Three-dimensional space network quality mapping and prediction method and system based on AI model

Abstract

The application discloses a three-dimensional space network quality mapping and prediction method and system based on an AI model, belongs to the technical field of communication network operation and maintenance, and aims to solve the technical problem of how to overcome the defects of data islands and response lag in communication network operation and maintenance and efficiently and accurately realize network operation and maintenance.The method comprises the following steps: constructing an inference model based on a Transformer-3D network model with a time-space attention mechanism introduced;outputting signal strength prediction values of each position coordinate in a three-dimensional space, position coordinates and occurrence probabilities of poor-quality areas based on the trained inference model;calculating three-dimensional space network quality values based on an improved Kriging interpolation algorithm with a building shelter factor introduced, generating a three-dimensional quality heat map, and marking poor-quality areas in the three-dimensional quality heat map;and generating a targeted network optimization strategy according to the poor-quality area information presented by the three-dimensional quality map.

Description

Technical Field

[0001] This invention relates to the field of communication network operation and maintenance technology, specifically to a three-dimensional spatial network quality mapping and prediction method and system based on AI models. Background Technology

[0002] Traditional network quality assessment relies on manual drive testing or fixed sensors, which has the following problems:

[0003] 1. Limited data dimension: It can only obtain local signal strength (RSRP / RSSI / SINR), lacking spatial continuity;

[0004] 2. Delayed response: The discovery of poor-quality areas relies on user complaints, and the average repair time is too long;

[0005] 3. Inadequate analysis: It fails to correlate multiple factors such as equipment status and environmental interference, resulting in insufficient accuracy in root cause localization.

[0006] While existing technologies such as drone inspections can improve efficiency, they do not solve the following core problems:

[0007] 1. Data silos: Data such as spectrum, images, and device logs are not deeply integrated;

[0008] 2. Lack of dynamic prediction: There is a lack of modeling for network quality degradation trends based on historical data;

[0009] 3. Insufficient 3D visualization: 2D maps cannot reflect the vertical distribution of signals, such as signal blind spots in high-rise buildings.

[0010] Traditional network quality assessment relies on manual drive testing or fixed sensors, which suffers from problems such as limited data dimensions, delayed response, and coarse analysis, making it difficult to achieve accurate and efficient network operation and maintenance. Overcoming the deficiencies of data silos and delayed response in communication network operation and maintenance, and achieving efficient and accurate network operation and maintenance, is a technical problem that needs to be solved. Summary of the Invention

[0011] The technical objective of this invention is to address the above-mentioned shortcomings by providing a three-dimensional spatial network quality mapping and prediction method and system based on AI models, in order to solve the technical problems of overcoming the defects of data silos and response delays in communication network operation and maintenance, and to achieve efficient and accurate network operation and maintenance.

[0012] In a first aspect, the present invention provides a three-dimensional spatial network quality mapping and prediction method based on an AI model, comprising the following steps:

[0013] Multimodal data acquisition: Drones equipped with specialized equipment perform inspection tasks in the network coverage area to be inspected, collecting spectrum data to reflect network performance, 3D point cloud data to reflect the spatial structure of the network environment, and positioning data to reflect the drone's geographical location. Timestamps and location coordinates are added to the spectrum data and point cloud data to form multimodal data with spatiotemporal correlation.

[0014] Data preprocessing: Multimodal data is preprocessed through data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling to obtain effective datasets and quality defect labeling information. The effective datasets include spatiotemporally aligned signal indicators, environmental features, and location coordinates, while the quality defect labeling information includes the location coordinates of the quality defect areas, the quality defect type, and the duration.

[0015] Inference and Prediction: Based on timestamps and location coordinates, historical valid datasets and historical status data of network devices are integrated into a sample set. An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect labeling information. Inference and prediction are performed based on the trained inference model. The predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the probability of occurrence are output as the inference results.

[0016] 3D quality map generation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm that incorporates building occlusion factors. Based on the effective dataset and inference results, the network quality analysis model is used to calculate the 3D spatial network quality values. Combined with dynamic layered rendering, a 3D quality heat map is generated, and poor quality areas are marked in the 3D quality heat map.

[0017] Network optimization: Based on the poor quality area information presented by the 3D quality map, a list of location coordinates to be optimized is displayed. The network resource information of the corresponding area is presented by associating the location coordinates, and targeted network optimization strategies are generated. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

[0018] Preferably, the drone is equipped with dedicated equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. Multimodal data acquisition includes the following operations:

[0019] Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment will carry out inspection tasks according to the planned routes at regular intervals.

[0020] Network performance data is collected by a spectrum analyzer mounted on a drone. The network performance data includes signal strength RSRP / RSSI / SINR, bit error rate, and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset that includes signal indicators and time.

[0021] By capturing images of the network environment using an RGB-D camera mounted on a drone, and calculating the three-dimensional spatial coordinates of each point in the network environment image based on depth sensing technology, a three-dimensional point cloud dataset is generated to identify obstacles including buildings and trees.

[0022] By acquiring geographic coordinates and timestamps in real time through the GPS module mounted on the drone, and adding timestamps and geographic coordinates to the spectrum dataset and point cloud dataset, a multimodal dataset with spatiotemporal dimensions is formed.

[0023] Preferably, data preprocessing includes the following operations:

[0024] Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. A Huffman tree is constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles, resulting in compressed point cloud data.

[0025] Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data;

[0026] Feature extraction: Using compressed point cloud data as input, environmental features are extracted through a CNN model; using smooth, interference-free spectral data as input, the temporal variation features of the signal are captured through an LSTM model, and the signal features are output.

[0027] Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset;

[0028] Quality defect annotation: Based on the preset quality defect judgment rules, it traverses the spatiotemporally aligned dataset, identifies and annotates the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect annotation information including geographic coordinates, quality defect type and duration.

[0029] As a preferred method, inference prediction includes the following operations:

[0030] Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference annotation information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set;

[0031] Model Construction: An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the occurrence probability as the inference result. The quality defect type corresponding to the quality defect area includes weak signal, network congestion and signal interference.

[0032] Inference and Prediction: The trained inference model is invoked to predict the signal strength values ​​of each coordinate point in three-dimensional space within a future predetermined time period based on historical valid datasets. The location coordinates and occurrence probability of poor quality areas are identified as the inference results output.

[0033] Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the drone coverage blind area.

[0034] As a preferred method, the generation of a 3D quality map includes the following operations:

[0035] Occlusion factor calculation: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space;

[0036] Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm incorporating building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients from the interpolation data source, interpolation calculations are performed across the entire 3D space using the network quality analysis model. The network quality values ​​for each coordinate point in the 3D space are output. The calculation formula for the network quality analysis model is expressed as follows:

[0037] ,

[0038] in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient;

[0039] Dynamic layered rendering: Based on the 3D spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple layers, and each layer is dynamically rendered using a corresponding color.

[0040] Poor quality region labeling: Combine the rendering results with the quality defect labeling information corresponding to the valid dataset to generate a 3D network quality heatmap.

[0041] Secondly, the present invention provides a three-dimensional spatial network quality mapping and prediction system based on an AI model, comprising a multimodal data acquisition module, a data preprocessing module, an inference and prediction module, a three-dimensional quality map generation module, and a network optimization module;

[0042] The multimodal data acquisition module is used to perform the following: by using a drone equipped with a special device to perform a patrol mission in the network coverage area to be inspected, it collects spectrum data to reflect network performance, three-dimensional point cloud data to reflect the spatial structure of the network environment, and positioning data to reflect the geographical location of the drone. It adds timestamps and location coordinates to the spectrum data and point cloud data to form multimodal data with spatiotemporal dimensions.

[0043] The data preprocessing module is used to perform the following: data preprocessing on multimodal data, including data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling, to obtain effective datasets and quality defect labeling information. The effective datasets include spatiotemporally aligned signal indicators, environmental features, and location coordinates, while the quality defect labeling information includes the location coordinates of the quality defect areas, the quality defect type, and the duration.

[0044] The inference prediction module is used to perform the following: merging historical valid datasets and historical status data of network devices into a sample set based on timestamps and location coordinates; constructing an inference model based on the Transformer-3D network model with a spatiotemporal attention mechanism; training the inference model based on the sample set and quality defect annotation information; performing inference prediction based on the trained inference model; and outputting the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area, and the probability of occurrence as the inference result.

[0045] The 3D quality map generation module is used to perform the following: construct a network quality analysis model based on an improved Kriging interpolation algorithm that incorporates building occlusion factors; calculate the 3D spatial network quality values ​​based on the effective dataset and inference results through the network quality analysis model; generate a 3D quality heatmap by combining dynamic layered rendering; and mark poor quality areas in the 3D quality heatmap.

[0046] The network optimization module performs the following: based on the poor quality area information presented by the 3D quality map, it displays a list of coordinates of locations to be optimized, and presents the network resource information of the corresponding area through the location coordinates, and generates targeted network optimization strategies. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

[0047] Preferably, the drone is equipped with dedicated equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. The multimodal data acquisition module is used to perform the following operations:

[0048] Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment will carry out inspection tasks according to the planned routes at regular intervals.

[0049] Network performance data is collected by a spectrum analyzer mounted on a drone. The network performance data includes signal strength RSRP / RSSI / SINR, bit error rate, and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset that includes signal indicators and time.

[0050] By capturing images of the network environment using an RGB-D camera mounted on a drone, and calculating the three-dimensional spatial coordinates of each point in the network environment image based on depth sensing technology, a three-dimensional point cloud dataset is generated to identify obstacles including buildings and trees.

[0051] By acquiring geographic coordinates and timestamps in real time through the GPS module mounted on the drone, and adding timestamps and geographic coordinates to the spectrum dataset and point cloud dataset, a multimodal dataset with spatiotemporal dimensions is formed.

[0052] Preferably, the data preprocessing module is used to perform the following operations:

[0053] Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. A Huffman tree is constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles, resulting in compressed point cloud data.

[0054] Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data;

[0055] Feature extraction: Using compressed point cloud data as input, environmental features are extracted through a CNN model; using smooth, interference-free spectral data as input, the temporal variation features of the signal are captured through an LSTM model, and the signal features are output.

[0056] Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset;

[0057] Quality defect annotation: Based on the preset quality defect judgment rules, it traverses the spatiotemporally aligned dataset, identifies and annotates the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect annotation information including geographic coordinates, quality defect type and duration.

[0058] Preferably, the inference and prediction module is used to perform the following operations:

[0059] Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference annotation information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set;

[0060] Model Construction: An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the occurrence probability as the inference result. The quality defect type corresponding to the quality defect area includes weak signal, network congestion and signal interference.

[0061] Inference and Prediction: The trained inference model is invoked to predict the signal strength values ​​of each coordinate point in three-dimensional space within a future predetermined time period based on historical valid datasets. The location coordinates and occurrence probability of poor quality areas are identified as the inference results output.

[0062] Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the drone coverage blind area.

[0063] Preferably, the 3D quality map generation module is used to perform the following operations:

[0064] Occlusion factor calculation: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space;

[0065] Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm incorporating building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients from the interpolation data source, interpolation calculations are performed across the entire 3D space using the network quality analysis model. The network quality values ​​for each coordinate point in the 3D space are output. The calculation formula for the network quality analysis model is expressed as follows:

[0066] ,

[0067] in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient;

[0068] Dynamic layered rendering: Based on the 3D spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple layers, and each layer is dynamically rendered using a corresponding color.

[0069] Poor quality region labeling: Combine the rendering results with the quality defect labeling information corresponding to the valid dataset to generate a 3D network quality heatmap.

[0070] The AI-based three-dimensional spatial network quality mapping and prediction method and system of the present invention have the following advantages: By integrating UAV multimodal data acquisition, three-dimensional spatial modeling, and AI model training and inference into a network quality mapping method, the efficiency of network quality analysis is improved, and the time for detecting quality defects is expected to be shortened from 48 hours to 2 hours, and the positioning accuracy is improved to 90%; network operation and maintenance costs are optimized, with an expected reduction of 70% in manual road testing, and a single UAV mission covering 10 kilometers, supporting decision-making on network optimization schemes. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] The invention will be further described below with reference to the accompanying drawings.

[0073] Figure 1 This is a flowchart of an AI model-based three-dimensional spatial network quality mapping and prediction method, as described in Example 1. Detailed Implementation

[0074] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0075] This invention provides a three-dimensional spatial network quality mapping and prediction method and system based on AI models, which is used to solve the technical problems of how to overcome the defects of data silos and response lag in communication network operation and maintenance, and to achieve efficient and accurate network operation and maintenance.

[0076] Example 1:

[0077] This invention provides a three-dimensional spatial network quality mapping and prediction method based on an AI model, comprising five steps: multimodal data acquisition, data preprocessing, inference prediction, three-dimensional quality map generation, and network optimization.

[0078] Step S100 Multimodal Data Acquisition: Using a drone equipped with dedicated equipment, an inspection mission is carried out in the network coverage area to be inspected. Spectrum data is collected to reflect network performance, three-dimensional point cloud data is collected to reflect the spatial structure of the network environment, and positioning data is collected to reflect the drone's geographical location. Timestamps and location coordinates are added to the spectrum data and point cloud data to form multimodal data with spatiotemporal dimensions.

[0079] As a specific implementation of multimodal data acquisition, the drone is equipped with dedicated equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. This step includes the following operations:

[0080] (1) Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment regularly perform inspection tasks according to the planned route;

[0081] (2) Network performance index data are collected by a spectrum analyzer mounted on the UAV. The network performance index data includes signal strength RSRP / RSSI / SINR, bit error rate and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset including signal index and time.

[0082] (3) By capturing network environment images with an RGB-D camera mounted on a drone, the three-dimensional spatial coordinates of each point in the network environment image are calculated based on depth sensing technology to generate a three-dimensional point cloud dataset for identifying obstacles including buildings and trees.

[0083] (4) Geographic coordinates and timestamps are obtained in real time by using the GPS module installed on the UAV, and timestamps and geographic coordinates are added to the spectrum dataset and point cloud dataset to form a multimodal dataset with spatiotemporal dimensions.

[0084] In this embodiment, this step involves collecting network performance metrics (signal strength RSRP / RSSI / SINR, bit error rate, network latency) using a spectrum analyzer to generate spectrum data consisting of signal metrics, time, and coordinates. An RGB-D camera captures images of the network environment, acquiring 3D point cloud data and identifying obstacles such as buildings and trees. A GPS module adds timestamps and geographic coordinates to each type of data. Based on the BeiDou grid coordinates, spectrum data, point cloud data, and location data from the same time and space are correlated and matched. The output consists of raw multimodal data: spectrum data consisting of signal metrics, time, and coordinates, and point cloud data consisting of environmental obstacles and spatial coordinates.

[0085] Step S200 Data Preprocessing: Perform data preprocessing on the multimodal data. Through data preprocessing, perform data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling operations to obtain a valid dataset and quality defect labeling information. The valid dataset includes spatiotemporally aligned signal indicators, environmental features, and location coordinates. The quality defect labeling information includes the location coordinates of the quality defect area, the quality defect type, and the duration.

[0086] As a specific implementation of data preprocessing, this step includes the following operations:

[0087] (1) Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. Huffman trees are constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles to obtain compressed point cloud data.

[0088] (2) Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data;

[0089] (3) Feature extraction: Compressed point cloud data is used as input, and environmental features are extracted through a CNN model. Smooth and interference-free spectrum data is used as input, and the temporal variation features of the signal are captured through an LSTM model to output signal features.

[0090] (4) Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset;

[0091] (5) Quality defect labeling: Based on the preset quality defect judgment rules, the spatiotemporally aligned dataset is traversed to identify and label the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect labeling information including geographic coordinates, quality defect type and duration.

[0092] In this embodiment, this step extracts obstacle features (height, density) from point cloud data using CNN, processes the temporal variation patterns of spectral data using LSTM, compresses the point cloud data volume using Huffman coding, and corrects signal fluctuations using Kalman filtering. Output: spatiotemporally aligned signal indicators + environmental features + coordinate data, as well as a filtered, interference-free, coordinate-bias-free effective dataset, with labeled areas of poor quality, serving as input for model training and prediction, and also as the basic sampling data source for interpolation calculations during 3D quality map generation.

[0093] Step S300 Inference Prediction: Based on timestamps and location coordinates, historical valid datasets and historical status data of network devices are integrated into a sample set. An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect labeling information. Inference prediction is performed based on the trained inference model, and the predicted signal strength values ​​of each location coordinate in three-dimensional space, the location coordinates of the quality defect area, and the occurrence probability are output as the inference result.

[0094] As a specific implementation of inference and prediction, this step includes the following operations:

[0095] (1) Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference labeling information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set;

[0096] (2) Model construction: The inference model is constructed based on the Transformer-3D network model with the introduction of spatiotemporal attention mechanism. The inference model is trained based on the sample set and the quality difference standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength value of each position coordinate in three-dimensional space, the position coordinates of the quality difference area and the occurrence probability as the inference result. The quality difference type corresponding to the quality difference area includes weak signal, network congestion and signal interference.

[0097] (3) Inference and prediction: The trained inference model is called to predict the signal strength value of each coordinate point in the three-dimensional space in the future time period based on the historical effective dataset, and the location coordinates and occurrence probability of the poor quality area are identified as the inference result output.

[0098] (4) Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the blind area covered by the UAV.

[0099] Step S400: Generation of 3D quality map: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm that incorporates building occlusion factors. Based on the effective dataset and inference results, the network quality values ​​in 3D space are calculated through the network quality analysis model. A 3D quality heat map is generated by combining dynamic layered rendering, and poor quality areas are marked in the 3D quality heat map.

[0100] As a specific implementation of generating a 3D quality map, this step includes the following operations:

[0101] (1) Calculation of occlusion factor: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space;

[0102] (2) Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm that incorporates building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients of the interpolation data source, interpolation is performed across the entire three-dimensional space using the network quality analysis model to output the network quality values ​​for each coordinate point in the three-dimensional space. The calculation formula of the network quality analysis model is expressed as:

[0103] ,

[0104] in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient;

[0105] (3) Dynamic layered rendering: Based on the three-dimensional spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple levels (the network quality is divided into three levels: poor, average, and good). Each level is dynamically rendered using the corresponding color (for example, the three levels are dynamically rendered using red, yellow, and green respectively).

[0106] In this embodiment, spatial interpolation modeling is used to generate a 3D quality map. The core function of Kriging interpolation is to generate a continuous and accurate 3D spatial network quality numerical field based on discrete sampling point data. Traditional Kriging interpolation can only perform spatial smoothing interpolation based on discrete sampling points, but it ignores the influence of physical obstacles in the communication network on signal obstruction, leading to significant deviations in the quality numerical calculations for tall buildings and complex terrain areas. Improved Kriging interpolation introduces a building obstruction factor to correct attenuation errors in the signal propagation path, making the generated 3D numerical field more closely match the actual network quality distribution.

[0107] (4) Labeling of poor quality areas: Combine the rendering results with the quality labeling information corresponding to the effective dataset to generate a three-dimensional network quality heat map, such as generating a red / yellow / green three-dimensional network quality heat map to clearly define the three-dimensional spatial distribution of the labeled poor quality areas.

[0108] Step S500 Network Optimization: Based on the poor quality area information presented by the 3D quality map, display a list of coordinates of locations to be optimized, and present the network resource information of the corresponding area through the location coordinates to generate a targeted network optimization strategy. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

[0109] The method in this embodiment generates a spatially continuous quality heat map by using a drone equipped with multiple sensors (spectrum analyzer, camera), achieving full-domain 3D mapping; it combines historical inspection data to predict areas with poor quality in the next 24 hours, achieving quality prediction; and it guides network operation and maintenance by visualizing the quality map and associating it with factors such as equipment status, parameters, interference, and engineering, achieving network optimization.

[0110] Example 2:

[0111] This invention discloses a three-dimensional spatial network quality mapping and prediction system based on an AI model, comprising a multimodal data acquisition module, a data preprocessing module, an inference and prediction module, a three-dimensional quality map generation module, and a network optimization module.

[0112] The multimodal data acquisition module is used to perform the following: by using a drone equipped with special equipment to perform inspection tasks in the network coverage area to be inspected, it collects spectrum data to reflect network performance, three-dimensional point cloud data to reflect the spatial structure of the network environment, and positioning data to reflect the drone's geographical location. It adds timestamps and location coordinates to the spectrum data and point cloud data to form multimodal data with spatiotemporal dimensions.

[0113] As a specific implementation of the multimodal data acquisition module, the UAV is equipped with dedicated equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. This step includes the following operations:

[0114] (1) Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment regularly perform inspection tasks according to the planned route;

[0115] (2) Network performance index data are collected by a spectrum analyzer mounted on the UAV. The network performance index data includes signal strength RSRP / RSSI / SINR, bit error rate and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset including signal index and time.

[0116] (3) By capturing network environment images with an RGB-D camera mounted on a drone, the three-dimensional spatial coordinates of each point in the network environment image are calculated based on depth sensing technology to generate a three-dimensional point cloud dataset for identifying obstacles including buildings and trees.

[0117] (4) Geographic coordinates and timestamps are obtained in real time by using the GPS module installed on the UAV, and timestamps and geographic coordinates are added to the spectrum dataset and point cloud dataset to form a multimodal dataset with spatiotemporal dimensions.

[0118] In this embodiment, the module collects network performance metrics (signal strength RSRP / RSSI / SINR, bit error rate, network latency) using a spectrum analyzer, forming spectrum data consisting of signal metrics, time, and coordinates. It also captures network environment images using an RGB-D camera, acquiring 3D point cloud data and identifying obstacles such as buildings and trees. A GPS module adds timestamps and geographic coordinates to each type of data, and based on the BeiDou grid coordinates, it correlates and matches spectrum data, point cloud data, and location data within the same time and space. The output includes raw multimodal data: spectrum data consisting of signal metrics, time, and coordinates, and point cloud data consisting of environmental obstacles and spatial coordinates.

[0119] The data preprocessing module is used to perform the following: data preprocessing on multimodal data, including data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling, to obtain a valid dataset and quality defect labeling information. The valid dataset includes spatiotemporally aligned signal indicators, environmental features, and location coordinates, while the quality defect labeling information includes the location coordinates of the quality defect area, the quality defect type, and the duration.

[0120] As a specific implementation of the data preprocessing module, this module is used to perform the following operations:

[0121] (1) Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. Huffman trees are constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles to obtain compressed point cloud data.

[0122] (2) Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data;

[0123] (3) Feature extraction: Compressed point cloud data is used as input, and environmental features are extracted through a CNN model. Smooth and interference-free spectrum data is used as input, and the temporal variation features of the signal are captured through an LSTM model to output signal features.

[0124] (4) Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset;

[0125] (5) Quality defect labeling: Based on the preset quality defect judgment rules, the spatiotemporally aligned dataset is traversed to identify and label the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect labeling information including geographic coordinates, quality defect type and duration.

[0126] In this embodiment, the module extracts obstacle features (height, density) from point cloud data using CNN, processes the temporal variation patterns of spectral data using LSTM, compresses the point cloud data volume using Huffman coding, and corrects signal fluctuations using Kalman filtering. Output: spatiotemporally aligned signal indicators + environmental features + coordinate data, as well as a filtered, interference-free, and coordinate-bias-free effective dataset, with labeled areas of poor quality, serving as input for model training and prediction, and also as the basic sampling data source for interpolation calculations during 3D quality map generation.

[0127] The inference prediction module performs the following actions: it integrates historical valid datasets and historical state data of network devices into a sample set based on timestamps and location coordinates; it constructs an inference model based on the Transformer-3D network model with a spatiotemporal attention mechanism; it trains the inference model based on the sample set and quality defect annotation information; it performs inference prediction based on the trained inference model; and it outputs the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area, and the probability of occurrence as the inference result.

[0128] As a specific implementation of the inference and prediction module, this module is used to perform the following operations:

[0129] (1) Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference labeling information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set;

[0130] (2) Model construction: The inference model is constructed based on the Transformer-3D network model with the introduction of spatiotemporal attention mechanism. The inference model is trained based on the sample set and the quality difference standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength value of each position coordinate in three-dimensional space, the position coordinates of the quality difference area and the occurrence probability as the inference result. The quality difference type corresponding to the quality difference area includes weak signal, network congestion and signal interference.

[0131] (3) Inference and prediction: The trained inference model is called to predict the signal strength value of each coordinate point in the three-dimensional space in the future time period based on the historical effective dataset, and the location coordinates and occurrence probability of the poor quality area are identified as the inference result output.

[0132] (4) Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the blind area covered by the UAV.

[0133] The 3D quality map generation module performs the following: it constructs a network quality analysis model based on an improved Kriging interpolation algorithm that incorporates building occlusion factors; it calculates the 3D spatial network quality values ​​based on the effective dataset and inference results through the network quality analysis model; it generates a 3D quality heatmap by combining dynamic layered rendering; and it marks poor quality areas in the 3D quality heatmap.

[0134] As a specific implementation of the 3D quality map generation module, this module is used to perform the following operations:

[0135] (1) Calculation of occlusion factor: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space;

[0136] (2) Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm that incorporates building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients of the interpolation data source, interpolation is performed across the entire three-dimensional space using the network quality analysis model to output the network quality values ​​for each coordinate point in the three-dimensional space. The calculation formula of the network quality analysis model is expressed as:

[0137] ,

[0138] in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient;

[0139] (3) Dynamic layered rendering: Based on the three-dimensional spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple levels (the network quality is divided into three levels: poor, average, and good). Each level is dynamically rendered using the corresponding color (for example, the three levels are dynamically rendered using red, yellow, and green respectively).

[0140] In this embodiment, spatial interpolation modeling is used to generate a 3D quality map. The core function of Kriging interpolation is to generate a continuous and accurate 3D spatial network quality numerical field based on discrete sampling point data. Traditional Kriging interpolation can only perform spatial smoothing interpolation based on discrete sampling points, but it ignores the influence of physical obstacles in the communication network on signal obstruction, leading to significant deviations in the quality numerical calculations for tall buildings and complex terrain areas. Improved Kriging interpolation introduces a building obstruction factor to correct attenuation errors in the signal propagation path, making the generated 3D numerical field more closely match the actual network quality distribution.

[0141] (4) Labeling of poor quality areas: Combine the rendering results with the quality labeling information corresponding to the effective dataset to generate a three-dimensional network quality heat map, such as generating a red / yellow / green three-dimensional network quality heat map to clearly define the three-dimensional spatial distribution of the labeled poor quality areas.

[0142] The network optimization module performs the following: based on the poor quality area information presented by the 3D quality map, it displays a list of coordinates of locations to be optimized, and presents the network resource information of the corresponding area through the location coordinates, and generates targeted network optimization strategies. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

[0143] In this embodiment, the system can execute the method disclosed in Embodiment 1 to realize three-dimensional spatial network quality mapping and prediction.

[0144] The above provides a detailed description of the AI-based three-dimensional spatial network quality mapping and prediction method and system provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for predicting the quality of three-dimensional spatial networks based on an AI model, characterized in that, Includes the following steps: Multimodal data acquisition: Drones equipped with specialized equipment perform inspection tasks in the network coverage area to be inspected, collecting spectrum data to reflect network performance, 3D point cloud data to reflect the spatial structure of the network environment, and positioning data to reflect the drone's geographical location. Timestamps and location coordinates are added to the spectrum data and point cloud data to form multimodal data with spatiotemporal correlation. Data preprocessing: Multimodal data is preprocessed through data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling to obtain effective datasets and quality defect labeling information. The effective datasets include spatiotemporally aligned signal indicators, environmental features, and location coordinates, while the quality defect labeling information includes the location coordinates of the quality defect areas, the quality defect type, and the duration. Inference and Prediction: Based on timestamps and location coordinates, historical valid datasets and historical status data of network devices are integrated into a sample set. An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect labeling information. Inference and prediction are performed based on the trained inference model. The predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the probability of occurrence are output as the inference results. 3D quality map generation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm that incorporates building occlusion factors. Based on the effective dataset and inference results, the network quality analysis model is used to calculate the 3D spatial network quality values. Combined with dynamic layered rendering, a 3D quality heat map is generated, and poor quality areas are marked in the 3D quality heat map. Network optimization: Based on the poor quality area information presented by the 3D quality map, a list of location coordinates to be optimized is displayed. The network resource information of the corresponding area is presented by associating the location coordinates, and targeted network optimization strategies are generated. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

2. The method for predicting and mapping the quality of three-dimensional spatial networks based on an AI model according to claim 1, characterized in that, The drone is equipped with specialized equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. Multimodal data acquisition includes the following operations: Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment will carry out inspection tasks according to the planned routes at regular intervals. Network performance data is collected by a spectrum analyzer mounted on a drone. The network performance data includes signal strength RSRP / RSSI / SINR, bit error rate, and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset that includes signal indicators and time. By capturing images of the network environment using an RGB-D camera mounted on a drone, and calculating the three-dimensional spatial coordinates of each point in the network environment image based on depth sensing technology, a three-dimensional point cloud dataset is generated to identify obstacles including buildings and trees. By acquiring geographic coordinates and timestamps in real time through the GPS module mounted on the drone, and adding timestamps and geographic coordinates to the spectrum dataset and point cloud dataset, a multimodal dataset with spatiotemporal dimensions is formed.

3. The method for predicting and mapping the quality of three-dimensional spatial networks based on an AI model according to claim 1, characterized in that, Data preprocessing includes the following operations: Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. A Huffman tree is constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles, resulting in compressed point cloud data. Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data; Feature extraction: Using compressed point cloud data as input, environmental features are extracted through a CNN model; using smooth, interference-free spectral data as input, the temporal variation features of the signal are captured through an LSTM model, and the signal features are output. Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset; Quality defect annotation: Based on the preset quality defect judgment rules, it traverses the spatiotemporally aligned dataset, identifies and annotates the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect annotation information including geographic coordinates, quality defect type and duration.

4. The method for predicting and mapping the quality of three-dimensional spatial networks based on an AI model according to claim 1, characterized in that, Inference and prediction include the following operations: Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference annotation information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set; Model Construction: An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the occurrence probability as the inference result. The quality defect type corresponding to the quality defect area includes weak signal, network congestion and signal interference. Inference and Prediction: The trained inference model is invoked to predict the signal strength values ​​of each coordinate point in three-dimensional space within a future predetermined time period based on historical valid datasets. The location coordinates and occurrence probability of poor quality areas are identified as the inference results output. Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the drone coverage blind area.

5. The method for predicting and mapping the quality of three-dimensional spatial networks based on an AI model according to claim 1, characterized in that, Generating a 3D quality map involves the following operations: Occlusion factor calculation: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space; Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm incorporating building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients from the interpolation data source, interpolation calculations are performed across the entire 3D space using the network quality analysis model. The network quality values ​​for each coordinate point in the 3D space are output. The calculation formula for the network quality analysis model is expressed as follows: ， in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient; Dynamic layered rendering: Based on the 3D spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple layers, and each layer is dynamically rendered using a corresponding color. Poor quality region labeling: Combine the rendering results with the quality defect labeling information corresponding to the valid dataset to generate a 3D network quality heatmap.

6. A three-dimensional spatial network quality mapping and prediction system based on an AI model, characterized in that, It includes a multimodal data acquisition module, a data preprocessing module, an inference and prediction module, a 3D quality map generation module, and a network optimization module; The multimodal data acquisition module is used to perform the following: by using a drone equipped with a special device to perform a patrol mission in the network coverage area to be inspected, it collects spectrum data to reflect network performance, three-dimensional point cloud data to reflect the spatial structure of the network environment, and positioning data to reflect the geographical location of the drone. It adds timestamps and location coordinates to the spectrum data and point cloud data to form multimodal data with spatiotemporal dimensions. The data preprocessing module is used to perform the following: data preprocessing on multimodal data, including data compression, anomaly filtering, feature extraction, spatiotemporal alignment, and quality defect labeling, to obtain effective datasets and quality defect labeling information. The effective datasets include spatiotemporally aligned signal indicators, environmental features, and location coordinates, while the quality defect labeling information includes the location coordinates of the quality defect areas, the quality defect type, and the duration. The inference prediction module is used to perform the following: merging historical valid datasets and historical status data of network devices into a sample set based on timestamps and location coordinates; constructing an inference model based on the Transformer-3D network model with a spatiotemporal attention mechanism; training the inference model based on the sample set and quality defect annotation information; performing inference prediction based on the trained inference model; and outputting the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area, and the probability of occurrence as the inference result. The 3D quality map generation module is used to perform the following: construct a network quality analysis model based on an improved Kriging interpolation algorithm that incorporates building occlusion factors; calculate the 3D spatial network quality values ​​based on the effective dataset and inference results through the network quality analysis model; generate a 3D quality heatmap by combining dynamic layered rendering; and mark poor quality areas in the 3D quality heatmap. The network optimization module performs the following: based on the poor quality area information presented by the 3D quality map, it displays a list of coordinates of locations to be optimized, and presents the network resource information of the corresponding area through the location coordinates, and generates targeted network optimization strategies. The network resource information includes device status, engineering parameters, coverage area, and operation and maintenance records.

7. The AI-based three-dimensional spatial network quality mapping and prediction system according to claim 6, characterized in that, The drone is equipped with specialized equipment including a spectrum analyzer, an RGB-D camera, and a GPS module. The multimodal data acquisition module is used to perform the following operations: Based on the coverage area of ​​the network to be inspected and the inspection requirements, drones equipped with special equipment will carry out inspection tasks according to the planned routes at regular intervals. Network performance data is collected by a spectrum analyzer mounted on a drone. The network performance data includes signal strength RSRP / RSSI / SINR, bit error rate, and network latency. The network performance data is serialized and stored in chronological order to form a spectrum dataset that includes signal indicators and time. By capturing images of the network environment using an RGB-D camera mounted on a drone, and calculating the three-dimensional spatial coordinates of each point in the network environment image based on depth sensing technology, a three-dimensional point cloud dataset is generated to identify obstacles including buildings and trees. By acquiring geographic coordinates and timestamps in real time through the GPS module mounted on the drone, and adding timestamps and geographic coordinates to the spectrum dataset and point cloud dataset, a multimodal dataset with spatiotemporal dimensions is formed.

8. The AI-based three-dimensional spatial network quality mapping and prediction system according to claim 6, characterized in that, The data preprocessing module is used to perform the following operations: Encoding compression: Statistical analysis is performed on the point cloud data in the multimodal dataset. A Huffman tree is constructed based on the frequency of data occurrence and encoding conversion is performed. The data volume is compressed while preserving the key geometric features of obstacles, resulting in compressed point cloud data. Data filtering: The spectrum data is filtered based on Kalman filtering to output smooth and interference-free spectrum data; Feature extraction: Using compressed point cloud data as input, environmental features are extracted through a CNN model; using smooth, interference-free spectral data as input, the temporal variation features of the signal are captured through an LSTM model, and the signal features are output. Spatiotemporal alignment: Based on timestamps and geographic coordinates, signal features, environmental features and geographic coordinates are associated to form a spatiotemporally aligned dataset; Quality defect annotation: Based on the preset quality defect judgment rules, it traverses the spatiotemporally aligned dataset, identifies and annotates the geographic coordinates, quality defect type and duration of the quality defect area, and outputs a valid dataset including signal indicators, environmental features and geographic coordinates, as well as quality defect annotation information including geographic coordinates, quality defect type and duration.

9. The AI-based three-dimensional spatial network quality mapping and prediction system according to claim 6, characterized in that, The inference and prediction module is used to perform the following operations: Sample set construction: Collect historical device status data of network devices, obtain historical valid datasets and corresponding quality difference annotation information, and merge historical device status data with historical valid datasets based on timestamps and location coordinates to form a sample set; Model Construction: An inference model is constructed based on the Transformer-3D network model with a spatiotemporal attention mechanism. The inference model is trained based on the sample set and quality defect standard information to obtain the trained inference model. The inference model is used to learn the correlation between signal strength and environmental features and device status based on the input sample set. It outputs the predicted signal strength values ​​at each location coordinate in three-dimensional space, the location coordinates of the quality defect area and the occurrence probability as the inference result. The quality defect type corresponding to the quality defect area includes weak signal, network congestion and signal interference. Inference and Prediction: The trained inference model is invoked to predict the signal strength values ​​of each coordinate point in three-dimensional space within a future predetermined time period based on historical valid datasets. The location coordinates and occurrence probability of poor quality areas are identified as the inference results output. Data output: The inference results are fed into the 3D quality map generation operation as a supplementary interpolation data source to correct the interpolation result deviation in the drone coverage blind area.

10. The AI-based three-dimensional spatial network quality mapping and prediction system according to claim 6, characterized in that, The 3D quality map generation module is used to perform the following operations: Occlusion factor calculation: Based on the concentrated environmental features of the effective data, extract the relevant parameters of the building to calculate the occlusion coefficient of the building corresponding to each coordinate point in the three-dimensional space; Improved Kriging Interpolation Calculation: A network quality analysis model is constructed based on an improved Kriging interpolation algorithm incorporating building occlusion factors. The effective dataset and inference results are aligned to form a complete interpolation data source. Based on the discrete sampling point signal indicators and building occlusion coefficients from the interpolation data source, interpolation calculations are performed across the entire 3D space using the network quality analysis model. The network quality values ​​for each coordinate point in the 3D space are output. The calculation formula for the network quality analysis model is expressed as follows: ， in, This represents a network quality value. Represents the weights of discrete sampling points. This represents the signal strength value at discrete sampling points. Indicates the occlusion effect coefficient. Indicates the building's shading coefficient; Dynamic layered rendering: Based on the 3D spatial network quality value obtained by interpolation, a quality grading threshold is set to divide the network quality into multiple layers, and each layer is dynamically rendered using a corresponding color. Poor quality region labeling: Combine the rendering results with the quality defect labeling information corresponding to the valid dataset to generate a 3D network quality heatmap.

Images