An AI-based method and system for visualizing urban planning

Abstract

This invention discloses an artificial intelligence-based method and system for visualizing urban planning, belonging to the field of data processing technology. The method includes: acquiring multi-source heterogeneous urban basic data and planning schemes to be evaluated; constructing a heterogeneous attribute spatial map through temporal denoising and feature matching; using the planning scheme as a priori excitation condition, inputting it along with the map into a multi-layer spatiotemporal perception graph neural network model with topology preservation loss, and outputting a spatiotemporal evolution prediction matrix tensor; constructing an adaptive weighted grid evaluation model based on multi-dimensional constraints such as perspective, predicted polarization gradient, and hardware load, and combining it with PID closed-loop monitoring of dynamic threshold, to reconstruct a high-efficiency urban visualization 3D chassis model with weighted allocation in real time. This invention effectively solves the problems of false connectivity distortion in traditional topology prediction and computational power collapse caused by rigid LOD distance pruning, achieving deep collaboration between highly realistic situational evolution and adaptive computational power.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to an artificial intelligence-based method and system for visualizing urban planning. Background Technology

[0002] With the implementation of smart city construction and the theory of all-time-domain digital twin cities, decision-making systems such as urban municipal planning and intelligent road network engineering evaluation urgently need to simulate and output the chain effects caused by the implementation of hypothetical urban planning intervention schemes in a digital twin computing environment. Decision-makers hope to not only see static building models visually, but also accurately predict the dynamic and temporal effects of engineering interventions on the surrounding environment, microclimate, and historical traffic flow patterns throughout the city.

[0003] However, current mainstream 3D GIS platforms and urban planning visualization systems suffer from the following extremely serious technical deficiencies when dealing with multidimensional evolution prediction and rendering of massive digital assets: First, existing environmental and traffic flow evolution prediction models generally suffer from the problem of predictive feature illusion caused by topological blind spots. Conventional prediction algorithms based on time series (RNN / LSTM) or ordinary graph neural networks (GCN), when faced with the highly complex road network base with non-Euclidean geometry within cities, often focus only on numerical approximation errors in their loss functions, lacking hard constraints on the physical adjacency rules. When conducting long-term time series extrapolations spanning multiple future time steps, extreme drastic changes in the characteristics of certain local nodes can easily tear apart their high-dimensional feature manifolds, leading to false connections such as crossing rivers and walls, or experiencing huge traffic flow jumps without physical road connections. Such spurious predictions, due to the inability to guarantee physical topological invariants, will directly lead to major errors in urban planning decisions.

[0004] Secondly, existing 3D graphics engines are extremely rigid in resource allocation, making them prone to rendering power crashes in large-scale prediction and visualization scenarios. The adaptive Level of Detail (LOD) culling component in traditional 3D rendering engines has relied on hard-coded spatial logic for decades—that is, it only depends on the absolute distance between the rendering camera's viewpoint coordinates and the geometric distance to the surface of the 3D object to reduce the number of polygons in the model. However, in urban planning simulation scenarios, this logic has a fatal shortsightedness: when a core disaster or congestion polarization area caused by a planning event happens to be far from the viewing angle, traditional engines will cullate it into a low-resolution mosaic, preventing users from seeing important core simulation details; conversely, some static buildings that are very close to the camera but have no parameter changes and no observation value are forcibly occupied by triangular facets in the video memory. This flaw, where the allocation of rendering computing power is severely deviated from the core focus of human observation, makes the system not only unreasonable and useless in visual evaluation, but also instantly drains the GPU's device buffer memory pool resources after superimposing hundreds of thousands of dynamically polarized macroscopic evolution particle sets, causing system crashes or interactive lag.

[0005] Finally, traditional city-level multidimensional predictive data display pipelines are too simplistic and crude, typically relying solely on CPU-based access to two-dimensional thermal plane maps to represent pollution or pedestrian density. This approach not only severely slows down the CPU's throughput but also lacks spatial depth and a nuanced experience of microscopic fluid dynamics. For spatial road sections with extreme abrupt change thresholds, there is often a lack of targeted, independent visual guidance, resulting in an excessively high barrier to intuitive comprehension and extremely poor perception for assessors facing a screen full of cluttered information.

[0006] In summary, to overcome the data illusion caused by the aforementioned topological distortion in prediction and the computational power misalignment and lack of expressiveness caused by the crude distance-guided LOD, it is urgent to propose a new generation of comprehensive visualization methods and integrated systems that delve into the control of the underlying features of multidimensional graph networks and achieve high-frequency change evolution inference and the joint breakthrough of the rendering engine architecture reconstructed by cutting-edge computer graphics pipelines.

[0007] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide an artificial intelligence-based method and system for visualizing urban planning, which can solve the technical problems existing in the prior art, such as data illusion caused by predicted topological spatial distortion, and computing power misalignment and OOM crash caused by rigidity of conventional distance LOD allocation.

[0009] To achieve the above objectives, in a first aspect, the present invention provides an artificial intelligence-based method for visualizing urban planning, comprising the following steps: Step S1: Obtain multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area; Step S2: Perform multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous urban basic data, and construct a heterogeneous attribute spatial map of the target urban area through time-series frequency denoising and node attribute mapping calculation. Step S3: Extract the intervention feature vector from the urban planning scheme data, and input the heterogeneous attribute spatial map and the intervention feature vector into the spatiotemporal perception graph neural network model to generate multi-dimensional spatiotemporal evolution feature data; wherein, the spatiotemporal perception graph neural network model is trained through a multi-task joint loss function, the loss function including mean squared error loss, topological physical connectivity preservation loss and weight decay regularization term; Step S4: Based on the constraints of the predicted gradient and the device hardware load, construct an adaptive level of detail weighted mesh evaluation model to generate an efficient dynamic urban visualization 3D chassis model. The adaptive level of detail weighted mesh evaluation model takes the predicted gradient and the device hardware load as inputs, calculates the comprehensive rendering weight W(i) of each spatial mesh unit, and dynamically adjusts the geometric detail resolution granularity of the 3D chassis model based on the weight. Step S5: Perform multimodal visual attribute mapping and rendering on the multidimensional spatiotemporal evolution feature data in the 3D chassis model, and output a panoramic display interface of the urban planning scheme.

[0010] Optionally, in step S2, the specific processing procedure for time-series frequency-average noise reduction of multi-source heterogeneous data includes: The local constraint window-based dynamic time warping algorithm (CDTW) is adopted to constrain the maximum offset of the global search path within the preset local search bandwidth. The optimal time alignment path, including extreme value constraints, is solved by dynamic programming to find the optimal time alignment path for the multiple time-series matrices generated by the inconsistent sampling frequencies of each sensor. Based on the penalty path benchmark, the baseline synchronous stationary timing characteristics of multi-source hardware data are derived.

[0011] Optionally, in step S2, the step of constructing the heterogeneous attribute spatial map of the target urban area specifically includes: The three-dimensional spatial topology data is abstracted into undirected nodes V representing urban functional entities and geometric edges E representing spatial relationships, forming a basic directed physical graph structure G(V,E). Calculate the Euclidean spatial straight-line distance between adjacent nodes V and the Shannon information entropy distribution features of the clustering distribution density of multi-source POI interest points, and form spatial-semantic dual composite attribute connection weights for each edge E by Hadamard dot product fusion. Finally, the traffic flow and micro-meteorological feature time series segments extracted from the benchmark synchronous stationary time series features are mapped to the time-varying state matrices of the corresponding nodes, and the heterogeneous attribute space map with dynamic weight bias parameters is jointly established.

[0012] Optionally, in step S3, the backpropagation optimization process of the multi-layer spatiotemporal perception graph neural network model is based on the multi-task joint loss function L. total The joint loss function is constructed as follows: ; Among them, L topo L is the topological physical connectivity preservation loss feature term. MSE Calculate the loss for the mean square error between the output prediction matrix of the last layer of the spatiotemporal-aware graph neural network model and the true next-period distribution time-series label tensor; L reg The weight decay L2 norm structure penalty regularization term is used to limit overfitting of the graph network; λ1 and λ2 are both adaptive equilibrium decay hyperparameters that limit gradient polarization.

[0013] Optionally, in step S4, the formula for calculating the comprehensive rendering weight W(i) of each central spatial mesh cell i at the current moment using the adaptive level of detail weighted mesh evaluation model includes the following mathematical composite constraints: ; Where, d i f represents the Euclidean perspective distance from the center of the current logical camera viewpoint to the core of the 3D bounding box of the i-th spatial mesh unit in the rendered image; dist (·) represents the nonlinear inverse proportional decay mapping function; f vis (·) represents the cosine similarity angle θ between the bounding box normal orientation and the line-of-sight projection observation vector. i Determined peak clipping factor for the view frustum; g i The variable represents the absolute value of the discrete partial derivative gradient of the environmental drastic change with respect to the i-th spatial grid cell extracted from the multi-dimensional spatiotemporal evolution feature data prediction matrix after removing the baseline steady state; Δt represents the cumulative offset elapsed rate of the logical rendering time step relative to the first starting point in the front-end prediction and inference process; C VRAM This represents the penalty factor for the proportion of real-time global remaining video memory or buffer memory usage obtained from the underlying graphics API interface; α, β, and γ are the adaptive adjustment coefficients of the bias associated with each positive and negative weight.

[0014] Optionally, step S4 further includes: dynamically monitoring the system's video memory load through a closed-loop feedback mechanism to obtain the dynamic frame rate threshold T(t) for smooth playback at the current time t, specifically expressed by the discrete calculation equation as follows: ; Wherein, e(t) is the digital time error signal component obtained by subtracting the current rendering output's true frame rate from the pre-set safe anti-lag target smooth frame rate; K p K i K d These are the gain proportional gain, cumulative integral fault tolerance coefficient, and differential predictive hysteresis regulator derivative coefficients defined within the PID digital anti-shake regulator; T base For the underlying scene, a minimum threshold parameter is reserved for the underlying scene's static forced drawing; e(t) 1) Represents the error signal component from the previous moment; This represents the cumulative sum of errors from the initial time to the current time t.

[0015] Optionally, in step S5, multimodal visual attribute mapping and rendering is performed on the multidimensional spatiotemporal evolution feature data in the three-dimensional chassis model, specifically including: For the static scalar prediction component matrix features extracted and separated from the multimodal prediction tensor, the system transforms its three-dimensional spatial grid position projection into scalar heat source absorption and scattering cross section coefficient weights. The system then applies the ray stepping algorithm in the GPU shader to iteratively approximate the line-of-sight interaction surface to form illumination accumulation and occlusion, and renders a three-dimensional diffraction micro-meteorological thermal landscape with semi-transparent high-density voxel clouds. For the traffic micro-dynamic vector trajectory feature set extracted by tensor, its spatiotemporal momentum field is reduced in dimension and written into a low-impedance 3D spatial vector texture map. At the same time, a curl noise function is enabled in the graphics backend, and the vortex flow partial derivative acceleration calculation field is generated by inputting the vector parameters. The position matrix and survival life dissipation state function of the hundreds of thousands of independent geometric polygonal particle systems are efficiently solved and reset in the concurrent computing shader independent thread tree cluster. For surface entities of spatial nodes that deviate from the upper limit of the specified safe and healthy values ​​due to extreme fluctuations, the truncation normal difference parameter of their GBuffer depth buffer is read and analyzed. The sharp and abnormal overlapping occlusion contours are extracted by using the Sobel convolution operator. A layer of constrained and controlled non-constant low-frequency sinusoidal trigonometric function time-varying driving term code block is colored and compounded within this edge to generate a bright edge periodic warning strobe effect based on the current viewing depth.

[0016] A second aspect of the present invention also provides an artificial intelligence-based urban planning visualization system, wherein the system has a built-in multi-layer bus data architecture and communication sub-interface, the system is used to implement the above-mentioned artificial intelligence-based urban planning visualization method, and the system comprises the following modules: The data acquisition module is configured to acquire multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area; The feature fusion module, connected to the data acquisition module, is used to perform multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous urban basic data, and construct a heterogeneous attribute spatial map of the target urban area through time-series frequency denoising and node attribute mapping calculation. The spatiotemporal prediction module, connected to the feature fusion module, is configured to extract intervention feature vectors from the urban planning scheme data, and input the heterogeneous attribute spatial map and the intervention feature vectors into the spatiotemporal perception graph neural network model to generate multi-dimensional spatiotemporal evolution feature data; wherein, the spatiotemporal perception graph neural network model is trained through a multi-task joint loss function, the loss function including mean squared error loss, topological physical connectivity preservation loss and weight decay regularization term; The rendering optimization module is connected to the spatiotemporal prediction module. Based on the constraints of the prediction gradient and the device hardware load, it constructs an adaptive level of detail weighted mesh evaluation model. The adaptive level of detail weighted mesh evaluation model takes the prediction gradient and the device hardware load as input, calculates the comprehensive rendering weight W(i) of each spatial mesh unit, and dynamically adjusts the geometric detail resolution granularity of the 3D chassis model based on the weight to generate an efficient dynamic urban visualization 3D chassis model. The visualization output module, connected to the rendering optimization module, is used to perform multimodal visual attribute mapping rendering on the multidimensional spatiotemporal evolution feature data in the three-dimensional chassis model, and output a panoramic display interface of the urban planning scheme. A third aspect of the present invention also provides a computer control device, including a storage medium, a processor assembly, and a data bus communication interface array, wherein: The storage medium internally partitions and burns non-volatile readable flash memory structure blocks and dynamic high-speed cache storage areas to record the underlying operating firmware of the control device and the application function instruction sequence set containing the logic system; When the processor component is powered on and invokes the above-mentioned application function instruction sequence set, it specifically implements the sequence of process operation steps of the artificial intelligence-based urban planning visualization method as defined in any one of claims 1 to 7.

[0017] A fourth aspect of the present invention also provides a computer-readable carrier medium with a non-volatile structure, wherein a special binary readable computer program assembly stream code segment with complete encoding that can be read and called by a hardware architecture is stored in its magnetic chip area or flash memory chip group; when the program assembly stream code segment is read, parsed and executed by the processing logic chip of its host system, it realizes the working steps of the above-mentioned artificial intelligence-based urban planning visualization display method.

[0018] Due to the adoption of the above-mentioned combination of invention patent features, the beneficial effects of this invention are summarized as follows: In existing technologies, conventional time series prediction is prone to distortion when encountering long-wavelength extrapolation. However, this invention uses a constrained windowed CDTW algorithm to clean up spikes in non-co-frequency data. Combined with a multi-task joint loss function embedded in the backpropagation of a pre-trained spatiotemporal perception network, this ensures that when extracting high-order polynomial features, the projection of high-dimensional nonlinear features into low-dimensional space is strictly constrained within the actual physical road network probability connection framework. This effectively avoids the AI ​​illusion that violates the real road network interconnection and discontinuity during the process of drastic data polarization and evolution. Ultimately, it provides municipal decision-making agencies with a set of accurate predictive evolution data that is both highly realistic and has extremely reliable and meticulous time series.

[0019] This invention utilizes a comprehensive rendering weight equation W(i) comprising four sets of superimposed factors to calculate the partial derivative gradient g of AI-predicted mutations. i Polarization amplification, and its interaction with distance attenuation, view frustum angle, and remaining penalty C from the graphics card warning. VRAM By applying composite weighting and combining it with the independent PID closed-loop control equation for real-time anti-vibration system monitoring, a dynamic card release limit T(t) is derived. This allows the rendering module to proactively identify the geographical areas affected by the most severe traffic flow or pollution caused by the intervention of the planning scheme during the execution of a high-precision digital twin simulation of the entire city. This overcomes the vertex overflow and collapse caused by the digital chassis during the simulation and display of ultra-large maps, ensuring that the simulation display combining macro and micro aspects of the entire city is always kept at the most efficient and intuitive level of evaluation.

[0020] This invention constructs an asynchronous concurrent multimodal pipeline based on GPUs, transferring scalar features to a pipeline module with an iterative approximation technology module featuring a ray-stepping algorithm. This generates a semi-transparent 3D micro-meteorological volumetric cloud display of thermal pollution with absorption and attenuation. For meshes that locally exceed warning thresholds, the Sobel convolution function is used to superimpose periodic bright and dark strobe light by reading and extracting the GBuffer depth. This achieves not only a microscopic reconstruction of the disorder of high-density macroscopic particles and the fluidity of their overall trajectories without increasing the CPU load or lengthening the buffer, but also allows reviewers to accurately pinpoint the secondary crises and traffic flow trends of urban planning projects after implementation, even through the 3D smoke screen and luminous warnings. This improves the clarity of the system's interactive analysis and the fault tolerance of the evaluation scheme. Attached Figure Description

[0021] Figure 1 A flowchart illustrating an artificial intelligence-based urban planning visualization method according to one embodiment of the present invention; Figure 2 A schematic diagram of the LOD comprehensive adaptive weight evaluation truncation mechanism under PID closed-loop monitoring in one embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0024] It should be understood that in the various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0025] It should be understood that in this invention, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0026] It should be understood that in this invention, "multiple" refers to two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, "and / or B" can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "Contains A, B, and C", "Contains A, B, and C" means that all three A, B, and C are contained; "Contains A, B, or C" means that one of A, B, and C is contained; "Contains A, B, and / or C" means that any one, two, or three of A, B, and C are contained.

[0027] It should be understood that in this invention, "B corresponding to A", "B corresponding to A", "A and B correspond", or "B and A correspond" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information. Matching A and B is defined as a similarity between A and B that is greater than or equal to a preset threshold.

[0028] Depending on the context, "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection."

[0029] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0030] like Figure 1 As shown, the artificial intelligence-based urban planning visualization method according to a preferred embodiment of the present invention includes the following steps: S1, acquire multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area; wherein, the multi-source heterogeneous urban basic data includes three-dimensional spatial topology data, historical traffic trajectory data and environmental micro-meteorological data; S2 involves performing multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous urban basic data. Through time-series frequency denoising and node attribute mapping calculation, a heterogeneous attribute spatial map of the target urban area containing static topological relationships and dynamic time-series feature dimensions is constructed. Steps S1 and S2 eliminate signal noise pollution caused by heterogeneous sampling of underlying sensors and achieve accurate frequency reduction and anchoring of multimodal data in multiple time dimensions.

[0031] S3, extract the intervention feature vector from the urban planning scheme data, and input the heterogeneous attribute spatial map and the intervention feature vector into the spatiotemporal perception graph neural network model to generate multi-dimensional spatiotemporal evolution feature data; wherein, the spatiotemporal perception graph neural network model is trained through a multi-task joint loss function, the loss function including mean squared error loss, topological physical connectivity preservation loss and weight decay regularization term; S4. Based on the constraints of the predicted gradient and the device hardware load, an adaptive level of detail weighted mesh evaluation model is constructed to generate an efficient dynamic urban visualization 3D chassis model. The adaptive level of detail weighted mesh evaluation model takes the predicted gradient and the device hardware load as inputs, calculates the comprehensive rendering weight W(i) of each spatial mesh unit, and dynamically adjusts the geometric detail resolution granularity of the 3D chassis model based on the weight. S5. In the reconstructed and generated efficient dynamic urban visualization 3D chassis model, a multi-path graphics buffer distribution rendering strategy is adopted. The depth buffer computing capability and spatial mapping shading pipeline of the computer graphics processing hardware GPU are used to perform asynchronous concurrent multimodal visual attribute mapping rendering on different types of prediction tensors in the multi-dimensional spatiotemporal evolution feature data. Finally, in the screen space post-processing stage, the panoramic urban planning scheme display interface is output according to the mixed transparency combination.

[0032] In one embodiment, the specific processing steps for time-series frequency-average noise reduction of multi-source heterogeneous data in step S2 include: A dynamic time warping algorithm (CDTW) based on local constraint band window (Sakoe-Chiba Band) is adopted to constrain the maximum offset of the global search path within a preset local search bandwidth. The optimal time alignment path, including extreme value constraints, is solved by dynamic programming for the multi-segment time matrix generated by the inconsistent sampling frequency of each sensor. Based on this penalized path benchmark, the benchmark synchronous stationary time series characteristics of multi-source hardware data are obtained.

[0033] In one embodiment, step S2, which involves constructing a heterogeneous attribute spatial map of the target urban area, specifically includes: The three-dimensional spatial topology data is abstracted into undirected nodes V representing urban functional entities and geometric edges E representing spatial relationships, forming a basic directed physical graph structure G(V,E). Calculate the Euclidean spatial straight-line distance between adjacent nodes V and the Shannon information entropy distribution features of the clustering distribution density of multi-source POI interest points, and form spatial-semantic dual composite attribute connection weights for each edge E by Hadamard dot product fusion. Finally, the traffic flow and micro-meteorological feature time series segments extracted from the stationary time series features are mapped to the time-varying state matrix of the corresponding nodes, and the heterogeneous attribute space map with dynamic weight bias parameters is jointly established.

[0034] In one embodiment, in step S3, the backpropagation optimization process of the pre-trained multi-layer spatiotemporal awareness graph neural network model is based on the multi-task joint loss function L. total The joint loss function is constructed as follows: ; Among them, L topo L is the topological physical connectivity preservation loss feature term. MSE Calculate the loss for the mean square error between the output prediction matrix of the last layer of the spatiotemporal-aware graph neural network model and the true next-period distribution time-series label tensor; L regThe weight decay L2 norm structure penalty regularization term is used to limit overfitting of the graph network; λ1 and λ2 are both adaptive equilibrium decay hyperparameters that limit gradient polarization.

[0035] In particular, L topo To preserve the loss feature term of physical connectivity of the topology, it uses the KL divergence mathematical formula to rigorously measure the prior difference between the high-dimensional nonlinear feature manifold of the graph node prediction output after mapping to the low-dimensional potential representation space and the adjacency probability distribution formed by the actual physical road network topology constraints. During backpropagation, it blocks and avoids the false connectivity of non-adjacent network node spatial illusions caused by drastic numerical evolution, and cuts off the spatiotemporal prediction artifact association that is easily triggered by conventional graph neural networks when subjected to drastic data changes.

[0036] In one embodiment, in step S4, the formula for calculating the comprehensive rendering weight W(i) of each central spatial mesh cell i at the current moment by the adaptive level of detail weighted mesh evaluation model includes the following mathematical composite constraints: ; Where, d i f represents the Euclidean perspective distance from the center of the current logical camera viewpoint to the core of the 3D bounding box of the i-th spatial mesh unit in the rendered image; dist (·) represents the nonlinear inverse proportional decay mapping function; f vis (·) represents the cosine similarity angle θ between the bounding box normal orientation and the line-of-sight projection observation vector. i Determined peak clipping factor for the view frustum; g i The variable represents the absolute value of the discrete partial derivative gradient of the environmental drastic change with respect to the i-th spatial grid cell extracted from the multi-dimensional spatiotemporal evolution feature data prediction matrix after removing the baseline steady state; Δt represents the cumulative offset elapsed rate of the logical rendering time step relative to the first starting point in the front-end prediction and inference process; C VRAM This represents the penalty factor for the proportion of real-time global remaining video memory or buffer memory usage obtained from the underlying graphics API interface; α, β, and γ are the adaptive adjustment coefficients of the bias associated with each positive and negative weight.

[0037] In one embodiment, in step S4, the system memory load is dynamically monitored through a closed-loop feedback mechanism to obtain the dynamic frame rate threshold T(t) for smooth playback at the current time t. The discrete calculation equation is specifically expressed as follows: ; Wherein, e(t) is the digital time error signal component obtained by subtracting the current rendering output's true frame rate from the pre-set safe anti-lag target smooth frame rate; K p K i K dThese are the gain proportional gain, cumulative integral fault tolerance coefficient, and differential predictive hysteresis regulator derivative coefficients defined within the PID digital anti-shake regulator; T base For the underlying scene, a minimum threshold parameter is reserved for the underlying scene's static forced drawing; e(t) 1) Represents the error signal component from the previous moment; This represents the cumulative sum of errors from the initial time to the current time t.

[0038] Specifically, such as Figure 2 As shown, the closed-loop control integral PID structure equation is used to monitor the actual FPS of the current output and subtract the error frame e(t) to obtain the most reasonable dynamic frame skipping threshold T(t) for smooth gameplay. At this time, the dramatically changed plot entity, which is activated and has its comprehensive weight increased, finally successfully enters the "Yes" branch and passes the detection comparison logic that is higher than the frame skipping threshold. Through the combined interference of the composite calculation framework and the underlying PID integral anti-shake detection, the underlying graphics display engine is endowed with the vision awareness of "intelligently focusing on the evolution core".

[0039] In one embodiment, the steps involved in S5 of asynchronous concurrent multimodal visual attribute mapping rendering using a multi-path graphics buffer distribution rendering strategy specifically include: Static environmental parameters are extracted from the multimodal spatiotemporal data tensor and a three-dimensional scalar field is generated through spatial interpolation algorithms. The static environmental parameters include PM2.5 concentration, noise level in decibels, and temperature and humidity. A ray stepping algorithm is implemented based on the CUDA parallel architecture. With the ray step size Δs=0.1m and the maximum number of iterations N=100 as constraints, the interaction integral between the ray and the atmospheric scattering medium is calculated to generate a thermal voxel cloud with an attenuation coefficient. A semi-transparent blending mode and voxelized LOD technology are used to achieve dynamic visualization of micro-meteorological thermal maps, with a voxel resolution ≥5cm³ and a rendering frame rate ≥30FPS. Spatiotemporal alignment and trajectory clustering were performed on traffic trajectory data to extract velocity field vector features. A low-impedance 3D vector texture (VTF) was constructed using the Houdini engine, and a modified Perlin noise function was used to generate a vortex flow field with a frequency f = 0.1 Hz and an amplitude A = 0.5 m / s². The NVIDIA Flex particle system was implemented in the GPU Compute Shader, with particle lifetimes T = 300 s and mass decay coefficient β = 0.01 s. -1 It supports real-time collision detection of millions of particles; Depth gradient calculation based on depth buffer: The Sobel operator kernel is set to [[-1,0,1],[-2,0,2],[-1,0,1]] to extract abnormal regions where the normal deviation Δn > 0.1 on the building surface. A low-frequency sinusoidal drive signal (frequency f = 2Hz, phase offset φ = π / 4) is generated using the finite-difference time-domain (FDTD) method. The periodic flickering of the edge halo is achieved through Shader Graph, with a duty cycle D = 50% and a brightness modulation range L = [0.8, 1.2]. When the user's eye tracking depth Z_eye < 50m, the transparency of the warning effect is gradually changed, α = 0.3 → 1.0, and a spatial audio alarm is played simultaneously, with a frequency range of 200-800Hz.

[0040] It should be noted that, in implementation, a concurrency mechanism for deep GPU scheduling is utilized to transfer scalar features to a pipeline with an iterative approximation technology module featuring a ray stepping algorithm, generating a semi-transparent 3D thermal pollution micro-meteorological volumetric cloud display with absorption and attenuation. Simultaneously, the travel vector tensor is extracted, and a low-cost 3D spatial vector texture map is passed to a concurrent independent computer thread shader. Combined with curled noise, this generates extremely smooth, fluid-like light trails to indicate the direction of movement. If traffic pollution is detected crossing a line, the occlusion depth of the GBuffer is measured to read the phase difference parameter of the cross-section, and edge Sobel convolution is used to delineate and calculate the outer triangular luminous contour function, stimulating a flicker effect to enhance visual alarm management performance. By connecting a pure GPU multimodal channel in parallel, the system can accommodate the parallel microscopic reconstruction simulation of millions of disordered dynamic trajectory particles with low frame loss under power-saving conditions without significant power consumption increase, thereby expanding the texture of the three-dimensional flow field of the city; it also enables review users to instantly penetrate massive amounts of redundant information and get straight to the point through the strobe effect with a strong visual thickness, greatly optimizing the accuracy of sensory analysis and the fault tolerance space of urban integrated governance decision-making.

[0041] In one embodiment, an artificial intelligence-based urban planning visualization system is provided. The system has a built-in multi-layer bus data architecture and communication sub-interfaces. The system is used to implement the aforementioned artificial intelligence-based urban planning visualization method and includes, but is not limited to, the following components: The data acquisition module is configured to acquire multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area. The multi-source heterogeneous data includes three-dimensional spatial topology data, historical traffic trajectory data and environmental micro-meteorological data. The feature fusion module, connected to the data acquisition module, is used to perform multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous data. Through time-series frequency denoising and node attribute mapping calculation, a heterogeneous attribute spatial map of the target urban area containing static topological relationships and dynamic time-series feature dimensions is constructed. The spatiotemporal prediction module, connected to the feature fusion module, is configured to extract intervention feature vectors from the urban planning scheme data and input them as prior stimulus conditions into a pre-trained spatiotemporal perception graph neural network model. The model includes a long and short time series global attention mechanism, which is used to calculate and output the spatiotemporal evolution prediction matrix of each graph node under multiple future projection time steps, and generate multi-dimensional spatiotemporal evolution feature data. The rendering optimization module is connected to the spatiotemporal prediction module. Based on the constraints of the prediction gradient and the device hardware load, it constructs an adaptive level of detail (LOD) weighted mesh evaluation model. Through a closed-loop feedback mechanism, it dynamically monitors the system memory load and adjusts the geometric detail resolution granularity of the spatial mesh unit in real time to generate an efficient dynamic urban visualization 3D chassis model. The visualization output module, connected to the rendering optimization module, performs multimodal visual attribute mapping rendering on the multi-dimensional spatiotemporal evolution feature data in the three-dimensional chassis model, and outputs a panoramic display interface of the urban planning scheme.

[0042] In one embodiment, a computer control device is provided, including a storage medium, a processor component, and a data bus communication interface array, wherein: The storage medium internally partitions and burns non-volatile readable flash memory structure blocks and dynamic high-speed cache storage areas to record the underlying operating firmware of the control device and the application function instruction sequence set containing the logic system; When the processor component is powered on and invokes the above-mentioned application function instruction sequence set, it specifically implements the sequence of process operation steps of the artificial intelligence-based urban planning visualization method as defined in any one of claims 1 to 7.

[0043] In one embodiment, a computer-readable carrier medium with a non-volatile structure is provided, wherein a special binary readable computer program assembly stream code segment is stored in its magnetic area or flash memory chip group and can be read and called by the hardware architecture; when the program assembly stream code segment is read, parsed and executed by the processing logic chip of its host system, it implements the working steps of the urban planning visualization display method based on artificial intelligence as described above.

[0044] In summary, the artificial intelligence-based urban planning visualization method and system proposed in this invention overturns the technical barriers of traditional 3D urban dashboards, which "only display static surfaces and cannot deeply integrate rigorous deductive logic and anti-collapse underlying computational control." It uses a CDTW-driven correction and cleaning mechanism and heterogeneous spatial maps as strong physical input surfaces; a spatiotemporal perception multi-layer graph neural network with an internally mounted KL divergence cross-boundary penalty feedback structure forms the deductive prediction axis, connected to a bottom-level PID calculus closed-loop computation mechanism that issues thresholds in real time, and an LOD adaptive reconstruction, elimination, and interception system that asynchronously sends independent parameters to the high-speed pipeline of the hardware and software GPU. This not only enables the generated city / provincial-level macro-twin deductive models to achieve high-fidelity prediction, but also reshapes the originally rigid traditional computer rendering engine into an intelligent entity with millisecond-level anti-jitter convergence capabilities. It guarantees absolutely safe, smooth, and lag-free operation, forging a robust underlying digital assessment weapon for comprehensive urban planning prediction and construction intervention sand table decision-making actions with practical resilience.

[0045] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0046] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0048] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0049] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for visualizing urban planning based on artificial intelligence, characterized in that, The method includes the following steps: Step S1: Obtain multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area; Step S2: Perform multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous urban basic data, and construct a heterogeneous attribute spatial map of the target urban area through time-series frequency denoising and node attribute mapping calculation. Step S3: Extract the intervention feature vector from the urban planning scheme data, and input the heterogeneous attribute spatial map and the intervention feature vector into the spatiotemporal perception graph neural network model to generate multi-dimensional spatiotemporal evolution feature data; wherein, the spatiotemporal perception graph neural network model is trained through a multi-task joint loss function, the loss function including mean squared error loss, topological physical connectivity preservation loss and weight decay regularization term; Step S4: Based on the constraints of the predicted gradient and the device hardware load, construct an adaptive level of detail weighted mesh evaluation model to generate an efficient dynamic urban visualization 3D chassis model. The adaptive level of detail weighted mesh evaluation model takes the predicted gradient and the device hardware load as inputs, calculates the comprehensive rendering weight W(i) of each spatial mesh unit, and dynamically adjusts the geometric detail resolution granularity of the 3D chassis model based on the weight. Step S5: Perform multimodal visual attribute mapping and rendering on the multidimensional spatiotemporal evolution feature data in the 3D chassis model, and output a panoramic display interface of the urban planning scheme.

2. The urban planning visualization method based on artificial intelligence according to claim 1, characterized in that, In step S2, the step of performing time-series frequency-average noise reduction on multi-source heterogeneous data includes: A dynamic time warping algorithm based on local constraints with a window is adopted to constrain the maximum offset of the global search path within a preset local search bandwidth. The optimal time alignment path, including extreme value constraints, is solved by dynamic programming to find the optimal time alignment path for the multiple time-series matrices generated by the inconsistent sampling frequencies of each sensor. Based on the penalty path benchmark, the baseline synchronous stationary timing characteristics of multi-source hardware data are derived.

3. The artificial intelligence-based urban planning visualization method according to claim 2, characterized in that, In step S2, the specific steps for constructing the heterogeneous attribute spatial map of the target urban area include: The three-dimensional spatial topology data of the multi-source heterogeneous urban basic data is abstracted into undirected nodes V representing urban functional entities and geometric edges E representing spatial relationships, forming a basic directed physical graph structure G(V,E). Calculate the Euclidean spatial straight-line distance between adjacent nodes V and the Shannon information entropy distribution features of the clustering distribution density of multi-source POI interest points, and form spatial-semantic dual composite attribute connection weights for each edge E by Hadamard dot product fusion. Finally, the traffic flow and environmental micro-meteorological feature time series segments extracted from the baseline synchronous stationary time series features are mapped to the time-varying state matrices of the corresponding nodes, and a heterogeneous attribute space map with dynamic weight bias parameters is jointly established.

4. The urban planning visualization method based on artificial intelligence according to claim 1, characterized in that, In step S3, the backpropagation optimization process of the multi-layer spatiotemporal perception graph neural network model is based on the multi-task joint loss function L. total The joint loss function is constructed as follows: ； Among them, L topo L is the topological physical connectivity preservation loss feature term. MSE Calculate the loss for the mean square error between the output prediction matrix of the last layer of the spatiotemporal-aware graph neural network model and the true next-period distribution time-series label tensor; L reg The weight decay L2 norm structure penalty regularization term is used to limit overfitting of the graph network; λ1 and λ2 are both adaptive equilibrium decay hyperparameters that limit gradient polarization.

5. The artificial intelligence-based urban planning visualization method according to claim 1, characterized in that, In step S4, the formula for calculating the comprehensive rendering weight W(i) of each central spatial mesh cell i at the current moment in the adaptive level of detail weighted mesh evaluation model includes the following mathematical composite constraints: ； Where, d i f represents the Euclidean perspective distance from the center of the current logical camera viewpoint to the core of the 3D bounding box of the i-th spatial mesh unit in the rendered image; dist (·) represents the nonlinear inverse proportional decay mapping function; f vis (·) represents the cosine similarity angle θ between the bounding box normal orientation and the line-of-sight projection observation vector. i Determined peak clipping factor for the view frustum; g i The variable represents the absolute value of the discrete partial derivative gradient of the environmental drastic change with respect to the i-th spatial grid cell extracted from the multi-dimensional spatiotemporal evolution feature data prediction matrix after removing the baseline steady state; Δt represents the cumulative offset elapsed rate of the logical rendering time step relative to the first starting point in the front-end prediction and inference process; C VRAM This represents the penalty factor for the proportion of real-time global remaining video memory or buffer memory usage obtained from the underlying graphics API interface; α, β, and γ are the adaptive adjustment coefficients of the bias associated with each positive and negative weight.

6. The artificial intelligence-based urban planning visualization method according to claim 5, characterized in that, Step S4 also includes: dynamically monitoring the system's video memory load through a closed-loop feedback mechanism to obtain the dynamic frame rate threshold T(t) for smooth playback at the current time t. The discrete calculation equation is specifically expressed as: ； Wherein, e(t) is the digital time error signal component obtained by subtracting the current rendering output's true frame rate from the pre-set safe anti-lag target smooth frame rate; K p K i K d These are the gain proportional gain, cumulative integral fault tolerance coefficient, and differential predictive hysteresis regulator derivative coefficients defined within the PID digital anti-shake regulator; T base The underlying scene is silently reserved for the underlying hardware threshold parameters; e(t-1) represents the error signal component of the previous time step; This represents the cumulative sum of errors from the initial time to the current time t.

7. The urban planning visualization method based on artificial intelligence according to claim 1, characterized in that, In step S5, multimodal visual attribute mapping and rendering are performed on the multidimensional spatiotemporal evolution feature data in the 3D chassis model, specifically including: For the static scalar prediction component matrix features extracted and separated from the multimodal prediction tensor, the system transforms its three-dimensional spatial grid position projection into scalar heat source absorption and scattering cross section coefficient weights. The system then applies the ray stepping algorithm in the GPU shader to iteratively approximate the line-of-sight interaction surface to form illumination accumulation and occlusion, and renders a three-dimensional diffraction micro-meteorological thermal landscape with semi-transparent high-density voxel clouds. For the traffic micro-dynamic vector trajectory feature set extracted by tensor, its spatiotemporal momentum field is reduced in dimension and written into a low-impedance 3D spatial vector texture map. At the same time, a curl noise function is enabled in the graphics backend, and the vortex flow partial derivative acceleration calculation field is generated by inputting the vector parameters. The position matrix and survival life dissipation state function of the hundreds of thousands of independent geometric polygonal particle systems are efficiently solved and reset in the concurrent computing shader independent thread tree cluster. For surface entities of spatial nodes that deviate from the upper limit of the specified safe and healthy values ​​due to extreme fluctuations, the truncation normal difference parameter of their GBuffer depth buffer is read and analyzed. The sharp and abnormal overlapping occlusion contours are extracted by using the Sobel convolution operator. A layer of constrained and controlled non-constant low-frequency sinusoidal trigonometric function time-varying driving term code block is colored and compounded within this edge to generate a bright edge periodic warning strobe effect based on the current viewing depth.

8. An artificial intelligence-based urban planning visualization system, characterized in that, The system has a built-in multi-layer bus data architecture and communication sub-interfaces. The system is used to implement the AI-based urban planning visualization method as described in any one of claims 1-7. The system comprises the following modules: The data acquisition module is configured to acquire multi-source heterogeneous urban basic data and urban planning scheme data to be evaluated for the target urban area; The feature fusion module, connected to the data acquisition module, is used to perform multi-dimensional feature alignment and fusion processing on the multi-source heterogeneous urban basic data, and construct a heterogeneous attribute spatial map of the target urban area through time-series frequency denoising and node attribute mapping calculation. The spatiotemporal prediction module, connected to the feature fusion module, is configured to extract intervention feature vectors from the urban planning scheme data, and input the heterogeneous attribute spatial map and the intervention feature vectors into the spatiotemporal perception graph neural network model to generate multi-dimensional spatiotemporal evolution feature data; wherein, the spatiotemporal perception graph neural network model is trained through a multi-task joint loss function, the loss function including mean squared error loss, topological physical connectivity preservation loss and weight decay regularization term; The rendering optimization module is connected to the spatiotemporal prediction module. Based on the constraints of the prediction gradient and the device hardware load, it constructs an adaptive level of detail weighted mesh evaluation model. The adaptive level of detail weighted mesh evaluation model takes the prediction gradient and the device hardware load as input, calculates the comprehensive rendering weight W(i) of each spatial mesh unit, and dynamically adjusts the geometric detail resolution granularity of the 3D chassis model based on the weight to generate an efficient dynamic urban visualization 3D chassis model. The visualization output module, connected to the rendering optimization module, is used to perform multimodal visual attribute mapping rendering on the multidimensional spatiotemporal evolution feature data in the three-dimensional chassis model, and output a panoramic display interface of the urban planning scheme.

9. A computer control device, characterized in that, This includes storage media, processor components, and a data bus communication interface array, among which: The storage medium internally partitions and burns non-volatile readable flash memory structure blocks and dynamic high-speed cache storage areas to record the underlying operating firmware of the control device and the application function instruction sequence set containing the logic system; When the processor component is powered on and invokes the above-mentioned application function instruction sequence set, it specifically implements the sequence of process operation steps of the artificial intelligence-based urban planning visualization method as defined in any one of claims 1 to 7.

10. A computer-readable carrier medium with a non-volatile structure, characterized in that: Its magnetic chip area or flash memory chip group stores a binary readable computer program assembly stream code segment that is fully encoded and can be read and called by the hardware architecture; when the program assembly stream code segment is read, parsed and executed by the processing logic chip of its host system, it implements the working steps of the artificial intelligence-based urban planning visualization display method covered by any one of the claims 1 to 7.

Images