CN-121984621-A - Urban digital twin on-line channel modeling method

CN121984621ACN 121984621 ACN121984621 ACN 121984621ACN-121984621-A

Abstract

The invention discloses an urban digital twin on-line channel modeling method, which relates to the technical field of cross fusion application of digital twin technology and wireless channel modeling, and comprises the steps of 1) obtaining geographic information system data of a target urban area, building a three-dimensional static model of a city through Unity software, carrying out equal proportion reconstruction and reflection surface parameter addition to enable electromagnetic characteristics of the model to be matched with real physical world, 2) collecting satellite remote sensing images, constructing a dataset of various urban scenes, training the dataset by adopting a Swin transducer to realize recognition and segmentation of urban communication scenes, 3) completing electromagnetic parameter matching of the communication scenes by combining a 6G universal channel model based on scene recognition and segmentation results, and simultaneously carrying out static electromagnetic reconstruction by utilizing a ray tracing technology of WIRELESS INSITE software, and 4) integrating a scene segmentation module, a channel characteristic prediction module and a three-dimensional urban model into the Unity platform. The method and the device can accurately capture the channel characteristic change in the complex urban scene.

Inventors

LI JUNLING
JIA WEI
WANG CHENGXIANG
ZHU CHAO
LUO SHENGHAN
YANG JINGWEN
HUANG CHEN

Assignees

东南大学
紫金山实验室

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. An urban digital twin on-line channel modeling method, comprising: S1, acquiring GIS data of a target city geographic information system, and constructing an initial city static model according to the geographic information data of the target city; s2, constructing an urban communication scene data set and training a scene segmentation model to form a scene segmentation module, wherein the scene segmentation module segments an urban scene to obtain a scene segmentation result; step S3, constructing a 6GPCM channel characteristic prediction module, wherein, And the 6GPCM channel characteristic prediction module is used for distributing corresponding electromagnetic parameters for different city scenes by combining a scene segmentation result with the 6G universal channel model 6GPCM and obtaining channel characteristic information, and simultaneously, carrying out static electromagnetic reconstruction on an initial city static model through a ray tracing technology to generate channel state information CSI.
2. The method for modeling an urban digital twin on-line channel according to claim 1, further comprising step S4, And S4, integrating the scene segmentation module, the 6GPCM channel characteristic prediction module and the initial city static model into a Unity platform.
3. The method for modeling an urban digital twin on-line channel according to claim 1, wherein step S1 comprises: and (3) importing Geographic Information System (GIS) data of the target city region, generating a basic three-dimensional model in Blender software through BlendeGIS plug-in units, carrying out equal proportion reconstruction and reflection surface material addition, and importing the reconstructed basic three-dimensional model into a Unity engine to form an initial city static model.
4. The method for modeling an urban digital twin on-line channel according to claim 1, wherein the scene segmentation model is a Swin transform segmentation model, and the step S2 of the scene segmentation module comprises the steps of collecting satellite remote sensing images, then labeling four types of urban communication scenes of factories, parking lots, residential areas and arterial roads in the satellite remote sensing images to construct a dataset, and training the Swin transform segmentation model to enable the Swin transform segmentation model to have the recognition and segmentation capabilities of the urban communication scenes, thereby forming the scene segmentation module.
5. The urban digital twin on-line channel modeling method according to claim 1, wherein in step S3, an initial urban static model is imported WIRELESS INSITE into software, and static electromagnetic reconstruction is performed by a ray tracing technology, so as to generate channel state information CSI.
6. The method for modeling urban digital twin on-line channels according to claim 1, wherein step S1 is specifically as follows: step S101, obtaining a basic three-dimensional model of a target city through a Blender plug-in, wherein the basic three-dimensional model comprises buildings, roads, streets, highways, water channels and terrain elements; Step S102, optimizing a basic three-dimensional model, reconstructing a building, a house and a road structure according to a preset proportion, and adding reflection surface parameters to the surface of the basic three-dimensional model to enable electromagnetic properties of a three-dimensional urban scene to be consistent with electromagnetic reflection and scattering characteristics of a real physical environment; And step S103, importing the optimized three-dimensional model into a Unity engine to complete the construction of the initial city static model.
7. The method for modeling urban digital twin on-line channels according to claim 4, wherein step S2 is specifically as follows: step S201, constructing a data set of satellite remote sensing images, wherein each satellite remote sensing image is marked with a scene type label and a segmentation mask, and the scene type is limited into four types of factories, parking lots, residential areas and main roads; Step S202, configuring Swin transducer model parameters, adopting a self-attention mechanism based on a sliding window, dividing an input satellite remote sensing image into 4X 4 non-overlapping patches through a patch dividing module, projecting patch characteristics to a preset dimension C through a linear embedding layer, and gradually reducing the number of patch tokens through hierarchical network processing to generate hierarchical characteristics; In step S203, the constructed dataset is used as input to train the Swin transform segmentation model, in the training process, the average intersection ratio mIoU, the average accuracy mAcc and the overall accuracy aAcc are used as evaluation indexes, so that the trained Swin transform segmentation model is ensured to be not lower than 0.77 in factory scene intersection ratio IoU and not lower than 0.83 in residential area scene intersection ratio IoU in four types of urban communication scene segmentation tasks, and the trained Swin transform segmentation model outputs scene segmentation results.
8. The method for modeling urban digital twin on-line channels according to claim 4, wherein step S3 is specifically as follows: Step S301, extracting a scene segmentation result output by the trained Swin transducer segmentation model, determining specific area ranges of four types of scenes of factories, parking lots, residential areas and main roads, and distributing corresponding electromagnetic parameters for different scenes; Step S302, constructing a channel impulse response CIR matrix based on the 6G universal channel model 6GPCM , wherein, Is the first The transmitting end and the first Channel impulse responses between the individual receiver antenna elements, As a function of the carrier frequency, The number of the receiving antennas and the transmitting antennas respectively, As a function of the instantaneous time variable, The simulation model is based on a 6G universal channel model and is used by WIRELESS INSITE software; step S303, importing WIRELESS INSITE the basic three-dimensional model added with the electromagnetic attribute into software, setting ray tracing parameters, and enabling a transmitting end to achieve elevation angle according to the arrival And azimuth of arrival The rays are emitted at intervals, the rays reach a receiving end after interaction with the surface of an object, an effective path is screened according to the arrival power, angular power spectral density, time delay power spectral density and root mean square delay spread channel characteristic data are generated, and static electromagnetic reconstruction is completed; in step S302 of the process, ; Wherein, the For the line of sight LoS component, As a non-line-of-sight NLoS component, Is that Time of day rice Factor, and 6GPCM require simultaneous calculation of delay power spectral density , , Wherein, the As a function of the frequency-dependent function, Is a natural logarithm lower base symbol, j is a complex unit, Is the frequency offset.
9. The urban digital twin on-line channel modeling method according to claim 2, wherein in step S4, a scene segmentation module, a 6GPCM channel characteristic prediction module and an initial urban static model are integrated in a Unity platform, so that real-time interaction of data of each module is ensured, and association mapping of scene segmentation results and channel characteristic data is realized; The Unity platform can realize the function of displaying channel characteristics, and specifically comprises the following steps: developing a thermodynamic diagram checking function module, supporting a user to select the Tx position and thermodynamic diagram type of a transmitting end, and displaying the overall received power distribution in a preselected urban scene in real time, wherein the thermodynamic diagram is smoothly switched along with an observation angle; Developing a channel information updating function module, setting sampling points in a wireless valley area of Unity, enabling the sampling points to move along a preset track or manually control the movement, feeding back coordinates of the sampling points and corresponding signal and delay information in real time in the moving process, and generating a channel characteristic chart; The scene recognition function module is developed, so that the matching of the city model and the segmentation result is realized, the translation and scaling operation under the overlook view angle is provided, the interactive viewing of the recognized scene is supported, the rapid navigation to the position of the transmitting end and the dynamic visualization of the Rx moving process of the receiving end are realized, and the user can conveniently compare the channel information of different scenes; Wherein the channel characteristic graph comprises a root mean square RMS delay spread, a frequency correlation function, and a time autocorrelation function, wherein the root mean square RMS delay spread Calculated by the following formula: ; Wherein, the As a function of the frequency-dependent function, To delay the power spectral density.
10. The method of modeling an urban digital twin on-line channel according to claim 8, wherein in step S303, the ray tracing parameters include that rays are emitted at a preset interval between an arrival elevation angle θ and an arrival azimuth angle φ, the number of allowed interactions between the rays and the surface of the object is not less than 3, the criterion of the effective paths is that the power reaching the receiving end is not less than-120 dBm, and the user selects the number of the effective paths to be displayed.

Description

Urban digital twin on-line channel modeling method Technical Field The invention relates to the technical field of cross fusion application of digital twin technology and wireless channel modeling, in particular to a city digital twin on-line channel modeling method based on a Swin Transformer. Background With the development of the 6G wireless communication technology to the large-scale, high-dynamic and full-scene coverage direction, the urban communication environment is faced with multiple challenges such as high-density population, complex infrastructure layout, dynamic traffic flow and the like, so that the wireless channel has the characteristics of remarkable multipath effect, complex signal scattering, severe channel characteristic time variation and the like, the training cost of the traditional channel acquisition and modeling method is extremely high, and the requirements of the 6G system on the high precision and the real-time performance of channel information are difficult to meet. The traditional channel modeling method mainly relies on field measurement and deterministic simulation, wherein the field measurement needs to deploy a large amount of hardware equipment, consumes manpower and material resources, is influenced by dynamic change of environment, has poor timeliness of measured data and limited coverage range, and the deterministic simulation (such as ray tracing) can accurately simulate signal propagation, but has high computational complexity and long time, and is difficult to realize online real-time channel prediction of a large-scale urban scene. Meanwhile, in the application of the existing digital twin in urban scenes, static multi-focusing is carried out on physical environments, intelligent recognition capability of communication scene semantics is lacking, accurate association mapping of scene-channel cannot be realized, and therefore the adaptability of a channel model to the actual environments is insufficient, and dynamic optimization and efficient operation and maintenance of a 6G network are difficult to support. In recent years, vision Transformer (ViT) series models show strong semantic segmentation capability in the field of computer vision, wherein a Swin transform can efficiently capture multi-scale features and spatial dependence in high-resolution city images through a hierarchical sliding window self-attention mechanism, so as to provide technical support for intelligent recognition of city scenes, and a 6G universal channel model (6 GPCM) can capture the time-frequency non-stationary characteristics of channels through geometric random modeling and output statistical channel features under multiple scenes and multiple frequency bands. However, currently, no technical solution has been provided to deeply integrate the scene segmentation capability of the Swin Transformer, the channel prediction capability of 6GPCM and the virtual multi-engraving capability of digital twin, so as to construct an integrated platform with the whole flow functions of environment reconstruction, scene recognition, channel prediction and online visualization, and the requirements of the 6G system for obtaining the channel information in real time, visualization and interaction in the urban complex scene cannot be met. The existing channel modeling method has the defects of large resource consumption, poor real-time performance and insufficient scene suitability. Disclosure of Invention The invention aims to solve the technical problems of overcoming the defects of the prior art and providing a city digital twin on-line channel modeling method based on a Swin converter, which integrates three-dimensional city reconstruction driven by GIS data, intelligent scene segmentation driven by the Swin converter, channel characteristic prediction driven by 6GPCM, combines a Unity platform to realize functional integration and visual interaction, fills the blank of the prior art in the field of digital twin+intelligent channel modeling, provides key technical support for optimizing a 6G city communication system, and realizes accurate prediction and on-line visual interaction of channel characteristics in a complex city scene. The invention adopts the following technical scheme for solving the technical problems: The invention provides an urban digital twin on-line channel modeling method, which comprises the following steps: S1, acquiring GIS data of a target city geographic information system, and constructing an initial city static model according to the geographic information data of the target city; s2, constructing an urban communication scene data set and training a scene segmentation model to form a scene segmentation module, wherein the scene segmentation module segments an urban scene to obtain a scene segmentation result; step S3, constructing a 6GPCM channel characteristic prediction module, wherein, And the 6GPCM channel characteristic prediction module is used for distributing corr