaller the standard deviation, the more accurate the prediction result.

[0150] The above is the modeling process of the relationship function f(p) between the business indicator q and the wireless parameter p in this embodiment. f(p) is updated with each online sampling process, making the posterior distribution more and more accurate until the computing resources are exhausted. In this way, the optimal parameter configuration for each business scenario can be quickly found within limited computing resources, ultimately improving the overall business indicator performance.

[0151] like Figure 6 As shown, the Bayesian parameter sampler 610 is used to determine the wireless parameter p that needs to be sampled and evaluated next time. The Bayesian parameter sampler 610 performs sampling under the probability distribution to find a wireless parameter p that can maximize the benefits brought by the online evaluation.

[0152] Specifically, the maximization of expected improvement (MAXI) method can be used to determine the radio parameter settings for the next sampling period. This method combines existing posterior distribution information to predict the potential improvement of a parameter value compared to the previous optimal value. By maximizing this expected improvement, the most valuable parameter configuration for evaluation can be determined. This method achieves a balance between exploring new parameter spaces and leveraging existing optimal solutions, gradually approaching the global optimal solution. With continuous online sampling and model updates, this MAXI strategy becomes increasingly accurate, ultimately finding the optimal radio parameter settings for each scenario.

[0153] Exemplarily, the expected improvement is calculated as follows:

[0154]

[0155]

[0156] Among them, μ n (p) represents the mean of the business indicator q of the n sample sets obtained in the historical iterative sampling process; f(p * ) represents the maximum value of the business indicator q in the n sample set; p * represents the wireless parameter value corresponding to the maximum service indicator q; ∈ represents the constant controlling the exploration rate; Φ represents the cumulative distribution function CDF; φ represents the probability density function PDF; EI p represents the expected improvement in the wireless parameter p.

[0157] At the same time, this embodiment uses parallel sampling technology to collect a large amount of sample data corresponding to different cells at the same time, and selects the expected improvement EI p The largest k wireless parameters p are used as the input parameters to be evaluated in the next online sampling optimization. By combining parallel sampling with the EI selection strategy, the time efficiency of online wireless parameter optimization can be greatly improved.

[0158] For example, Figure 7 This is an exemplary flowchart of the parameter online optimization module provided in the present disclosure. The specific steps may include:

[0159] Step S701: Build an optimized data set using the wireless parameter configuration and service indicator data of the existing network cells. When the online parameter optimization module is started, the configured wireless parameters and corresponding service indicators of the existing network cells under the same scenario category are input and the optimized data set is formed through the data recovery module.

[0160] Step S702: The Bayesian parameter optimizer fits the probability distribution model between the wireless parameters and the service indicators based on the optimization data set, and updates the distribution function of the Gaussian process.

[0161] Step S703: Determine whether the computing resources are exhausted. If the computing resources are not exhausted, execute step S704; if the computing resources are exhausted, execute step S708.

[0162] Step S704: The Bayesian parameter sampler samples k wireless parameters with the greatest expected improvement.

[0163] After the Bayesian parameter optimizer is updated, the Bayesian parameter sampler randomly samples from the parameter value set, calculates the expected improvement of each sampled parameter value, and selects the k parameter values ​​with the largest expected improvement as the parameter values ​​to be evaluated.

[0164] Step S705: Randomly sample k*a cells in the same scenario.

[0165] K*a cells are randomly sampled from existing network cells under the same scenario category, where a may be determined according to the number of cells. In this embodiment, a is tentatively set to 10.

[0166] Step S706: The parameter delivery module randomly selects parameter values ​​to be evaluated and delivers them to the sampling cells.

[0167] Step S707: The data recovery module records the time when the parameter value is issued, as well as the business indicators before and after the issuance, to form a sample and add it to the optimized data set. Return to step S702 and execute until the computing resources are exhausted.

[0168] Step S708: The Bayesian parameter optimizer calculates the target parameter combination.

[0169] The Bayesian parameter optimizer calculates the expected service indicator estimates for all candidate wireless parameter sets. The wireless parameter with the highest expected estimate is the target parameter combination for this scenario and service category.

[0170] Step S709: Send the target parameter combination to the existing network cell.

[0171] Based on this, by adopting a parameter online optimization solution that combines parallel sampling and maximum expected improvement, not only can parameter optimization be performed separately for different scenario categories, thereby greatly improving optimization efficiency, but it also ensures the targetedness and effectiveness of parameter settings in different scenarios, which helps to improve overall business quality.

[0172] One or more technical solutions provided in the exemplary embodiments of the present disclosure can increase the dimension of data analysis by preprocessing multimodal data such as wireless communication data, road test data, and satellite map tile data, enhance the readability of data analysis results, and form a more comprehensive feature representation of multimodal scenarios.

[0173] In addition, a multimodal scene recognition method that integrates standardized wireless network communication data, road test raster data and semantic maps solves the input feature limitations caused by wireless cell scene recognition technology relying solely on communication data and ignoring two-dimensional signal distribution and geographic environment information, thereby improving the granularity and accuracy of wireless cell scene recognition.

[0174] In addition, by adopting a parameter online optimization solution that combines parallel sampling and maximum expected improvement, not only can parameter optimization be performed separately for different scenario categories, thereby greatly improving optimization efficiency, but it also ensures the targetedness and effectiveness of parameter settings in different scenarios, helping to improve overall business quality.

[0175] Therefore, the wireless parameter optimization method provided in the exemplary embodiment of the present disclosure not only solves the problem of the uniqueness of wireless cell analysis data, but also can perform parameter optimization for different scenario categories, thereby greatly improving the optimization efficiency.

[0176] Based on the above embodiments, the present disclosure also provides a wireless parameter optimization method, Figure 8 This is a flow chart of a wireless parameter optimization method provided by an exemplary embodiment of the present disclosure, such as Figure 8 As shown, the method may include the following steps:

[0177] Step S810: Acquire multimodal data corresponding to a plurality of cells to be optimized. The multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions, and geographical environment of the cells to be optimized.

[0178] In this embodiment, multimodal data refers to various data sources used to comprehensively describe the characteristics of the cell to be optimized, and may include the following types:

[0179] Wireless network communication data, such as the cell's uplink and downlink throughput, number of user connections per second, and average CQI (Channel Quality Indicator) value, reflects the cell's wireless network communication quality. For example, fields describing the wireless network communication status of the cell to be optimized can be found in Table 1 above.

[0180] Two-dimensional signal distribution data, such as RSRP (Reference Signal Received Power) and RSRQ (Reference Signal Received Quality) heat maps within a cell, can intuitively demonstrate the coverage and interference conditions within the cell.

[0181] In addition, it can also include relevant data on the geographical environment of the community, such as population density distribution, road conditions and building distribution. These environmental feature data help to more accurately identify the actual usage scenarios of the community.

[0182] Step S820: Input the multiple multimodal data into a scene recognition model to determine the target scene categories corresponding to the multiple cells to be optimized. The scene recognition model includes a scene coding sub-model and a scene clustering module. The scene coding sub-model is used to encode the multimodal data to obtain scene codes, and the scene clustering module is used to determine the target scene categories of the cells to be optimized based on the scene codes.

[0183] In an embodiment, multimodal data is input into a scene recognition model for processing, potential feature information of the multimodal data is extracted through a scene encoding sub-model, and then a clustering module is used to divide this feature information into different scene categories, thereby determining the target scene category to which the cell to be optimized belongs.

[0184] The scene recognition model can be composed of two parts: a scene encoding sub-model and a scene clustering module. The scene encoding sub-model is used to extract features and encode the input multimodal data, compressing the original multimodal data into a more compact scene encoding vector. The encoding process can use deep learning technology to capture the potential feature information in the data.

[0185] The scene clustering module receives scene coding vectors from the scene coding sub-model and then uses unsupervised clustering algorithms such as K-means or DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to classify these coding vectors into different scene categories. This cluster analysis automatically identifies the target scene category for the cell to be optimized.

[0186] Step S830: Based on the target scenario categories corresponding to the cells to be optimized, the parameter optimization module performs parameter optimization on the multiple cells to be optimized in parallel to obtain target parameter configurations corresponding to the multiple cells to be optimized.

[0187] In the embodiment, the parameter optimization module adopts a parallel processing method to perform parameter optimization for each cell to be optimized and obtain the corresponding target parameter configuration.

[0188] Specifically, for each cell to be optimized, the parameter optimization module first reads the target scenario category information for the cell. Based on the target scenario category, the parameter optimization module activates the corresponding parameter optimization algorithm, such as a genetic algorithm or simulated annealing, to automatically adjust and optimize the cell's key wireless parameters.

[0189] During the optimization process, the goal can be to improve cell coverage, capacity, or overall performance. The algorithm will iterate through multiple rounds to find the optimal parameter configuration. During the parallel optimization process, each cell to be optimized will receive a set of target parameter configurations tailored to its target scenario.

[0190] This scenario-recognition-based parallel parameter optimization method can fully utilize the characteristics of different cell scenarios and tailor the optimal parameter configuration for each cell. Compared with the traditional one-size-fits-all optimization method, it can significantly improve the overall network performance.

[0191] Based on this, multimodal data describing the wireless network communication conditions, two-dimensional signal distribution, and geographical environment of the cell to be optimized is introduced into the scene recognition process, thereby more comprehensively describing the actual operating environment of the wireless cell and improving the accuracy of scene recognition and the subsequent parameter optimization effect.

[0192] Based on the above embodiment, in another embodiment provided by the present disclosure, the multimodal data includes wireless network communication data, drive test data, and satellite map tile data, and the above wireless parameter optimization method may further include:

[0193] Performing dimensionless or vector processing on wireless network communication data to obtain standardized wireless network communication data;

[0194] Performing rasterization processing on the drive test data to obtain drive test raster data;

[0195] Perform semantic segmentation on satellite map tile data to obtain a semantic map.

[0196] For example, wireless network communication data can generally be divided into numerical data or categorical data. For numerical fields, standardization can be used to perform dimensionless processing so that the data conforms to a data distribution with a mean of 0 and a standard deviation of 1. This step can be expressed as the above formula (1). For categorical fields, the data can be vectorized using one-hot encoding. This step can be expressed as the above formula (2).

[0197] For example, the drive test data can be rasterized to obtain the drive test grid data. First, the longitude and latitude coordinates of the measurement points are mapped onto a two-dimensional plane using a pseudo-Mercator projection to form a grid coordinate system. This step can be expressed as the above formula (3).

[0198] In this embodiment, the zoom level z can be set to 20, indicating a grid size of 25*25 meters. A signal distribution feature map is constructed by selecting grids within a 600-meter range centered on the wireless cell location. Each pixel in the signal distribution feature map represents a grid, and the pixel value represents the signal measurement value of that grid.

[0199] Furthermore, for multiple measurement points within the same grid cell, the average of their signal strengths can be used as the representative measurement value of the grid to obtain the drive test grid data. The drive test grid data can be used to represent the signal strength corresponding to each grid. This step can be expressed as the above formula (4).

[0200] For example, satellite map tile data can be used to describe the geographic environment of a cell and its surroundings. A semantic segmentation network can be used to perform semantic recognition and segmentation on the input satellite map tile data to generate a semantic map. In subsequent scene recognition, important geographic features such as roads and buildings can be extracted from the semantic map, providing valuable geographic descriptions for wireless planning.

[0201] Based on this, wireless communication data can be dimensionlessly or vectorized to obtain standardized wireless network communication data reflecting network performance. Drive test data can be rasterized to obtain geographic features such as signal coverage. Satellite imagery can be semantically segmented to obtain semantic features of geographic environments such as roads and buildings. This multi-source and multi-modal data fusion provides richer input features for subsequent parameter optimization, thereby improving the overall effectiveness of wireless parameter planning and optimization.

[0202] Based on the above embodiment, in another embodiment provided by the present disclosure, the scene coding sub-model includes a first autoencoder, a second autoencoder, and a third autoencoder. The above step S820 may specifically include:

[0203] Using a first autoencoder to perform dimensionality reduction and data encoding processing on the standardized wireless network communication data to obtain first encoded data;

[0204] Using a second autoencoder to perform data encoding processing on the drive test grid data to obtain second encoded data;

[0205] Using a third autoencoder to perform data encoding processing on the semantic map to obtain third encoded data;

[0206] The first coded data, the second coded data, and the third coded data are concatenated to obtain a scene code.

[0207] In an embodiment, the network structure of the first autoencoder may include FC-64, FC-32, FC-8, FC-32 and FC-64, with the number of neurons being 64, 32, 8, 32, and 64 respectively, and ReLu being used as the activation function. The network structure of the first autoencoder is symmetrical front to back.

[0208] Exemplarily, the training process of the first autoencoder for encoding wireless network communication data may include:

[0209] First, the standardized wireless network communication data is input into the input layer of the first autoencoder, and then passes through the FC-64, FC-32 and FC-8 encoder parts in sequence, gradually reducing the data dimension from high dimension to low dimension, and finally obtaining an 8-dimensional feature vector.

[0210] Then, the 8-dimensional feature vector is input into the decoder part of the first autoencoder, and passes through FC-32 and FC-64 in sequence, and finally restored to the original high-dimensional input data.

[0211] During the training process, the parameters of the autoencoder are optimized by minimizing the difference between the input data and the restored data, so that it can effectively extract the key features of the wireless communication data.

[0212] In order to reduce the amount of computation for subsequent processing, the output of the third hidden layer (FC-8) of the first autoencoder can be selected as the encoding result of the wireless communication data, with a dimension of 8, which is recorded as the first encoded data H 1 .

[0213] For example, for the road test grid data, it can be first structured into image data of size 24*24*1, and then the image data is encoded using a second autoencoder containing a 7-layer neural network. The second autoencoder is a convolutional autoencoder, and the specific network structure of the second autoencoder may include:

[0214] The first layer is the convolution layer Conv3-4, which contains four 3*3 convolution kernels with a step size of 2;

[0215] The second convolutional layer, Conv3-8, contains eight 3x3 convolution kernels with a stride of 2. After two layers of convolution, the image data size is reduced to 1 / 4 of its original size. If it is directly connected to a fully connected layer, the input dimension of the fully connected layer will be very high due to the small size of the feature map, resulting in a large number of parameters and prone to overfitting. Therefore, a Flatten layer is added after the second convolutional layer to flatten the two-dimensional feature map into a one-dimensional vector. This significantly reduces the input dimension of the fully connected layer, reduces the number of parameters, and improves the network's generalization ability.

[0216] The third layer is the fully connected layer FC-64, with 64 neurons;

[0217] The fourth layer is the fully connected layer FC-8, with 8 neurons;

[0218] The fifth layer is a fully connected layer FC-64 with 64 neurons. A Reshape layer is added after the fifth fully connected layer to restore the feature vector to the format of a two-dimensional feature map.

[0219] The sixth layer is the transposed convolution layer Deconv3-8, which contains 8 3*3 convolution kernels with a step size of 2;

[0220] The seventh layer is the transposed convolution layer Deconv3-4, which contains four 3*3 convolution kernels with a step size of 2.

[0221] Exemplarily, the training process of the second autoencoder encoding the drive test grid data may include:

[0222] First, the road test grid data is input into the encoder part of the convolutional autoencoder, and then through the convolution of the first and second layers and the pooling of the third and fourth layers, the 8-dimensional feature vector H is extracted. 2 .

[0223] Then, the 8-dimensional feature vector is input into the decoder part. After the fifth layer of pooling, the feature vector is restored to the format of a two-dimensional feature map. After the sixth and seventh layers of transposed convolution, it is restored to the original 24*24*1 image data.

[0224] During the training process, the parameters of the autoencoder are optimized by minimizing the difference between the input image and the restored image, enabling it to efficiently extract the key feature data of the road test raster data.

[0225] Since the H output of the fourth layer (FC-8) 2 It not only reduces the data dimension but also retains important features, and can better extract and characterize the key feature information of the road test raster data. It can also be better integrated with the first encoded data and the third encoded data to form a more comprehensive multimodal scene representation.

[0226] Therefore, the H output of the fourth layer (FC-8) of the second autoencoder can be 2 The second encoded data is used as the drive test grid data.

[0227] For example, for a semantic map, it can be first structured into 24*24*1 image data, and then encoded using a third autoencoder containing a 7-layer neural network. The third autoencoder is a convolutional autoencoder, and its specific network structure and training process are the same as those of the second autoencoder. Please refer to the above for details.

[0228] Since the output dimension of the fourth layer (FC-8) is 8, it not only reduces the data dimension but also retains important features. It can better extract and represent the key feature information of the semantic map data. It can also be better integrated with the first encoded data and the second encoded data to form a more comprehensive multimodal scene representation.

[0229] Therefore, the H output of the fourth layer (FC-8) of the third autoencoder can be 3 The third encoded data is the drive test grid data.

[0230] Exemplarily, the training of the first, second and third autoencoders can all use the mean square error loss function, and the mean square error loss function can be found in the above formula (5).

[0231] After obtaining the encoding results of the standardized wireless network communication data, drive test grid data, and semantic map, the three types of encoding results can be spliced ​​into a vector with a dimension of 24, and this vector is used as the cell scene code H. This step can be expressed as the above formula (6).

[0232] Based on this, the first encoded data, second encoded data and third encoded data of the 8-dimensional feature not only retain the core feature information of the original data and provide high-quality input features for multimodal scene recognition, but the dimensionality reduction operation can also greatly reduce the computational complexity of subsequent processing, which is helpful for subsequent machine learning clustering processing and improves the efficiency and accuracy of clustering.

[0233] Based on the above embodiment, in another embodiment provided by the present disclosure, the above step S820 may specifically include:

[0234] Obtain the number of clusters that maximizes the average silhouette coefficient, and determine the number of clusters as the target number of clusters k;

[0235] Use the k-means clustering algorithm to perform cluster analysis on the scene codes and determine the cluster categories corresponding to the scene codes;

[0236] The cluster category corresponding to the scene code is determined as the target scene category of the cell to be optimized.

[0237] In the embodiment, the silhouette coefficient method is used to determine the optimal number of clusters k.

[0238] Specifically, the number of clusters, k, is first given. The K-Means model is then trained using scene coding data from all cells to obtain clustering results corresponding to all cells. Next, z samples are randomly sampled from all cells, and the average silhouette coefficient of these z samples is calculated. This average silhouette coefficient is used as the silhouette coefficient corresponding to the number of clusters, k. Finally, the number of clusters, k, with the largest silhouette coefficient, is selected as the optimal number of clusters and used to train the K-Means model to obtain the target scene categories corresponding to the cells to be optimized.

[0239] For example, the silhouette coefficient calculation formula can be expressed as the above formulas (7) and (8). Taking the average value of the silhouette coefficient as the silhouette coefficient corresponding to the target cluster number k, the calculation formula can be expressed as the above formula (9).

[0240] For example, the training process of the K-Means model may specifically include:

[0241] First, k cluster centers are randomly initialized. Then, for each sample point, its distance to each cluster center is calculated and assigned to the cluster center closest to it. Furthermore, the center of each cluster is recalculated, representing the average value of all sample points in that cluster. This "assign-update" process is repeated until the cluster center no longer changes or the maximum number of iterations is reached, ultimately yielding the scene categories corresponding to all cells.

[0242] After obtaining the target cluster number k, the k-means clustering algorithm is used to perform cluster analysis on the scene codes to determine the cluster categories corresponding to the scene codes.

[0243] Specifically, the k-means clustering algorithm is used to perform cluster analysis on the scene code vectors, and these scene codes are divided into K different cluster categories according to the distance similarity between the vectors.

[0244] Finally, the cluster category corresponding to the scene code is determined as the target scene category of the cell to be optimized.

[0245] Based on this, the optimal number of clusters is determined by the silhouette coefficient method, and clustering is performed based on the K-Means algorithm. This can effectively divide the scene classification in the scene data of the cell to be optimized, ensure targeted parameter optimization for different scene conditions, and improve the overall optimization effect.

[0246] Based on the above embodiment, in another embodiment provided by the present disclosure, performing semantic segmentation processing on satellite map tile data to obtain a semantic map may specifically include:

[0247] Input satellite map tile data into the semantic segmentation model to obtain the initial semantic map;

[0248] The initial semantic map is downsampled to obtain a semantic map.

[0249] In this embodiment, a high-resolution satellite image within a certain range (e.g., 500 meters) is selected with the wireless cell location as the center to obtain a satellite map tile data image with a resolution of 512*512. Then, the SegFormer semantic segmentation network can be used to perform semantic recognition and segmentation on the input satellite map tile data image to obtain an initial semantic map with the same resolution.

[0250] In subsequent scene recognition, important geographic element information such as roads and buildings can be extracted from the semantic map, providing valuable geographic environment description for wireless planning.

[0251] Furthermore, by downsampling, the resolution of the initial semantic map can be reduced from 512*512 to 24*24 to obtain a semantic map. This semantic map maintains the same resolution as the rasterized drive test data mentioned above, facilitating subsequent feature fusion with other data sources (such as wireless network data and drive test data).

[0252] Based on this, wireless communication data can be dimensionlessly or vectorized to obtain standardized wireless network communication data reflecting network performance. Drive test data can be rasterized to obtain geographic features such as signal coverage. Satellite imagery can be semantically segmented to obtain semantic features of geographic environments such as roads and buildings. This multi-source and multi-modal data fusion provides richer input features for subsequent parameter optimization, thereby improving the overall effectiveness of wireless parameter planning and optimization.

[0253] Based on the above embodiment, in another embodiment provided by the present disclosure, the above step S830 may specifically include:

[0254] Obtain the current parameter configuration information, current service indicator data, and target parameter optimization model corresponding to the target scenario category of the cell to be optimized;

[0255] The current parameter configuration information and the current business indicator data are input into the target parameter optimization model to obtain the target parameter configuration corresponding to the cell to be optimized; the target parameter optimization model includes a Bayesian parameter optimizer and a Bayesian parameter sampler; the Bayesian parameter optimizer is used to fit the distribution relationship between the current parameter configuration and the current business indicator data, and based on the distribution relationship, respectively determine the business indicator estimation expectations corresponding to multiple wireless parameter configurations to be evaluated, and determine the wireless parameter configuration to be evaluated with the maximum business indicator estimation expectation as the target parameter configuration; the Bayesian parameter sampler is used to determine the wireless parameter configuration to be evaluated according to a preset sampling rule.

[0256] In the embodiment, it is necessary to obtain the current parameter configuration information and current service indicator data of the cell to be optimized, and determine the corresponding target parameter optimization model according to the target scenario category of the cell to be optimized.

[0257] For example, the parameter optimization problem can be modeled as finding the optimal value of the corresponding parameter under the condition of maximizing the service quality index. The optimization problem can be expressed as the above formula (10).

[0258] This target parameter optimization model can include a Bayesian parameter optimizer and a Bayesian parameter sampler. The Bayesian parameter optimizer is used to find optimal parameter values ​​by fitting the data distribution and perform iterative optimization, while the Bayesian parameter sampler can efficiently explore the parameter space with limited computing resources, thereby quickly approaching the optimal parameter solution.

[0259] The Bayesian parameter optimizer is responsible for modeling the relationship function q = f(p) between the service indicator q and the wireless parameter p. The input data of the Bayesian parameter optimizer is the wireless parameter value, and the output data is the estimated value of the service indicator.

[0260] Exemplarily, a Gaussian process can be used to design an optimizer, and the relationship function f(p) between the service indicator q and the wireless parameter p can be expressed as the above formula (11).

[0261] When using the Bayesian parameter optimizer and sampler for online parameter optimization, we first collect the previous n historical parameter value samples and their corresponding business metrics. With this data, we can use it to estimate the posterior probability distribution of the n+1th sample. The posterior distribution can be understood as the probability or likelihood of a parameter value in the n+1th sample, given the previous n samples.

[0262] For example, the posterior distribution of the n+1th sample can be expressed as the above formula (12).

[0263] By constructing a Bayesian parameter optimizer, we can obtain the posterior distribution corresponding to any wireless parameter configuration, and from the posterior distribution we can obtain the expected service indicator μ under the wireless parameters. n+1 (p) and its standard deviation σ n+1 The expected value of a business indicator represents the estimated value of the business indicator, which is the objective function to be maximized. The standard deviation of the business indicator represents the uncertainty of this estimated value. The smaller the standard deviation, the more accurate the prediction result.

[0264] The above is the modeling process of the relationship function f(p) between the business indicator q and the wireless parameter p in this embodiment. f(p) is updated with each online sampling process, making the posterior distribution more and more accurate until the computing resources are exhausted. In this way, the optimal parameter configuration for each business scenario can be quickly found within limited computing resources, ultimately improving the overall business indicator performance.

[0265] For example, the Bayesian parameter sampler can use the maximization expected improvement method to determine the radio parameters for the next sampling period. This method combines existing posterior distribution information to predict the potential improvement compared to the previous optimal value if a parameter value is selected for evaluation. By maximizing this expected improvement, the most valuable parameter configuration for evaluation can be determined. This balance is achieved between exploring new parameter space and leveraging existing optimal solutions, gradually approaching the global optimal solution. With continuous online sampling and model updates, this maximum expected improvement strategy becomes increasingly accurate, ultimately finding the optimal radio parameter settings for each scenario.

[0266] Exemplarily, the calculation formulas for the expected improvement can be expressed as the above formulas (12) and (13).

[0267] At the same time, this embodiment uses parallel sampling technology to collect a large amount of sample data corresponding to different cells at the same time, and selects the expected improvement EI p The largest k wireless parameters p are used as the input parameters to be evaluated in the next online sampling optimization. By combining parallel sampling with the EI selection strategy, the time efficiency of online wireless parameter optimization can be greatly improved.

[0268] Based on this, by adopting a parameter online optimization solution that combines parallel sampling and maximum expected improvement, not only can parameter optimization be performed separately for different scenario categories, thereby greatly improving optimization efficiency, but it also ensures the targetedness and effectiveness of parameter settings in different scenarios, which helps to improve overall business quality.

[0269] Based on the above embodiment, in another embodiment provided by the present disclosure, the above wireless parameter optimization method may further include:

[0270] Construct initial parameter optimization models corresponding to different scenario categories and obtain training sample data corresponding to different scenario categories; the training sample data includes wireless parameter configuration and service indicator data corresponding to multiple existing network cells under the same scenario category;

[0271] The training sample data are input into the corresponding initial parameter optimization model in parallel, and the distribution relationship between wireless parameters and service indicators is fitted through the Bayesian parameter optimizer;

[0272] The Bayesian parameter sampler uses a maximum expectation improvement method to determine the wireless parameter configuration to be evaluated in multiple existing cells of the same scenario category based on distribution relationships.

[0273] The wireless parameter configuration to be evaluated is distributed to existing cells in the same scenario category, and the distribution time of the wireless parameter configuration to be evaluated and the service indicator data before and after the distribution are collected;

[0274] The delivery time of the wireless parameter configuration to be evaluated and the service indicator data before and after the delivery are stored in the training sample data;

[0275] The Bayesian parameter optimizer performs iterative training based on the updated training sample data until the iteration end condition is met;

[0276] When the iteration end condition is met, the trained initial parameter optimization model is determined as the target parameter optimization model.

[0277] In the embodiment, it is first necessary to construct initial parameter optimization models for different scenario categories. The training data of the model includes the wireless parameter configurations and corresponding service indicator data of multiple existing cells in the same scenario.

[0278] Based on the training data, each initial optimization model will use the Bayesian parameter optimizer to fit the distribution relationship between wireless parameters and business indicators.

[0279] The Bayesian parameter sampler then uses the maximization of expected improvement method to determine the wireless parameter configurations to be evaluated for similar scenarios based on the distribution relationship. The wireless parameter configurations to be evaluated are then distributed to the corresponding live cells for testing. The distribution time and service indicator data before and after the distribution are recorded and added to the training sample data.

[0280] The Bayesian parameter optimizer fits the distribution relationship again based on the updated training sample data. By continuously updating the training data, the optimizer can iteratively optimize the distribution relationship model and improve the accuracy and effect of parameter optimization.

[0281] Once the preset iteration end condition is met, the initial parameter optimization model is determined as the final target parameter optimization model, which can be used to guide parameter optimization in actual cells.

[0282] Based on this, the Bayesian parameter sampler and Bayesian parameter optimizer can fully utilize existing network data to adaptively train parameter optimization models for different scenarios. The trained parameter optimization models can be applied to wireless parameter optimization in different scenarios.

[0283] One or more technical solutions provided in the exemplary embodiments of the present disclosure can increase the dimension of data analysis and enhance the readability of data analysis results by preprocessing multimodal data such as wireless communication data, road test data, and satellite map tile data, thereby forming a more comprehensive feature representation of multimodal scenarios.

[0284] In addition, a multimodal scene recognition method that integrates standardized wireless network communication data, road test raster data and semantic maps solves the input feature limitations caused by wireless cell scene recognition technology relying solely on communication data and ignoring two-dimensional signal distribution and geographic environment information, thereby improving the granularity and accuracy of wireless cell scene recognition.

[0285] In addition, by adopting a parameter online optimization solution that combines parallel sampling and maximum expected improvement, not only can parameter optimization be performed separately for different scenario categories, thereby greatly improving optimization efficiency, but it also ensures the targetedness and effectiveness of parameter settings in different scenarios, helping to improve overall business quality.

[0286] Therefore, the wireless parameter optimization method provided in the exemplary embodiment of the present disclosure not only solves the problem of the uniqueness of wireless cell analysis data, but also can perform parameter optimization for different scenario categories, thereby greatly improving the optimization efficiency.

[0287] The above mainly introduces the solutions provided by the exemplary embodiments of the present disclosure. It is understandable that in order to implement the above functions, the electronic device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed herein, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present disclosure.

[0288] The exemplary embodiments of the present disclosure can divide the functional units of the electronic device according to the above method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the exemplary embodiments of the present disclosure is schematic and is only a logical functional division. In actual implementation, there may be other division methods.

[0289] In the case of dividing each functional module according to each function, an exemplary embodiment of the present disclosure provides a wireless parameter optimization device, which may be a server or a chip applied to a server. Figure 9 This is a schematic block diagram of the functional modules of a wireless parameter optimization device provided by an exemplary embodiment of the present disclosure. Figure 9 As shown, the wireless parameter optimization device 900 includes:

[0290] The data acquisition module 910 is used to acquire multimodal data corresponding to a plurality of cells to be optimized; the multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions, and geographical environment of the cells to be optimized;

[0291] A data processing module 920 is configured to input the plurality of multimodal data into a scene recognition model to determine target scene categories corresponding to the plurality of cells to be optimized. The scene recognition model includes a scene coding submodel and a scene clustering module. The scene coding submodel is configured to perform data coding processing on the multimodal data to obtain scene codes. The scene clustering module is configured to determine the target scene category of the cell to be optimized based on the scene codes.

[0292] The data processing module 920 is further configured to perform parameter optimization on the multiple cells to be optimized in parallel in the parameter optimization module based on the target scenario categories corresponding to the cells to be optimized, so as to obtain target parameter configurations corresponding to the multiple cells to be optimized.

[0293] In another embodiment provided by the present disclosure, the multimodal data includes wireless network communication data, drive test data and satellite map tile data. The data processing module 920 is further used to perform dimensionless or vectorized processing on the wireless network communication data to obtain standardized wireless network communication data; perform rasterization processing on the drive test data to obtain drive test raster data; and perform semantic segmentation processing on the satellite map tile data to obtain a semantic map.

[0294] In another embodiment provided by the present disclosure, the scene coding sub-model includes a first autoencoder, a second autoencoder and a third autoencoder; the data processing module 920 is further used to use the first autoencoder to perform dimensionality reduction and data encoding processing on the standardized wireless network communication data to obtain first encoded data; use the second autoencoder to perform data encoding processing on the road test raster data to obtain second encoded data; use the third autoencoder to perform data encoding processing on the semantic map to obtain third encoded data; and splice the first encoded data, the second encoded data and the third encoded data to obtain the scene coding.

[0295] In another embodiment provided by the present disclosure, the data processing module 920 is further used to obtain the number of clusters that makes the average silhouette coefficient reach the maximum value, and determine the number of clusters as the target number of clusters k; use the k-means clustering algorithm to perform cluster analysis on the scene coding to determine the cluster category corresponding to the scene coding; and determine the cluster category corresponding to the scene coding as the target scene category of the cell to be optimized.

[0296] In another embodiment provided by the present disclosure, the data processing module 920 is further configured to input the satellite map tile data into a semantic segmentation model to obtain an initial semantic map; and perform downsampling processing on the initial semantic map to obtain a semantic map.

[0297] In another embodiment provided by the present disclosure, the data processing module 920 is also used to obtain the current parameter configuration information of the cell to be optimized, the current business indicator data and the target parameter optimization model corresponding to the target scenario category; the current parameter configuration information and the current business indicator data are input into the target parameter optimization model to obtain the target parameter configuration corresponding to the cell to be optimized; the target parameter optimization model includes a Bayesian parameter optimizer and a Bayesian parameter sampler; the Bayesian parameter optimizer is used to fit the distribution relationship between the current parameter configuration and the current business indicator data, and based on the distribution relationship, respectively determine the business indicator estimation expectations corresponding to multiple wireless parameter configurations to be evaluated, and determine the wireless parameter configuration to be evaluated with the largest business indicator estimation expectation as the target parameter configuration; the Bayesian parameter sampler is used to determine the wireless parameter configuration to be evaluated according to a preset sampling rule.

[0298] In another embodiment provided by the present disclosure, the data processing module 920 is further used to respectively construct initial parameter optimization models corresponding to different scenario categories and obtain training sample data corresponding to different scenario categories; the training sample data includes wireless parameter configurations and service indicator data corresponding to multiple existing cells under the same scenario category; the training sample data are respectively input into the corresponding initial parameter optimization model in parallel, and the distribution relationship between wireless parameters and service indicators is fitted by the Bayesian parameter optimizer; the Bayesian parameter sampler adopts a maximum expectation improvement method to determine the wireless parameter configuration to be evaluated in multiple existing cells of the same scenario category according to the distribution relationship; the wireless parameter configuration to be evaluated is sent to the existing cells of the same scenario category, and the sending time of the wireless parameter configuration to be evaluated and the service indicator data before and after the sending are collected; the sending time of the wireless parameter configuration to be evaluated and the service indicator data before and after the sending are stored in the training sample data; the Bayesian parameter optimizer performs iterative training based on the updated training sample data until the iteration end condition is met; when the iteration end condition is met, the trained initial parameter optimization model is determined as the target parameter optimization model.

[0299] Figure 10 This is a schematic block diagram of a chip provided as an example in the present disclosure. Figure 10 As shown, the chip 1000 includes one or more (including two) processors 1001 and a communication interface 1002. The communication interface 1002 can support the server to perform the data sending and receiving steps in the above method, and the processor 1001 can support the server to perform the data processing steps in the above method.

[0300] Optional, such as Figure 10 As shown, the chip 1000 also includes a memory 1003. The memory 1003 may include a read-only memory and a random access memory, and provides operation instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory (NVRAM).

[0301] In some embodiments, as Figure 10As shown, the processor 1001 performs corresponding operations by calling the operation instructions stored in the memory (the operation instructions may be stored in the operating system). The processor 1001 controls the processing operations of any one of the terminal devices, and the processor may also be called a central processing unit (CPU). The memory 1003 may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory 1003 may also include NVRAM. For example, in an application, the memory, the communication interface, and the memory are coupled together through a bus system, wherein the bus system may include a power bus, a control bus, and a status signal bus in addition to a data bus. However, for the sake of clarity, in Figure 10 Various buses are labeled as bus system 1004.

[0302] The methods disclosed in the above embodiments of the present disclosure can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor may be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The methods, steps, and logic block diagrams disclosed in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in conjunction with the embodiments of the present disclosure can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0303] The exemplary embodiments of the present disclosure further provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being configured to cause the electronic device to perform a method according to an exemplary embodiment of the present disclosure when executed by the at least one processor.

[0304] Exemplary embodiments of the present disclosure further provide a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to perform a method according to an embodiment of the present disclosure.

[0305] Exemplary embodiments of the present disclosure further provide a computer program product, including a computer program, wherein when the computer program is executed by a processor of a computer, it is used to cause the computer to perform the method according to the embodiment of the present disclosure.

[0306] Figure 11 The structural block diagram of an electronic device provided as an example of the present disclosure will now be described as a structural block diagram of an electronic device 1100 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0307] like Figure 11 As shown, the electronic device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. Various programs and data required for the operation of the electronic device 1100 can also be stored in the RAM 1103. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0308] Multiple components within electronic device 1100 are connected to I / O interface 1105, including an input unit 1106, an output unit 1107, a storage unit 1108, and a communication unit 1109. Input unit 1106 can be any type of device capable of inputting information into electronic device 1100. Input unit 1106 can receive input numeric or character information and generate key input signals related to user settings and / or function control of the electronic device. Output unit 1107 can be any type of device capable of presenting information and may include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. Storage unit 1108 may include, but is not limited to, a magnetic disk or an optical disk. Communication unit 1109 allows electronic device 1100 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks and may include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver and / or chipset, such as a Bluetooth™ device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0309] The computing unit 1101 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1101 performs the various methods and processes described above. Each of the methods described above can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1108. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 1100 via the ROM 1102 and / or the communication unit 1109.

[0310] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0311] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0312] As used in this disclosure, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and / or device (e.g., a magnetic disk, an optical disk, a memory, a programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0313] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0314] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0315] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

[0316] In the above embodiments, they can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the process or function described in the embodiment of the present disclosure is performed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user device, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program or instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, a hard disk, or a tape; it can also be an optical medium, such as a digital video disc (DVD); it can also be a semiconductor medium, such as a solid state drive (SSD).

[0317] Although the present disclosure has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present disclosure. Accordingly, this specification and the drawings are merely illustrative of the present disclosure as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present disclosure. Obviously, those skilled in the art may make various modifications and variations to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, the present disclosure is intended to include such modifications and variations if they fall within the scope of the claims of the present disclosure and their equivalents.

Claims

1. A wireless parameter optimization method, characterized in that: The method comprises: Acquire multimodal data corresponding to a plurality of cells to be optimized, wherein the multimodal data is used to describe the wireless network communication conditions, two-dimensional signal distribution conditions, and geographical environment of the cells to be optimized, and the multimodal data includes wireless network communication data, drive test data, and satellite map tile data; Performing dimensionless or vector processing on wireless network communication data to obtain standardized wireless network communication data; Performing rasterization processing on the drive test data to obtain drive test raster data; Perform semantic segmentation on satellite map tile data to obtain a semantic map; Inputting the plurality of multimodal data into a scene recognition model respectively to determine target scene categories corresponding to the plurality of cells to be optimized; the scene recognition model includes a scene coding submodel and a scene clustering module, the scene coding submodel is used to perform data coding processing on the multimodal data to obtain scene codes, and the scene clustering module is used to determine the target scene category of the cell to be optimized based on the scene codes; The scene coding sub-model includes a first autoencoder, a second autoencoder, and a third autoencoder; and performing data coding processing on the multimodal data to obtain scene coding includes: Using the first autoencoder to perform dimensionality reduction and data encoding processing on the standardized wireless network communication data to obtain first encoded data; Using the second autoencoder to perform data encoding processing on the drive test grid data to obtain second encoded data; Using the third autoencoder to perform data encoding processing on the semantic map to obtain third encoded data; performing splicing processing on the first coded data, the second coded data, and the third coded data to obtain the scene code; Based on the target scenario categories respectively corresponding to the cells to be optimized, the parameter optimization module performs parameter optimization on the multiple cells to be optimized in parallel to obtain target parameter configurations respectively corresponding to the multiple cells to be optimized.

2. The method according to claim 1, characterized in that The determining, according to the scene code, a target scene category of the cell to be optimized includes: Obtaining the number of clusters that maximizes the average silhouette coefficient, and determining the number of clusters as the target number of clusters k; Performing cluster analysis on the scene code using a k-means clustering algorithm to determine a cluster category corresponding to the scene code; The cluster category corresponding to the scene code is determined as the target scene category of the cell to be optimized.

3. The method according to claim 1, characterized in that The step of performing semantic segmentation processing on the satellite map tile data to obtain a semantic map includes: Inputting the satellite map tile data into a semantic segmentation model to obtain an initial semantic map; Downsampling is performed on the initial semantic map to obtain a semantic map.

4. The method according to claim 1, wherein The performing parameter optimization on the multiple cells to be optimized in parallel in the parameter optimization module based on the target scenario categories respectively corresponding to the cells to be optimized to obtain target parameter configurations respectively corresponding to the multiple cells to be optimized includes: Obtaining current parameter configuration information, current service indicator data, and a target parameter optimization model corresponding to the target scenario category of the cell to be optimized; The current parameter configuration information and the current business indicator data are input into a target parameter optimization model to obtain a target parameter configuration corresponding to the cell to be optimized; the target parameter optimization model includes a Bayesian parameter optimizer and a Bayesian parameter sampler; the Bayesian parameter optimizer is used to fit the distribution relationship between the current parameter configuration and the current business indicator data, and based on the distribution relationship, respectively determine the business indicator estimation expectations corresponding to multiple wireless parameter configurations to be evaluated, and determine the wireless parameter configuration to be evaluated with the maximum business indicator estimation expectation as the target parameter configuration; the Bayesian parameter sampler is used to determine the wireless parameter configuration to be evaluated according to a preset sampling rule.

5. The method according to claim 4, characterized in that The method further comprises: Constructing initial parameter optimization models corresponding to different scenario categories and obtaining training sample data corresponding to different scenario categories; the training sample data includes wireless parameter configuration and service indicator data corresponding to multiple existing network cells under the same scenario category; Inputting the training sample data into the corresponding initial parameter optimization model in parallel respectively, and fitting the distribution relationship between wireless parameters and service indicators through the Bayesian parameter optimizer; The Bayesian parameter sampler adopts a maximum expectation improvement method to determine the wireless parameter configuration to be evaluated in multiple existing network cells of the same scenario category according to the distribution relationship; Send the wireless parameter configuration to be evaluated to the existing network cells of the same scenario category, and collect the sending time of the wireless parameter configuration to be evaluated and the service indicator data before and after the sending; Storing the time of issuing the wireless parameter configuration to be evaluated and the service indicator data before and after issuing into the training sample data; The Bayesian parameter optimizer performs iterative training based on the updated training sample data until an iteration end condition is met; When the iteration end condition is met, the trained initial parameter optimization model is determined as the target parameter optimization model.

6. A wireless parameter optimization device, characterized in that: The device comprises: A data acquisition module is used to acquire multimodal data corresponding to multiple cells to be optimized; the multimodal data is used to describe the wireless network communication status, two-dimensional signal distribution and geographical environment of the cells to be optimized, and the multimodal data includes wireless network communication data, drive test data and satellite map tile data; Performing dimensionless or vector processing on wireless network communication data to obtain standardized wireless network communication data; Performing rasterization processing on the drive test data to obtain drive test raster data; Perform semantic segmentation on satellite map tile data to obtain a semantic map; a data processing module, configured to input the plurality of multimodal data into a scene recognition model respectively, and determine the target scene categories corresponding to the plurality of cells to be optimized; the scene recognition model includes a scene coding submodel and a scene clustering module, the scene coding submodel is configured to perform data coding processing on the multimodal data to obtain a scene code, and the scene clustering module is configured to determine the target scene category of the cell to be optimized based on the scene code; The scene coding sub-model includes a first autoencoder, a second autoencoder, and a third autoencoder; and performing data coding processing on the multimodal data to obtain scene coding includes: Using the first autoencoder to perform dimensionality reduction and data encoding processing on the standardized wireless network communication data to obtain first encoded data; Using the second autoencoder to perform data encoding processing on the drive test grid data to obtain second encoded data; Using the third autoencoder to perform data encoding processing on the semantic map to obtain third encoded data; performing splicing processing on the first coded data, the second coded data, and the third coded data to obtain the scene code; The data processing module is further used to perform parameter optimization on the multiple cells to be optimized in parallel in the parameter optimization module based on the target scenario categories corresponding to each cell to be optimized, so as to obtain the target parameter configurations corresponding to the multiple cells to be optimized.

7. An electronic device, characterized in that: include: at least one processor; a memory for storing the at least one processor-executable instruction; The at least one processor is configured to execute the instructions to implement the steps of the method according to any one of claims 1 to 5.

8. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images