Wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling

By combining 3D point cloud modeling and graph convolutional neural networks, the problem of quantifying 3D spatial signal attenuation and interference intensity in mobile communication networks has been solved, achieving efficient multi-base station collaborative optimization and precise network coverage adjustment, and improving operation and maintenance efficiency and intelligence level.

CN120786404APending Publication Date: 2025-10-14XINDA IND CHANGSHA

Patent Information

Application Number
CN202510989490.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-17
Publication Date
2025-10-14

AI Technical Summary

Technical Problem

Existing technologies in mobile communication networks are unable to accurately characterize signal attenuation and multipath effects inside buildings, find it difficult to quantify the coupling interference intensity of multiple base stations in three-dimensional space, and lack a global perspective on the coordinated adjustment of multiple base stations. This leads to inefficient network optimization and difficulty in accurately locating coverage issues in deeply obscured areas.

Method used

Using three-dimensional point cloud modeling, graph convolutional neural network and multi-objective search optimization technology, a spatial topology map is constructed. Spatial semantic features are extracted through graph convolution operations. The interference influencing factor map is combined to perform adaptive adjustment of multi-base station parameters, and the optimization results are displayed in a three-dimensional visual way.

Benefits of technology

It has achieved accurate modeling and collaborative optimization of wireless network coverage quality in complex building environments, improved the accuracy of problem identification and the efficiency of multi-base station collaborative parameter adjustment, and significantly enhanced the level of network optimization intelligence and operation and maintenance response speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120786404A_ABST
    Figure CN120786404A_ABST
Patent Text Reader

Abstract

The invention discloses a wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling, and the method comprises the following steps: S1, collecting room and base station information, and carrying out the preprocessing; s2, constructing a three-dimensional point cloud model, and extracting the position relation of each room; s3, constructing a spatial topological graph by taking the rooms as nodes, wherein an edge weight is a spatial distance between the nodes; s4, performing multi-layer graph convolution operation on the spatial topological graph, and calculating the probability of each room serving as an abnormal point; s5, constructing an interference influence factor map, and carrying out interference intensity modeling on a plurality of base station antennas covering abnormal points; s6, performing multi-target optimization search, and adjusting the transmitting power, the direction angle and the downward inclination angle of each antenna in the associated base station; and S7, mapping the parameter adjustment suggestion to the three-dimensional point cloud model, and outputting an intelligent order sending execution path of the target room. According to the invention, wireless coverage abnormity identification and antenna parameter collaborative optimization are realized, and the signal quality and the operation and maintenance efficiency are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of communication network optimization, and in particular to a wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling. Background Art

[0002] Amid the rapid evolution of mobile communication networks, coordinated coverage between outdoor macrocells, indoor microcells, and various distributed antennas has become the mainstream networking model. Network planning and optimization technologies are also gradually shifting from traditional experience-driven approaches to data-driven ones. However, existing technologies still rely on two-dimensional geographic information and statistical traffic data to infer cell coverage boundaries. Operations and maintenance engineers use network management indicators and a small amount of drive test data to identify issues such as weak coverage, interference, and overlapping coverage. They then optimize the network by manually fine-tuning base station power, antenna azimuth, or downtilt. This approach fails to accurately characterize signal attenuation and multipath effects within buildings and struggles to quantify the intensity of coupled interference between multiple base stations in three dimensions. As network scenarios expand from two-dimensional to three-dimensional, angular occlusion between neighboring cells, vertically layered coverage, and signal leakage between floors can all lead to complex co-channel interference. Traditional two-dimensional simulation models significantly increase errors when predicting vertical coverage. Furthermore, existing methods typically optimize at the granularity of a single cell, lacking a global perspective on coordinated adjustments across multiple base stations and failing to achieve an optimal balance between coverage and interference.

[0003] With the prevalence of 5G massive MIMO, wide-angle beamforming, and high-density site deployment, the number of adjustable RF parameters in the network has increased significantly: each beam element includes both azimuth and elevation angles, as well as multiple levels of transmit power and dynamic tilt. However, these parameters are highly coupled, and adjusting them individually often leads to a negative deterioration of other metrics. Existing algorithms often use genetic search, particle swarm search, or greedy strategies based on KPI thresholds. These algorithms can converge in scenarios with low parameter dimensionality and a small number of sites, but are prone to local optima in high-dimensional parameter spaces. More importantly, these optimization algorithms typically only process statistical coverage information at the base station level and lack spatial semantic features down to the room level. The accuracy of coverage predictions in deeply obscured areas such as basements, mezzanines, and inner corridors is less than 60%, resulting in a high number of complaints and a passive O&M model of "fix wherever there are complaints."

[0004] To alleviate complaints, some operations teams have incorporated MR big data from mobile terminals, using machine learning models to predict weak coverage. However, the vertical accuracy of MR data within buildings is less than one floor, making it difficult to pinpoint specific apartment types. Furthermore, the lack of analysis of the interaction between antenna beams and building structures results in significant discrepancies between predictions and real-world scenarios. Furthermore, algorithms based on single-source data cannot fully integrate user complaints, wireless measurements, antenna physical parameters, and building structure information, often overlooking the complex relationship between user perception and physical propagation.

[0005] At the visualization level, existing network optimization platforms mostly rely on GIS floor plans or simple 3D cylinder diagrams to display coverage heat maps. These platforms are unable to demonstrate signal shielding paths caused by complex indoor apartment layouts, nor can they visually demonstrate changes in energy distribution between floors or apartment layouts after antenna adjustments. Engineers often need to switch between multiple software programs and manually compare KPI curves with on-site survey results, resulting in long optimization cycles and low accuracy. Furthermore, the disconnect between the dispatch process and network optimization results makes it difficult to implement site adjustment recommendations in a timely manner, missing the optimal optimization window.

[0006] Therefore, how to provide a wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0007] One purpose of the present invention is to propose a wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling. The present invention makes full use of three-dimensional point cloud modeling, graph convolutional neural network and multi-objective search optimization technology, and describes in detail the anomaly recognition, interference modeling, parameter adaptive adjustment and three-dimensional visualization dispatching process for indoor and outdoor stereo coverage scenarios. It has the advantages of high positioning accuracy, strong interference suppression, fast optimization efficiency and intuitive visualization.

[0008] A method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect the room location information and base station antenna information and perform preprocessing;

[0010] S2. Based on the preprocessed location information, a 3D point cloud model is constructed to extract the location relationship of each room and bind the room number with historical signal data and user complaint frequency.

[0011] S3. Build a spatial topology graph with rooms as nodes. The edge weights are the spatial distances between nodes. The node features are composed of the room's signal index, 3D spatial coordinates, and base station antenna features.

[0012] S4. Build a graph convolutional neural network, perform multi-layer graph convolution operations on the spatial topology graph, extract spatial semantic features, calculate the probability of each room being an outlier, and output an anomaly classification label;

[0013] S5. Combining anomaly classification labels with spatial semantic features, we construct an interference impact factor map and model the interference intensity of multiple base station antennas covering the anomaly point.

[0014] S6. Perform a multi-objective optimization search on the interference impact factor map, adjust the transmit power, azimuth angle, and downtilt angle of each antenna in the associated base station, and generate parameter adjustment suggestions for the target room.

[0015] S7. Map the parameter adjustment suggestions to the 3D point cloud model, display the optimization results in a 3D visualization, and output the intelligent dispatch execution path of the target room.

[0016] Optionally, the location information includes signal indicators, room numbers and three-dimensional space coordinates, and the antenna information includes longitude and latitude positions, base station height, azimuth angle, downtilt angle, and transmission power.

[0017] Optionally, the preprocessing includes missing value filling, outlier removal, timestamp alignment, coordinate format unification, signal filtering and denoising, base station parameter analysis and data standardization.

[0018] Optionally, the S2 specifically includes:

[0019] S21, extract the three-dimensional space coordinates of each room from the pre-processed position information to form a room set Q = {(x i ,y i ,z i )}, where x i 、y i 、z i Represent the coordinate values ​​of room i in the horizontal, longitudinal and vertical directions respectively;

[0020] S22. Based on the room set Q, a three-dimensional point cloud model is constructed using spatial density clustering. The coordinates of each room are mapped to spatial point cloud nodes. The building, floor, and apartment structure information of the room are extracted through spatial voxel partitioning.

[0021] S23, the historical signal data s i (t) is bound to the corresponding coordinate point (x i ,y i ,z i ), the s i (t) represents the signal strength value of room i at time t;

[0022] S24, user complaint frequency c i Bind to the corresponding coordinate point (x i ,y i ,z i ), the c i represents the cumulative number of complaints for room i per unit time;

[0023] S25, taking each coordinate point in the 3D point cloud model as a benchmark, and combining the corresponding historical signal data and user complaint frequency, construct and output the node feature vector fi =[x i ,y i ,z i ,s i (t),c i ].

[0024] Optionally, the S3 specifically includes:

[0025] S31, take each room in the 3D point cloud model as a graph node, and define the node set as V = {v1, v2, ..., v n}, where v i represents the node corresponding to room i, and n represents the number of nodes;

[0026] S32, according to any two nodes v i 、v j The corresponding coordinate point (x i ,y i ,z i )、(x j ,y j ,z j ), calculate the Minkowski distance between nodes, and use the Minkowski distance as the edge e ij The edge weights constitute the edge set E;

[0027] S33. Construct a spatial topology graph G = (V, E) using the node set V and the edge set E to represent the three-dimensional spatial connection relationship between the rooms;

[0028] S34. For each node v i , get the base station number b with the shortest distance i , extract base station number b i The corresponding antenna azimuth angle θ i , downtilt angle φ i , transmit power p i , construct the baseline antenna characteristic vector a i =[θ i ,φ i ,p i ];

[0029] S35, the baseline antenna characteristic vector a i and the node feature vector f i Splice to form a node representation vector h i , used for graph convolution operations.

[0030] Optionally, the S4 specifically includes:

[0031] S41, based on node set V and node v i The node representation vector h i , define the initial feature matrix H(0) It is the result of vertical concatenation of all node features, with dimensions of n rows and 8 columns;

[0032] S42. Calculate the adjacency matrix A, add unit values ​​to the main diagonal to obtain the matrix A+I, and then calculate the node degree matrix D, where the i-th diagonal element of the node degree matrix is ​​the sum of all elements in the i-th row in A+I, and define the normalized adjacency matrix as

[0033] S43. Build a multi-channel graph convolutional network and use the following propagation formula to update the feature matrix in each layer:

[0034]

[0035] Among them, H (l) represents the feature matrix of the lth layer, H (l+1) represents the feature matrix of the l+1th layer, represents the weight matrix of the k-th channel l-th layer, represents the corresponding bias matrix, Represents the kth channel attention weight, satisfying all channels The sum represents 1, K represents the number of convolution channels, represents the normalized adjacency matrix of the kth channel, Sigmoid(·) and ReLU(·) represent different nonlinear activation functions respectively;

[0036] S44, extract the spatial semantic features, and transform the feature matrix H (l) Combined with the feature means of all the first l+1 layers, an abnormality probability matrix is ​​constructed:

[0037]

[0038] Among them, P represents the abnormal probability matrix, W (p) represents the main discriminant weight matrix, b (p) represents the main discriminant bias matrix, Represents the column-wise average result of all feature matrices from layer 0 to layer l, W (q) represents the auxiliary discriminant weight matrix, b (q) represents the auxiliary discriminant bias matrix, ⊙ represents the element-by-element multiplication operation, tanh(·) represents the hyperbolic tangent function, and Softmax(·) represents the classification normalization function, which is used to output the probability that each row belongs to each class;

[0039] S45. Analyze the abnormal probability vector of each node and select the location of the maximum value as the abnormal classification label output of the node.

[0040] Optionally, the abnormal probability matrix P is composed of abnormal probability vectors corresponding to all nodes, combined with H(l) It forms a nonlinear expression structure with the feature mean of all the first l+1 layers, and uses element-wise multiplication to enhance the ability to express feature differences. It also outputs the abnormal probability vector corresponding to the node through the Softmax function to realize the association modeling of node spatial semantic features and abnormal probability.

[0041] Optionally, the abnormal classification labels include three types: signal attenuation, collaborative interference and spatial occlusion. The signal attenuation type corresponds to a room where the signal strength value is lower than a preset threshold and there is no significant interference source. The collaborative interference type corresponds to a situation where the superimposed signal strength of antennas from multiple directions exceeds the interference threshold. The spatial occlusion type corresponds to a situation where there is a physical occlusion structure between the room and the main service antenna and the antenna direction angle deviates from the direction of the line connecting the center of the room.

[0042] Optionally, the S5 specifically includes:

[0043] S51. According to the abnormal classification label, let the label set be where l i Represents the abnormal type number of room i, which includes signal attenuation, collaborative interference and spatial occlusion. The abnormal node set is defined as where u i represents the abnormal node corresponding to room i;

[0044] S52. Assume that the number set of base stations in the space is Each base station The three-dimensional space coordinates are (x j ,y j ,z j ), the baseline antenna characteristics are: transmit power P j , azimuth angle θ j , downtilt angle φ j ;

[0045] S53, according to any abnormal node u i The three-dimensional space coordinates (x i ,y i ,z i ), computing node u i With base station b j Manhattan distance d ij , filter out ij <T d The base station number, where T d is the distance threshold;

[0046] S54. For each abnormal node u i , traverse u i Covered base station number set Construct the edges in the interference factor graph and calculate thei ,b j )’s interference intensity value:

[0047]

[0048] s max (t)=max(s k (t)|k=1,2,...,n);

[0049]

[0050] Among them, I ij Indicates base station b j For abnormal node u i The interference intensity, P j Indicates base station b j Current antenna transmit power, G ij Indicates base station b j Pointing to abnormal node u i Directional gain, α ij Indicates base station b j Pointing to abnormal node u i The angle in the horizontal projection direction, β ij Indicates abnormal node u i Relative base station b j The pitch angle, ∈ is a small positive constant to prevent the denominator from being zero, s i (t) represents the signal strength value of room i at time t, s max (t) represents the maximum signal strength of all room nodes at time t, λ is the abnormal interference enhancement coefficient, which is used to adjust the signal distortion sensitivity, arctan(·) represents the inverse tangent function, max(·) represents the maximum value function, and x i ,y i ,z i They represent abnormal nodes u respectively i The horizontal, vertical, and height coordinates, x j ,y j ,z j Represent base station b j The horizontal, vertical and height coordinates of , n represents the number of nodes;

[0051] S55, all edges (u i ,b j ) and the corresponding interference intensity I ij Construct weighted edges to generate an interference impact factor graph. The nodes in the graph are room nodes and base station nodes, and the edge weight is the corresponding interference intensity I ij , used in the multi-base station collaborative optimization process.

[0052] Optionally, the S6 specifically includes:

[0053] S61, extract all edges (v i ,b j ), build interference optimization target set Among them I ij Indicates base station b j For abnormal node u i Interference intensity, s i (t) represents the signal strength value of room i at time t, l i Indicates the exception type number of room i;

[0054] S62, define each base station b j The parameter vector is where ΔP j is the transmit power adjustment, Δθ j is the azimuth adjustment amount, Δφ j is the downtilt angle adjustment amount;

[0055] S63. Construct a multi-objective optimization function:

[0056]

[0057] in, Indicates base station b after parameter adjustment j For abnormal node u i The interference intensity value, Indicates abnormal node u after parameter adjustment i The received signal enhancement value is calculated based on the re-estimation of the directional gain function. It represents the coverage loss value of the base station to other normal nodes after parameter adjustment. ω1, ω2, and ω3 are the weight coefficients of interference suppression, signal enhancement, and coverage balance. is the set of abnormal nodes, Represents the parameter vector obtained when the function reaches its minimum value

[0058] S64. Under the above multi-objective optimization function, a heuristic search method is used to generate a parameter adjustment solution set, and the optimal parameter combination that meets the optimization goal is extracted, which is recorded as

[0059] S65. Combine the optimal parameters The corresponding base station number, target room number, and adjustment result are written into the parameter suggestion table to generate a structured antenna adjustment suggestion output.

[0060] The beneficial effects of the present invention are:

[0061] First, this invention constructs a room-by-room 3D point cloud model, incorporating the physical parameters and spatial locations of base station antennas. This overcomes the inability of traditional 2D planning methods to accurately describe a building's internal structure and signal propagation paths, effectively improving the accuracy of the fit between signal coverage and building structure. By combining historical signal data with user complaint data in 3D space, high-resolution spatial mapping of network quality is achieved, enabling more precise identification of problem areas.

[0062] Secondly, the present invention introduces a graph convolutional neural network to learn the constructed spatial topology map, integrating the spatial location of room nodes, signal indicators, and antenna information to accurately model the nonlinear relationship between coverage quality and network parameters, and accurately output abnormal rooms and their corresponding classification labels. At the same time, it constructs an interference impact factor map and combines azimuth angle, downtilt angle, and transmit power parameters to support multi-base station, multi-parameter, and multi-objective collaborative optimization. This overcomes the shortcomings of traditional methods, which have difficulty handling high-dimensional parameter spaces and are prone to falling into local optimality, significantly improving the intelligence and global optimality of network adjustments.

[0063] Finally, this method maps the optimization results into a three-dimensional point cloud model and visually displays the impact of each base station adjustment on the room's signal strength, helping operations personnel quickly understand signal behavior after parameter changes. Furthermore, it generates intelligent dispatch paths based on spatial coordinates, achieving a closed-loop linkage between optimization, execution, and scheduling. This streamlines the entire process from problem identification to on-site execution, significantly shortening operation and maintenance response time and improving the automation level and execution efficiency of network optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1 This is a flow chart of the wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling proposed by the present invention;

[0066] Figure 2 A schematic diagram of constructing a three-dimensional point cloud model of the wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling proposed in the present invention;

[0067] Figure 3 This is a schematic diagram of the spatial topology diagram construction and feature combination of the wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling proposed by the present invention. DETAILED DESCRIPTION

[0068] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0069] refer to Figure 1-3 , a wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling, comprising the following steps:

[0070] S1. Collect the room location information and base station antenna information and perform preprocessing;

[0071] S2. Based on the preprocessed location information, a 3D point cloud model is constructed to extract the location relationship of each room and bind the room number with historical signal data and user complaint frequency.

[0072] S3. Build a spatial topology graph with rooms as nodes. The edge weights are the spatial distances between nodes. The node features are composed of the room's signal index, 3D spatial coordinates, and base station antenna features.

[0073] S4. Build a graph convolutional neural network, perform multi-layer graph convolution operations on the spatial topology graph, extract spatial semantic features, calculate the probability of each room being an outlier, and output an anomaly classification label;

[0074] S5. Combining anomaly classification labels with spatial semantic features, we construct an interference impact factor map and model the interference intensity of multiple base station antennas covering the anomaly point.

[0075] S6. Perform a multi-objective optimization search on the interference impact factor map, adjust the transmit power, azimuth angle, and downtilt angle of each antenna in the associated base station, and generate parameter adjustment suggestions for the target room.

[0076] S7. Map the parameter adjustment suggestions to the 3D point cloud model, display the optimization results in a 3D visualization, and output the intelligent dispatch execution path of the target room.

[0077] The present invention achieves accurate modeling and collaborative optimization of wireless network coverage quality in complex building environments by constructing a full-process network optimization method from data acquisition, three-dimensional modeling, spatial graph construction, graph neural reasoning, interference modeling, parameter optimization to three-dimensional visualization, improves the accuracy of problem identification and the efficiency of multi-base station collaborative parameter adjustment, and significantly enhances the intelligent level of network optimization.

[0078] In this embodiment, the location information includes signal indicators, room numbers and three-dimensional space coordinates, and the antenna information includes latitude and longitude positions, base station height, azimuth angle, downtilt angle, and transmission power.

[0079] This invention clarifies the spatial data composition of rooms and antennas. By combining signal indicators, three-dimensional coordinates and antenna parameters, the subsequent modeling process has a stronger physical correlation in spatial semantic expression and signal propagation modeling, effectively enhancing the algorithm model's modeling ability for propagation paths and obstacle identification.

[0080] In this embodiment, the preprocessing includes missing value filling, outlier removal, timestamp alignment, coordinate format unification, signal filtering and denoising, base station parameter analysis and data standardization.

[0081] By introducing data preprocessing methods such as missing value filling, outlier removal, time alignment and format unification, the present invention ensures the integrity and consistency of input data for three-dimensional modeling and subsequent calculation processes, thereby improving model stability and processing robustness.

[0082] In this embodiment, S2 specifically includes:

[0083] S21, extract the three-dimensional space coordinates of each room from the pre-processed position information to form a room set Q = {(x i ,y i ,z i )}, where x i 、y i 、z i Represent the coordinate values ​​of room i in the horizontal, longitudinal and vertical directions respectively;

[0084] S22. Based on the room set Q, a three-dimensional point cloud model is constructed using spatial density clustering. The coordinates of each room are mapped to spatial point cloud nodes. The building, floor, and apartment structure information of the room are extracted through spatial voxel partitioning.

[0085] S23, the historical signal data s i (t) is bound to the corresponding coordinate point (x i ,y i ,z i ), the s i (t) represents the signal strength value of room i at time t;

[0086] S24, user complaint frequency c i Bind to the corresponding coordinate point (x i ,y i ,z i ), the c i represents the cumulative number of complaints for room i per unit time;

[0087] S25, taking each coordinate point in the 3D point cloud model as a benchmark, and combining the corresponding historical signal data and user complaint frequency, construct and output the node feature vector f i =[xi ,y i ,z i ,s i (t),c i ].

[0088] The present invention uses spatial density clustering and voxel partitioning methods to construct a three-dimensional point cloud model, explicitly mapping structural information such as rooms, floors, and apartment types to a spatial grid, and combining signal and complaint information to complete high-dimensional spatial feature fusion, thereby improving the degree of refinement of spatial modeling and problem location capabilities.

[0089] In this embodiment, S3 specifically includes:

[0090] S31, take each room in the 3D point cloud model as a graph node, and define the node set as V = {v1, v2, ..., v n}, where v i represents the node corresponding to room i, and n represents the number of nodes;

[0091] S32, according to any two nodes v i 、v j The corresponding coordinate point (x i ,y i ,z i )、(x j ,y j ,z j ), calculate the Minkowski distance between nodes, and use the Minkowski distance as the edge e ij The edge weights constitute the edge set E;

[0092] S33. Construct a spatial topology graph G = (V, E) using the node set V and the edge set E to represent the three-dimensional spatial connection relationship between the rooms;

[0093] S34. For each node v i , get the base station number b with the shortest distance i , extract base station number b i The corresponding antenna azimuth angle θ i , downtilt angle φ i , transmit power p i , construct the baseline antenna characteristic vector a i =[θ i ,φ i ,p i ];

[0094] S35, the baseline antenna characteristic vector a i and the node feature vector f i Splice to form a node representation vector h i , used for graph convolution operations.

[0095] The present invention constructs a spatial topology graph based on the three-dimensional spatial relationship between rooms, introduces the Minkowski distance as the edge weight and integrates the antenna direction and power characteristics, realizing the multi-dimensional integration of structure and signal, enabling the graph neural network to have stronger generalization ability and precision performance in the process of spatial semantic expression.

[0096] In this embodiment, the S4 specifically includes:

[0097] S41, based on node set V and node v i The node representation vector h i , define the initial feature matrix H (0) It is the result of vertical concatenation of all node features, with dimensions of n rows and 8 columns;

[0098] S42. Calculate the adjacency matrix A, add unit values ​​to the main diagonal to obtain the matrix A+I, and then calculate the node degree matrix D, where the i-th diagonal element of the node degree matrix is ​​the sum of all elements in the i-th row in A+I, and define the normalized adjacency matrix as

[0099] S43. Build a multi-channel graph convolutional network and use the following propagation formula to update the feature matrix in each layer:

[0100]

[0101] Among them, H (l) represents the feature matrix of the lth layer, H (l+1) represents the feature matrix of the l+1th layer, represents the weight matrix of the k-th channel l-th layer, represents the corresponding bias matrix, Represents the kth channel attention weight, satisfying all channels The sum represents 1, K represents the number of convolution channels, represents the normalized adjacency matrix of the kth channel, Sigmoid(·) and ReLU(·) represent different nonlinear activation functions respectively;

[0102] S44, extract the spatial semantic features, and transform the feature matrix H (l) Combined with the feature means of all the first l+1 layers, an abnormality probability matrix is ​​constructed:

[0103]

[0104] Among them, P represents the abnormal probability matrix, W (p) represents the main discriminant weight matrix, b (p) represents the main discriminant bias matrix, Represents the column-wise average result of all feature matrices from layer 0 to layer l, W(q) represents the auxiliary discriminant weight matrix, b (q) represents the auxiliary discriminant bias matrix, ⊙ represents the element-by-element multiplication operation, tanh(·) represents the hyperbolic tangent function, and Softmax(·) represents the classification normalization function, which is used to output the probability that each row belongs to each class;

[0105] S45. Analyze the abnormal probability vector of each node and select the location of the maximum value as the abnormal classification label output of the node.

[0106] The multi-channel graph convolution structure designed in this paper integrates node features and spatial structures. Through feature mean aggregation, dual-branch nonlinear mapping and abnormal probability output mechanism, it improves the recognition accuracy of different types of abnormal signal areas and provides high-confidence classification labels for subsequent optimization.

[0107] In this embodiment, the abnormal probability matrix P is composed of the abnormal probability vectors corresponding to all nodes, combined with H (l) It forms a nonlinear expression structure with the feature mean of all the first l+1 layers, and uses element-wise multiplication to enhance the ability to express feature differences. It also outputs the abnormal probability vector corresponding to the node through the Softmax function to realize the association modeling of node spatial semantic features and abnormal probability.

[0108] The present invention constructs an interactive mechanism between the anomaly probability matrix and the multi-layer feature average vector, and enhances the expression of spatial feature differences through element-by-element multiplication, making the anomaly recognition results more sensitive and discriminative, effectively supporting the accurate classification and interpretation of complex coverage problems.

[0109] In this embodiment, the abnormal classification labels include three types: signal attenuation, collaborative interference, and spatial occlusion. The signal attenuation type corresponds to a room where the signal strength value is lower than a preset threshold and there is no significant interference source. The collaborative interference type corresponds to a room where the superimposed signal strength of antennas from multiple directions exceeds the interference threshold. The spatial occlusion type corresponds to a room where there is a physical occlusion structure between the room and the main service antenna and the antenna direction angle deviates from the direction of the line connecting the center of the room.

[0110] By dividing abnormal areas into three categories: signal attenuation, collaborative interference, and spatial occlusion, and combining specific signal states with physical relationship conditions, the present invention provides an operational abnormality root cause judgment mechanism for subsequent optimization models, effectively supporting differentiated processing strategies for different types of problems.

[0111] In this embodiment, the S5 specifically includes:

[0112] S51. According to the abnormal classification label, let the label set be where l iRepresents the abnormal type number of room i, which includes signal attenuation, collaborative interference and spatial occlusion. The abnormal node set is defined as where u i represents the abnormal node corresponding to room i;

[0113] S52. Assume that the number set of base stations in the space is Each base station The three-dimensional space coordinates are (x j ,y j ,z j ), the baseline antenna characteristics are: transmit power P j , azimuth angle θ j , downtilt angle φ j ;

[0114] S53, according to any abnormal node u i The three-dimensional space coordinates (x i ,y i ,z i ), computing node u i With base station b j Manhattan distance d ij , filter out ij <T d The base station number, where T d is the distance threshold;

[0115] S54. For each abnormal node u i , traverse u i Covered base station number set Construct the edges in the interference factor graph and calculate the i ,b j )’s interference intensity value:

[0116]

[0117] s max (t)=max(s k (t)|k=1,2,...,n);

[0118]

[0119] Among them, I ij Indicates base station b j For abnormal node u i The interference intensity, P j Indicates base station b j Current antenna transmit power, G ij Indicates base station b j Pointing to abnormal node u i Directional gain, αij denotes the base station b j points to the abnormal node u i the included angle in the horizontal projection direction, β ij denotes the abnormal node u i relative to the base station b j the pitch angle, ∈ is a small positive constant to prevent the denominator from being zero, s i (t) represents the signal strength value of the room i at time t, s max (t) represents the maximum value of the signal strength of all room nodes at time t, λ is an abnormal interference enhancement coefficient for adjusting the signal distortion sensitivity, arctan(·) represents the inverse tangent function, max(·) represents the maximum value function, x i , y i , z i respectively represent the horizontal, vertical and height coordinates of the abnormal node u i , x j , y j , z j respectively represent the horizontal, vertical and height coordinates of the base station b j , and n represents the number of nodes.

[0120] S55, all edges (u i , b j ) and the corresponding interference intensity I ij are constructed into weighted edges to generate an interference influence factor graph, the nodes in the graph are room nodes and base station nodes, and the edge weight is the corresponding interference intensity I ij , which is used in the process of multi-base station cooperative optimization.

[0121] The present application constructs an interference influence factor graph and introduces direction gain, angle deviation and signal strength to model the interference intensity, so that the cooperative interference identification not only has angle interpretation, but also quantifies the interference relationship between the base station and the room, thereby establishing a precise foundation for parameter optimization.

[0122] In the embodiment, the S6 specifically comprises:

[0123] S61, all edges (v i , b j ) in the interference influence factor graph are extracted to construct an interference optimization target set wherein I ij represents the interference intensity of the base station b j to the abnormal node u i , s i (t) represents the signal strength value of the room i at time t, and l i represents the abnormal type number of the room i.

[0124] S62, define each base station b j The parameter vector is where ΔP j is the transmit power adjustment, Δθ j is the azimuth adjustment amount, Δφ j is the downtilt angle adjustment amount;

[0125] S63. Construct a multi-objective optimization function:

[0126]

[0127] in, Indicates base station b after parameter adjustment j For abnormal node u i The interference intensity value, Indicates abnormal node u after parameter adjustment i The received signal enhancement value is calculated based on the re-estimation of the directional gain function. It represents the coverage loss value of the base station to other normal nodes after parameter adjustment. ω1, ω2, and ω3 are the weight coefficients of interference suppression, signal enhancement, and coverage balance. is the set of abnormal nodes, Represents the parameter vector obtained when the function reaches its minimum value

[0128] S64. Under the above multi-objective optimization function, a heuristic search method is used to generate a parameter adjustment solution set, and the optimal parameter combination that meets the optimization goal is extracted, which is recorded as

[0129] S65. Combine the optimal parameters The corresponding base station number, target room number, and adjustment result are written into the parameter suggestion table to generate a structured antenna adjustment suggestion output.

[0130] The present invention jointly models the three elements of interference suppression, signal enhancement and coverage maintenance through a multi-objective optimization function, and generates an efficient and convergent antenna adjustment solution set based on heuristic search, thereby realizing the joint tuning of multiple antenna parameters. The output parameter recommendations are characterized by high accuracy and strong adaptability.

[0131] Example 1:

[0132] To verify the feasibility of this invention, we applied it to optimizing wireless communication coverage in a densely populated area of ​​high-rise residential buildings. This area has a high building density, close spacing between buildings, and a typical floor count of 20 or more. Network complaints are frequent, with users reporting issues such as unstable network signals, high rates of dropped voice calls, slow download speeds, and video playback lag. Existing optimization solutions primarily rely on empirical adjustments based on MR data and complaint hotspots. However, the actual coverage improvement is limited, and complaints persist even after adjustments.

[0133] In this scenario, the present invention first collected the building structure data, three-dimensional spatial coordinate information, historical signal strength records, and user complaint frequency data of all rooms in the area. After preprocessing the unified coordinate format and eliminating invalid data, a three-dimensional point cloud model containing 8,320 room nodes was constructed. Combined with spatial density analysis, the system automatically identified the building hierarchy and household type affiliation, forming a three-dimensional spatial distribution map. Subsequently, a spatial topology map was constructed based on the relative positions between room nodes and base station antennas. A total of 162,450 spatial edge connections were generated in the map, and Manhattan distance was used as the edge weight to achieve accurate modeling of building spacing and signal diffusion distance.

[0134] During the graph convolutional neural network analysis phase, the model performed a three-layer graph convolution on the spatial topology map based on initial features derived from signal strength, complaint frequency, 3D coordinates, and antenna parameters. The model ultimately output anomaly probabilities and classification labels for each room. 2,376 abnormal rooms, representing 28.5%, were identified. Classification results showed that signal attenuation accounted for 47.2%, collaborative interference accounted for 35.9%, and spatial occlusion accounted for 16.9%. The model accurately identified multiple problematic scenarios, including antenna downtilt angles exceeding 20 degrees from the center of the target room, antenna main lobes being blocked, or excessively overlapped.

[0135] Based on the modeling results of the interference impact factor map, the system calculated the interference strength and constructed an optimization objective function for 142 base station antennas in the abnormal room coverage. Through a heuristic multi-objective optimization process, 563 candidate parameter solutions were automatically generated. Ultimately, 42 solutions were selected that met the optimal conditions for signal enhancement, interference suppression, and coverage, and a list of recommended antenna parameter adjustments was formed.

[0136] By adjusting the transmit power to no more than ±3dBm, the azimuth angle to no more than ±15 degrees, and the downtilt angle to be fine-tuned within the range of ±5 degrees, the 42 parameter change plans that were finally implemented were successfully implemented, and a periodic network performance comparison evaluation was conducted in the region. The evaluation data showed that the overall RSRP increased by an average of 3.7dB, the SINR increased by an average of 4.1dB, the user voice drop rate decreased by 62.8%, and the number of complaint tickets decreased by 51.3%. In particular, in multiple high-rise buildings identified as spatial obstructions, the RSRP increased by up to 8.9dB and the SINR increased by up to 9.4dB, verifying the accuracy of the structural modeling and the effectiveness of collaborative optimization of the present invention in complex spatial scenarios.

[0137] To further improve O&M efficiency, this invention also projects optimization suggestions into the building point cloud structure through 3D visualization. This, combined with the dispatch system, directly generates an inspection route map for each target room. Engineers can now execute and verify parameters using a visual interface. Survey feedback indicates that the dispatch execution cycle has been shortened from an average of 4.3 days to 2.1 days, a 51.2% reduction. The following table compares key indicators after implementing this invention (for selected rooms):

[0138] Table 1 Comparison of network performance and complaints before and after optimization in typical rooms

[0139]

[0140]

[0141] This example fully verifies the feasibility, visual expression capabilities, and collaborative tuning effects of the present invention in actual complex building environments. It outperforms traditional manual experience-based parameter tuning methods in multiple dimensions, providing an engineering-level verification foundation for future-oriented automated intelligent network optimization.

[0142] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling, characterized in that: The steps include: S1. Collect the room location information and base station antenna information and perform preprocessing; S2. Based on the preprocessed location information, a 3D point cloud model is constructed to extract the location relationship of each room and bind the room number with historical signal data and user complaint frequency. S3. Build a spatial topology graph with rooms as nodes. The edge weights are the spatial distances between nodes. The node features are composed of the room's signal index, 3D spatial coordinates, and base station antenna features. S4. Build a graph convolutional neural network, perform multi-layer graph convolution operations on the spatial topology graph, extract spatial semantic features, calculate the probability of each room being an outlier, and output an anomaly classification label; S5. Combining anomaly classification labels with spatial semantic features, we construct an interference impact factor map and model the interference intensity of multiple base station antennas covering the anomaly point. S6. Perform a multi-objective optimization search on the interference impact factor map, adjust the transmit power, azimuth angle, and downtilt angle of each antenna in the associated base station, and generate parameter adjustment suggestions for the target room. S7. Map the parameter adjustment suggestions to the 3D point cloud model, display the optimization results in a 3D visualization, and output the intelligent dispatch execution path of the target room.

2. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The location information includes signal indicators, room numbers and three-dimensional space coordinates, and the antenna information includes longitude and latitude positions, base station height, azimuth angle, downtilt angle and transmission power.

3. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The preprocessing includes missing value filling, outlier removal, timestamp alignment, coordinate format unification, signal filtering and denoising, base station parameter analysis and data standardization.

4. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The S2 specifically includes: S21, extract the three-dimensional space coordinates of each room from the pre-processed position information to form a room set Q = {(x i ,y i ,z i )}, where x i 、y i 、z i Represent the coordinate values ​​of room i in the horizontal, longitudinal and vertical directions respectively; S22. Based on the room set Q, a three-dimensional point cloud model is constructed using spatial density clustering. The coordinates of each room are mapped to spatial point cloud nodes. The building, floor, and apartment structure information of the room are extracted through spatial voxel partitioning. S23, the historical signal data s i (t) is bound to the corresponding coordinate point (x i ,y i ,z i ), the s i (t) represents the signal strength value of room i at time t; S24, user complaint frequency c i Bind to the corresponding coordinate point (x i ,y i ,z i ), the c i represents the cumulative number of complaints for room i per unit time; S25, taking each coordinate point in the 3D point cloud model as a benchmark, and combining the corresponding historical signal data and user complaint frequency, construct and output the node feature vector f i =[x i ,y i ,z i ,s i (t),c i ].

5. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The S3 specifically includes: S31, take each room in the 3D point cloud model as a graph node, and define the node set as V = {v1, v2, ..., v n }, where v i represents the node corresponding to room i, and n represents the number of nodes; S32, according to any two nodes v i 、v j The corresponding coordinate point (x i ,y i ,z i )、(x j ,y j ,z j ), calculate the Minkowski distance between nodes, and use the Minkowski distance as the edge e ij The edge weights constitute the edge set E; S33. Construct a spatial topology graph G = (V, E) using the node set V and the edge set E to represent the three-dimensional spatial connection relationship between the rooms; S34. For each node v i , get the base station number b with the shortest distance i , extract base station number b i The corresponding antenna azimuth angle θ i , downtilt angle φ i , transmit power p i , construct the baseline antenna characteristic vector a i =[θ i ,φ i ,p i ]; S35, the baseline antenna characteristic vector a i and node feature vector f i Splice to form a node representation vector h i , used for graph convolution operations.

6. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The S4 specifically includes: S41, based on node set V and node v i The node representation vector h i , define the initial feature matrix H (0) It is the result of vertical concatenation of all node features, with dimensions of n rows and 8 columns; S42. Calculate the adjacency matrix A, add unit values ​​to the main diagonal to obtain the matrix A+I, and then calculate the node degree matrix D, where the i-th diagonal element of the node degree matrix is ​​the sum of all elements in the i-th row in A+I, and define the normalized adjacency matrix as S43. Build a multi-channel graph convolutional network and use the following propagation formula to update the feature matrix in each layer: Among them, H (l) represents the feature matrix of the lth layer, H (l+1) represents the feature matrix of the l+1th layer, represents the weight matrix of the k-th channel l-th layer, represents the corresponding bias matrix, Represents the kth channel attention weight, satisfying all channels The sum represents 1, K represents the number of convolution channels, represents the normalized adjacency matrix of the kth channel, Sigmoid(·) and ReLU(·) represent different nonlinear activation functions respectively; S44, extract the spatial semantic features, and transform the feature matrix H (l) Combined with the feature means of all the first l+1 layers, an abnormality probability matrix is ​​constructed: Among them, P represents the abnormal probability matrix, W (p) represents the main discriminant weight matrix, b (p) represents the main discriminant bias matrix, Represents the column-wise average result of all feature matrices from layer 0 to layer l, W (q) represents the auxiliary discriminant weight matrix, b (q) represents the auxiliary discriminant bias matrix, ⊙ represents the element-by-element multiplication operation, tanh(·) represents the hyperbolic tangent function, and Softmax(·) represents the classification normalization function, which is used to output the probability that each row belongs to each class; S45. Analyze the abnormal probability vector of each node and select the location of the maximum value as the abnormal classification label output of the node.

7. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 6, characterized in that: The abnormal probability matrix P is composed of the abnormal probability vectors corresponding to all nodes, combined with H (l) It forms a nonlinear expression structure with the feature mean of all the first l+1 layers, and uses element-wise multiplication to enhance the ability to express feature differences. It also outputs the abnormal probability vector corresponding to the node through the Softmax function to realize the association modeling of node spatial semantic features and abnormal probability.

8. The wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling according to claim 6 is characterized in that: The abnormal classification labels include three types: signal attenuation, collaborative interference, and spatial occlusion. The signal attenuation type corresponds to a room where the signal strength value is lower than a preset threshold and there is no significant interference source. The collaborative interference type corresponds to a situation where the superimposed signal strength of antennas from multiple directions exceeds the interference threshold. The spatial occlusion type corresponds to a situation where there is a physical occlusion structure between the room and the main service antenna and the antenna direction angle deviates from the direction of the line connecting the center of the room.

9. The method for connecting and coordinating wireless network antenna base stations based on three-dimensional modeling according to claim 1, characterized in that: The S5 specifically includes: S51. According to the abnormal classification label, let the label set be where l i Represents the abnormal type number of room i, which includes signal attenuation, collaborative interference and spatial occlusion. The abnormal node set is defined as where u i represents the abnormal node corresponding to room i; S52. Assume that the number set of base stations in the space is Each base station The three-dimensional space coordinates are (x j ,y j ,z j ), the baseline antenna characteristics are: transmit power P j , azimuth angle θ j , downtilt angle φ j ; S53, according to any abnormal node u i The three-dimensional space coordinates (x i ,y i ,z i ), computing node u i With base station b j Manhattan distance d ij , filter out ij <T d The base station number, where T d is the distance threshold; S54. For each abnormal node u i , traverse u i Covered base station number set Construct the edges in the interference factor graph and calculate the i ,b j )’s interference intensity value: Among them, I ij Indicates base station b j For abnormal node u i The interference intensity, P j Indicates base station b j Current antenna transmit power, G ij Indicates base station b j Pointing to abnormal node u i Directional gain, α ij Indicates base station b j Pointing to abnormal node u i The angle in the horizontal projection direction, β ij Indicates abnormal node u i Relative base station b j The pitch angle, ∈ is a small positive constant to prevent the denominator from being zero, s i (t) represents the signal strength value of room i at time t, s max (t) represents the maximum signal strength of all room nodes at time t, λ is the abnormal interference enhancement coefficient, which is used to adjust the signal distortion sensitivity, arctan(·) represents the inverse tangent function, max(·) represents the maximum value function, and x i ,y i ,z i They represent abnormal nodes u respectively i The horizontal, vertical, and height coordinates, x j ,y j ,z j Represent base station b j The horizontal, vertical and height coordinates of , n represents the number of nodes; S55, all edges (u i ,b j ) and the corresponding interference intensity I ij Construct weighted edges to generate an interference impact factor graph. The nodes in the graph are room nodes and base station nodes, and the edge weight is the corresponding interference intensity I ij , used in the multi-base station collaborative optimization process.

10. The wireless network antenna base station connection and collaborative optimization method based on three-dimensional modeling according to claim 1, characterized in that: The S6 specifically includes: S61, extract all edges (v i ,b j ), build interference optimization target set Among them I ij Indicates base station b j For abnormal node u i Interference intensity, s i (t) represents the signal strength value of room i at time t, l i Indicates the exception type number of room i; S62, define each base station b j The parameter vector is where ΔP j is the transmit power adjustment, Δθ j is the azimuth adjustment amount, Δφ j is the downtilt angle adjustment amount; S63. Construct a multi-objective optimization function: in, Indicates base station b after parameter adjustment j For abnormal node u i The interference intensity value, Indicates abnormal node u after parameter adjustment i The received signal enhancement value is calculated based on the re-estimation of the directional gain function. It represents the coverage loss value of the base station to other normal nodes after parameter adjustment. ω1, ω2, and ω3 are the weight coefficients of interference suppression, signal enhancement, and coverage balance. is the set of abnormal nodes, Represents the parameter vector obtained when the function reaches its minimum value S64. Under the above multi-objective optimization function, a heuristic search method is used to generate a parameter adjustment solution set, and the optimal parameter combination that meets the optimization goal is extracted, which is recorded as S65. Combine the optimal parameters The corresponding base station number, target room number, and adjustment result are written into the parameter suggestion table to generate a structured antenna adjustment suggestion output.

Citation Information

Patent Citations

  • Wireless network three-dimensional coverage optimization method

    CN112752268A

  • Network deployment strategy determination method and device and computer equipment

    CN119172764A

  • Systems and Methods for Measurement Integration and Accurate Digital Twin Creation

    US20240302516A1

Cited By

  • UAV positioning and joint deployment and resource allocation method based on radar

    CN121254254A