An electromagnetic distribution prediction method, system and terminal based on an environmental medium model

Abstract

This invention relates to an electromagnetic distribution prediction method, system, and terminal based on an environmental medium model, aiming to achieve efficient dynamic allocation of electromagnetic computing resources for multiple types of media in complex spaces. The prediction method includes: addressing the hierarchical problem of computational complexity caused by the non-uniformity of medium spatial distribution and electromagnetic parameters; constructing a frequency-dependent three-dimensional medium model by collecting multi-source geometric and electromagnetic parameter data, quantifying regional complexity, and automatically generating regional computational requirement labels by combining adaptive thresholds and spatial aggregation algorithms; based on a differentiated computational configuration scheme, high-complexity regions employ full-wave fine-grained solutions, while low-complexity regions use empirical approximation models, and field strength smoothing is performed at regional boundaries. This method effectively improves resource allocation efficiency and ensures the continuity and accuracy of global electromagnetic field prediction.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic distribution prediction and spatial region complexity adaptive modeling technology, and in particular to an electromagnetic distribution prediction method, system and terminal based on an environmental medium model. Background Technology

[0002] Current electromagnetic distribution prediction technology is widely used in complex spatial scenarios such as wireless communication, radar systems, antenna layout, and indoor and outdoor coverage analysis. Mainstream methods are based on ray tracing, full-wave numerical simulation, and empirical modeling. By discretizing the target space in three dimensions and invoking prior medium parameter models, quantitative predictions of electromagnetic wave propagation, scattering, and attenuation characteristics are achieved. As environmental structures and medium types become increasingly complex, this field is trending towards using methods such as three-dimensional modeling, adaptive parameter adjustment, and multi-source data fusion to balance prediction accuracy and computational efficiency in large-scale, multi-scenario applications.

[0003] Existing prediction methods struggle to accurately identify the actual propagation complexity of each region during spatial discretization and region partitioning. Most rely on coarse partitioning based on material type and simple geometric properties, resulting in insufficient accuracy in high-precision regions and wasted computational resources in low-precision regions. Furthermore, ray tracing and full-wave algorithms typically employ a globally uniform computational dimension, lacking the ability to adaptively adjust to different environmental regions. At complex boundaries in multi-media environments and at heterogeneous mesh connections, the lack of continuous optimization easily leads to discrepancies and artifacts in the predicted field strength. Moreover, reliance on manually set thresholds and static model parameters makes it difficult to automatically update partitioning criteria and algorithm configurations according to environmental changes, limiting the model's versatility and intelligence.

[0004] To address the aforementioned technical deficiencies, there is an urgent need in this field for an electromagnetic distribution prediction method based on multiple environmental media and spatial geometric features. This method should be able to achieve quantitative scoring of regional complexities, dynamic partitioning, and adaptive allocation of computational dimensions. It should also employ heterogeneous computational models for hierarchical prediction based on the actual needs of different regions, and ensure the continuity and physical consistency of the global field distribution through boundary optimization, thereby improving environmental adaptability, computational efficiency, and prediction accuracy. Summary of the Invention

[0005] This application provides an electromagnetic distribution prediction method, system, and terminal based on an environmental medium model, aiming to solve one of the problems or issues of the prior art mentioned in the background.

[0006] Firstly, this application provides an electromagnetic distribution prediction method based on an environmental medium model, specifically including: Obtain geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space. The electromagnetic parameters include at least three of the following: dielectric constant, loss tangent, and conductivity, and generate an original media dataset. Based on the original medium dataset, perform precise assignment of medium electromagnetic parameters, map the assigned parameters to a three-dimensional geometric mesh in the target space, and construct a three-dimensional model of the medium environment. The geometric complexity features and electromagnetic wave propagation complexity features of each sub-region are extracted using the three-dimensional model of the medium environment, and the above features are input into a preset weighted summation logic to calculate and generate a total complexity score. The total complexity score is compared with a preset threshold. If the score is higher than the preset threshold, it is marked as a high computational demand region. If the score is lower than or equal to the preset threshold, it is marked as a low computational demand region, and the region division result is generated. Based on the region partitioning results with computational dimension requirement labels, a spatially adaptive electromagnetic computation dimensionality reduction strategy is executed. For regions with low computational requirements, simplified computational dimensions are allocated and empirical statistical models are adopted. For regions with high computational requirements, refined computational dimensions are allocated and full-wave algorithms are adopted to generate differentiated computational configuration schemes. Electromagnetic distribution prediction calculations are performed in parallel according to the differentiated calculation configuration scheme, and linear interpolation is applied at the boundary of adjacent regions to smooth the transition, so as to eliminate the discontinuity of field strength caused by abrupt changes in calculation dimension and generate continuous electromagnetic field distribution prediction data across the entire domain.

[0007] As one possible implementation, the method involves extracting the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and inputting these features into a preset weighted summation logic to calculate and generate a total complexity score. Specifically, this includes: Based on the three-dimensional model of the medium environment, the surface curvature change rate and geometric structure discreteness of the scatterer in each sub-region of the target space are subjected to gridded scanning processing to generate a set of geometric complexity feature vectors characterizing the complexity of the spatial morphology. Using the geometric complexity feature vector set as a spatial constraint, and combining the frequency-related dielectric constant and conductivity distribution data in the three-dimensional model of the medium environment, a multi-path ray tracing attenuation simulation operation is performed to generate an electromagnetic wave propagation complexity feature matrix that reflects beam penetration loss and scattering intensity. Based on the statistical distribution characteristics of the electromagnetic wave propagation complexity feature matrix and the geometric complexity feature vector set, the entropy weight method is used to dynamically calculate the weight coefficients of each feature dimension in order to generate an adaptive weight configuration parameter set that adapts to the dynamic changes of the current environmental medium. Based on the adaptive weight configuration parameter group, a weighted fusion calculation operation is performed on the geometric complexity feature vector set and the electromagnetic wave propagation complexity feature matrix to generate an initial complexity evaluation value sequence that quantitatively represents the comprehensive computational difficulty of each sub-region. The initial complexity evaluation value sequence is processed by applying a nonlinear normalization mapping function to unify the dimensions, so as to eliminate the influence of differences in the magnitude of different features and finally generate a standardized list of total complexity scores.

[0008] As one possible implementation, the method of performing a spatially adaptive electromagnetic computation dimensionality reduction strategy based on the region partitioning results with computational dimensionality requirement labels allocates simplified computational dimensions and adopts an empirical statistical model to regions with low computational requirements, and allocates refined computational dimensions and adopts a full-wave algorithm to regions with high computational requirements, thereby generating differentiated computational configuration schemes, specifically including: The region segmentation results with computational dimension requirement labels are parsed and processed to extract the set of regions with high computational requirement and the set of regions with low computational requirement. A full-wave algorithm mapping index is constructed based on the set of regions with high computational requirement, and an empirical statistical model mapping index is constructed based on the set of regions with low computational requirement, thereby generating an initial algorithm matching list containing the correspondence between region categories and candidate algorithms. Based on the set of high computational demand regions in the initial algorithm matching list, a fine computational dimension allocation operation is performed. The sub-region meshes within the set of high computational demand regions are densified using a 3D mesh refinement algorithm to generate a fine computational mesh dataset with high-resolution geometric features. The fine computational mesh dataset is then bound to the full-wave algorithm mapping index to generate a high-precision computational configuration unit for high computational demand regions. Based on the set of low computational demand regions in the initial algorithm matching list, a simplified computational dimension allocation operation is performed. The ray tracing order reduction algorithm is used to sparsify the sub-region grids within the set of low computational demand regions to generate a simplified computational grid dataset with low-resolution geometric features. The simplified computational grid dataset is then bound to the empirical statistical model mapping index to generate a fast approximate computational configuration unit for low computational demand regions. Boundary coupling parameter calculation processing is performed on the high-precision computing configuration unit and the fast approximation computing configuration unit. The field intensity gradient features of the high-precision computing configuration unit and the fast approximation computing configuration unit at the adjacent boundary are extracted by the linear interpolation algorithm to generate a smooth transition weight matrix for eliminating the dimension abrupt effect. The smooth transition weight matrix is ​​then injected into the high-precision computing configuration unit and the fast approximation computing configuration unit respectively, thereby generating an optimized region computing configuration package containing a boundary smoothing mechanism. The final configuration scheme integration operation is performed based on the optimized regional computing configuration package containing the boundary smoothing mechanism. Parallel task scheduling logic is used to encapsulate the high-precision computing configuration unit and the fast approximation computing configuration unit in the optimized regional computing configuration package into independent computing threads to generate differentiated computing configuration schemes adapted to the characteristics of the global space.

[0009] As one possible implementation, the frequency-related electromagnetic parameters are obtained by filling the parameter gaps between discrete frequency points using a linear interpolation algorithm.

[0010] As one possible implementation, the three-dimensional geometric mesh is generated using an adaptive mesh generation algorithm, and the three-dimensional geometric mesh is a non-uniform mesh.

[0011] As one possible implementation, in the three-dimensional model of the medium environment, the electromagnetic parameters at the boundaries of adjacent meshes of different medium types are corrected by a weighted average interpolation algorithm.

[0012] As one possible implementation, the preset threshold is dynamically generated by an adaptive threshold calculation algorithm based on the extreme value characteristics and distribution density characteristics of the total complexity score.

[0013] As one possible implementation, in the region partitioning results, grid cells with the same computational dimension requirement label and spatially adjacent are merged into independent spatial subdomains.

[0014] As one possible implementation, the topological structure of the interface between adjacent regions with different computational dimension requirement labels in the region partitioning result is optimized by morphological filtering.

[0015] As one possible implementation, the electromagnetic field distribution prediction data is generated by fusing the field strength at the boundary between high-computation-demand regions and low-computation-demand regions after weighted correction using a linear interpolation algorithm.

[0016] Secondly, this application provides an electromagnetic distribution prediction system based on an environmental medium model, including: a data acquisition module, a three-dimensional model construction module for the medium environment, a complexity calculation module, a region division module, a spatial adaptive electromagnetic calculation module, and an electromagnetic distribution prediction module; The data acquisition module is used to acquire geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space, and generate the original media dataset. The medium environment 3D model construction module performs precise medium electromagnetic parameter assignment operations based on the original medium dataset, maps the assigned parameters to the 3D geometric mesh of the target space, and constructs the medium environment 3D model. The complexity calculation module is used to extract the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and input the above features into the preset weighted summation logic to calculate and generate the total complexity score. The region division module is used to compare the total complexity score with a preset threshold. If the score is higher than the preset threshold, it is marked as a region with high computational demand; if the score is lower than or equal to the preset threshold, it is marked as a region with low computational demand. The spatial adaptive electromagnetic computing module, based on the partitioning results of the region partitioning module, executes a spatially adaptive electromagnetic computing dimensionality reduction strategy. It allocates simplified computing dimensions and adopts an empirical statistical model for regions with low computing demand, and allocates refined computing dimensions and adopts a full-wave algorithm for regions with high computing demand. The electromagnetic distribution prediction module performs electromagnetic distribution prediction calculations based on an electromagnetic computation dimensionality reduction strategy. It also applies linear interpolation to smooth the transition at the boundaries of adjacent regions to eliminate the discontinuity of field strength caused by abrupt changes in computational dimension, thereby generating continuous electromagnetic field distribution prediction data across the entire domain.

[0017] Thirdly, this application provides a terminal, including: At least one processor; Memory used to store at least one processor-executable instruction; In this embodiment, at least one processor is configured to execute instructions to implement the electromagnetic distribution prediction method based on the environmental medium model proposed in this application.

[0018] Compared with the prior art, the beneficial effects of this application are as follows: 1. This application proposes an electromagnetic distribution prediction method based on an environmental medium model. This method discretizes the target space grid into several sub-regions, calculates the total complexity evaluation value of each region based on the geometric complexity and electromagnetic propagation environment complexity characteristics of each sub-region, and dynamically divides high and low computational demand regions according to a preset threshold. Low complexity regions are solved quickly by using empirical models or low-order ray tracing; high complexity regions are solved by using full-wave numerical algorithms or high-order ray tracing to ensure local computational accuracy. By dynamically distinguishing the intensity of computational resource input, the method solves the problem of overcomputation and insufficient accuracy in unified modeling, and reduces the computational burden of the system while ensuring overall accuracy.

[0019] 2. This application proposes an electromagnetic distribution prediction method based on an environmental medium model. By using differentiated computational dimension allocation and boundary smoothing mechanism, the spatial continuity and physical consistency of the field distribution are guaranteed, releasing the acceleration potential of the simplified model in low-complexity regions. It is particularly suitable for electromagnetic propagation environments with significant spatial heterogeneity, such as urban blocks, indoor multi-rooms, and vegetation cover.

[0020] 3. This application proposes an electromagnetic distribution prediction method based on an environmental medium model. By using a parallel execution architecture, the prediction cycle is further shortened, supporting application requirements that are sensitive to response speed, such as emergency communication deployment, spectrum management and scheduling, and navigation and obstacle avoidance of unmanned systems. It provides an efficient and adaptive computational implementation framework for electromagnetic sensing in complex environments. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating the electromagnetic distribution prediction method based on an environmental medium model in an embodiment of this application. Figure 2 This is a flowchart illustrating step S4 in an embodiment of this application; Figure 3 This is a flowchart illustrating step S6 in an embodiment of this application. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0023] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] This application aims to provide an electromagnetic distribution prediction method, system, and terminal based on an environmental medium model. The prediction method relies on multi-type medium electromagnetic characteristic models and parameter generation mechanisms to obtain the geometric configuration, spatial distribution, and frequency-related electromagnetic parameters of each medium in the target space. The target space grid is then discretized into several sub-regions. The geometric complexity and electromagnetic propagation environment complexity features of each sub-region are extracted, and a weighted scoring function integrating these two types of features is established to solve for the total complexity evaluation value of the sub-region. High and low computational demand regions are dynamically divided according to a preset threshold. Low-complexity regions are solved quickly using empirical models or low-order ray tracing. High-complexity regions use full-wave numerical algorithms or high-order ray tracing to ensure local computational accuracy. Linear interpolation and weighted averaging are used at the boundaries of high and low regions to achieve a smooth transition, ensuring continuous field quantities without abrupt changes. Finally, predictions for each sub-region are completed in parallel based on a parallel computing architecture, and the results are integrated to obtain a globally consistent high-fidelity electromagnetic distribution prediction map.

[0025] Firstly, this application provides a method for predicting electromagnetic distribution based on an environmental medium model, such as... Figure 1 As shown, it specifically includes: S1: Obtain geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space. The electromagnetic parameters include at least three of the following: dielectric constant, loss tangent, and conductivity, to generate the original media dataset.

[0026] As an example, step S1 specifically includes the following sub-steps: S1.1: Collect and preprocess multi-source heterogeneous geometric data in the target space, and use a three-dimensional reconstruction algorithm to extract the shape, position coordinates and surface curvature variation features of the scatterer to generate a standardized geometric feature dataset.

[0027] It receives multi-source heterogeneous raw geometric structure data within the target space, including LiDAR point clouds, oblique photogrammetry 3D models, and CAD design drawings, as the initial input for geometric feature extraction.

[0028] Coordinate system unification and denoising preprocessing are performed on multi-source heterogeneous data. Environmental noise points are filtered out using a statistical outlier removal algorithm. The data collected by different sensors are registered to a unified global Cartesian coordinate system through an iterative nearest point algorithm to generate a spatially aligned fused point cloud dataset.

[0029] Surface reconstruction is performed based on the fused point cloud dataset. A closed triangular mesh model is constructed using the Poisson surface reconstruction algorithm to repair holes and non-manifold edges in the mesh, thereby generating a continuous three-dimensional geometric surface model with a complete topological structure.

[0030] Curvature analysis is performed on the three-dimensional geometric surface model. The principal curvature of each mesh vertex is calculated using the principle of differential geometry, and the Gaussian curvature is obtained by product.

[0031] Based on the calculated Gaussian curvature distribution, the geometric abrupt change regions of the scatterer are identified, and the shape contour features, spatial coordinates, and surface curvature change gradient of the scatterer are extracted to form a local geometric feature descriptor.

[0032] The local geometric feature descriptor is normalized by performing a normalization mapping, which normalizes the discrete feature values ​​to a preset interval, eliminates the difference in dimensions, and generates a normalized geometric structure feature dataset.

[0033] Through the above chain-like derivation process, the multi-source raw data from the previous step is transformed into a standardized geometric structure feature dataset containing precise morphological and positional information, achieving the expected technical effect of providing high-precision geometric constraints for subsequent media type identification.

[0034] For example, in an electromagnetic simulation scenario of an urban street, LiDAR point cloud data with a resolution of 0.05 meters was collected. Using a statistical outlier removal algorithm, with a neighborhood count of 30 and a standard deviation factor of 2.0, floating noise points in the air were filtered out. A Poisson reconstruction algorithm was employed, with an octree depth of 10, to generate a closed triangular mesh. The curvature of the mesh vertices was calculated. For building corners, the principal curvature k1 was 0.8, k2 was 0.1, and the calculated Gaussian curvature K was 0.08, marking it as a high curvature feature region; for flat walls, the K value was close to 0. The XYZ coordinates and K values ​​of all vertices were extracted and normalized to the [0,1] interval, generating a geometric structure feature dataset containing 100,000 mesh cells, significantly improving the geometric positioning accuracy of subsequent medium matching.

[0035] S1.2: Based on the geometric structure feature dataset, perform medium type identification processing, use spectrum analysis technology to match the material category corresponding to each geometric unit, and call the frequency-related electromagnetic parameter database to retrieve the values ​​of dielectric constant, loss tangent and conductivity to generate a candidate set of medium electromagnetic parameters with frequency labels.

[0036] It receives a standardized dataset of geometric structural features, extracts the three-dimensional spatial coordinate index and surface normal vector distribution information of each geometric unit, and uses them as spatial constraints for medium identification.

[0037] The surface reflectivity data of the geometric unit is subjected to spectral decomposition. The time-domain reflection signal is converted into a frequency-domain complex impedance spectrum using fast Fourier transform to obtain the amplitude attenuation and phase shift characteristics at different frequency points.

[0038] Construct a material fingerprint library based on dielectric response characteristics to store standard curves of the complex dielectric constant of common building materials as a function of frequency, including reference values ​​of electromagnetic parameters for concrete, glass, metal and composite panels.

[0039] The least squares matching algorithm is used to calculate the sum of squared residuals between the measured frequency domain complex impedance spectrum and the standard curve in the material fingerprint database as the matching distance. The material fingerprint database is traversed, and the material category with the smallest matching distance D is selected as the initial medium type identifier of the current geometric unit, and the corresponding confidence score is recorded.

[0040] For low-confidence regions, multi-feature fusion verification is performed. By combining geometric curvature threshold and spatial neighborhood medium consistency constraint, the abnormal classification results are corrected and isolated noise points are eliminated.

[0041] The frequency-related electromagnetic parameter database is invoked, and the discrete dielectric constant, loss tangent, and conductivity values ​​of the material within the target operating frequency band are retrieved based on the determined medium type identifier.

[0042] The retrieved discrete electromagnetic parameters are bound to the spatial index of the geometric unit to generate a structured data record containing a medium type label and the corresponding frequency point parameter value.

[0043] Through the above-mentioned spectrum matching and database retrieval processing methods, the geometric structural features of the previous step are transformed into a candidate set of electromagnetic parameters of the medium with frequency labels, thereby achieving accurate mapping of the physical properties of the medium.

[0044] For example, in an indoor office setting, the operating frequency band is set to 2.4GHz to 5.8GHz, with a step size of 0.2GHz. Spectral analysis is performed on the wall geometry to obtain complex impedance data at 18 frequency points. A material fingerprint database is constructed, including concrete, gypsum board, and tempered glass. The matching distance D between the measured impedance and the standard curve for concrete is calculated to be 0.05, 0.32 for gypsum board, and 0.45 for glass. The element is determined to be made of concrete with a confidence level of 95%. A database search reveals that concrete at 2.4GHz has a dielectric constant of 6.5, a loss tangent of 0.02, and a conductivity of 0.01 S / m. This parameter set is written into the element attribute field to generate candidate entries.

[0045] S1.3: Perform dynamic frequency interpolation calculation on the candidate set of dielectric electromagnetic parameters, and use a linear interpolation algorithm to fill the parameter gaps between discrete frequency points to generate a complete dielectric electromagnetic parameter sequence in the continuous frequency domain.

[0046] The system receives a candidate set of dielectric electromagnetic parameters with frequency labels generated by S1.2. This dataset contains the dielectric constant, loss tangent, and conductivity values ​​at discrete sampling points. The frequency dimension of the candidate set is sorted and its integrity is verified. Interpolation intervals between adjacent known frequency points are identified, and a frequency domain mapping model based on linear relationships is constructed. For each intermediate frequency point to be interpolated, the upper and lower boundary frequency nodes of its interval and their corresponding electromagnetic parameter values ​​are located to determine the reference data pair for interpolation calculation. A piecewise linear interpolation algorithm is used to calculate the weighting coefficients of each electromagnetic parameter based on the relative position ratio of the target frequency within the interval, achieving continuous derivation of parameter values. Specifically, for any target frequency f, its corresponding dielectric constant ε(f) is calculated using the following formula: in, and For adjacent known frequency points, and To correspond to the dielectric constant, linear interpolation calculations for the loss tangent and conductivity are performed simultaneously to ensure consistency and physical rationality of the variation trends of multiple physical parameters in the frequency domain. The calculated continuous frequency domain parameter sequence is merged and recombined with the original discrete data to eliminate discontinuities caused by parameter gaps, generating a complete dielectric electromagnetic parameter sequence covering the entire operating frequency band. Through dynamic frequency interpolation, the discrete parameter candidate set from the previous step is transformed into a complete dielectric electromagnetic parameter sequence in the continuous frequency domain, achieving a smooth transition and high-precision restoration of frequency response characteristics in electromagnetic simulation.

[0047] For example, when modeling the electromagnetic properties of concrete media in the 1GHz to 3GHz frequency band, the database only provides parameters for three discrete points: 1GHz, 2GHz, and 3GHz. The dielectric constant is 6.5 at 1GHz, 6.2 at 2GHz, and 5.9 at 3GHz. The dielectric constant at 1.5GHz needs to be obtained. Substituting into the above formula, f1=1, f2=2, ε1=6.5, ε2=6.2, and the target frequency f=1.5. The calculated ε(1.5)=6.5+(1.5-1) / (2-1)×(6.2-6.5)=6.35. Similarly, other frequency points and loss parameters are calculated, generating a continuous parameter sequence with a step size of 0.1GHz, significantly improving the frequency response fitting accuracy in subsequent ray tracing simulations.

[0048] S1.4: Based on the geometric structure feature dataset and the complete medium electromagnetic parameter sequence, perform spatial mapping association operation, and use spatial indexing technology to assign electromagnetic parameters to the corresponding geometric grid cells to generate intermediate state data of medium spatial distribution features with physical attribute annotations.

[0049] This process receives a geometric feature dataset and a complete dielectric electromagnetic parameter sequence, and constructs a 3D mesh mapping framework based on a spatial hash index. The target space is uniformly voxelized, establishing a direct mapping between voxel coordinates and memory addresses to form an initial spatial index table. Each scatterer surface mesh cell in the geometric feature dataset is traversed, and its centroid coordinates and bounding box extent are calculated. The spatial index table is queried based on the centroid coordinates to determine the target voxel index ID to which the mesh cell belongs. The complete dielectric electromagnetic parameter sequence is retrieved, and the dielectric constant, loss tangent, and conductivity values ​​matching the material type of the mesh cell are extracted. The extracted electromagnetic parameter values ​​are encapsulated into attribute vectors and written to the attribute storage area of ​​the corresponding voxel index ID. For complex geometric cells spanning multiple voxels, the centroid coordinate interpolation method is used to calculate the weight distribution of each vertex within adjacent voxels. Electromagnetic parameters are discretely allocated to all relevant voxels covering the geometric cell according to the weight ratio, ensuring the spatial continuity of physical properties. Blank voxels without assigned electromagnetic parameters are detected, and nearest-neighbor interpolation is performed based on the dielectric type identifier of neighboring voxels to eliminate data gaps. Through spatial mapping and association operations, the discrete geometric and parametric data from the previous step are transformed into intermediate data of medium spatial distribution characteristics with physical attribute annotations, achieving precise alignment between geometric topology and electromagnetic properties, and providing a structured data foundation for subsequent 3D model construction.

[0050] For example, for an indoor scene with dimensions of 10m × 10m × 5m, a voxel resolution of 0.1m is set, generating a 100 × 100 × 50 spatial index grid. Geometric mesh data of the concrete wall is collected, with its centroid coordinates at (2.5, 3.0, 2.5). The index table is queried to locate the voxel ID (25, 30, 25). The dielectric constant of the concrete at a frequency of 2.4GHz is read from the electromagnetic parameter sequence: 5.5, loss tangent 0.02, and conductivity 0.01 S / m. This triplet parameter is written into the attribute field of voxel (25, 30, 25). For the metal table leg spanning the voxel boundary, the centroid weights of its vertices within the four adjacent voxels are calculated to be 0.6, 0.2, 0.1, and 0.1, respectively. The conductivity is then allocated proportionally to 5.8 × 10⁻⁶ voxels. 7 After interpolation and padding, all voxels possess complete electromagnetic properties, resulting in a compact data structure and significantly improved query efficiency.

[0051] S1.5: Perform format encapsulation and integrity verification on intermediate state data of spatial distribution characteristics of the medium with physical attribute annotations, and integrate geometric information and electromagnetic parameter information using a data serialization protocol to generate the original medium dataset.

[0052] S2: Based on the original medium dataset, perform precise assignment of medium electromagnetic parameters, map the assigned parameters to the three-dimensional geometric mesh of the target space, and construct a three-dimensional model of the medium environment.

[0053] As an example, step S2 specifically includes the following sub-steps: S2.1: Discretize the multi-type medium geometric structure data contained in the original medium dataset, and use an adaptive mesh generation algorithm to generate a non-uniform three-dimensional geometric mesh covering the target space, so as to output an initial three-dimensional geometric mesh model with spatial topological relationships.

[0054] It should be noted that the 3D model of the medium environment is a structured data model composed of a 3D geometric mesh and its associated electromagnetic parameter attribute fields. Its architecture includes two layers: a mesh topology layer and a physical attribute layer. The mesh topology layer stores the geometric coordinates, spatial adjacency relationships, and boundary topology information of the mesh cells, supporting spatial indexing and region partitioning. The physical attribute layer stores the medium type identifier of each mesh cell and its corresponding dielectric constant, loss tangent, and conductivity values, supporting subsequent electromagnetic wave propagation calculations. The model's input consists of the geometric structure data and frequency-related electromagnetic parameters from the original medium dataset. The output is a 3D meshed data field covering the entire target space, with each mesh cell possessing complete electromagnetic parameter attributes. This model serves as the fundamental data carrier for electromagnetic distribution prediction calculations in subsequent steps, providing a unified geometric-electromagnetic parameter associated data environment for ray tracing, full-wave solution, and complexity feature extraction.

[0055] Receive the original medium dataset containing the spatial distribution characteristics of the medium generated in step S1, and extract the boundary coordinates and topological connection data of the geometric structures of various types of media.

[0056] Bounding box calculations are performed over the entire target space to determine the minimum bounding cuboid range. Based on the ratio between electromagnetic wavelength and the minimum feature size of the medium, the basic resolution parameters for the initial mesh generation are set.

[0057] An octree spatial partitioning algorithm is used to recursively divide the target space, subdividing the spatial unit into eight sub-cubes step by step until the preset geometric accuracy threshold or maximum recursion depth limit is met.

[0058] During the recursive partitioning process, it is detected in real time whether each sub-cube contains a medium interface or high curvature geometric features. If a geometric discontinuity is detected, the sub-cube is forced to be further subdivided to capture details.

[0059] For homogeneous media regions without complex geometric features, recursive subdivision is stopped and adjacent idle nodes are merged to generate sparse coarse grid cells to reduce data redundancy.

[0060] For mesh elements near the medium interface, unstructured tetrahedral partitioning is performed, and the Delaunay triangulation algorithm is used to optimize the mesh shape quality, ensuring that the interior angles of the mesh elements meet the numerical computation stability requirements.

[0061] Establish an adjacency list data structure between grid cells, record the face, edge, and vertex indices of each grid cell and the identifiers of its neighboring cells, and construct a complete spatial topological relationship network.

[0062] Perform consistency checks on the generated 3D geometric mesh, and repair topological errors such as dangling nodes, overlapping cells, and non-manifold edges to ensure the geometric closure and logical integrity of the mesh model.

[0063] By using adaptive mesh generation and unstructured optimization, the original discrete geometric data is transformed into an initial three-dimensional geometric mesh model with high-precision topological relationships, realizing the accurate discretization representation of complex medium boundaries and providing a standardized spatial carrier for subsequent electromagnetic parameter mapping.

[0064] S2.2: Based on the centroid coordinates of each grid cell in the initial three-dimensional geometric grid model, perform a medium spatial distribution matching operation and use a spatial index query mechanism to map the medium type identifier in the original medium dataset to the corresponding grid cell to generate a preliminary attribute grid model with medium type identifier.

[0065] Receive the initial 3D geometric mesh model generated by S2.1, extract the centroid spatial coordinates and bounding box boundary information of each mesh cell, and construct a spatial index structure for fast retrieval.

[0066] Based on the multi-type media geometry definitions in the original media dataset, the minimum bounding rectangle volume of each media entity is calculated using the axis-aligned bounding box (AABB) algorithm, and a spatial index tree of media objects is established.

[0067] Traverse each grid cell in the initial 3D geometric grid model and obtain its centroid coordinate vector, which serves as the key value for spatial location query.

[0068] Using the constructed medium space index tree, perform an inclusion check operation on the centroid coordinates of the current grid cell to determine whether the coordinate point falls within the bounding box of any medium entity.

[0069] If the centroid coordinates fall within the area surrounded and overlapping by multiple media entities, further calculations using ray casting or symbolic distance field (SDF) are employed to accurately determine the topological relationship between the centroid and the boundaries of each media, thus identifying the unique media type identifier.

[0070] The determined medium type identifier is written into the attribute field of the corresponding grid cell, thus completing the mapping association from geometric space to medium semantic space.

[0071] Through the aforementioned spatial index query and topology determination processing methods, the initial three-dimensional geometric mesh model is transformed into a preliminary attribute mesh model with a clear medium type identifier, realizing the discretization and accurate assignment of medium distribution information, and providing an accurate material classification basis for subsequent electromagnetic parameter filling.

[0072] For example, in an architectural interior scene containing concrete walls and glass windows, the initial mesh model is divided into 10cm × 10cm × 10cm cube units. A medium spatial index based on R-tree is established, with the concrete walls defined as medium ID01 and the glass windows as medium ID02. For the centroid of the mesh unit at coordinates (5.05, 3.05, 2.05), a quick R-tree search reveals that it is located within the wall's bounding box. Further calculation shows that the signed distance from this point to the wall surface is positive, and since it is not within the glass window's bounding box, this mesh unit is marked as ID01. For mesh units located at wall corners, if the centroid is close to both the wall and air regions, SDF calculation confirms that it is located inside the wall, and it is also marked as ID01. In the final generated preliminary attribute mesh model, all meshes are assigned a unique medium ID, ensuring the accuracy of subsequent electromagnetic parameter assignments and avoiding calculation errors caused by ambiguous medium attribution.

[0073] S2.3: For each grid cell in the preliminary attribute grid model with dielectric type identifier, perform frequency-related electromagnetic parameter assignment operations, and fill the attribute fields of the corresponding grid with dielectric constant, loss tangent and conductivity according to the preset dielectric electromagnetic parameter lookup table to generate a physical attribute grid model with complete scalar electromagnetic parameters.

[0074] For each grid cell in the preliminary attribute grid model with a medium type identifier, its centroid coordinates and associated medium type identifier are read and used as index keys for assigning electromagnetic parameters.

[0075] Based on the preset frequency operating point, the discrete frequency electromagnetic parameter set corresponding to the medium type is retrieved from the original medium dataset. This set includes measured or simulated data points of dielectric constant, loss tangent, and conductivity.

[0076] To address the mismatch between the current simulation frequency and discrete data points, the same linear interpolation logic as in S1.3 is used to calculate the dielectric constant, loss tangent, and conductivity values ​​at the target frequency. This constructs all the scalar parameters required for a complete complex dielectric constant model, ensuring the physical continuity of the parameters in the frequency domain.

[0077] The calculated dielectric constant, loss tangent, and conductivity values ​​are filled into the attribute fields of the corresponding grid cells according to the predefined data structure, thus completing the instantiation of the physical attributes of a single grid.

[0078] Traverse all non-uniform three-dimensional geometric mesh units in the target space, and repeat the above search, interpolation and assignment operations until the entire space is covered.

[0079] By performing precise mapping of frequency-related electromagnetic parameters grid by grid, the preliminary attribute grid model with medium type identifier generated in the previous step is transformed into a physical attribute grid model with complete scalar electromagnetic parameters, realizing the digital reconstruction of the physical properties of the electromagnetic environment and providing high-precision underlying data support for subsequent complexity scoring.

[0080] For example, in an indoor Wi-Fi signal propagation scenario with an operating frequency of 2.4 GHz, the mesh cell G125 is identified as concrete material. The database contains two reference frequencies: 2.0 GHz and 3.0 GHz, with corresponding dielectric constants of 5.5 and 5.2, respectively. The calculation is: ε(2.4) = 5.5 + ((2.4 - 2.0) / (3.0 - 2.0)) × (5.2 - 5.5) = 5.5 + 0.4 × (-0.3) = 5.38. Similarly, the loss tangent at this frequency is calculated to be 0.024, and the conductivity is 0.015 S / m. These values ​​are then written into the attribute fields of G125. The same operation is performed on approximately 100,000 mesh cells across the entire domain, generating a 3D model containing complete electromagnetic parameters. This significantly improves the accuracy of subsequent ray tracing algorithms in calculating reflection and refraction at the medium boundary, eliminating the field strength prediction jump phenomenon caused by parameter discretization.

[0081] S2.4: Based on a physical property grid model with complete scalar electromagnetic parameters, perform parameter smoothing transition processing at the medium interface, and use a weighted average interpolation algorithm to perform gradient correction on the electromagnetic parameters at the boundary of adjacent different medium types of grids, so as to generate a three-dimensional model of the medium environment that eliminates numerical abrupt changes and has continuous physical field characteristics.

[0082] Obtain a physical property mesh model with complete scalar electromagnetic parameters, and identify the set of interface mesh elements where the medium type changes abruptly.

[0083] For the identified interface grid cells, the dielectric constant, loss tangent and conductivity values ​​of their adjacent grid cells are extracted to construct a local parameter gradient field.

[0084] Based on the local parameter gradient field, the Euclidean distance between adjacent grid centers is calculated as a spatial weighting factor to determine the contribution of each neighboring node to the boundary node.

[0085] An inverse distance weighted interpolation algorithm is used to calculate the electromagnetic parameters of adjacent grids by weighting them according to the spatial weight factor, thereby generating smooth transition parameter values ​​at the boundary.

[0086] The smoothed dielectric constant of the interface mesh element k is calculated using the following formula: in, interface mesh The smooth dielectric constant, For adjacent grid sets, For the first Spatial weights of adjacent grids, For the first The original dielectric constant of each adjacent grid.

[0087] Replace the abrupt change parameters at the interface in the original mesh model with the calculated smooth transition parameter values, and perform a full-field consistency check.

[0088] By using weighted average interpolation, the results of the previous step are transformed into a three-dimensional model of the medium environment that eliminates numerical abrupt changes and possesses continuous physical field characteristics, thereby improving the stability of boundary field strength calculation in electromagnetic simulation.

[0089] For example, for the mesh cells at the concrete-air interface, the dielectric constant of adjacent concrete meshes (5.0) and air meshes (1.0) are extracted. The spatial weighting factor is set to the reciprocal of the distance, resulting in a smoothed dielectric constant of 3.0 for the interface. This process eliminates the abrupt change in dielectric constant from 5.0 to 1.0, reducing the error in calculating the refraction angle during ray tracing across the boundary and significantly improving the accuracy of electromagnetic wave propagation prediction in complex media environments.

[0090] S3: Extract the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and input the above features into the preset weighted summation logic to calculate and generate the total complexity score.

[0091] As an example, step S3 specifically includes the following sub-steps: S3.1: Based on the three-dimensional model of the medium environment, the surface curvature change rate and geometric structure discreteness of the scatterer in each sub-region of the target space are processed by gridding to generate a geometric complexity feature vector set that characterizes the complexity of the spatial morphology.

[0092] It should be noted that a sub-region is a set of continuous grid cells obtained after spatially discretizing the target space. The sub-regions are spatially adjacent and share boundaries through grid topology relationships, serving as the basic spatial units for complexity feature extraction.

[0093] The system receives a 3D model of the receiving medium environment and extracts the geometric topology data of the grid cells in each sub-region. It then performs a discretized grid traversal on the scatterer surface of each sub-region within the target space, calculates the cosine of the angle between the normal vectors of adjacent grid cells, and constructs a local curvature tensor matrix. Based on this local curvature tensor matrix, it uses Gaussian curvature and mean curvature algorithms to quantify the surface curvature of each grid cell, generating a curvature distribution heatmap.

[0094] Gradient operations are performed on the curvature distribution heatmap to calculate the rate of change of curvature in the spatial dimension, identify curvature abrupt boundary, and generate a surface curvature change rate feature vector. The spatial distribution density of grid cells within sub-regions is statistically analyzed, and the ratio of the number of grid nodes to the number of connected edges per unit volume is calculated to characterize the discreteness of the geometric structure. The Shannon entropy concept is introduced to evaluate the uniformity of grid node distribution, and a geometric structure discreteness index is calculated to reflect the irregularity of the scatterer's morphology. The surface curvature change rate feature vector and the geometric structure discreteness index are concatenated and fused to form a multidimensional geometric complexity feature descriptor. Principal component analysis is performed on the multidimensional geometric complexity feature descriptor for dimensionality reduction, extracting the top K principal components with the highest contribution rates to eliminate feature redundancy.

[0095] By using a gridded scanning process, the physical property grid from the previous step is transformed into a set of geometric complexity feature vectors that characterize the complexity of spatial morphology, thereby achieving a quantitative mapping of geometric morphology to computational difficulty.

[0096] For example, for a sub-region of urban building clusters, the grid resolution is set to 0.5 meters. The calculated Gaussian curvature of a certain building's exterior wall grid cell is 0.02, the average curvature is 0.15, and the gradient value of the rate of change of curvature is 0.08. The grid node density of this sub-region is 120 nodes / cubic meter, the edge ratio is 3.5, and the geometric discretization entropy is 1.2. Through principal component analysis, the first three principal components are extracted to form a geometric complexity feature vector [0.85, 0.62, 0.41]. This vector accurately reflects the complex geometric details of the building facade, providing a precise spatial constraint basis for subsequent electromagnetic propagation simulations and significantly improving the accuracy of region division.

[0097] S3.2: Using the geometric complexity feature vector set as a spatial constraint, and combining the frequency-related dielectric constant and conductivity distribution data in the three-dimensional model of the medium environment, perform multi-path ray tracing attenuation simulation to generate an electromagnetic wave propagation complexity feature matrix that reflects beam penetration loss and scattering intensity.

[0098] The system receives the geometric complexity feature vector set generated by S3.1 and inputs it into the ray tracing engine as spatial constraints. Combining the dielectric constant and conductivity distribution data of each mesh element in the 3D model of the medium environment, it initializes a multipath ray emission source. It then sets the ray emission angle step size and the maximum reflection threshold to construct an initial ray beam set.

[0099] For each ray path, a grid-by-grid traversal operation is performed, calculating the refraction angle and reflection coefficient of the ray at the interface of the medium using Snell's law. The energy distribution ratio at the interface is derived using Fresnel's equations, updating the ray's propagation direction and amplitude intensity. For the ray portion that penetrates the medium, its propagation distance within the current grid is calculated.

[0100] Based on frequency-dependent conductivity and loss tangent parameters, an exponential decay model is used to calculate the absorption loss of rays inside the medium, and the field strength attenuation factor of a single path is obtained.

[0101] The total path loss of the ray from the emission point to the current grid center is calculated cumulatively, including free-space diffusion loss, interface reflection loss, and medium absorption loss. The number of all ray paths falling into the target sub-region grid and their average remaining energy are counted. The path density is multiplied by the reciprocal of the average energy to generate a local complexity index characterizing the scattering intensity and penetration difficulty of the region.

[0102] Traverse all sub-regions within the target space, repeating the ray tracing and loss accumulation process described above to construct an electromagnetic wave propagation complexity feature matrix covering the entire region. The element values ​​of this matrix directly reflect the degree of obstruction to electromagnetic wave propagation and the significance of multipath effects in the corresponding region.

[0103] By performing multipath ray tracing attenuation simulation, geometric constraints and medium electromagnetic parameters are transformed into electromagnetic wave propagation complexity feature matrices that quantify beam penetration loss and scattering intensity. This enables accurate measurement of the complexity of the electromagnetic propagation environment in each region, providing a physical basis for subsequent dynamic weight allocation.

[0104] For example, in the 2.4 GHz band, the total number of ray emissions is set to 10,000, and the maximum number of reflections is 5. For a concrete wall region with a dielectric constant of 4.5 and a conductivity of 0.01 S / m, the calculated average path loss is 45 dB, and the ray penetration is less than 10%. The electromagnetic wave propagation complexity feature matrix element values ​​corresponding to this region are significantly higher than those in the free space region, accurately identifying the region with high computational requirements.

[0105] S3.3: Based on the statistical distribution characteristics of the electromagnetic wave propagation complexity feature matrix and the geometric complexity feature vector set, the entropy weight method is used to dynamically calculate the weight coefficients of each feature dimension in order to generate an adaptive weight configuration parameter set that adapts to the dynamic changes of the current environmental medium.

[0106] Receive the geometric complexity feature vector set generated by S3.1 and the electromagnetic wave propagation complexity feature matrix generated by S3.2, and construct the original feature data matrix containing multidimensional feature indicators of all sub-regions.

[0107] The original feature data matrix is ​​subjected to standardization preprocessing. The range transformation method is used to eliminate the dimensional differences and map the values ​​of each feature dimension to the [0,1] interval to generate a dimensionless standard feature matrix.

[0108] The information entropy value of each feature dimension is calculated based on the standard feature matrix, and then the difference coefficient is obtained. The initial weight of each dimension is obtained through normalization processing, specifically using the standard calculation process of the entropy weight method. In this embodiment, the preset weighted summation logic refers to the weighted fusion logic based on the entropy weight method.

[0109] An environmental medium dynamic adjustment factor is introduced, and the initial weights are fine-tuned according to the variance of the dielectric loss tangent distribution at the current frequency, thereby increasing the weight ratio of electromagnetic propagation characteristics in the high-loss dielectric region.

[0110] The fine-tuned weight combination is encapsulated into an adaptive weight configuration parameter group, which reflects in real time the different sensitivities of the current environmental medium characteristics to geometric and electromagnetic propagation complexity.

[0111] By dynamically calculating using the entropy weight method and correcting for media characteristics, the feature data from the previous step is transformed into an adaptive weight configuration parameter set that adapts to the dynamic changes in the environment, thereby enabling the scoring function to accurately respond to media heterogeneity.

[0112] S3.4: Based on the adaptive weight configuration parameter group, perform weighted fusion calculation operation on the geometric complexity feature vector set and the electromagnetic wave propagation complexity feature matrix to generate an initial complexity evaluation value sequence that quantitatively represents the comprehensive computational difficulty of each sub-region.

[0113] The system receives an adaptive weight configuration parameter set, a geometric complexity feature vector set, and an electromagnetic wave propagation complexity feature matrix as input data. For each discretized grid cell in the target space, its corresponding geometric complexity scalar value and electromagnetic wave propagation complexity scalar value are extracted to construct a two-dimensional feature input vector for that cell. Based on the geometric dimension weight coefficients and propagation dimension weight coefficients determined in the adaptive weight configuration parameter set, multiplication operations are performed on the extracted geometric complexity scalar value and electromagnetic wave propagation complexity scalar value to generate weighted geometric component values ​​and weighted propagation component values. The weighted geometric component values ​​and weighted propagation component values ​​of the same grid cell are linearly superimposed and summed to calculate the initial unnormalized complexity evaluation value of that cell. The above weighted fusion calculation logic is repeated for all grid cells in the target space until the entire spatial range is covered. Through the weighted fusion calculation processing method, the adaptive weights and multi-dimensional feature data from the previous step are transformed into a sequence of initial complexity evaluation values ​​that quantitatively represent the comprehensive computational difficulty of each sub-region, achieving a precise quantitative assessment of the electromagnetic computing resource requirements of each region.

[0114] For example, in an electromagnetic environment model of a city block, the total number of grid cells is set to 10,000. For the i-th grid cell, the surface curvature change rate feature value Gi is extracted from the geometric complexity feature vector set as 0.85, and the multipath attenuation factor feature value Pi is extracted from the electromagnetic wave propagation complexity feature matrix as 0.62. Based on the adaptive weight configuration parameter set calculated by the entropy weight method, the geometric dimension weight coefficient wG is determined to be 0.4, and the propagation dimension weight coefficient wP is determined to be 0.6. Weighted fusion calculation is performed: the weighted geometric component is 0.4 × 0.85 equals 0.34, and the weighted propagation component is 0.6 × 0.62 equals 0.372. The two components are added together to obtain the initial complexity evaluation value Ci of the i-th grid cell as 0.34 + 0.372 equals 0.712. The same operation is performed on all 10,000 grid cells to generate an initial complexity evaluation value sequence of length 10,000. This sequence accurately reflects the combined impact of building density and material loss in different neighborhoods on the difficulty of electromagnetic calculations, significantly improving the accuracy of subsequent area division.

[0115] S3.5: Apply a nonlinear normalization mapping function to the initial complexity evaluation value sequence to unify the dimensions, so as to eliminate the influence of differences in the magnitude of different features and finally generate a standardized list of total complexity scores.

[0116] S4: The total complexity score is compared with a preset threshold. If the score is higher than the preset threshold, it is marked as a high-computation-demand region; if the score is lower than or equal to the preset threshold, it is marked as a low-computation-demand region, thus generating the region division results. Figure 2 As shown.

[0117] As an example, step S4 specifically includes the following sub-steps: S4.1: Perform statistical distribution analysis on the total complexity score to extract score extreme value features and distribution density features. Use an adaptive threshold calculation algorithm to dynamically generate the optimal segmentation threshold based on the score extreme value features and distribution density features to obtain a dynamic judgment benchmark value for distinguishing the difficulty level of the calculation.

[0118] The system receives a list of standardized total complexity scores generated by S3.5 as the input data source for dynamic threshold calculation. Histogram statistical distribution analysis is performed on the score list, setting the initial bin size to the square root of the total number of scores, and constructing a frequency distribution histogram to quantify the clustering and dispersion of the scores. The global maximum and minimum values ​​in the histogram are extracted as extreme value features, and the peak position and half-width at half-maximum (WHM) of the distribution curve are calculated as distribution density features, forming a statistical feature vector characterizing the current environmental complexity distribution. Based on the statistical feature vector, the Otsu's algorithm (maximum between-class variance method) is used to iteratively search for the optimal split point, dividing the score set into two subsets: high complexity and low complexity. During the iteration process, all possible threshold candidate values ​​are traversed, and for each candidate threshold, the within-class variance and between-class variance of the two subsets are calculated. The between-class variance is calculated using the following formula: in, This is the inter-class variance (used to measure the degree of separation between two classes). and These are the probability weights for the two subsets, respectively. and These are the means of the two subsets, respectively.

[0119] The candidate value that maximizes the inter-class variance is selected as the initial optimal threshold. A distribution density correction factor is introduced to fine-tune the initial optimal threshold based on the skewness and kurtosis of the score distribution. When the distribution exhibits long-tail characteristics, it is shifted towards high-density regions to retain more complex details. The corrected dynamic judgment benchmark value is output.

[0120] By using an adaptive threshold calculation algorithm, the standardized score from the previous step is transformed into a dynamic judgment benchmark value that distinguishes the difficulty level of the calculation, thereby achieving an adaptive adjustment effect of the threshold as the environmental medium changes dynamically.

[0121] For example, an analysis was performed on a list of score values ​​containing 1000 grid cells, with a bin number of 32. The statistical results showed that the extreme score range was 0.12 to 0.98, with a peak value at 0.45. Iterative calculation using the Otsu algorithm revealed that the inter-class variance reached its maximum value of 0.045 when the threshold was 0.52. Considering the right skewness of the distribution, a correction factor of 0.03 was introduced, ultimately determining the dynamic judgment benchmark value to be 0.55. This threshold effectively separates the complex scattering region from the free space region, avoiding misjudgments caused by a fixed threshold.

[0122] S4.2: Perform point-by-point comparison processing on the total complexity score based on the dynamic judgment benchmark value, and use logical judgment operators to mark the score value higher than the dynamic judgment benchmark value as a high complexity state and the score value lower than or equal to the dynamic judgment benchmark value as a low complexity state, so as to generate an initial region classification dataset containing binary complexity state labels.

[0123] Receive the dynamic judgment benchmark value generated by S4.1 and the list of standardized total complexity scores output by S3, and construct point-by-point comparison logic to generate the initial region classification dataset.

[0124] Traverse the index of each grid cell in the three-dimensional geometric grid of the target space, and extract the scalar complexity score value corresponding to the current grid cell from the list of standardized total complexity score values.

[0125] The logical judgment operator is invoked to compare the extracted scalar complexity score with the dynamic judgment benchmark value to determine the electromagnetic computation difficulty level of the current grid cell.

[0126] If the scalar complexity score is significantly greater than the dynamic judgment benchmark, the grid cell is determined to be in a state of high electromagnetic wave scattering or strong dielectric loss, and is assigned a high complexity state identifier.

[0127] If the scalar complexity score is less than or equal to the dynamic judgment benchmark value, the grid cell is determined to be in a low electromagnetic wave interaction state and is assigned a low complexity state identifier.

[0128] The judgment result is mapped to a binary complexity state identifier, where the high complexity state identifier is defined as the logical true value 1, and the low complexity state identifier is defined as the logical false value 0.

[0129] The generated binary complexity state identifier is written into the metadata attribute field of the corresponding grid cell, forming a structured data record containing spatial location coordinates and complexity state identifier.

[0130] Repeat the above comparison and marking operations for all grid cells until all discretized grid cells in the target space are covered, ensuring that no region is missed.

[0131] By using a point-by-point comparison process, the continuous scoring data from the previous step is transformed into discrete binary complexity state identifier data, enabling a preliminary quantitative classification of the regional computing demand level and providing a basic input for subsequent spatial connectivity analysis.

[0132] For example, in electromagnetic prediction for urban micro base station coverage scenarios, the dynamic judgment benchmark is set to 0.65. For grid cell G128 located in a densely built-up area, its standardized total complexity score C is 0.82. Since 0.82 is greater than 0.65, the logical judgment operator outputs a true value, marking G128 as a high-complexity state 1. For grid cell G459 located in an open park area, its score C is 0.31. Since 0.31 is less than 0.65, the logical judgment operator outputs a false value, marking G459 as a low-complexity state 0. The resulting initial region classification dataset accurately distinguishes between high-reflectivity building clusters and low-loss open areas, significantly improving the efficiency of subsequent region aggregation.

[0133] S4.3: Perform spatial connectivity analysis on the initial region classification dataset to identify discrete high-complexity state clusters and continuous low-complexity state clusters. Use the region aggregation algorithm to merge grid cells with the same complexity state identifier and spatial adjacency into independent spatial subdomains to generate a set of candidate high-computation-demand regions and a set of candidate low-computation-demand regions with clear geometric boundaries.

[0134] Receive an initial region classification dataset containing binary complexity state identifiers, which consists of grid cell indices and their corresponding high / low complexity labels.

[0135] Based on the topological connectivity of a 3D geometric mesh, a mesh adjacency matrix is ​​constructed, defining two mesh cells as spatially adjacent if they share a face, edge, or vertex.

[0136] A breadth-first search algorithm is used to traverse all grid cells marked as high complexity. Starting from any unvisited high complexity grid, all its spatially adjacent neighbor grids that are also marked as high complexity are recursively searched.

[0137] Encapsulate a set of connected high-complexity grids into an independent high-complexity state cluster and assign a unique cluster ID until all high-complexity grids have been accessed.

[0138] Perform the same connectivity analysis on the grid cells marked as low-complexity states to identify continuous clusters of low-complexity states, forming a preliminary spatial subdomain partition.

[0139] Calculate the volume ratio and surface area-volume ratio of each high-complexity state cluster, remove isolated noise clusters whose volume is less than the preset minimum physical scale threshold, and flip their labels to low complexity to smooth out local anomalies.

[0140] The outer boundary mesh surface of each retained cluster is extracted using the convex hull algorithm or the AlphaShapes algorithm to generate a set of candidate high computational demand regions with well-defined geometric boundaries.

[0141] Similarly, a set of candidate low-computation-requirement regions is generated to ensure that the entire space is covered without overlap.

[0142] By using spatial connectivity analysis and region aggregation processing, discrete grid classification results are transformed into a set of spatial subdomains with topological continuity, realizing a geometrically structured representation of the computational demand region and providing clear boundary constraints for subsequent differentiated algorithm allocation.

[0143] For example, in a scenario with a total of 100,000 target space grids, 35,000 high-complexity grids were initially classified. A minimum physical scale threshold of 0.5 cubic meters was set. After performing a breadth-first search, 12 major high-complexity clusters and 45 small, isolated clusters were identified. The 45 small clusters had an average volume of 0.1 cubic meters, below the threshold, so their labels were corrected to low complexity and they were merged into neighboring low-complexity regions. The remaining 12 high-complexity clusters had their boundaries extracted using the AlphaShapes algorithm, generating a set of candidate high-computational-demand regions. The final output consisted of 12 high-demand regions and 1 low-demand region surrounding the rest of the space. The region boundaries were clear, without fragmentation, significantly improving the efficiency of subsequent full-wave algorithm loading.

[0144] S4.4: Allocate full-wave algorithm calculation dimension labels based on the candidate high computational demand region set and allocate empirical statistical model calculation dimension labels based on the candidate low computational demand region set. Use the label mapping mechanism to write the calculation dimension labels into the metadata attributes of the corresponding spatial subdomain to generate preliminary region division results with calculation dimension demand labels.

[0145] The algorithm receives sets of candidate high computational demand regions and sets of candidate low computational demand regions, and analyzes the grid topology and corresponding total complexity score of each spatial subdomain. It constructs an algorithm label mapping dictionary, defining the full-wave algorithm label identifier as "HighDim_Tag" and the empirical statistical model label identifier as "LowDim_Tag," establishing the correspondence between computational dimensions and solver types. Full-wave algorithms include the finite-difference time-domain method or the finite element method.

[0146] Traverse each independent spatial subdomain in the candidate high computational demand region set and extract the unique index ID of all grid cells within the subdomain. Write the "HighDim_Tag" into the metadata attribute field "CalcMethod" of the aforementioned grid cell to mark that the region requires high-precision full-wave electromagnetic field solving, ensuring accurate capture of complex scattering and diffraction effects.

[0147] Traverse each independent spatial subdomain in the candidate low computational demand region set and extract the unique index ID of all grid cells within the subdomain. Write the "LowDim_Tag" into the metadata attribute field "CalcMethod" of the aforementioned grid cell to mark the region as suitable for low-order ray tracing or statistical approximation calculations, thereby reducing computational power consumption in non-critical regions.

[0148] A regional boundary grid index table is established to identify adjacent grid cell pairs with different labels, and their spatial adjacency is recorded to support subsequent boundary smoothing. The attribute data of all labeled grid cells are integrated to generate a structured preliminary regional division result containing geometric coordinates, medium parameters, and computational dimension labels.

[0149] By using a label mapping mechanism, the region classification results from the previous step are transformed into grid attribute data with explicit computational strategy instructions, enabling differentiated pre-allocation of computational resources in the spatial domain and providing a standardized input interface for subsequent parallel solutions.

[0150] For example, in the simulation of an urban microcell scene, the candidate set of high computational demand areas includes five dense building cluster subdomains, with a total of 200,000 grid cells. The system traverses these subdomains and uniformly assigns the "CalcMethod" field the value "HighDim_Tag". The candidate set of low computational demand areas includes open streets and parks, with a total of 800,000 grid cells, and the system assigns the "CalcMethod" field the value "LowDim_Tag". In the final preliminary region division result, all 1,000,000 grid cells carry clear computational dimension labels, with high-demand areas accounting for 20% and low-demand areas accounting for 80%. This data structure directly supports the parallel invocation of the full-wave algorithm and empirical model in step S5, avoiding computational redundancy caused by globally unified high-precision calculations and significantly improving the simulation efficiency in large-scale scenes.

[0151] S4.5: Perform boundary smoothing on the preliminary region partitioning results with computational dimension requirement labels to eliminate region fragmentation caused by threshold jumps. Optimize the interface topology of adjacent regions with different computational dimension requirement labels using morphological filtering operations to finally generate the region partitioning results.

[0152] Receive preliminary region partitioning results with computational dimension requirement labels. These results include the complexity status identifiers of each grid cell in the target space and the corresponding computational dimension labels.

[0153] The initial region division results are processed by a three-dimensional morphological opening operation. The structural element is defined as a 3x3x3 cube kernel, and all mesh cells marked as low computational demand regions are traversed.

[0154] Identify and remove isolated low-computation-demand grid clusters, and re-mark disconnected regions with a volume smaller than a preset threshold as high-computation-demand states to eliminate noise points caused by local score fluctuations.

[0155] Perform three-dimensional morphological closing operations on the dataset after opening operations to fill the tiny voids inside regions with high computational demands, ensuring the integrity and continuity of the topological structure of highly complex regions.

[0156] Extract the set of interface meshes between label regions of different computational dimensions, and construct a boundary adjacency matrix to represent the spatial topological relationship between adjacent meshes.

[0157] For the grid cells at the interface, calculate the ratio of high computational demand labels to low computational demand labels in their surrounding neighborhood and generate boundary smoothing weight coefficients.

[0158] A majority voting mechanism combined with a distance-weighted algorithm is used to reclassify the transition meshes with ambiguous labels at the interface, so that the boundary direction conforms to the natural trend of medium geometric changes.

[0159] By using the morphological filtering and boundary optimization methods described above, the preliminary partitioning results are transformed into final region partitioning results with smooth topological boundaries, thereby achieving the expected technical effect of eliminating region fragmentation and improving the convergence of subsequent electromagnetic calculations.

[0160] For example, in the simulation of an urban canyon scene, the initial partitioning generates a large number of isolated low-complexity meshes. The structuring element size is set to 0.5 meters, and after performing an opening operation, isolated low-complexity clusters with a volume less than 0.125 cubic meters are removed. A closing operation fills the tiny gaps within high-complexity building clusters. For boundary meshes, if 80% of the neighbors of a low-complexity mesh are high-complexity, it is remarked as high-complexity. In the final generated region partitioning result, the boundaries between high / low demand regions are smooth and continuous, without jagged fragmentation, significantly improving the efficiency of mesh generation in step S5. This makes the coupling interface between the full-wave algorithm and the statistical model clear and unambiguous, avoiding the fragmentation of computational resources caused by region fragmentation.

[0161] S5: Based on the region partitioning results with computational dimension requirement labels, a spatially adaptive electromagnetic computation dimensionality reduction strategy is executed. Regions with low computational requirements are allocated simplified computational dimensions and empirical statistical models are used, while regions with high computational requirements are allocated refined computational dimensions and full-wave algorithms are used to generate differentiated computational configuration schemes.

[0162] As an example, step S5 includes the following sub-steps: S5.1: The region segmentation results with computational dimension requirement labels are parsed and processed to extract the set of regions with high computational requirement and the set of regions with low computational requirement. A full-wave algorithm mapping index is constructed based on the set of regions with high computational requirement, and an empirical statistical model mapping index is constructed based on the set of regions with low computational requirement, thereby generating an initial algorithm matching list containing the correspondence between region categories and candidate algorithms.

[0163] It should be noted that the empirical statistical model refers to a mathematical model that uses a pre-built library of statistical parameters to quickly approximate the electromagnetic wave propagation loss. This model does not require solving Maxwell's equations; instead, it estimates the field strength using a ray tracing reduction algorithm and statistical distribution parameters.

[0164] Receive the region division result data with computational dimension requirement labels, and parse the metadata fields to extract the set of region identifiers with high computational requirement and the set of region identifiers with low computational requirement.

[0165] Perform spatial index traversal on the set of identifiers for regions with high computational demands to obtain the unique ID sequence and boundary topology information of each subdomain in the 3D mesh.

[0166] Based on the interface specification of the full-wave algorithm solver, an algorithm mapping structure containing mesh ID, medium parameter pointer, and boundary condition definition is constructed.

[0167] Inject the unique ID sequence of the high computational demand region into the algorithm mapping structure to establish a one-way reference relationship from the spatial subdomain to the full-wave algorithm instance.

[0168] Parallel scanning is performed on the set of low computational demand regions to extract the corresponding sparse grid coordinates and simplified media property parameters.

[0169] Based on the input requirements of the empirical statistical model, a dimensionality reduction calculation configuration object is generated, which includes ray tracing order constraints and statistical distribution parameters.

[0170] The coordinate information of regions with low computational requirements is bound to the dimensionality reduction calculation configuration object via key-value binding to form an index table for fast approximate calculation.

[0171] The full-wave algorithm mapping structure and the empirical statistical model index table are merged, and the data stream is reorganized according to the spatial location order.

[0172] Through the above mapping and binding process, the region division result of the previous step is transformed into an initial algorithm matching list containing the correspondence between region categories and candidate algorithms, thereby achieving precise matching between computing resources and region complexity.

[0173] For example, in an urban microcell scenario, the high-computation-demand area comprises three densely built-up subdomains with grid IDs 101-150. The full-wave algorithm mapping structure is set to the finite-time difference method with a time step of 0.1 ns. The low-computation-demand area comprises two open plaza subdomains with grid IDs 201-250. The empirical statistical model is configured in dual-ray tracing mode, with a maximum reflection count limit of 2. In the generated initial algorithm matching list, IDs 101-150 point to high-precision solver instances, and IDs 201-250 point to fast statistical model instances, ensuring that subsequent computation threads can directly call the corresponding algorithm configurations, significantly improving system scheduling efficiency.

[0174] S5.2: Based on the set of high computational demand regions in the initial algorithm matching list, perform a fine computational dimension allocation operation. Use a three-dimensional mesh refinement algorithm to encrypt the sub-region meshes within the set of high computational demand regions to generate a fine computational mesh dataset with high-resolution geometric features. Then bind the fine computational mesh dataset with the full-wave algorithm mapping index to generate a high-precision computational configuration unit for high computational demand regions.

[0175] The initial algorithm matching list is parsed to extract the set of regions with high computational requirements and their corresponding spatial boundary coordinates and medium property distribution data.

[0176] Based on the geometric topology within a set of regions with high computational demands, the points of abrupt curvature changes on the surface of the scatterer and the locations of medium interfaces are identified and marked as priority nodes for mesh refinement.

[0177] For the marked priority nodes, the octree subdivision algorithm is applied to perform local mesh subdivision, recursively splitting the original coarse mesh unit into multiple sub-units until the preset minimum mesh size threshold is met.

[0178] During the mesh splitting process, the side length ratio of the sub-units is dynamically adjusted according to the relationship between the electromagnetic wave wavelength and the dielectric constant of the medium to ensure that the phase change error in each sub-unit is controlled within the allowable range.

[0179] Perform geometric consistency checks on the generated sub-cells, eliminate dangling nodes and non-manifold edges, and generate a topologically connected high-resolution fine-grained computational grid dataset.

[0180] Establish a bidirectional association between the fine computational grid dataset and the full-wave algorithm mapping index, and bind the physical attribute fields of the grid cells to the input interface of the full-wave solver.

[0181] By combining 3D mesh refinement with the full-wave algorithm, the region division results from the previous step are transformed into finely configured computational units with high-resolution geometric features, achieving the expected technical effect of high-precision solution of electromagnetic fields in highly complex regions.

[0182] For example, in a complex indoor environment with a frequency of 2.4 GHz, the computationally demanding area includes metal furniture and concrete walls. The minimum mesh size is set to λ / 10, approximately 12.5 mm. A three-level octree subdivision is performed on the metal edges with a radius of curvature less than 50 mm, generating a tetrahedral mesh with a side length of 6.25 mm. The calculated wavenumber k is approximately 50.27 rad / m, and the relative permittivity ε... r A value of 4.5 was used to verify that the mesh size met the phase error requirements. Finally, a fine dataset containing 500,000 mesh cells was generated and bound to the FDTD solver to ensure that the field strength prediction error was less than 0.5 dB.

[0183] S5.3: Based on the set of low computational demand regions in the initial algorithm matching list, perform a simplified computational dimension allocation operation. Use the ray tracing order reduction algorithm to sparsify the sub-region grids within the set of low computational demand regions to generate a simplified computational grid dataset with low-resolution geometric features. Then bind the simplified computational grid dataset with the empirical statistical model mapping index to generate a fast approximate computational configuration unit for low computational demand regions.

[0184] The algorithm first obtains a set of low-computation-demand regions from the initial matching list and analyzes the spatial topology and electromagnetic parameter distribution of each subdomain. For the 3D geometric mesh within these low-computation-demand regions, it performs a curvature threshold-based mesh simplification process, removing redundant mesh nodes with surface curvature change rates below a preset threshold, retaining key geometric feature points, and generating a preliminary sparse mesh skeleton. The preliminary sparse mesh skeleton undergoes connectivity testing to repair mesh breaks or non-manifold edges caused by node removal. The Laplace smoothing algorithm is then used to optimize the coordinate positions of the remaining nodes, ensuring the regularity of the mesh cell shape and outputting a simplified mesh model with complete geometric topology. Based on the simplified mesh model, a ray-tracing order reduction index is constructed, mapping the original high-density mesh to low-resolution voxel cells. Each voxel cell aggregates the average dielectric constant and conductivity of all original meshes within it, generating an equivalent dielectric parameter matrix. An empirical statistical model mapping mechanism is used to bind the equivalent dielectric parameter matrix to the spatial coordinates of the simplified mesh model, establishing a direct index relationship from spatial location to rapidly approximate calculation parameters. By combining the ray tracing reduction algorithm with an empirical statistical model, the low computational demand region from the previous step is transformed into a simplified computational grid dataset with low-resolution geometric features and corresponding fast approximate computational configuration units, thereby achieving the expected technical effect of reducing computational dimensionality and improving simulation efficiency.

[0185] For example, in a subdomain of 10m × 10m × 5m within a low computational demand region, the initial mesh size is 100,000. A curvature threshold of 0.05 is set, and 80,000 redundant nodes with curvature less than this value are removed, retaining 20,000 critical nodes. After 50 iterations of Laplacian smoothing, a simplified mesh model containing 5,000 tetrahedral elements is generated. The average dielectric constant of 3.5 and conductivity of 0.01 S / m from the original mesh are assigned to the corresponding voxel elements to construct an equivalent dielectric parameter matrix. The resulting fast approximate computational configuration reduces the ray tracing computation in this region by 90%, significantly improving the computational speed for overall electromagnetic distribution prediction.

[0186] S5.4: Perform boundary coupling parameter calculation processing on the high-precision computing configuration unit and the fast approximation computing configuration unit. Use a linear interpolation algorithm to extract the field strength gradient features of the high-precision computing configuration unit and the fast approximation computing configuration unit at the adjacent boundary to generate a smooth transition weight matrix to eliminate the dimension abrupt effect. Then, inject the smooth transition weight matrix into the high-precision computing configuration unit and the fast approximation computing configuration unit respectively to generate an optimized region computing configuration package containing a boundary smoothing mechanism.

[0187] Obtain the grid node coordinates and initial physical property parameters of the high-precision calculation configuration unit and the fast approximate calculation configuration unit at adjacent boundaries, and construct the boundary coupling data input set.

[0188] Perform spatial topology mapping on the boundary-coupled data input set, identify the set of shared mesh patches at the interface between high-computation-demand regions and low-computation-demand regions, and determine the index sequence of boundary nodes to be processed.

[0189] Based on the boundary node index sequence, the field intensity gradient vectors of the high-precision side grid nodes and the low-precision side grid nodes are extracted, denoted as follows: and .

[0190] A linear interpolation algorithm is used to construct the field intensity distribution model within the boundary transition zone, and a distance weighting coefficient is defined. The calculation formula is as follows: in, This is the Euclidean distance from the current node to the boundary of the high-precision region. This is the starting distance of the transition zone. This is the distance at which the transition zone ends.

[0191] Using distance weighting coefficients The field intensity gradients on both sides are weighted and fused to generate a smooth transition field intensity gradient. The calculation formula is: Based on smooth transition field intensity gradient The reverse derivation generates a smooth transition weight matrix used to correct the original computational configuration. .

[0192] Smooth transition weight matrix The boundary constraints of the high-precision calculation configuration unit and the fast approximate calculation configuration unit are injected respectively, and the iterative solution parameters of each configuration unit are updated.

[0193] By using the above linear interpolation and gradient fusion processing method, the discrete region configuration of the previous step is transformed into an optimized region calculation configuration package that includes a boundary smoothing mechanism, thereby achieving the expected technical effect of eliminating dimensional abrupt change effects and ensuring the continuity of electromagnetic fields.

[0194] For example, the transition zone width is set to 0.5 meters, and the distance between the boundary nodes of the high-precision area is... Distance between low-precision region boundary nodes For an intermediate node 0.2 meters from the high-precision boundary, the weight coefficient α is calculated as (0.2-0) / (0.5-0) = 0.4. If the high-precision side gradient... Low-precision lateral gradient (10,0,0) V / m²) If the gradient is (2,0,0)V / m², then the fused gradient... =0.4×(10,0,0)+0.6×(2,0,0)=(5.2,0,0)V / m². The weight matrix generated based on this gradient significantly reduces the field strength jump at the boundary, significantly improves the smoothness of the spliced ​​electromagnetic field distribution curve, and effectively eliminates numerical artifacts caused by differences in algorithm dimensionality.

[0195] S5.5: Based on the optimized regional computing configuration package including the boundary smoothing mechanism, the final configuration scheme integration operation is performed. The high-precision computing configuration unit and the fast approximation computing configuration unit in the optimized regional computing configuration package are encapsulated into independent computing threads using parallel task scheduling logic to generate differentiated computing configuration schemes that adapt to the characteristics of the entire spatial domain, thereby realizing the dynamic optimal allocation of electromagnetic computing resources in regions of different complexity.

[0196] S6: Based on the differentiated computing configuration scheme, electromagnetic distribution prediction calculations are performed in parallel, and linear interpolation is applied at the boundaries of adjacent regions to smooth the transition, thereby eliminating the discontinuity of field strength caused by abrupt changes in computing dimensions, thus generating continuous electromagnetic field distribution prediction data across the entire domain; such as Figure 3 As shown.

[0197] As an example, step S6 specifically includes the following sub-steps: S6.1: Based on the high computational demand area identifier in the differentiated computing configuration scheme, the full-wave algorithm solver is called to perform refined field strength solution on the three-dimensional model of the medium environment in the high computational demand area, so as to generate a high-precision local electromagnetic field distribution dataset with high spatial resolution.

[0198] This paper analyzes differentiated computing configuration schemes, extracts the identifiers of high-computation-demand regions and their corresponding fine-grained computing grid datasets and full-wave algorithm mapping indices. It loads dielectric constant, loss tangent, and conductivity distribution data of high-computation-demand regions from the 3D model of the medium environment to construct a local Maxwell's equations solution domain. Boundary conditions are initialized on the fine-grained computing grid dataset, and the smooth transition weight matrix from adjacent low-computation-demand regions is injected as a second type of boundary condition into the edge grid nodes of the high-demand region. The finite-difference time-domain method or the finite element method is used as the core solver for the full-wave algorithm, and the time step is set to satisfy the Courant stability condition to ensure numerical convergence.

[0199] At each time step, all grid cells in the region with high computational requirements are traversed, and the electric and magnetic field components are iteratively updated according to Maxwell's curl equation. The loss term caused by the conductivity of the medium is considered simultaneously to correct the field strength attenuation.

[0200] The above field strength update process is iteratively executed until the preset maximum simulation duration or field strength convergence threshold is reached. The electric field amplitude, phase, and power density data of each grid node at the final moment are extracted and assembled into a high spatial resolution three-dimensional field strength distribution matrix. Coordinate alignment processing is performed on the field strength distribution matrix to ensure consistency with the spatial index of the global grid system.

[0201] By employing a refined solution method based on the full-wave algorithm, the configuration parameters from the previous step are transformed into a high-precision local electromagnetic field distribution dataset with high spatial resolution, thereby enabling accurate reconstruction of the electromagnetic wave propagation characteristics within complex media regions.

[0202] For example, targeting high-computational-demand areas within urban building complexes, the mesh size is set to 0.05 meters, the dielectric constant to 4.5, and the conductivity to 0.01 S / m. The finite-difference time-domain method is employed, with a time step of 8.33 e. -11 The time step is set to satisfy the stability condition. After 5000 time step iterations, the electric field strength converges, and the output three-dimensional electric field distribution data with a resolution of 0.05 meters is obtained. The field strength prediction error is controlled within 2dB, which significantly improves the prediction accuracy under complex scattering environments.

[0203] S6.2: Based on the low-computation-demand area identifier in the differentiated computing configuration scheme, the empirical statistical model solver is called to quickly approximate the field strength of the three-dimensional model of the medium environment in the low-computation-demand area, so as to generate a low-precision local electromagnetic field distribution dataset with low computational overhead.

[0204] We analyze differentiated computing configuration schemes and extract the set of regions with low computing requirements and their associated simplified computing grid datasets and empirical statistical model mapping indexes.

[0205] Read the topology and electromagnetic properties of the grid cells in the simplified computational grid dataset, and identify the geometric center coordinates and adjacent connections of each grid cell.

[0206] Based on the empirical statistical model mapping index, a pre-trained ray tracing order reduction algorithm parameter library is loaded, and the field strength propagation attenuation coefficient and scattering phase correction factor are initialized.

[0207] For each source point in a region with low computational requirements, a sparse ray emission operation is performed to generate a main propagation path ray beam based on a preset angle sampling interval.

[0208] The propagation distance of the ray in the simplified grid is calculated. Combining the conductivity of the medium and the loss tangent, the field strength amplitude at the receiving point is calculated using an exponential decay and inverse distance proportional model.

[0209] Traverse all valid ray paths, accumulate the field intensity vectors generated by multipath effects, and apply phase interference correction terms to handle constructive and destructive interference.

[0210] The calculated discrete field strength values ​​are mapped back to simplified computational grid nodes, and the field strength distribution in areas not directly covered by the grid is filled by inverse distance weighted interpolation.

[0211] The local field strength data after filling is smoothed and filtered to suppress numerical oscillations caused by sparse sampling, thereby generating a low-precision local electromagnetic field distribution dataset.

[0212] By using an empirical statistical model solver to quickly approximate the field strength in regions with low computational requirements, the grid parameters from the previous step are transformed into a low-precision local electromagnetic field distribution dataset with low computational overhead, thus achieving efficient utilization of computing resources.

[0213] For example, an open plaza area in an urban microcell scenario is designated as a low-computational-demand region. This region has a 5-meter grid size and uses a hybrid air-concrete model as the medium. A ray emission interval of 15 degrees is set, emitting a total of 24 main rays. The medium attenuation constant α is set to 0.02 dB / m. For the receiving grid 100 meters from the source point, the initial field strength E0 is 1 V / m. The calculated single-path field strength amplitude is approximately 0.01 V / m. Considering multipath superposition, the final generated local electromagnetic field distribution dataset covers all grid nodes in the plaza area. The computation time is only 5% of that of the full-wave algorithm, and the field strength trend matches the measured data, significantly improving the overall simulation efficiency.

[0214] S6.3: Using the spatial coordinate indices of the high-precision local electromagnetic field distribution dataset and the low-precision local electromagnetic field distribution dataset, extract the set of boundary grid nodes at the intersection of adjacent regions, and read the initial field strength values ​​corresponding to the boundary grid node sets in the high-precision local electromagnetic field distribution dataset and the low-precision local electromagnetic field distribution dataset respectively, so as to generate a feature list of boundary nodes containing bilateral field strength information.

[0215] The system receives high-precision and low-precision local electromagnetic field distribution datasets and extracts their grid node index information in the spatial coordinate system. It then traverses the adjacent interfaces between high-computation-demand and low-computation-demand regions in the target space, identifying the common set of grid nodes located at these interfaces. Finally, it performs topological sorting on the identified boundary grid node sets and establishes a mapping table between node IDs and spatial coordinates.

[0216] Based on the mapping table, the complex electric field intensity value corresponding to each boundary node, including both real and imaginary amplitudes, is retrieved from the high-precision local electromagnetic field distribution dataset. Simultaneously, the complex electric field intensity value corresponding to the same boundary node is retrieved from the low-precision local electromagnetic field distribution dataset. The retrieved high-precision and low-precision side field intensity values ​​are paired to form bilateral field intensity data pairs. Each pair of field intensity data is appended with a node spatial location identifier and its corresponding region label to construct a structured data record. All structured data records of the boundary nodes are aggregated to generate a boundary node feature list containing bilateral field intensity information.

[0217] It should be noted that linear interpolation smooth transition processing refers to weighted correction of the field strength values ​​of each grid node at the boundary of adjacent regions, so that the field strength value changes continuously and gradually from the high-precision side to the low-precision side within the transition zone, rather than interpolating the grid geometric parameters or the electromagnetic parameters of the medium. Specifically, the fusion ratio of high-precision field strength values ​​and low-precision field strength values ​​is controlled by a spatial position weighting coefficient to achieve a smooth transition between field strength amplitude and phase.

[0218] By using the spatial index matching and data extraction processing methods described above, the discrete local field strength data generated in the previous step is transformed into a structured feature list with boundary correspondence, thereby providing accurate two-sided field strength input data for subsequent linear interpolation smooth transition and ensuring the continuity of the global electromagnetic field splicing.

[0219] For example, in a micro base station coverage scenario in a certain city, the full-wave algorithm results with a 0.5-meter grid precision are used in areas with high computational demand, while the empirical model results with a 5-meter grid precision are used in areas with low computational demand. 120 common boundary nodes are identified at the interface between the two. For the boundary point with node ID N45, its spatial coordinates are (x=102.5, y=200.3, z=15.0). The field strength at this point is read as E from the high-precision dataset. high=0.85+0.12jV / m, the field strength at this point is read as E from the low-precision dataset. low =0.82+0.15jV / m. (The coordinates, E...) high E low Each region label is encapsulated into a single record. The same operation is performed on all 120 nodes, generating a list of boundary node features containing 120 records. This list fully preserves the field strength difference information at the interface, providing a quantitative basis for eliminating field strength discontinuities caused by abrupt dimensional changes, and significantly improving the data accuracy of subsequent smoothing processes.

[0220] S6.4: Based on the bilateral field strength information in the feature list of boundary nodes, execute the linear interpolation smooth transition algorithm to calculate the weighted field strength correction value of the boundary grid node set in the transition zone, so as to generate a smooth transition field strength correction dataset to eliminate the field strength discontinuity caused by the abrupt change in computational dimension.

[0221] The high-precision and low-precision side field strength values ​​in the feature list of boundary nodes are obtained as the input data source for the linear interpolation smooth transition algorithm.

[0222] Determine the normalized distance coefficient of each grid node in the transition zone relative to the region boundary. The coefficient ranges from 0 to 1 and represents the relative position weight of the node in the transition zone.

[0223] A linear weighting function based on normalized distance coefficients is constructed to perform weighted fusion calculation of high-precision and low-precision side field strength values.

[0224] The smooth transition field strength correction value for any node within the transition zone is calculated using the following formula: in, This is the correction value for the smooth transition field strength at position x within the transition zone. For position Normalized distance weighting coefficients at the location, These are the boundary field strength values ​​in a high-precision local electromagnetic field distribution dataset. These are the boundary field strength values ​​in a low-precision local electromagnetic field distribution dataset.

[0225] Traverse all nodes in the boundary node set, calculate the corresponding normalized distance weight coefficient α(x) based on its spatial coordinates, and substitute it into the above formula to solve for the field strength correction value of each node.

[0226] The calculated field strength correction values ​​for all nodes are encapsulated in spatial index order to generate a smooth transition field strength correction dataset.

[0227] By using linear weighted interpolation, the bilateral field strength information from the previous step is transformed into a continuously and gradually changing smooth transition field strength correction dataset, thereby achieving the expected technical effect of eliminating the field strength discontinuity caused by abrupt changes in computational dimension.

[0228] For example, in a complex urban micro-cell environment, a 2-meter-long transition zone is selected at the boundary between a region with high computational demand (full-wave algorithm) and a region with low computational demand (ray tracing). This transition zone is discretized into 20 grid nodes with a step size of 0.1 meters. A high-precision lateral field strength E is set at the boundary. high The low-precision side field strength E is 10V / m. low The value is 8V / m. For a node 0.5 meters from the boundary of the high-precision region, its normalized distance weighting coefficient α(x) is calculated to be 0.75 (assuming the weight decreases linearly with distance). Substituting into the formula, we calculate: E(0.5) = 0.75 × 10 + (1 - 0.75) × 8 = 7.5 + 2.0 = 9.5V / m. The same calculation is performed on all 20 nodes, generating a smooth transition field strength correction dataset containing 20 corrected field strength values. This dataset ensures that the field strength change from 10V / m to 8V / m is spatially linear and continuous, effectively eliminating field strength jump artifacts caused by algorithm switching and significantly improving the physical consistency of the global electromagnetic field distribution prediction.

[0229] S6.5: Based on the smooth transition field strength correction dataset, the field strength numerical fusion processing is performed on the splicing edge of the high-precision local electromagnetic field distribution dataset and the low-precision local electromagnetic field distribution dataset, and the fused local data is integrated into the global grid structure of the target space to generate global continuous and artifact-free electromagnetic field distribution prediction data.

[0230] It receives a smooth transition field strength correction dataset, a high-precision local electromagnetic field distribution dataset, and a low-precision local electromagnetic field distribution dataset as input data sources.

[0231] Based on the spatial topology index of the global grid structure, all grid cells to be merged within the target space are traversed to identify the transition zone grid set located at the boundary between high computing demand region and low computing demand region.

[0232] For any grid node within the transition zone, extract its initial complex field strength values ​​from the high-precision local electromagnetic field distribution dataset and from the low-precision local electromagnetic field distribution dataset.

[0233] The weighted fusion coefficient corresponding to the grid node is read from the smooth transition field strength correction dataset. This coefficient represents the confidence weight of the transition from the high-dimensional full-wave solution to the low-dimensional statistical solution.

[0234] Using the same linear weighted fusion logic as S6.4, the weighted fusion coefficients in the smooth transition field strength correction dataset are applied to the complex field strength values ​​on the high-precision side and the low-precision side to calculate the fused field strength of each node in the transition zone.

[0235] A phase consistency check is performed on the calculated complex values ​​of the fused field strength to ensure that the continuity between the real and imaginary parts conforms to the boundary condition constraints of Maxwell's equations.

[0236] Write the verified fused field strength values ​​into the attribute fields of the corresponding nodes in the global grid structure, overwriting the original single-dimensional calculation results.

[0237] For non-transition zone regions with high computational demands, the field strength values ​​of high-precision local electromagnetic field distribution datasets are directly mapped to the global grid structure.

[0238] For non-transition zone regions with low computational requirements, the field strength values ​​of low-precision local electromagnetic field distribution datasets are directly mapped to the global grid structure.

[0239] Perform a full-domain grid data integrity scan to detect the presence of unassigned void nodes or singularities with abnormal values.

[0240] The detected abnormal nodes are repaired using the neighboring node average value filling method to ensure the topological integrity of the global electromagnetic field data.

[0241] By using the above-mentioned linear interpolation fusion and global mapping processing method, the locally differentiated field strength data generated in the previous step is transformed into global continuous electromagnetic field distribution prediction data without artifacts. This achieves seamless splicing of multi-scale calculation results in the spatial domain, eliminates the field strength discontinuity caused by abrupt changes in the calculation dimension, and significantly improves the overall accuracy and physical reality of complex electromagnetic environment prediction.

[0242] For example, in an urban micro-cell scene containing a mixture of concrete walls and free space, the target space is divided into a 100x100x50 three-dimensional mesh. The finite element method is used to calculate areas with high computational demand (near the walls), while the ray tracing method is used to calculate areas with low computational demand (open areas). Within a transition zone 2 meters from the wall surface, a typical mesh node is selected, with its high-precision field strength E... high The low-precision field strength E is 3.5 + 0.2jV / m. low The value is 3.2 + 0.1jV / m. Based on the smooth transition weight matrix generated in S6.4, the weighted fusion coefficient α for this node is 0.6. Calculation: E final =0.6×(3.5+0.2j)+(1-0.6)×(3.2+0.1j)=(2.1+0.12j)+(1.28+0.04j)=3.38+0.16jV / m, where Efinal This is the final field strength after fusion. This value is written into the global grid. After full-field verification, the rate of change of the field strength gradient at the splicing point decreased from 15% before processing to less than 2%, effectively eliminating reflection artifacts and ensuring the physical continuity of the electromagnetic distribution prediction data.

[0243] Secondly, this application provides an electromagnetic distribution prediction system based on an environmental medium model, including: a data acquisition module, a three-dimensional model construction module for the medium environment, a complexity calculation module, a region division module, a spatial adaptive electromagnetic calculation module, and an electromagnetic distribution prediction module; The data acquisition module is used to acquire geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space, and generate the original media dataset. The medium environment 3D model construction module performs precise medium electromagnetic parameter assignment operations based on the original medium dataset, maps the assigned parameters to the 3D geometric mesh of the target space, and constructs the medium environment 3D model. The complexity calculation module is used to extract the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and input the above features into the preset weighted summation logic to calculate and generate the total complexity score. The region division module is used to compare the total complexity score with a preset threshold. If the score is higher than the preset threshold, it is marked as a region with high computational demand; if the score is lower than or equal to the preset threshold, it is marked as a region with low computational demand. The spatial adaptive electromagnetic computing module, based on the partitioning results of the region partitioning module, executes a spatially adaptive electromagnetic computing dimensionality reduction strategy. It allocates simplified computing dimensions and adopts an empirical statistical model for regions with low computing demand, and allocates refined computing dimensions and adopts a full-wave algorithm for regions with high computing demand. The electromagnetic distribution prediction module performs electromagnetic distribution prediction calculations based on an electromagnetic computation dimensionality reduction strategy. It also applies linear interpolation to smooth the transition at the boundaries of adjacent regions to eliminate the discontinuity of field strength caused by abrupt changes in computational dimension, thereby generating continuous electromagnetic field distribution prediction data across the entire domain.

[0244] Thirdly, this application provides a terminal, including: At least one processor; Memory used to store at least one processor-executable instruction; In this embodiment, at least one processor is configured to execute instructions to implement the electromagnetic distribution prediction method based on the environmental medium model proposed in this application.

[0245] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0246] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the element or object preceding “comprising” or “including” encompasses the element or object listed following “comprising” or “including” and its equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0247] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting electromagnetic distribution based on an environmental medium model, characterized in that, include: Obtain geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space. The electromagnetic parameters include at least three of the following: dielectric constant, loss tangent, and conductivity, and generate an original media dataset. Based on the original medium dataset, perform precise assignment of medium electromagnetic parameters, map the assigned parameters to a three-dimensional geometric mesh in the target space, and construct a three-dimensional model of the medium environment. The geometric complexity features and electromagnetic wave propagation complexity features of each sub-region are extracted using the three-dimensional model of the medium environment, and the above features are input into a preset weighted summation logic to calculate and generate a total complexity score. The total complexity score is compared with a preset threshold. If the score is higher than the preset threshold, it is marked as a high computational demand region. If the score is lower than or equal to the preset threshold, it is marked as a low computational demand region, and the region division result is generated. Based on the region partitioning results with computational dimension requirement labels, a spatially adaptive electromagnetic computation dimensionality reduction strategy is executed. For regions with low computational requirements, simplified computational dimensions are allocated and empirical statistical models are adopted. For regions with high computational requirements, refined computational dimensions are allocated and full-wave algorithms are adopted to generate differentiated computational configuration schemes. Electromagnetic distribution prediction calculations are performed in parallel according to the differentiated calculation configuration scheme, and linear interpolation is applied at the boundary of adjacent regions to smooth the transition, so as to eliminate the discontinuity of field strength caused by abrupt changes in calculation dimension and generate continuous electromagnetic field distribution prediction data across the entire domain.

2. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The process of extracting the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and inputting these features into a preset weighted summation logic to calculate and generate a total complexity score, specifically includes: Based on the three-dimensional model of the medium environment, the surface curvature change rate and geometric structure discreteness of the scatterer in each sub-region of the target space are subjected to gridded scanning processing to generate a set of geometric complexity feature vectors characterizing the complexity of the spatial morphology. Using the geometric complexity feature vector set as a spatial constraint, and combining the frequency-related dielectric constant and conductivity distribution data in the three-dimensional model of the medium environment, a multi-path ray tracing attenuation simulation operation is performed to generate an electromagnetic wave propagation complexity feature matrix that reflects beam penetration loss and scattering intensity. Based on the statistical distribution characteristics of the electromagnetic wave propagation complexity feature matrix and the geometric complexity feature vector set, the entropy weight method is used to dynamically calculate the weight coefficients of each feature dimension in order to generate an adaptive weight configuration parameter set that adapts to the dynamic changes of the current environmental medium. Based on the adaptive weight configuration parameter group, a weighted fusion calculation operation is performed on the geometric complexity feature vector set and the electromagnetic wave propagation complexity feature matrix to generate an initial complexity evaluation value sequence that quantitatively represents the comprehensive computational difficulty of each sub-region. The initial complexity evaluation value sequence is processed by applying a nonlinear normalization mapping function to unify the dimensions, so as to eliminate the influence of differences in the magnitude of different features and finally generate a standardized list of total complexity scores.

3. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The method employs a spatially adaptive electromagnetic computation dimensionality reduction strategy based on the region partitioning results labeled with computational dimensionality requirements. This strategy allocates simplified computational dimensions and uses empirical statistical models to regions with low computational requirements, while allocating refined computational dimensions and using a full-wave algorithm to regions with high computational requirements, generating differentiated computational configuration schemes. Specifically, this includes: The region segmentation results with computational dimension requirement labels are parsed and processed to extract the set of regions with high computational requirement and the set of regions with low computational requirement. A full-wave algorithm mapping index is constructed based on the set of regions with high computational requirement, and an empirical statistical model mapping index is constructed based on the set of regions with low computational requirement, thereby generating an initial algorithm matching list containing the correspondence between region categories and candidate algorithms. Based on the set of high computational demand regions in the initial algorithm matching list, a fine computational dimension allocation operation is performed. The sub-region meshes within the set of high computational demand regions are densified using a 3D mesh refinement algorithm to generate a fine computational mesh dataset with high-resolution geometric features. The fine computational mesh dataset is then bound to the full-wave algorithm mapping index to generate a high-precision computational configuration unit for high computational demand regions. Based on the set of low computational demand regions in the initial algorithm matching list, a simplified computational dimension allocation operation is performed. The ray tracing order reduction algorithm is used to sparsify the sub-region grids within the set of low computational demand regions to generate a simplified computational grid dataset with low-resolution geometric features. The simplified computational grid dataset is then bound to the empirical statistical model mapping index to generate a fast approximate computational configuration unit for low computational demand regions. Boundary coupling parameter calculation processing is performed on the high-precision computing configuration unit and the fast approximation computing configuration unit. The field intensity gradient features of the high-precision computing configuration unit and the fast approximation computing configuration unit at the adjacent boundary are extracted by the linear interpolation algorithm to generate a smooth transition weight matrix for eliminating the dimension abrupt effect. The smooth transition weight matrix is ​​then injected into the high-precision computing configuration unit and the fast approximation computing configuration unit respectively, thereby generating an optimized region computing configuration package containing a boundary smoothing mechanism. The final configuration scheme integration operation is performed based on the optimized regional computing configuration package containing the boundary smoothing mechanism. Parallel task scheduling logic is used to encapsulate the high-precision computing configuration unit and the fast approximation computing configuration unit in the optimized regional computing configuration package into independent computing threads to generate differentiated computing configuration schemes adapted to the characteristics of the global space.

4. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The frequency-related electromagnetic parameters are obtained by filling the parameter gaps between discrete frequency points using a linear interpolation algorithm.

5. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The three-dimensional geometric mesh is generated using an adaptive mesh generation algorithm, and the three-dimensional geometric mesh is a non-uniform mesh.

6. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, In the three-dimensional model of the medium environment, the electromagnetic parameters at the boundaries of adjacent meshes of different medium types are corrected by a weighted average interpolation algorithm.

7. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The preset threshold is dynamically generated by an adaptive threshold calculation algorithm based on the extreme value characteristics and distribution density characteristics of the total complexity score.

8. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, In the region partitioning results, grid cells with the same computational dimension requirement label and spatially adjacent are merged into independent spatial subdomains.

9. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, In the region partitioning results, the interface topology of adjacent regions with different computational dimension requirement labels is optimized by morphological filtering.

10. The electromagnetic distribution prediction method based on an environmental medium model according to claim 1, characterized in that, The electromagnetic field distribution prediction data is generated by fusing the field strength at the boundary between high-computation-demand regions and low-computation-demand regions after weighted correction using a linear interpolation algorithm.

11. An electromagnetic distribution prediction system based on an environmental medium model, characterized in that, include: The module includes: data acquisition module, medium environment 3D model construction module, complexity calculation module, region division module, spatial adaptive electromagnetic calculation module, and electromagnetic distribution prediction module. The data acquisition module is used to acquire geometric structure data and frequency-related electromagnetic parameters of multiple types of media in the target space, and generate the original media dataset. The medium environment 3D model construction module performs precise medium electromagnetic parameter assignment operations based on the original medium dataset, maps the assigned parameters to the 3D geometric mesh of the target space, and constructs the medium environment 3D model. The complexity calculation module is used to extract the geometric complexity features and electromagnetic wave propagation complexity features of each sub-region using the three-dimensional model of the medium environment, and input the above features into the preset weighted summation logic to calculate and generate the total complexity score. The region division module is used to compare the total complexity score with a preset threshold. If the score is higher than the preset threshold, it is marked as a region with high computational demand; if the score is lower than or equal to the preset threshold, it is marked as a region with low computational demand. The spatial adaptive electromagnetic computing module, based on the partitioning results of the region partitioning module, executes a spatially adaptive electromagnetic computing dimensionality reduction strategy. It allocates simplified computing dimensions and adopts an empirical statistical model for regions with low computing demand, and allocates refined computing dimensions and adopts a full-wave algorithm for regions with high computing demand. The electromagnetic distribution prediction module performs electromagnetic distribution prediction calculations based on an electromagnetic computation dimensionality reduction strategy. It also applies linear interpolation to smooth the transition at the boundaries of adjacent regions to eliminate the discontinuity of field strength caused by abrupt changes in computational dimension, thereby generating continuous electromagnetic field distribution prediction data across the entire domain.

12. A terminal, characterized in that, include: At least one processor; Memory used to store at least one processor-executable instruction; In this embodiment, at least one processor is configured to execute instructions to implement the electromagnetic distribution prediction method based on the environmental medium model as described in any one of claims 1-10.

Images