Automatic generation method of urban spatial structure based on parameter iteration and machine learning

Abstract

The application discloses a method for automatically generating urban space structure based on parameter iteration and machine learning, and comprises the following steps: S1, loading basic information; S2, generating urban green corridors; S3, constructing a case library; S4, generating urban development corridors and urban centers; S5, generating urban key blocks; S6, generating urban general blocks; and S7, generating an urban space structure space graph. The method for automatically generating urban space structure identifies urban green corridors based on a parameter model, generates urban development corridors and urban centers through machine learning, further extracts and divides urban key development blocks and general blocks, and integrates to form a basic urban space structure. The automatically generating method can directly generate urban space structure planning drawings in national territory space planning, urban design compilation and urban special planning, and can be directly applied to urban and rural planning construction and management.

Description

Technical Field

[0001] This invention relates to the field of urban planning, specifically to an automatic method for generating urban spatial structures based on parameter iteration and machine learning. Background Technology

[0002] With the continuous advancement of machine learning technology in urban design, its application has provided significant assistance in urban spatial site selection, model scheme comparison, spatial data mining, pattern recognition, data classification, feature selection, and extraction, representing an important direction for the future development of urban planning. For urban design at different scales, extracting urban spatial structure rules and clarifying the basic spatial structure of the target city are crucial prerequisites. Since urban spatial structure is a diverse combination of natural, material, social, and economic elements, once determined, it plays a vital guiding role in the urban design process. Therefore, mapping the urban spatial structure is a key task for urban and rural planners in the early stages.

[0003] Existing methods for automatically generating urban spatial structures fall into two categories. One is based on traditional methods, relying on intuitive experience to determine the primary and secondary corridors and their corresponding spatial structures. This method lacks reliable objective data support. The other is based on multi-source big data, coupling data from mobile phone signaling, population flow, Points of Interest (POIs), and urban land use information. It generates kernel density images and uses statistical clustering methods to identify urban spatial structures. This method dynamically identifies urban spatial structures from the perspective of material and information flow, but data acquisition costs are high, and experimental subjects are mostly urban clusters or megacities. Its results serve quantitative research on urban space, with limited impact on morphological design in land spatial planning, urban design, and urban special planning. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic urban spatial structure generation method based on parameter iteration and machine learning. By relying on planning discipline theory, the method identifies primary green corridors in the city based on a parameter model, and then generates urban development corridors and city centers through machine learning. Furthermore, it extracts and divides key development areas and general areas of the city, and integrates them to form the basic urban spatial structure. At the same time, the automatic urban spatial structure generation method can directly generate urban spatial structure planning drawings in land spatial planning, urban design preparation, and urban special planning, and can be directly applied to urban and rural planning, construction and management.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] An automatic generation method for urban spatial structure based on parameter iteration and machine learning, the automatic generation method includes the following steps:

[0007] S1: Loading Basic Information

[0008] Based on the open data platform and government portal website, land use information data files of the target area are obtained, including urban administrative boundary data, road grade data, urban land use data, and urban DEM data. The data is cleaned and vectorized through the geographic information platform to obtain the urban land use information database C1.

[0009] S2: Generate urban green corridors

[0010] By extracting mountain and water system reservoirs from remote sensing images, a minimum resistance model is constructed by selecting indicators. Using the mountain reservoir as the source, a cost path tool is used to simulate ecological corridors, and the combined calculation results are used as urban green corridors.

[0011] S3: Case Library Construction

[0012] By collecting urban spatial structure cases through open-source websites, spatial structures are labeled on a geographic information platform based on urban road and boundary information, including primary and secondary development corridors and primary and secondary centers, forming a case study library.

[0013] S4: Generating Urban Development Corridors and Urban Centers

[0014] By defining the network structure and using the pix2pix deep learning model for iterative training, a Shapefile file representing the core and axes of the urban spatial structure is finally generated.

[0015] S5: Generate key urban areas

[0016] Urban areas dominated by land use types A1, M, and S are identified as urban administrative, industrial, and transportation sectors. Key urban sectors are extracted and generated through cluster analysis.

[0017] S6: Generate general city blocks

[0018] The city axis, the city core, and the key urban areas are selected and reversed to obtain other plots. The plot with land use G1 among the other plots is used as the generator. The crystal growth algorithm is used to form the crystal growth region of each generator under a given velocity V. The general urban area is formed by image vectorization and boundary curve shaping.

[0019] S7: Generate a spatial map of the urban spatial structure

[0020] The system integrates and generates an urban structural system encompassing "green corridors + development corridors + urban centers + key and general sectors." After processing with image editing software, it generates urban spatial structure planning drawings, which are then displayed on exhibition equipment.

[0021] Furthermore, the specific operation process for loading the S1 basic information is as follows:

[0022] Step S1-1: Obtain city data. Obtain the latest administrative division JPG file and the land use map of the target city's control plan through the government portal website. Import the files into the geospatial information platform and use the spatial correction tool to perform georegistration of the files based on the known point coordinates.

[0023] Step S1-2: Load the boundary and road network, create a polygon feature layer and manually extract the administrative boundary of the target city to obtain a shapefile file containing geographic coordinate information of the city's administrative boundary. Using the reclassification function, use the road representative color and other colors in the land use map file as a binary classification standard, binarize the image colors, and use the Arcscan tool to vectorize the obtained file to obtain the center line of the road network of the target city's control plan.

[0024] Step S1-3: Define road levels and assign road width attributes to expressways, arterial roads, secondary arterial roads, and branch roads according to the "GB50220-95 Urban Road Traffic Planning and Design Specification".

[0025] Step S1-4: Load land use information based on the road network in step S1-3, and vectorize the JPG format file of the land use control plan in step S1-2. The file contains land use information of each plot, and the land use type is accurate to the middle class, to obtain a Shapefile file with specified RGB color blocks.

[0026] Steps S1-5: Acquire remote sensing information. Obtain urban DEM elevation data and Normalized Difference Vegetation Index (NDVI) data from the geospatial data open website. All extracted data are at a resolution of 30m. Unify the coordinate system of the data, crop and stitch them together to form a 250m*250m raster file.

[0027] Step S1-6: Collect the data from steps S1-3, S1-4 and S1-5 into the geospatial information platform, use the city boundary obtained in step S1-1 as the base mask, and use the "extract by mask" function to perform cropping to obtain the urban land use information database C1 based on the city boundary, and input it into the DS1522+NAS ​​memory.

[0028] Furthermore, the generation steps of the S2 urban green corridor include:

[0029] Step S2-1: Combining the remote sensing information from steps S1-5, identify topographic and geomorphological features. Based on the Normalized Difference Vegetation Index (NDVI), extract land cover types of forest and grassland with an area ≥20 km². 2The land use was divided into mountain patches, and on-site calibration was performed using drones to ultimately obtain several morphologically independent mountain patches. Hydrological analysis was conducted based on DEM data to extract river networks and the boundaries between mountain ranges and river systems, constructing a mountain range database M(M1-M2). n ), water system reservoir R (R1-R n ).

[0030] Step S2-2: Construct an ecological resistance surface model, select land use type, DEM, NDVI, and slope as resistance factors, and in the geographic information platform, reclassify the land use map obtained in step S1-1 according to construction land, unused land, cultivated land, forest land, and water system, and assign values ​​of 5, 4, 3, 2, and 1 respectively.

[0031] Step S2-3: Create a cost statistics raster. Reclassify and match the raster classification standards of land use map, elevation (DEM), slope, and normalized vegetation index with resistance values ​​to obtain a resistance level distribution raster for each resistance factor. Then, superimpose the resistance level distribution raster according to the weight to obtain the cost statistics raster.

[0032] Steps S2-4: Simulate ecological corridors. Input cost statistics raster into the geographic information platform, use the cost distance tool to calculate cumulative resistance, obtain the cumulative resistance value of each area in the target city, and use image recognition tools to identify the mountain patch library (M1-M2). n The geometric center of the ecological source point is used as the ecological source point. The cost path tool is used to calculate the minimum cumulative resistance path from each ecological source point to other ecological source points in turn. The calculation results are displayed on the specified layer.

[0033] Step S2-5: Extract urban green corridors. Divide the cumulative resistance value into 3 levels using the equal interval method. Extract the ecological corridors that are completely in the first level range (i.e., the range with the smallest cumulative resistance value) as urban green corridors. Use the raster to surface tool to extract urban green corridors and output them as shapefile files.

[0034] Furthermore, the operation process for building the S3 case library is as follows:

[0035] Step S3-1: Collect urban spatial structure map cases to form a case library S1, which contains no less than 200 urban master plan maps with an image resolution higher than 150dpi and a single image size of less than 10m, and input them into DS1522+NAS ​​storage.

[0036] Step S3-2: Perform structural annotation on the case library. Use vector drawing tools on the geographic information platform to annotate the cases. Integrate all annotated case Shapefile files into the urban spatial structure case library S and input it into the DS1522+NAS ​​storage.

[0037] Furthermore, the steps for generating the S4 urban development corridor and urban center include:

[0038] Step S4-1: Data format conversion. Convert the Shapefile format files in the urban spatial structure case library S from step S3-2 into JPG format images that the algorithm can recognize, thus constructing the urban spatial structure case library S(S1~S2). n The conversion method involves using Python Geopandas third-party library functions to batch convert the planar map base sample data into an image format that Pix2pix can read, according to the same ratio.

[0039] Step S4-2: Define the network structure using the Pix2pix deep convolutional neural network model. In this model, the generator is based on the "U-Net" architecture, and the discriminator uses the "PatchGAN" classifier. The formula is as follows:

[0040]

[0041]

[0042]

[0043] Step S4-3: Targeting the urban spatial structure case library S(S1~S) n All categories in the model are trained using gradient descent. The fluctuations in the generator and discriminator loss functions during training are observed for each parameter. The learning rate and iteration count are adjusted and optimized. The optimal values ​​for each category are determined by comparing training time and generation results. Finally, a generative algorithm model library G(G1~G2) for generating planar images of urban spatial structure maps is constructed. n The learning rate is a tuning parameter in the optimization algorithm, which determines the step size in each iteration to minimize the loss function. The number of iterations is the number of loops during the iteration process. The gradient descent formula is:

[0044]

[0045] Where η is the learning rate, and i represents the i-th data point. The weight parameter w changes by a certain amount in each iteration.

[0046] Step S4-4: Input the shapefile of the design plot outline and road network of the target city (taking Wuxi Xiyanhu Ecological Science and Technology City as the design case) into the corresponding algorithm model G in step S4-3 to generate a JPG file of the spatial structure planar shape of the city. The city plot outline and road network file is a pure black color block with a scale of 1:2000, a resolution of 100dpi, and a size of 300mm*300mm.

[0047] Step S4-5: Urban spatial structure extraction. The image generated in S4-4 is read in RGB format. Then, the urban development corridors and urban centers are extracted by the different values ​​of the colors in the three channels. Spatial structures with the same color are found and the other parts are converted to white. The image is then converted to grayscale. The findContours function is used to detect line segments or points in the image. Each detected color block is encapsulated into a polygon. At the same time, the primary and secondary development corridors and primary and secondary urban centers are determined according to the correspondence between color values ​​and urban structures. This yields vector data containing the urban spatial structure.

[0048] The operation utilizes the DGX-2 calculator device with a 24-core processor and 1.5TB of memory, along with a Huawei OceanStor6800V5 storage device, for data processing and storage.

[0049] Furthermore, the generation process for the key urban blocks of S5 includes:

[0050] Step S5-1: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify the urban area dominated by land use A1 as the urban administrative block. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban administrative block L-. A1， The plots within the cluster are labeled L- A1。

[0051] Step S5-2: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify urban areas with land use primarily M as urban industrial zones. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban industrial zone L. M， The plots within the cluster are labeled L- M .

[0052] Step S5-3: In the Shapefile file containing road network and land use information of the target urban design plot generated in Step S1-4, identify the urban area with land use mainly in S3 as the urban transportation zone, and mark the key urban transportation zone as L- S3。

[0053] Step S5-4: Integrate and generate Wuxi Xiyanhu urban key area L- A1 L- M L- S3 .

[0054] Furthermore, the generation process of the general blocks of the S6 city includes the following steps:

[0055] Step S6-1: Obtain other land parcels. Extract the land parcels that are designated as urban axes, urban cores, and key urban sectors in Steps S4-4 and S5-4 using the ultra-deep computer and invert their selection to obtain other land parcels.

[0056] Step S6-2: Delineate general urban blocks. For the other plots obtained in step S6-1, use the crystal growth algorithm to delineate general urban blocks.

[0057] Step S6-3: Integrate the feature files and input the shapefiles of urban green corridors, urban main development corridors, urban secondary development corridors, urban main centers, urban secondary centers, urban key blocks and urban general blocks generated in steps S2-5, S4-6, S5-4 and S6-2 into the urban land use information database C1.

[0058] Furthermore, the steps for generating the urban spatial structure map include:

[0059] Step S7-1: Export files. Use Arcscan to extract the centerline shapefiles of urban green corridors, major urban development corridors, and secondary urban development corridors. Export all centerline shapefiles, point shapefiles of major and secondary urban centers, area shapefiles of key and general urban areas, and urban land use map raster as AI file groups.

[0060] Step S7-2: Generate an urban spatial structure system map. By inputting the image into the image editing software, generate the centerline arrow, center point coil, and area boundary, and adjust parameters such as line type, line width, and color to finally generate an urban spatial structure system map covering "green corridors + development corridors + urban center + key and general sectors".

[0061] Step S7-3: Export the results. Export the resulting JPG drawing from step S7-2 and display it on the display device.

[0062] Furthermore, the generator operation involves inputting urban spatial structure (development corridors and city centers) samples into an ultra-deep computer and generating images with similar feature distributions based on their characteristic patterns. The discriminator operation involves taking the city boundary and the generated image or the real urban spatial structure map as a new sample pair as input, determining whether this sample pair is a correct mapping from the city boundary to the real spatial structure, and outputting a probability value to identify the authenticity of the generated image. The network structure adopts the Patch GAN concept, dividing the generated result into multiple fixed-size patch images and inputting them into the discriminator network.

[0063] Furthermore, the delineation rules are as follows:

[0064] Determine the generators: Utilize image analysis to identify G1-type land use (park land) in other plots, identify the geometric center point of the land use, and output a generator library S = {p1, p2... p...} containing coordinate information. n The format is a point feature shapefile.

[0065] Establishing an algorithmic model on an ultra-deep computing platform: Let S = {p1, p2, ... p...} n Let} be the set of points on the L1 plane, and C be the target city database obtained in steps S1-6. Freedom of entry and exit from the transportation network is allowed at any point on C. Given a speed of v on the transportation network, and a speed of 1 and a distance of L1 in other areas of the plane, Reg will... gc (pi)={p|d gc (p,p i ) <d gc (p,p j The region Reg defined by )),i≠j} gc (p i ) is called the generator P i The typical urban crystal growth area, in which d gc (p,p i ) represents the distance from point p to generator p. i The shortest time for city distance is calculated, and the results for different regions are represented by specified RGB color blocks (RGB:254,225,142). Finally, a color block JPG file containing several crystal growth regions is generated.

[0066] All generated color patches of crystal growth regions are integrated, the images are vectorized through a geographic information platform, the boundary curves are formed using the approxPolyDP function, and the extracted shapefile is used as a general urban area.

[0067] The beneficial effects of this invention are:

[0068] 1. The automatic generation method of this invention, through the construction of a resistance model and the understanding of the characteristics and laws of urban green corridors by parametric models, automatically generates urban ecological corridors, providing a foundation and support for the intelligent generation design of urban spatial structures;

[0069] 2. The automatic generation method of this invention utilizes the pix2pix deep convolutional neural network model to build an algorithm model for generating urban structural planar images. The entire algorithm model can generate multiple schemes in a short time. Compared with the traditional planning and design process, it can help designers reduce mechanical and repetitive drawing work and provide inspiration and decision support for them in the conceptual design stage of the scheme.

[0070] 3. The automatic generation method of this invention uses OpenCV to vectorize images generated by machine learning, and establishes a technical method for vectorizing the two-dimensional planar scheme output by image learning and adding target information, providing technical support for the study of urban structure. Attached Figure Description

[0071] The invention will now be further described with reference to the accompanying drawings.

[0072] Figure 1 This is a flowchart of the automatic generation method steps of the present invention;

[0073] Figure 2 This is the urban green corridor map in the automatic generation method of this invention;

[0074] Figure 3 This is the urban spatial structure land parcel distribution map in the automatic generation method of this invention;

[0075] Figure 4 This is the key urban area map in the automatic generation method of this invention;

[0076] Figure 5 This is the general urban block map in the automatic generation method of this invention;

[0077] Figure 6 This is the urban spatial structure system diagram in the automatic generation method of this invention. Detailed Implementation

[0078] 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.

[0079] Automatic generation methods for urban spatial structure based on parameter iteration and machine learning, such as Figure 1 As shown, the automatic generation method includes the following steps:

[0080] S1: Basic information loading involves acquiring land use information data files for the target area based on an open data platform and government portal website. This includes urban administrative boundary data, road grade data, urban land use data, and urban DEM data. The data is then cleaned and vectorized using a geographic information platform to obtain the urban land use information database C1. The specific operation process for basic information loading is as follows:

[0081] Step S1-1: Obtain city data. Obtain the latest administrative division JPG file and the land use map of the target city's control plan through the government portal website. Import the files into the geospatial information platform and use the spatial correction tool to perform georegistration of the files based on the known point coordinates.

[0082] Step S1-2: Load the boundary and road network, create a polygon feature layer and manually extract the administrative boundary of the target city to obtain a shapefile file containing geographic coordinate information of the city's administrative boundary. Using the reclassification function, use the road representative color and other colors in the land use map file as a binary classification standard, binarize the image colors, and use the Arcscan tool to vectorize the obtained file to obtain the center line of the road network of the target city's control plan.

[0083] Step S1-3: Define road levels and assign road width attributes to expressways, arterial roads, secondary arterial roads, and branch roads according to the "GB50220-95 Urban Road Traffic Planning and Design Specification".

[0084] Step S1-4: Load land use information based on the road network in step S1-3, and vectorize the JPG format file of the land use control plan in step S1-2. The file contains land use information of each plot, and the land use type is accurate to the middle class, to obtain a Shapefile file with specified RGB color blocks.

[0085] Steps S1-5: Obtain remote sensing information. Obtain urban DEM elevation data and Normalized Difference Vegetation Index (NDVI) data from the Geospatial Data Open Website (http: / / www.gscloud.cn / ; https: / / ladsweb.Nascom.Nasa.gov / search / ). All extracted data are at a resolution of 30m. Unify the coordinate system of the data, crop and stitch them together to form a 250m*250m raster file.

[0086] Step S1-6: Collect the data from steps S1-3, S1-4 and S1-5 into the geospatial information platform, use the city boundary obtained in step S1-1 as the base mask, and use the "extract by mask" function to perform cropping to obtain the urban land use information database C1 based on the city boundary, and input it into the DS1522+NAS ​​memory.

[0087] S2: Generating Urban Green Corridors. Mountain and water system reservoirs are extracted from remote sensing images. A minimum resistance model is constructed using selected indicators. With the mountain reservoir as the source, a cost path tool is used to simulate ecological corridors. The combined calculation results are used as urban green corridors. The steps for generating urban green corridors include:

[0088] Step S2-1: Combining the remote sensing information from steps S1-5, identify topographic and geomorphological features. Based on the Normalized Difference Vegetation Index (NDVI), extract land cover types of forest and grassland with an area ≥20 km². 2 The land is designated as mountain patches. If two or more mountain patches share a common edge, they are considered as one patch. The patches are then corrected on-site using drones to ultimately obtain several morphologically independent mountain patches.

[0089] Hydrological analysis was performed based on DEM data to extract river networks and the boundaries of mountain ranges and river systems, and a mountain range database M(M1-M2) was constructed. n ), water system reservoir R (R1-R n ).

[0090] Step S2-2: Construct an ecological resistance surface model. Select land use type, DEM, NDVI, and slope as resistance factors. In the geographic information platform, reclassify the land use map obtained in Step S1-1 according to construction land, unused land, cultivated land, forest land, and water system, assigning values ​​of 5, 4, 3, 2, and 1 respectively. Reclassify the DEM elevation data obtained in Step S1-5 using the 1-5 gradient assignment method. According to the order of elevation from largest to smallest, divide it into 5 levels using the natural breakpoint method, assigning values ​​of 5, 4, 3, 2, and 1 respectively. The DEM elevation data obtained in steps S1-5 are reclassified using the 1-5 gradient assignment method. According to the slope from largest to smallest, they are divided into 5 levels using the natural breakpoint method, and assigned values ​​of 5, 4, 3, 2, and 1 respectively. The Normalized Difference Vegetation Index (NDVI) data obtained in steps S1-5 are also reclassified using the 1-5 gradient assignment method. According to the slope from smallest to largest, they are divided into 5 levels using the natural breakpoint method, and assigned values ​​of 5, 4, 3, 2, and 1 respectively. Principal component analysis is performed on the four types of factors to obtain the corresponding weights of the factors.

[0091] Step S2-3: Create a cost statistics raster. Reclassify the land use map, elevation (DEM), slope, and normalized difference vegetation index (NDE) rasters according to the classification standards in Table 1 and match them with resistance values ​​to obtain a resistance level distribution raster for each resistance factor. Then, superimpose the resistance level distribution rasters according to their weights to obtain the cost statistics raster. Table 1 is as follows:

[0092] Table 1

[0093]

[0094] Steps S2-4: Simulate ecological corridors. Input cost statistics raster into the geographic information platform, use the cost distance tool to calculate cumulative resistance, obtain the cumulative resistance value of each area in the target city, and use image recognition tools to identify the mountain patch library (M1-M2). nThe geometric center of the ecological source point is used as the ecological source point. The cost path tool is used to calculate the minimum cumulative resistance path from each ecological source point to other ecological source points in turn. The calculation results are displayed as grid lines with specified RGB colors (RGB: 76, 166, 76) on a specified layer. All paths are merged as potential ecological corridors. A buffer analysis with a radius of 500m is performed on the potential ecological corridors, and ecological corridors with a distance ≤1000m are merged.

[0095] Step S2-5: Extract urban green corridors. Using the equal interval method, divide the cumulative resistance value into three levels. Extract the ecological corridors that fall entirely within the first level range (i.e., the range with the lowest cumulative resistance value) as urban green corridors. Figure 2 As shown, the urban green corridor is extracted using the raster to polygon tool and output as a shapefile file.

[0096] S3: Case Study Library Construction. Case studies of urban spatial structures are collected from open-source websites. Spatial structures are then labeled on a geographic information platform based on urban road and boundary information, including primary and secondary development corridors and primary and secondary centers, forming a case study library. The operational process for building the case study library is as follows:

[0097] Step S3-1: Collect urban spatial structure map examples (file format is jpg) to form a case library S1, which contains no less than 200 urban master plan maps, with image resolution higher than 150dpi and single image size less than 10m, and input them into DS1522+NAS ​​storage.

[0098] Step S3-2: Perform structural annotation on the case library. Using vector drawing tools on the geographic information platform, mark all plots of land within 80 meters outward from the centerline of the main development corridor of the case as red (RGB: 255, 0, 0), plots within 40 meters outward from the centerline of the secondary development corridor as yellow (RGB: 255, 255, 0), plots within a 100-meter radius circle of the main urban center point (with a circular area exceeding 50% of the plot area) as purple (RGB: 255, 0, 255), and plots within a 50-meter radius circle of the secondary center point (with a circular area exceeding 50% of the plot area) as blue (RGB: 0, 0, 255). Integrate all annotated case Shapefile files into the urban spatial structure case library S and input it into the DS1522+NAS ​​storage device.

[0099] S4: Generate urban development corridors and city centers. By defining the network structure and using the pix2pix deep learning model for iterative training, the core and axis shapefiles of the urban spatial structure are finally generated. The steps for generating urban development corridors and city centers include:

[0100] Step S4-1: Data format conversion. Convert the Shapefile format files in the urban spatial structure case library S from step S3-2 into JPG format images that the algorithm can recognize, thus constructing the urban spatial structure case library S(S1~S2). n The conversion method involves using Python Geopandas third-party library functions to batch convert the planar map base sample data into an image format that Pix2pix can read, according to the same ratio. The specific steps are to mark the planar map base sample data with outline labels and color data labels, integrate it into the graphic visualization information platform, and output it as a 256-pixel * 256-pixel JPG format image containing urban spatial structure information, where the urban spatial structure information is information line segments or points with specified RGB values.

[0101] Step S4-2: Define the network structure using the Pix2pix deep convolutional neural network model. In this model, the generator is based on the "U-Net" architecture, and the discriminator uses the "PatchGAN" classifier. The formula is as follows:

[0102]

[0103]

[0104]

[0105] The generator operates by inputting urban spatial structure (development corridors and city centers) samples into an ultra-deep computer and generating images with similar feature distributions based on their characteristic patterns. The discriminator operates by taking the city boundary and the generated image or the real urban spatial structure map as a new sample pair as input, judging whether this sample pair is a correct mapping from the city boundary to the real spatial structure, and outputting a probability value to identify the authenticity of the generated image. The network structure adopts the Patch GAN idea, dividing the generated result into multiple fixed-size patch images and inputting them into the discriminator network.

[0106] Step S4-3: Targeting the urban spatial structure case library S(S1~S) n All categories in the model are trained using gradient descent. The fluctuations in the generator and discriminator loss functions during training are observed for each parameter. The learning rate and iteration count are adjusted and optimized. The optimal values ​​for each category are determined by comparing training time and generation results. Finally, a generative algorithm model library G(G1~G2) for generating planar images of urban spatial structure maps is constructed. n The learning rate is a tuning parameter in the optimization algorithm, which determines the step size in each iteration to minimize the loss function; the number of iterations is the number of loops during the iteration process; the gradient descent formula is:

[0107]

[0108] Where η is the learning rate, and i represents the i-th data point. The weight parameter w changes by a certain amount in each iteration.

[0109] Step S4-4: Input the shapefile of the design plot outline and road network of the target city (taking Wuxi Xiyanhu Ecological Science and Technology City as the design case) into the corresponding algorithm model G in step S4-3 to generate a JPG file of the spatial structure planar shape of the city. The city plot outline and road network file is a pure black color block with a scale of 1:2000, a resolution of 100dpi, and a size of 300mm*300mm.

[0110] Step S4-5: Urban spatial structure extraction. The image generated in S4-4 is read in RGB format. Urban development corridors and city centers are extracted based on the different color values ​​across the three channels. Spatial structures with the same color are identified, and other parts are converted to white. The image is then converted to grayscale. The `findContours` function is used to detect line segments or points in the image. Each detected color patch is encapsulated into a polygon. Simultaneously, primary and secondary development corridors and city centers are determined based on the correspondence between color values ​​and urban structures. This yields vector data containing the urban spatial structure, such as... Figure 3 As shown.

[0111] The operation utilizes the DGX-2 calculator device with a 24-core processor and 1.5TB of memory, along with a Huawei OceanStor6800V5 storage device, for data processing and storage.

[0112] S5: Generate key urban blocks, identify urban areas dominated by A1, M, and S land uses as urban administrative, industrial, and transportation blocks, and extract and generate key urban blocks through cluster analysis. The process of generating key urban blocks includes:

[0113] Step S5-1: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify urban areas dominated by A1 land use as urban administrative blocks. The identification criteria are that the proportion of A1 function exceeds 75% of the plot area and the plot area exceeds 0.5 hectares. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban administrative block L-. A1 Fill the plots within the cluster with pink (RGB: 255, 128, 191) and mark them as L-. A1 .

[0114] Step S5-2: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify urban areas dominated by land use M as urban industrial zones. The identification criteria are that the proportion of function M exceeds 75% of the plot area and the plot area exceeds 0.5 hectares. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban industrial zone L-. M Fill the plots within the cluster with brown (RGB: 77, 57, 0) and label them L-. M .

[0115] Step S5-3: In the Shapefile file containing road network and land use information of the target urban design plot generated in Step S1-4, identify the urban area dominated by S3 land use as the urban transportation sector. The identification criteria are that the S3 function accounts for more than 75% of the plot area and is a major urban transportation hub (airport, train station, port), thus being designated as the key urban transportation sector L- S3， Fill the selected plots with gray (RGB: 51, 51, 51) and mark them as L- S3 .

[0116] Step S5-4: Integrate and generate Wuxi Xiyanhu urban key area L- A1 L- M L- S3 ,like Figure 4 As shown.

[0117] S6: Generate general urban blocks. This involves selecting and reversing the selection of plots along the urban axis, in the urban core, and key urban blocks to obtain other plots. Using plots with land use designation G1 as generators, a crystallization growth algorithm is used to form the crystallization growth region of each generator under a given velocity V. Through image vectorization and boundary curve shaping, general urban blocks are formed. The generation process for general urban blocks includes the following steps:

[0118] Step S6-1: Obtain other land parcels. Extract the land parcels that are designated as urban axes, urban cores, and key urban sectors in Steps S4-4 and S5-4 using the ultra-deep computer and invert their selection to obtain other land parcels.

[0119] Step S6-2: Delineate general urban zoning blocks. For the other plots obtained in Step S6-1, use the crystallization growth algorithm to delineate general urban zoning blocks. The delineation rules are as follows:

[0120] Determine the generators: Utilize image analysis to identify G1-type land use (park land) in other plots, identify the geometric center point of the land use, and output a generator library S = {p1, p2... p...} containing coordinate information. n The format is a point feature shapefile.

[0121] Establishing an algorithmic model on an ultra-deep computing platform: Let S = {p1, p2, ... p...} n Let} be the set of points on the L1 plane, and C be the target city database obtained in steps S1-6. Freedom of entry and exit from the transportation network is allowed at any point on C. Given a speed of v on the transportation network, and a speed of 1 and a distance of L1 in other areas of the plane, Reg will... gc (pi)={p|d gc (p,p i ) <d gc (p,p j The region Reg defined by )),i≠j} gc (p i ) is called the generator P i The typical urban crystal growth area, in which d gc (p,p i ) represents the distance from point p to generator p. i The shortest time for city distance is calculated, and the results for different regions are represented by specified RGB color blocks (RGB:254,225,142). Finally, a color block JPG file containing several crystal growth regions is generated.

[0122] All generated color patches representing crystal growth regions are integrated, and the image is vectorized using a geographic information platform. The boundary curves are then extracted using the approxPolyDP function and extracted as shapefiles to represent general urban blocks. Figure 5 As shown.

[0123] Step S6-3: Integrate the feature files and input the shapefiles of urban green corridors, urban main development corridors, urban secondary development corridors, urban main centers, urban secondary centers, urban key blocks and urban general blocks generated in steps S2-5, S4-6, S5-4 and S6-2 into the urban land use information database C1.

[0124] S7: Generate a spatial map of the urban spatial structure, integrating and generating an urban structural system encompassing "green corridors + development corridors + urban center + key and general areas." After processing with image editing software, generate urban spatial structure planning drawings, which are then displayed on exhibition equipment. The steps for generating the urban spatial structure map include:

[0125] Step S7-1: Export files. Use Arcscan to extract the centerline shapefiles of urban green corridors, major urban development corridors, and secondary urban development corridors. Export all centerline shapefiles, point shapefiles of major and secondary urban centers, area shapefiles of key and general urban areas, and urban land use map raster as AI file groups.

[0126] Step S7-2: Generate an urban spatial structure system map. By inputting the image into the image editing software, generate the centerline arrow, center point coil, and area boundary, and adjust parameters such as line type, line width, and color to finally generate an urban spatial structure system map covering "green corridors + development corridors + urban center + key and general sectors".

[0127] Step S7-3: Export the results. Export the resulting JPG image from step S7-2 for display on a 55-inch or larger screen with a resolution of 1920×1080. The final exported image is shown below. Figure 6 As shown.

[0128] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0129] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An automatic urban spatial structure generation method based on parameter iteration and machine learning, characterized in that, The automatic generation method includes the following steps: S1: Loading Basic Information Based on the open data platform and government portal website, land use information data files of the target area are obtained, including urban administrative boundary data, road grade data, urban land use data, and urban DEM data. The data is cleaned and vectorized through the geographic information platform to obtain the urban land use information database C1. S2: Generate urban green corridors Mountain and water system reservoirs are extracted from remote sensing images. Indicators are selected to construct a minimum resistance model. Using the mountain reservoir as the source, cost path tools are used to simulate ecological corridors. The combined calculation results are used as urban green corridors. S3: Case Library Construction Urban spatial structure cases are collected through open-source websites, and spatial structures are labeled on a geographic information platform based on urban road and boundary information, including primary and secondary development corridors and primary and secondary centers, to form a case study library. S4: Generating Urban Development Corridors and Urban Centers By defining the network structure and using the pix2pix deep learning model for iterative training, the core and axis shapefile of the urban spatial structure is finally generated. S5: Generate key urban areas The urban areas dominated by land use A1, M, and S were identified as urban administrative, industrial, and transportation sectors, respectively. Key urban sectors were extracted and generated through cluster analysis. S6: Generate general city tectonic blocks The city axis, city core and key urban blocks are selected and reversed to obtain other blocks. The block with land use G1 among the other blocks is used as the generator. The crystal growth algorithm is used to form the crystal growth region of each generator under the given speed V. The general urban blocks are formed by image vectorization and boundary curve shaping. S7: Generate a spatial map of the urban spatial structure The system integrates and generates an urban structural system encompassing "green corridors + development corridors + urban centers + key and general sectors." After processing with image editing software, it generates urban spatial structure planning drawings, which are then displayed on exhibition equipment.

2. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 1, characterized in that, The specific operation procedure for loading the S1 basic information is as follows: Step S1-1: Obtain city data. Obtain the latest administrative division JPG file and land use map of the target city through the government portal website. Import the files into the geospatial information platform and use the spatial correction tool to perform georegistration of the files based on the known point coordinates. Step S1-2: Load the boundary and road network, create a surface feature layer and manually extract the administrative boundary of the target city to obtain a shapefile file containing geographic coordinate information of the city's administrative boundary. Using the reclassification function, the road representative color and other colors in the land use map file are used as the binary classification standard. The image color is binarized. The obtained file is vectorized using the Arcscan tool to obtain the center line of the control plan road network of the target city. Step S1-3: Define road levels and assign road width attributes of expressway, arterial road, secondary arterial road and local road to the control plan road network level in accordance with the "GB50220-95 Urban Road Traffic Planning and Design Specification". Step S1-4: Load land use information based on the road network in step S1-3, and vectorize the JPG format file of the land use control plan in step S1-2. The file contains land use information of each plot, and the land use type is accurate to the middle class, to obtain a Shapefile file with specified RGB color blocks. Steps S1-5: Obtain remote sensing information. Obtain urban DEM elevation data and Normalized Difference Vegetation Index (NDVI) data from the geospatial data open website. All extracted data are at 30m resolution. Unify the coordinate system of the data, crop and stitch them together to form a 250m resolution. 250m raster file; Step S1-6: Collect the data from steps S1-3, S1-4 and S1-5 into the geospatial information platform, use the city boundary obtained in step S1-1 as the base mask, and use the "extract by mask" function to perform cropping to obtain the urban land use information database C1 based on the city boundary, and input it into the DS1522+NAS ​​memory.

3. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 2, characterized in that, The steps for generating the S2 urban green corridor include: Step S2-1: Combining the remote sensing information from steps S1-5, identify topographic and geomorphological features. Based on the normalized vegetation index, extract land cover types as forest and grassland, with an area ≥ 20 km². 2 The land use was divided into mountain patches, and on-site calibration was performed using drones to obtain several morphologically independent mountain patches. Hydrological analysis was conducted based on DEM data to extract river networks and the boundaries between mountain ranges and river systems, thus constructing a mountain range database M1-M2. n Reservoirs R1-R n ; Step S2-2: Construct an ecological resistance surface model, select land use type, DEM, NDVI, and slope as resistance factors, and in the geographic information platform, reclassify the land use map obtained in step S1-1 according to construction land, unused land, cultivated land, forest land, and water system, and assign values ​​of 5, 4, 3, 2, and 1 respectively. Step S2-3: Create a cost statistics raster. Reclassify and match the raster classification standards of land use map, DEM elevation, slope, and normalized vegetation index with resistance values ​​to obtain the resistance level distribution raster of each resistance factor. Then, superimpose the resistance level distribution raster according to the weight to obtain the cost statistics raster. Steps S2-4: Simulate ecological corridors. Input cost statistics raster into the geographic information platform, use the cost distance tool to calculate cumulative resistance, obtain the cumulative resistance value of each area in the target city, and use image recognition tools to identify mountain patch databases. The geometric center is taken as the ecological source point. The cost path tool is used to calculate the minimum cumulative resistance path from each ecological source point to other ecological source points in turn. The calculation results are displayed on the specified layer. Step S2-5: Extract urban green corridors. Divide the cumulative resistance value into 3 levels using the equal interval method. Extract the ecological corridors that are completely in the first level range as urban green corridors. Use the raster to surface tool to extract urban green corridors and output them as shapefile files. The first level range is the range with the minimum cumulative resistance value.

4. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 3, characterized in that, The operation process for building the S3 case library is as follows: Step S3-1: Collect urban spatial structure map cases to form a case library S1, which contains no less than 200 urban master plan maps with an image resolution higher than 150dpi and a single image size of less than 10m, and input them into DS1522+NAS ​​storage. Step S3-2: Perform structural annotation on the case library. Use vector drawing tools on the geographic information platform to annotate the center lines of the main development corridors, the center lines of the secondary development corridors, the main urban center point, and the secondary center point of the cases. Integrate all annotated case Shapefile files into the urban spatial structure case library S and input it into the DS1522+NAS ​​storage device.

5. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 4, characterized in that, The steps for generating the S4 urban development corridor and urban center include: Step S4-1: Data format conversion. Batch convert the Shapefile format files in the urban spatial structure case library S from step S3-2 into JPG format images that the algorithm can recognize, thus constructing the urban spatial structure case library. ~ The conversion method involves using Python Geopandas third-party library functions to batch convert the planar map base sample data into an image format that Pix2pix can read, according to the same ratio. Step S4-2: Define the network structure using the Pix2pix deep convolutional neural network model. In this model, the generator is based on the "U-Net" architecture, and the discriminator uses the "PatchGAN" classifier. The formula is as follows: ； Step S4-3: Targeting the Urban Spatial Structure Case Library ~ All categories were trained using gradient descent. The fluctuations of the generator and discriminator loss functions during training were observed for each parameter. The learning rate and iteration count were adjusted and optimized. The optimal values ​​for each category were determined by comparing training time and generation results. Finally, a generative algorithm model library for generating planar images of urban spatial structure maps was constructed. ~ Wherein, the learning rate is the tuning parameter in the optimization algorithm, which determines the step size in each iteration to minimize the loss function; the number of iterations is the number of loops during the iterative operation; the gradient descent formula is: Where η is the learning rate, i represents the i-th data point, and the weight parameter w changes by the magnitude of each iteration; Step S4-4: Input the target city's design plot outline and road network shapefile into the corresponding algorithm model G in step S4-3 to generate a JPG file of the city's spatial structure plan view. The city plot outline and road network file is a pure black area, 1:2000 scale, 100dpi resolution, 300mm. 300mm JPG file; Step S4-5: Urban spatial structure extraction. Read the image generated in S4-4 in RGB format, and then extract the urban development corridors and urban centers by the different values ​​of the colors in the three channels. Find the spatial structures with the same color and convert the other parts to white. Then convert the image to grayscale. Then use the findContours function to detect the image line segments or points. Encapsulate each detected color block into a polygon. At the same time, determine the primary and secondary development corridors and primary and secondary urban centers according to the correspondence between color values ​​and urban structures. Thus, vector data containing urban spatial structure can be obtained. The operation utilizes the DGX-2 calculator device with a 24-core processor and 1.5TB of memory, along with a Huawei OceanStor 6800 V5 storage device, for data processing and storage.

6. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 5, characterized in that, The process for generating the key urban blocks in S5 includes: Step S5-1: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify the urban area with land use mainly in A1 as the urban administrative block. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban administrative block L-A1. ， The plots within the cluster are labeled L-A1. ； Step S5-2: In the Shapefile file containing road network and land use information of the target urban design plots generated in Step S1-4, identify urban areas with land use M as the main urban industrial sector. Perform cluster analysis on the selected plots and adjust the distance range until only one obvious cluster appears as the key urban industrial sector LM. ， The plots within the cluster are marked as LM; Step S5-3: In the Shapefile file containing road network and land use information of the target urban design plot generated in Step S1-4, identify the urban area with S3 land use as the main urban transportation zone, and mark the key urban transportation zone as L-S3. ； Step S5-4: Integrate and generate key urban blocks L-A1, LM, and L-S3.

7. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 6, characterized in that, The generation process of the general blocks of the S6 city includes the following steps: Step S6-1: Obtain other land parcels. Extract the land parcels in Steps S4-4 and S5-4 that are designated as urban axes, urban cores, and key urban sectors using the ultra-deep computer and invert their selection to obtain other land parcels. Step S6-2: Delineate general urban blocks. For the other plots obtained in step S6-1, use the crystal growth algorithm to delineate general urban blocks. Step S6-3: Integrate the feature files and input the shapefiles of urban green corridors, urban main development corridors, urban secondary development corridors, urban main centers, urban secondary centers, urban key blocks and urban general blocks generated in steps S2-5, S4-6, S5-4 and S6-2 into the urban land use information database C1.

8. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 6, characterized in that, The steps for generating the urban spatial structure map include: Step S7-1: Export files. Use Arcscan to extract the centerline shapefiles of urban green corridors, major urban development corridors, and secondary urban development corridors. Export all centerline shapefiles, point shapefiles of major and secondary urban centers, area shapefiles of key and general urban areas, and urban land use map raster as AI file groups. Step S7-2: Generate an urban spatial structure system map. By inputting the image into the image editing software, generate the centerline arrow, center point coil and area boundary, and adjust parameters such as line type, line width, and color to finally generate an urban spatial structure system map covering "green corridors + development corridors + urban center + key and general sectors". Step S7-3: Export the results. Export the resulting JPG drawing from step S7-2 and display it on the display device.

9. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 5, characterized in that, The generator operation steps are as follows: inputting development corridor and urban center samples into the ultra-deep computer, and generating images with similar feature distributions based on their characteristic patterns; the discriminator operation steps are as follows: inputting a new sample pair consisting of the urban boundary and the generated image or the real urban spatial structure map, judging whether the sample pair is a correct mapping from the urban boundary to the real spatial structure, and outputting a probability value to identify the authenticity of the generated image. The network structure adopts the Patch GAN concept, which divides the generated results into multiple fixed-size patch images and inputs them into the discriminant network.

10. The method for automatically generating urban spatial structure based on parameter iteration and machine learning according to claim 7, characterized in that, The delineation rules are as follows: Determine the generators: Utilize image analysis to identify parkland in other plots, pinpoint the geometric center of each plot, and output a generator library S={p1,p2...p...} containing coordinate information. n The format is a point feature shapefile file; Establishing an algorithmic model on an ultra-deep computing platform: Let S = {p1, p2... p...} n Let} be the set of points on the L1 plane, and C be the target city database obtained in steps S1-6. Freedom of entry and exit from the traffic network is allowed at any point on C. Given a speed of v on the traffic network, and a speed of 1 and a distance of L1 in other areas of the plane, Reg will... gc (pi) = {p|d gc (p,p i ) <d gc (p,p) j The region Reg defined by )),i≠j} gc (p) i ) is called the generator P i The typical urban crystal growth area, in which d gc (p,p i ) represents the distance from point p to generator p. i The shortest time for city distance, the generation results for different regions, and finally the color block JPG file containing several crystal growth regions; All generated color patches of crystal growth regions are integrated, the images are vectorized through a geographic information platform, the boundary curves are formed using the approxPolyDP function, and the extracted shapefile is used as a general urban area.

Images