Urban potential updating area identification and division method driven by artificial intelligence

Abstract

The invention discloses an artificial intelligence-driven urban potential updating area identification and division method, which comprises the following steps of: firstly, constructing an index system according to three dimensions of building space, industry form composition and crowd vitality by using the collected building number, land function data, POI (Point of Interest) industry form data, LBS (Location Based Service) crowd data, high-definition remote sensing image data and road network data; the method comprises the following steps: carrying out hierarchical classification identification on an update object through a natural discontinuity point method and a K-Means clustering algorithm, carrying out intelligent division and optimization on a city update unit through a DBSCAN clustering algorithm and a reinforcement learning model of a graph network structure database, and finally carrying out division and boundary optimization on a city update area. According to the invention, automatic identification and division of potential updating objects, updating units and updating areas of the city can be realized, and digital and intelligent application conversion of city planning work is promoted.

Description

Technical Field

[0001] This invention belongs to the field of urban design application transformation, specifically involving an artificial intelligence-driven method for identifying and delineating potential urban renewal areas. Background Technology

[0002] As urban development enters the stock renewal phase, accurately identifying potential renewal areas has become a key task in urban planning. Traditional methods mainly rely on manual on-site surveys and static indicator analysis, which suffer from insufficient data integration, limited data types, and rigid boundary delineation. On the other hand, although recent research has attempted to introduce machine learning, significant bottlenecks remain: deep learning models require a large number of labeled samples, while urban renewal case data is scarce. Therefore, there is an urgent need to develop an urban renewal area identification method that integrates multi-source big data, adaptive spatial clustering, and intelligent boundary optimization to solve the three core problems of data fragmentation, crude identification, and inaccurate boundaries, and to promote the transformation of urban planning from experience-driven to AI-driven.

[0003] The AI-driven method for identifying and delineating potential urban renewal areas overcomes the limitations of traditional, crude identification methods. Through a dual intelligent identification mechanism of natural discontinuity grading and DBSCAN density convergence, it significantly reduces the subjective bias of manual threshold setting and solves problems such as misjudgment of scattered plots. Furthermore, relying on a dual-database driven model of target city database and case urban renewal unit database, the reinforcement learning model has the ability to continuously iterate and optimize, providing a new generation of technological engine for digital governance of urban planning. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an artificial intelligence-driven method for identifying and delineating potential urban renewal areas. This method enables intelligent fusion and dynamic updating of multi-source heterogeneous urban data; breaks through the limitations of traditional manual thresholds and establishes a hierarchical and classification dual-track identification mechanism; generates renewal area boundaries that combine spatial continuity and feasibility, ultimately supporting the digital closed-loop management of the entire renewal decision-making process; and fundamentally solves the long-standing contradiction in the field of urban renewal between insufficient utilization of multi-source data and one-sided assessment of renewal potential.

[0005] The objective of this invention can be achieved through the following technical solutions: An AI-driven method for identifying and delineating potential urban renewal areas includes the following steps: Step S1: Construction and Update of Urban Multi-Source Big Data Unit Case Database Collect existing building data (including vector morphology and age information), land use data, POI business data, LBS population data, high-resolution remote sensing imagery data, and road network data for the target city. Standardize data units, eliminate redundant data, and construct a multi-source large database for the target city. Collect data from case urban renewal units and construct a multi-source large database for these units.

[0006] Step S2: Hierarchical and intelligent identification of potential urban renewal objects 2.1 Based on the target city multi-source big data constructed in step one, an indicator system is constructed from three dimensions—building space, business format composition, and population vitality—using the smallest land plot as the unit, and the indicators are quantitatively calculated for all land plots.

[0007] 2.2 Based on the numerical natural discontinuity method of the indicator system, potential urban renewal objects are initially identified by land use as the unit, and the potential urban renewal objects are classified.

[0008] 2.3 The DBSCAN clustering algorithm is used to intelligently cluster the potential update objects initially identified, and further classify the potential update objects to establish a hierarchical classification database of urban potential update objects.

[0009] Step S3: Intelligent Division and Optimization of Potential Urban Renewal Units 3.1 Based on the multi-source big data database of potential urban renewal units in the case city constructed in step one, the urban renewal units in the database are divided into three types of orientation: function-oriented, space-oriented, and population-oriented, based on three dimensions of indicators: business characteristics, architectural characteristics, and population characteristics.

[0010] 3.2 By combining geospatial clustering and concave hull algorithms, scattered potential urban renewal objects are initially divided into potential urban renewal units, and a preliminary graph network structure database is constructed. Each renewal object within a potential urban renewal unit is treated as a node, each node has its own attribute list, and the spatial relationships between adjacent renewal objects are used as connecting lines, thus constructing a preliminary graph network structure database. Furthermore, the same method is used to construct a graph network structure database for the urban renewal unit data case library.

[0011] 3.3 Using three target types—function-oriented, space-oriented, and population-oriented—as the guiding principles, a preliminary graph network structure database and a case database are constructed using reinforcement learning models. Through multiple rounds of automatic adjustment and iteration of the graph network structure (each iteration randomly adds or removes a node from the graph network structure to change the composition of potential regeneration units), the graph network structure of the potential regeneration units is until it conforms to one of the three target types: function-oriented, space-oriented, or population-oriented. Finally, a database of urban potential regeneration unit partitioning results is constructed.

[0012] Step S4: Delineation and Revision of Potential Urban Renewal Zones 4.1 Based on the urban potential renewal unit division result database constructed in step three, spatially adjacent urban potential renewal units are identified as the same urban potential renewal area.

[0013] 4.2 Based on the data of urban main and secondary roads and administrative divisions, the boundary of the potential urban renewal area is automatically corrected by the edge morphology optimization algorithm to obtain the final result of the urban potential renewal area division.

[0014] 4.3 The 3D modeling engine is used to classify and output the drawings of potential urban renewal areas and visualize them, and then output them to the interactive display platform.

[0015] Step S5: Automatic output and interactive display of urban potential redevelopment zone identification and delineation results. To create a holographic projection sand table, interactively display the urban renewal area identification and delineation scheme driven by artificial intelligence, the required equipment includes a three-dimensional physical sand table, a digital holographic projector, and a gesture recognition device.

[0016] Further, step S1 unifies data dimensions and removes redundant data; specifically, it includes: first, unifying the spatial reference by converting all vector data to the CGCS2000 coordinate system and resampling raster data to a resolution of 0.5 meters; then, normalizing the values ​​by performing Min-Max standardization on continuous variables such as LBS population heat values ​​and building volume ratio, using the formula: X_norm = (X - X_min) / (X_max - X_min), where X_norm is the standardized value, X is the original value, X_min is the minimum value of all data to be processed, and X_max is the maximum value of all data to be processed; the operation of removing redundant data specifically includes: spatial redundancy removal by deleting micro-plots with an area less than 10㎡ and geometrically overlapping features with a tolerance of less than 0.1 meters for duplicate nodes; attribute redundancy cleaning by deleting entire columns of fields with a missing rate greater than 30% and filling missing values ​​with the mode for discrete fields.

[0017] Furthermore, step S2 constructs an indicator system from three dimensions: architectural space, business format composition, and population vitality, and performs quantitative calculations of the indicators for all land parcels; the specific list is as follows. , The method of grading by numerical natural discontinuity of the indicator system initially identifies potential redevelopment objects based on land use units and then grades these potential redevelopment objects. Specifically, this includes adding a grading attribute field to each potential redevelopment object land use unit to record its specific potential level. In GIS, based on the comprehensive potential evaluation value calculated by the indicator system, the vector surface data of all land use units within the target area are statistically graded; the number of grades is set to 3 to 5, with each grade corresponding to a potential interval. Land use units belonging to the highest 1 to 2 potential intervals are automatically identified and marked as potential redevelopment objects.

[0018] The DBSCAN clustering algorithm is used to intelligently cluster the initially identified potential update objects, further classifying them. Specifically, this includes: establishing a hierarchical classification database for urban potential update objects; first, converting the spatial centroid coordinates of the potential update objects to planar coordinates in a projected coordinate system as input data; dynamically determining the neighborhood search radius Eps using a kernel density estimation algorithm, where Eps = 0.5 × (mean KDE peak distance); then setting the minimum number of clustering units MinPts = max(3, total number of units × 0.1%); finally, generating clusters for dense areas with a spatial distance less than Eps and a unit number ≥ MinPts, marking them as core update areas; marking units that do not meet the density threshold but are adjacent to the core area as edge update areas; and marking the remaining discrete units as isolated update areas. A classification coding field is generated for each cluster, storing the three-level classification identifiers for core area / edge area / isolated area.

[0019] Further, step S3 categorizes urban renewal units in the database into three types—function-oriented, space-oriented, and population-oriented—based on three dimensions: architectural spatial characteristics, business format characteristics, and population vitality characteristics. Specifically, this includes: firstly, extracting the following characteristic indicators from the multi-source database of potential urban renewal units constructed in step S1: architectural spatial characteristics, business format characteristics, and population vitality characteristics. The data for each urban renewal unit is represented as follows: , Among them, It is an update unit Attribute vectors, It is an indicator dimension; Then, after standardizing all indicators, a clustering algorithm was used to divide the update units into three types of orientations; the K-Means clustering algorithm was used, with a set number of clusters. These correspond to space-oriented, function-oriented, and population-oriented clusters, respectively; based on the clustering results, each update unit is assigned to its corresponding type-oriented cluster. , in, and These are the minimum and maximum values ​​of the j-th feature, respectively. This refers to the standardized value, which ranges from [0, 1]. Indicates the first Each urban renewal unit This indicates the total number of urban renewal units. Indicates the first The attribute vector of each update unit, Indicates the center of the j-th cluster; Step S3 uses a combination of geospatial clustering and concave hull algorithms to initially divide scattered potential urban renewal objects into potential urban renewal units and construct a preliminary graph network structure database; specifically, it includes: first, determining the geographic coordinates of each potential renewal object. and the property list of each object , For the first The attribute vector of a potential update object. For attribute vectors The specific content; then the DBSCAN algorithm is used to cluster the potential update objects, and... Set as the distance threshold, representing the maximum distance between objects. Set as the minimum number of points threshold, representing the minimum number of objects required to form a cluster; if a certain object If the number of points in the neighborhood is less than the minimum number of points threshold, then The core point is defined, and the core point and its density-reachable points are grouped into the same cluster; , in, Indicates the distance threshold Interior and Point The number of adjacent points; Then, the concave hull algorithm is used to generate boundaries for each clustering result, given a set of points. The concave hull is a containment structure. For each point, the smallest convex polygon is used to generate the concave hull boundary using the QuickHull or Graham scan algorithm; finally, each potential update object is treated as a node. Define the connection lines between nodes based on spatial relationships. As edges, use an undirected graph This represents a graph network, where V is the set of nodes and E is the set of edges. Step S3, guided by three target types—space-oriented, function-oriented, and population-oriented—constructs a preliminary graph network structure database and a case database using a reinforcement learning model. This is achieved through multiple rounds of automatic adjustment and iteration of the graph network structure. Specifically, this includes: first, designing the reinforcement learning model; and second, based on the graph network structure described in step S3... The attribute list for each node is as follows The current unit's type-oriented score is Randomly select a node from outside the graph network. Add a node to the current cell, and randomly remove a node from the current cell. Then, the reward is calculated based on the degree of matching between the graph network structure and the goal orientation; finally, multiple rounds of adjustments and iterations are performed; the initial state is set to the graph network generated in step S3.2. In each iteration, an action is randomly selected to generate a new graph network. Then calculate the new matching score. The reinforcement learning model is updated based on the reward function; when the matching score of the graph network... Once convergence is achieved or a preset threshold is reached, adjustment and iteration will cease. , , Where R is the reward value; It is the matching score between the graph network G and the goal-oriented approach; These are represented as the current and next steps in the graph network structure, respectively.

[0020] Furthermore, step S4 automatically corrects the boundaries of potential urban renewal areas using an edge morphology optimization algorithm. Specifically, this includes: firstly, merging road buffer zones by using the center lines of main and secondary urban roads as a reference to generate dynamic width buffer zones; secondly, performing spatial intersection calculations between the polygons of potential renewal areas and the administrative boundaries through administrative division topology verification; and thirdly, performing the following for renewal units that cross administrative boundaries: if the cross-area ratio is <15%, they are merged according to the administrative region with the largest area; if the cross-area ratio is ≥15%, they are divided into independent units along the administrative boundary; and finally, performing morphology optimization correction on the map patches. For micro-void areas with an area of ​​500-5000㎡ generated by the correction, automatic filling is performed based on neighborhood analysis, and a topology check report is generated.

[0021] Furthermore, step S5 establishes a holographic projection sand table to interactively display the urban renewal area identification and delineation scheme driven by artificial intelligence. The required equipment includes a 3D physical sand table, a digital holographic projector, and a gesture recognizer. Specifically, it includes: using ArcGIS software to construct a 3D physical sand table to realistically restore the urban terrain, building layout, and spatial structure as a basic display platform; based on Unity or Unreal Engine, combined with a holographic projection plugin, using digital holographic projection to accurately project the identification and delineation results of the urban renewal area onto the 3D sand table. The projected content includes the boundary of the renewal unit, type guidance information, and dynamically changing optimization schemes. Installing a gesture recognition device (Leap Motion or Kinect) captures user gestures (including clicking, dragging, zooming, rotating, etc.) to enable interactive operations with the holographic sand table. Users can zoom and rotate the sand table view using gestures, select update units to view detailed attribute information, switch between different division schemes, and dynamically simulate the evolution of the update area. Finally, the system automatically extracts the division results from the database and generates display content, including spatial range, type labels, and attribute data. It also supports exporting as images, videos, or spatial data files for use in urban planning meetings, public participation, and educational research. The beneficial effects of this invention are:

[0022] 1. This invention overcomes the bottlenecks of data silos and experience-based dependence in traditional update identification by employing a dual-database co-construction mechanism and a three-dimensional indicator system for quantification. It performs spatial benchmark unification, temporal benchmark conversion, and attribute structured association on heterogeneous data such as building vector morphology, POI business types, and LBS population data, eliminating the dimensional barriers between multi-source data. Simultaneously, it achieves a substantial improvement in identification accuracy through cross-validation of three-dimensional indicators of building space, business type composition, and population vitality, avoiding single-dimensional misjudgments. Furthermore, a case study city database provides a typological reference benchmark, driving the self-optimization of the identification logic.

[0023] 2. This invention achieves a revolutionary reconstruction of the intelligent aggregation paradigm for update units, pioneering a graph network reinforcement learning optimization framework that solves the problems of morphological fragmentation and functional mismatch in update unit partitioning. Reinforcement learning drives aggregation, establishing a reward function guided by three types: function, space, and population. Through iterative addition and subtraction of nodes, the unit structure converges to the optimal type matching state. Organic boundary generation, automatic merging of spatially adjacent units, and triple processing using edge morphology optimization algorithms achieve a paradigm shift from "mechanical aggregation" to "intelligent growth" in unit partitioning.

[0024] 3. This invention achieves a closed-loop digital process and intelligent decision support. Through a 3D interactive system and a reinforcement learning iterative mechanism, an evolvable planning and decision-making platform is constructed. A holographic decision-making sandbox system is built, with a 3D physical sandbox base and dynamic holographic projection overlay. Gesture recognition supports real-time scheme segmentation and attribute querying. A case library with continuous optimization capabilities is built; new data automatically triggers graph network retraining, and boundary optimization algorithm parameters self-adjust. This effect forms a complete intelligent closed loop of "recognition-optimization-display-iteration," empowering dynamic adjustment of planning. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a hierarchical classification and identification pattern diagram of potential urban renewal objects; Figure 3 It is a network structure pattern diagram of potential urban renewal units; Figure 4 This is a diagram illustrating the intelligent division and optimization model for potential urban renewal units. Figure 5 It is a diagram illustrating the delineation and revision model of potential urban renewal areas; Figure 6 It is an automatic output and interactive display map of the results of identifying and delineating potential urban renewal areas. Detailed Implementation

[0026] 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.

[0027] An AI-driven method for identifying and delineating potential urban regeneration areas, such as Figure 1-6 As shown, it includes the following steps: I. Collect existing building data (including vector morphology and age information), land use data, POI business data, LBS population data, high-resolution remote sensing imagery data, and road network data for the target city. Standardize data units, eliminate redundant data, and construct a multi-source large database for the target city. Collect data from case urban renewal units and construct a multi-source large database for these units.

[0028] The unified data units and elimination of redundant data include: firstly, unifying the spatial reference by converting all vector data to the CGCS2000 coordinate system and resampling raster data to a 0.5-meter resolution; secondly, normalizing the numerical values ​​by performing Min-Max standardization on continuous variables such as LBS population heat values ​​and building volume ratio, using the formula: X_norm = (X - X_min) / (X_max - X_min), where X_norm is the standardized value, X is the original value, X_min is the minimum value of all data to be processed, and X_max is the maximum value of all data to be processed; and thirdly, eliminating redundant data includes: spatial redundancy removal by deleting micro-plots with an area less than 10㎡ and geometrically overlapping features with a tolerance of less than 0.1 meters; and attribute redundancy cleaning by deleting entire columns of fields with a missing rate greater than 30% and filling missing values ​​in discrete fields using the mode.

[0029] Second, based on the target city's multi-source big data constructed in Step One, an indicator system is built using the smallest land parcel as the unit, encompassing three dimensions: building space, business format composition, and population vitality. The indicators for all land parcels are then quantitatively calculated. Based on the numerical natural discontinuity method of the indicator system, potential urban renewal objects are initially identified at the land use level and classified. The DBSCAN clustering algorithm is then used to intelligently cluster the initially identified potential renewal objects, further classifying them and establishing a hierarchical classification database of urban potential renewal objects.

[0030] The report describes an indicator system constructed from three dimensions: architectural space, business format composition, and population activity, to quantify the indicators for all land parcels. A detailed list is provided below. , The method of grading by numerical natural discontinuity of the indicator system initially identifies potential land use units as redevelopment objects and then grades them. Specifically, this includes adding a grading attribute field to each potential land use unit to record its specific potential level. In GIS, based on the comprehensive potential evaluation value calculated by the indicator system, the vector surface data of all land use units within the target area are statistically graded; the number of grades is set to 3 to 5, with each grade corresponding to a potential interval. Land use units belonging to the highest 1 to 2 potential intervals are automatically identified and marked as potential redevelopment objects.

[0031] The DBSCAN clustering algorithm is used to intelligently cluster the initially identified potential update objects, further classifying them. Specifically, this involves establishing a hierarchical classification database for urban potential update objects. First, the spatial centroid coordinates of the potential update objects are converted to planar coordinates in a projected coordinate system as input data. The neighborhood search radius Eps is dynamically determined using a kernel density estimation algorithm, where Eps = 0.5 × (mean KDE peak distance). Then, the minimum number of clustering units, MinPts = max(3, total number of units × 0.1%), is set. Finally, dense areas with a spatial distance less than Eps and a unit number ≥ MinPts are generated into clusters and marked as core update areas. Units that do not meet the density threshold but are adjacent to the core area are marked as edge update areas, and the remaining discrete units are marked as isolated update areas. A classification coding field is generated for each cluster, storing the three-level classification identifiers for core area / edge area / isolated area.

[0032] Third, based on the multi-source big data database of potential urban renewal units constructed in Step One, the urban renewal units in the database are divided into three types based on three dimensions: business characteristics, architectural characteristics, and population characteristics: function-oriented, space-oriented, and population-oriented. Using a combination of geospatial clustering and concave hull algorithms, scattered potential urban renewal objects are initially divided into potential urban renewal units, and a preliminary graph network structure database is constructed. Each renewal object within a potential urban renewal unit is treated as a node, each node has its own attribute list, and the spatial relationships between adjacent renewal objects are used as connecting lines, thus constructing a preliminary graph network structure database. Furthermore, the same method is used to construct a graph network structure database for the urban renewal unit case study database. With three target types—function-oriented, space-oriented, and population-oriented—a reinforcement learning model is constructed to create a preliminary graph network structure database and a case database. Through multiple rounds of automatic adjustment and iteration of the graph network structure (each iteration randomly adds or removes a node from the graph network structure to change the composition of potential regeneration units), the graph network structure of the potential regeneration units is until it conforms to one of the three target types. Finally, a database of urban potential regeneration unit partitioning results is constructed.

[0033] Based on three dimensions—architectural spatial characteristics, business format characteristics, and population vitality characteristics—urban renewal units in the database are categorized into three orientations: function-oriented, space-oriented, and population-oriented. Specifically, this involves extracting the following characteristic indicators from the multi-source database of potential urban renewal units constructed in step S1: architectural spatial characteristics, business format characteristics, and population vitality characteristics. The data for each urban renewal unit is represented as follows: , Among them, It is an update unit Attribute vectors, It is an indicator dimension; Then, after standardizing all indicators, a clustering algorithm was used to divide the update units into three types of orientations; the K-Means clustering algorithm was used, with a set number of clusters. These correspond to space-oriented, function-oriented, and population-oriented clusters, respectively; based on the clustering results, each update unit is assigned to its corresponding type-oriented cluster. , in, and These are the minimum and maximum values ​​of the j-th feature, respectively. This refers to the standardized value, which ranges from [0, 1]. Indicates the first Each urban renewal unit This indicates the total number of urban renewal units. Indicates the first The attribute vector of each update unit, Indicates the center of the j-th cluster; By combining geospatial clustering and concave hull algorithms, scattered potential urban regeneration objects are initially divided into potential urban regeneration units, and a preliminary graph network structure database is constructed. Specifically, this includes: first, determining the geographic coordinates of each potential regeneration object. and the property list of each object , For the first The attribute vector of a potential update object. For attribute vectors The specific content; then the DBSCAN algorithm is used to cluster the potential update objects, and... Set as the distance threshold, representing the maximum distance between objects. Set as the minimum number of points threshold, representing the minimum number of objects required to form a cluster; if a certain object If the number of points in the neighborhood is less than the minimum number of points threshold, then The core point is defined, and the core point and its density-reachable points are grouped into the same cluster; , in, Indicates the distance threshold Interior and Point The number of adjacent points; Then, for each clustering result, a concave hull algorithm is used to generate the boundary. Given a point set S, the concave hull is the smallest convex polygon containing all points in S. The concave hull boundary is generated using either the QuickHull or Graham scan algorithm. Finally, each potential update object is treated as a node. Define the connection lines between nodes based on spatial relationships. As edges, use an undirected graph This represents a graph network, where V is the set of nodes and E is the set of edges. With spatially oriented, functionally oriented, and population-oriented as the target orientations, a preliminary graph network structure database and a case database reinforcement learning model are constructed. This is achieved through multiple rounds of automatic adjustment and iteration of the graph network structure. Specifically, this includes: first, designing the reinforcement learning model; and second, based on the graph network structure described in step S3... The attribute list for each node is as follows The current unit's type-oriented score is Randomly select a node from outside the graph network. Add a node to the current cell, and randomly remove a node from the current cell. Then, the reward is calculated based on the degree of matching between the graph network structure and the goal orientation; finally, multiple rounds of adjustments and iterations are performed; the initial state is set to the graph network generated in step S3.2. In each iteration, an action is randomly selected to generate a new graph network. Then calculate the new matching score. The reinforcement learning model is updated based on the reward function; when the matching score of the graph network... Once convergence is achieved or a preset threshold is reached, adjustment and iteration will cease. , , Where R is the reward value; It is the matching score between the graph network G and the goal-oriented approach; These are represented as the current and next steps in the graph network structure, respectively.

[0034] Fourth, based on the urban potential renewal unit delineation database constructed in step three, spatially adjacent urban potential renewal units are identified as belonging to the same urban potential renewal area. According to urban main and secondary road data and administrative division data, the boundaries of urban potential renewal areas are automatically corrected using an edge morphology optimization algorithm, resulting in the final urban potential renewal area delineation results. The urban potential renewal area scheme drawings are then categorized, output, and visualized using a 3D modeling engine, and displayed on an interactive display platform.

[0035] The algorithm automatically corrects the boundaries of potential urban renewal areas using edge morphology optimization. Specifically, this involves: first, merging road transition zones by generating dynamic width buffer zones based on the centerlines of main and secondary urban roads; second, performing spatial intersection calculations between the polygons of potential renewal areas and administrative boundaries through administrative division topology verification; and third, for renewal units crossing administrative boundaries, merging them according to the largest administrative region if the cross-regional area ratio is less than 15%, and dividing them into independent units along the administrative boundary if the cross-regional area ratio is ≥15%. Finally, morphological optimization correction is performed on the map features. For micro-void areas with areas between 500-5000㎡ resulting from the correction, automatic filling is performed based on neighborhood analysis, and a topology check report is generated.

[0036] 5. Establish a holographic projection sand table to interactively demonstrate the urban renewal area identification and delineation scheme driven by artificial intelligence. The required equipment includes a three-dimensional physical sand table, a digital holographic projector, and a gesture recognition device.

[0037] The establishment of a holographic projection sand table to interactively showcase AI-driven urban renewal area identification and delineation schemes requires equipment including a 3D physical sand table, a digital holographic projector, and a gesture recognition device. Specifically, this involves using ArcGIS software to construct a 3D physical sand table that realistically recreates the city's topography, building layout, and spatial structure, serving as the basic display platform. Based on Unity or Unreal Engine, and combined with a holographic projection plugin, digital holographic projection accurately projects the identification and delineation results of the urban renewal area onto the 3D sand table. The projected content includes update unit boundaries, type guidance information, and dynamically changing optimization schemes. A gesture recognition device (Leap Motion or Kinect) is installed to capture user gestures (including clicking, dragging, zooming, and rotating) to enable interactive operations with the holographic sand table. Users can zoom and rotate the sand table view, select update units to view detailed attribute information, switch between different delineation schemes, and dynamically simulate the evolution of the renewal area. Finally, the system automatically extracts the delineation results from the database and generates display content, including spatial extent, type labels, and attribute data. It also supports exporting as image, video, or spatial data files for use in urban planning meetings, public participation, and educational research.

[0038] Example The technical solution of this invention will be described in detail below using a certain area in Nanjing as an example.

[0039] (1) Taking a certain area in Nanjing as an example, existing building data (including vector morphology and age information), land use data, POI business data, LBS population data, high-definition remote sensing image data, and road network data were collected. Data dimensions were standardized, redundant data was eliminated, and a multi-source big data database for the area in Nanjing was constructed. Data from the case study urban renewal unit was collected, and a multi-source big data database for the case study urban renewal unit was constructed. Specifically, this included: First, the spatial benchmark was unified by converting all vector data to the CGCS2000 coordinate system, and raster data was resampled to a resolution of 0.5 meters. Then, the numerical values ​​were normalized by performing Min-Max standardization on continuous variables such as LBS population heat values ​​and building volume ratio, using the formula: X_norm = (X - X_min) / (X_max - X_min), where X_norm is the standardized value, X is the original value, X_min is the minimum value of all data to be processed, and X_max is the maximum value of all data to be processed. Next, redundant data was removed, specifically including: spatial redundancy removal by deleting micro-plots with an area less than 10 square meters and geometrically overlapping features with a tolerance of less than 0.1 meters; and attribute redundancy cleaning by deleting entire columns of fields with a missing value rate greater than 30% and filling missing values ​​in discrete fields using the mode.

[0040] (2) Based on the constructed multi-source big data of a certain area in Nanjing, an indicator system was constructed from three dimensions—building space, business format composition, and population vitality—using the smallest land parcel as the unit, and the indicators of all land parcels were quantitatively calculated. According to the numerical natural discontinuity method of the indicator system, potential urban renewal objects were initially identified based on land use, and these potential renewal objects were classified. The DBSCAN clustering algorithm was used to intelligently cluster the initially identified potential renewal objects, further classifying them and establishing a hierarchical classification database of potential urban renewal objects in Nanjing. Specifically, this includes: An indicator system was constructed from three dimensions—architectural space, business format composition, and population vitality—to quantify the indicators for all land parcels in Nanjing. The specific list is as follows. , The method of grading by numerical natural discontinuity of the indicator system initially identifies potential land use units as redevelopment objects and then grades them. Specifically, this includes adding a grading attribute field to each potential land use unit to record its specific potential level. In GIS, based on the comprehensive potential evaluation value calculated by the indicator system, the vector surface data of all land use units within the target area of ​​Nanjing City are statistically graded; the number of grades is set to 3 to 5, with each grade corresponding to a potential interval. Land use units belonging to the highest 1 to 2 potential intervals are automatically identified and marked as potential redevelopment objects.

[0041] The DBSCAN clustering algorithm is used to intelligently cluster the initially identified potential update objects, further classifying them. Specifically, this involves establishing a hierarchical classification database for urban potential update objects. First, the spatial centroid coordinates of the potential update objects are converted to planar coordinates in a projected coordinate system as input data. The neighborhood search radius Eps is dynamically determined using a kernel density estimation algorithm, where Eps = 0.5 × (mean KDE peak distance). Then, the minimum number of clustering units, MinPts = max(3, total number of units × 0.1%), is set. Finally, dense areas with a spatial distance less than Eps and a unit number ≥ MinPts are generated into clusters and marked as core update areas. Units that do not meet the density threshold but are adjacent to the core area are marked as edge update areas, and the remaining discrete units are marked as isolated update areas. A classification coding field is generated for each cluster, storing the three-level classification identifiers for core area / edge area / isolated area.

[0042] (3) Based on the constructed multi-source big data of potential urban renewal units in a certain area of ​​Nanjing, the urban renewal units in the database are divided into three types of orientation: function-oriented, space-oriented, and population-oriented, using three dimensions: business characteristics, architectural characteristics, and population characteristics. Using a combination of geospatial clustering and concave hull algorithms, scattered potential urban renewal objects are initially divided into potential urban renewal units, and a preliminary graph network structure database is constructed. Each renewal object within a potential urban renewal unit is used as a node, each node has its own attribute list, and the spatial relationships between adjacent renewal objects are used as connecting lines, thus constructing a preliminary graph network structure database. Furthermore, the same method is used to construct a graph network structure database for the urban renewal unit data case library. With three target types—function-oriented, space-oriented, and population-oriented—a reinforcement learning model is constructed to create a preliminary graph network structure database and a case database. Through multiple rounds of automatic adjustment and iteration of the graph network structure (each iteration randomly adds or removes a node from the graph network structure to change the composition of potential regeneration units), the graph network structure of the potential regeneration units is until it conforms to one of the three target types. Finally, a database of urban potential regeneration unit partitioning results is constructed.

[0043] Specifically, the urban renewal units in the database are categorized into three orientations—functional, spatial, and population-oriented—based on three dimensions: architectural spatial characteristics, business format characteristics, and population vitality characteristics. This includes: firstly, extracting the following characteristic indicators from the multi-source database of potential urban renewal units constructed in step S1: architectural spatial characteristics, business format characteristics, and population vitality characteristics. The data for each urban renewal unit is represented as follows: , Among them, It is an update unit Attribute vectors, It is an indicator dimension; Then, after standardizing all indicators, a clustering algorithm was used to divide the update units into three types of orientations; the K-Means clustering algorithm was used, with a set number of clusters. These correspond to space-oriented, function-oriented, and population-oriented clusters, respectively; based on the clustering results, each update unit is assigned to its corresponding type-oriented cluster. , in, and These are the minimum and maximum values ​​of the j-th feature, respectively. This refers to the standardized value, which ranges from [0, 1]. Indicates the first Each urban renewal unit This indicates the total number of urban renewal units. Indicates the first The attribute vector of each update unit, Indicates the center of the j-th cluster; By combining geospatial clustering and concave hull algorithms, scattered potential urban regeneration objects are initially divided into potential urban regeneration units, and a preliminary graph network structure database is constructed. Specifically, this includes: first, determining the geographic coordinates of each potential regeneration object. and the property list of each object , For the first The attribute vector of a potential update object. For attribute vectors The specific content; then the DBSCAN algorithm is used to cluster the potential update objects, and... Set as the distance threshold, representing the maximum distance between objects. Set as the minimum number of points threshold, representing the minimum number of objects required to form a cluster; if a certain object If the number of points in the neighborhood is less than the minimum number of points threshold, then The core point is defined, and the core point and its density-reachable points are grouped into the same cluster; , in, Indicates the distance threshold Interior and Point The number of adjacent points; Then, for each clustering result, a concave hull algorithm is used to generate the boundary. Given a point set S, the concave hull is the smallest convex polygon containing all points in S. The concave hull boundary is generated using either the QuickHull or Graham scan algorithm. Finally, each potential update object is treated as a node. Define the connection lines between nodes based on spatial relationships. As edges, use an undirected graph This represents a graph network, where V is the set of nodes and E is the set of edges. With spatially oriented, functionally oriented, and population-oriented as the target orientations, a preliminary graph network structure database and a case database reinforcement learning model are constructed. This is achieved through multiple rounds of automatic adjustment and iteration of the graph network structure. Specifically, this includes: first, designing the reinforcement learning model; and second, based on the graph network structure described in step S3... Each node's attribute list is as follows: The current unit's type-oriented score is Randomly select a node from outside the graph network. Add a node to the current cell, and randomly remove a node from the current cell. Then, the reward is calculated based on the degree of matching between the graph network structure and the goal orientation; finally, multiple rounds of adjustments and iterations are performed; the initial state is set to the graph network generated in step S3.2. In each iteration, an action is randomly selected to generate a new graph network. Then calculate the new matching score. The reinforcement learning model is updated based on the reward function; when the matching score of the graph network... Once convergence is achieved or a preset threshold is reached, adjustment and iteration will cease. , , Where R is the reward value; It is the matching score between the graph network G and the goal-oriented approach; These are represented as the current and next steps in the graph network structure, respectively.

[0044] (4) Based on the constructed database of potential urban renewal unit division results for a certain area in Nanjing, spatially adjacent potential urban renewal units are identified as the same potential urban renewal area. According to the data of urban main and secondary roads and administrative division data, the boundary of the potential urban renewal area is automatically corrected through the edge morphology optimization algorithm to obtain the final division result of the potential urban renewal area. The 3D modeling engine is used to classify and output the drawings of the potential urban renewal area scheme and visualize them, and output them to the interactive display platform.

[0045] The process involves automatically correcting the boundaries of potential urban renewal areas using an edge morphology optimization algorithm. Specifically, this includes: first, merging road buffer zones by generating a dynamic width buffer zone based on the centerlines of main and secondary roads; second, performing spatial intersection calculations between the polygons of potential renewal areas and administrative boundaries through administrative division topology verification; and third, for renewal units crossing administrative boundaries, the following steps are applied: if the cross-boundary area ratio is <15%, they are merged according to the administrative region with the largest area; if the cross-boundary area ratio is ≥15%, they are divided into independent units along the Nanjing municipal administrative boundary. Finally, morphological optimization correction is performed on the map features. For micro-void areas with areas between 500-5000㎡ resulting from the correction, automatic filling is performed based on neighborhood analysis, and a topology check report is generated. (5) Establish a holographic projection sand table to interactively display the identification and division scheme of an urban renewal area in a certain area of ​​Nanjing City driven by artificial intelligence. The required equipment includes a three-dimensional physical sand table, a digital holographic projector and a gesture recognition device.

[0046] The establishment of a holographic projection sand table to interactively display the identification and delineation schemes of urban renewal areas driven by artificial intelligence requires equipment including a 3D physical sand table, a digital holographic projector, and a gesture recognizer. Specifically, this involves using ArcGIS software to construct a 3D physical sand table that realistically recreates the urban topography, building layout, and spatial structure of a certain area in Nanjing, serving as the basic display platform. Based on Unity or Unreal Engine, and combined with a holographic projection plugin, digital holographic projection is used to accurately project the identification and delineation results of the urban renewal area onto the 3D sand table. The projected content includes the boundaries of the update units, type guidance information, and dynamically changing optimization schemes. A gesture recognizer (Leap Motion or Kinect) is installed to capture user gestures (including clicking, dragging, zooming, rotating, etc.) to enable interactive operations with the holographic sand table. Users can use gestures to zoom and rotate the sand table view, select update units to view detailed attribute information, switch between different delineation schemes, and dynamically simulate the evolution process of the update area. Ultimately, the system automatically extracts the division results from the database and generates display content, including spatial extent, type labeling, and attribute data. It also supports exporting as images, videos, or spatial data files for use in urban planning meetings, public participation, and educational research.

Claims

1. An AI-driven method for identifying and delineating potential urban renewal areas, characterized in that, Includes the following steps: Step S1: Construction and Update of Urban Multi-Source Big Data Unit Case Database Collect current building data, land use data, POI business data, LBS population data, high-definition remote sensing image data, and road network data of the target city, unify data dimensions, eliminate redundant data, and construct a multi-source big data database for the target city; collect data from case urban renewal units and construct a multi-source big data database for case urban renewal units. Step S2: Hierarchical and intelligent identification of potential urban renewal objects 2.1 Based on the target city multi-source big data constructed in step S1, an indicator system is constructed from three dimensions—building space, business format composition, and population vitality—using the smallest land plot as the unit, and the indicators are quantitatively calculated for all land plots. 2.2 Based on the numerical natural discontinuity method of the indicator system, potential urban renewal objects are initially identified by land use as the unit, and the potential urban renewal objects are classified. 2.3 The DBSCAN clustering algorithm is used to intelligently cluster the potential update objects initially identified, further classify the potential update objects, and establish a hierarchical classification database of urban potential update objects; Step S3: Intelligent Division and Optimization of Potential Urban Renewal Units 3.1 Based on the multi-source big data database of potential urban renewal units in the case city constructed in step S1, the urban renewal units in the database are divided into three types of orientation: function-oriented, space-oriented, and population-oriented, based on three dimensions of indicators: business characteristics, architectural characteristics, and population characteristics. 3.2 By combining geospatial clustering algorithm and concave hull algorithm, scattered potential urban renewal objects are initially divided into potential urban renewal units, and a preliminary graph network structure database is constructed. Using each renewal object within a potential urban renewal unit as a node, with each node having its own attribute list, and the spatial relationships between adjacent renewal objects as connecting lines, a preliminary graph network structure database is constructed; the same method is used to construct a graph network structure database for the urban renewal unit data case library. 3.3 Taking three types of target orientations—function-oriented, space-oriented, and population-oriented—as the guiding principles, a preliminary graph network structure database and a case database of graph network structure databases are constructed using reinforcement learning models. Through multiple rounds of automatic adjustment and iteration of the graph network structure, the graph network structure of potential regeneration units is adjusted until it conforms to one of the three orientations: function-oriented, space-oriented, or population-oriented. Finally, a database of urban potential regeneration unit division results is constructed. Step S4: Delineation and Revision of Potential Urban Renewal Zones 4.1 Based on the urban potential renewal unit division result database constructed in step S3, spatially adjacent urban potential renewal units are identified as the same urban potential renewal area; 4.2 Based on urban main and secondary road data and administrative division data, the boundary of the potential urban renewal area is automatically corrected by the edge morphology optimization algorithm to obtain the final urban potential renewal area division result; 4.3 The 3D modeling engine is used to classify and output the drawings of potential urban renewal areas and visualize them, which are then exported to an interactive display platform; Step S5: Automatic output and interactive display of urban potential redevelopment zone identification and delineation results. To create a holographic projection sand table, interactively display the urban renewal area identification and delineation scheme driven by artificial intelligence, the required equipment includes a three-dimensional physical sand table, a digital holographic projector, and a gesture recognition device.

2. The method for identifying and delineating potential urban renewal areas driven by artificial intelligence according to claim 1, characterized in that step S1 unifies data dimensions and clears redundant data, specifically including: firstly, unifying the spatial benchmark by converting all vector data to the CGCS2000 coordinate system and resampling raster data to 0.5-meter resolution; then normalizing the numerical values ​​by performing Min-Max standardization on continuous variables such as LBS population heat values ​​and building volume ratio, using the formula: X_norm = (X - X_min) / (X_max - X_min), where X_norm is the standardized value, X is the original value, X_min is the minimum value of all data to be processed, and X_max is the maximum value of all data to be processed; the operation of clearing redundant data specifically includes: spatial redundancy removal by deleting micro-patterns with an area less than 10㎡ and geometrically overlapping elements with a tolerance of less than 0.1 meters for duplicate nodes; attribute redundancy cleaning by deleting entire columns of fields with a missing rate greater than 30% and filling missing values ​​with the mode for discrete fields.

3. The method for identifying and delineating potential urban renewal areas driven by artificial intelligence according to claim 2, characterized in that step S2 constructs an indicator system from three dimensions: architectural space, business format composition, and population vitality, and performs quantitative calculations of indicators for all land parcels; the specific list is as follows. ， The method of grading by numerical natural discontinuity of the indicator system initially identifies potential redevelopment objects based on land use units and grades these potential redevelopment objects. Specifically, this includes: adding a grading attribute field to each potential redevelopment object land use unit to record its specific potential level; statistically grading all land use unit vector surface data in the target area based on the comprehensive potential evaluation value calculated by the indicator system in GIS; the number of grading levels is set to 3 to 5, with each level corresponding to a potential interval, and land use units belonging to the highest 1 to 2 potential intervals are automatically identified and marked as potential redevelopment objects. The DBSCAN clustering algorithm is used to intelligently cluster the initially identified potential update objects, further classifying them. Specifically, this includes: establishing a hierarchical classification database for potential urban update objects; first, converting the spatial centroid coordinates of the potential update objects into planar coordinates in the projected coordinate system as input data; dynamically determining the neighborhood search radius Eps using a kernel density estimation algorithm, where Eps = 0.5 × (mean KDE peak distance); then setting the minimum number of clustering units MinPts = max(3, total number of units × 0.1%); finally, generating clusters for dense areas with a spatial distance less than Eps and a unit number ≥ MinPts, marking them as core update areas; marking units that do not meet the density threshold but are adjacent to the core area as edge update areas; and marking the remaining discrete units as isolated update areas; generating a classification coding field for each cluster to store the three-level classification identifiers of core area / edge area / isolated area.

4. The method for identifying and dividing potential urban renewal areas driven by artificial intelligence according to claim 3, characterized in that step S3 divides urban renewal units in the database into three types of orientation: function-oriented, space-oriented, and population-oriented, based on three dimensions: architectural spatial characteristics, business format characteristics, and population vitality characteristics; specifically, it includes: firstly, extracting the following characteristic indicators from the multi-source big data database of potential urban renewal units constructed in step S1: architectural spatial characteristics, business format characteristics, and population vitality characteristics, with the data for each urban renewal unit represented as follows: ， Among them, It is an update unit Attribute vectors, It is an indicator dimension; Then, after standardizing all indicators, a clustering algorithm was used to divide the update units into three types of orientations; the K-Means clustering algorithm was used, with a set number of clusters. These correspond to space-oriented, function-oriented, and population-oriented clusters, respectively; based on the clustering results, each update unit is assigned to its corresponding type-oriented cluster. ， in, and These are the minimum and maximum values ​​of the j-th feature, respectively. This refers to the standardized value, which ranges from [0, 1]. Indicates the first Each urban renewal unit This indicates the total number of urban renewal units. Indicates the first The attribute vector of each update unit, Indicates the center of the j-th cluster; Step S3 uses a combination of geospatial clustering and concave hull algorithms to initially divide scattered potential urban renewal objects into potential urban renewal units and construct a preliminary graph network structure database; specifically, it includes: first, determining the geographic coordinates of each potential renewal object. and the property list of each object , For the first The attribute vector of a potential update object. For attribute vectors The specific content; then the DBSCAN algorithm is used to cluster the potential update objects, and... Set as the distance threshold, representing the maximum distance between objects. Set as the minimum number of points threshold, representing the minimum number of objects required to form a cluster; if a certain object If the number of points in the neighborhood is less than the minimum number of points threshold, then The core point is defined, and the core point and its density-reachable points are grouped into the same cluster; ， in, Indicates the distance threshold Interior and Point The number of adjacent points; Then, the concave hull algorithm is used to generate boundaries for each clustering result, given a set of points. The concave hull is a containment structure. For each point, the smallest convex polygon is used to generate the concave hull boundary using the QuickHull or Graham scan algorithm; finally, each potential update object is treated as a node. Define the connection lines between nodes based on spatial relationships. As edges, use an undirected graph This represents a graph network, where V is the set of nodes and E is the set of edges. Step S3, guided by three target types—space-oriented, function-oriented, and population-oriented—constructs a preliminary graph network structure database and a case database using a reinforcement learning model. This involves multiple rounds of automatic adjustment and iteration of the graph network structure. Specifically, this includes: first, designing the reinforcement learning model; and second, based on the graph network structure described in step S3... The attribute list for each node is as follows The current unit's type-oriented score is Randomly select a node from outside the graph network. Add a node to the current cell, and randomly remove a node from the current cell. Then, the reward is calculated based on the degree of matching between the graph network structure and the goal orientation; finally, multiple rounds of adjustments and iterations are performed; the initial state is set to the graph network generated in step S3.

2. In each iteration, an action is randomly selected to generate a new graph network. Then calculate the new matching score. The reinforcement learning model is updated based on the reward function; when the matching score of the graph network... Once convergence is achieved or a preset threshold is reached, adjustment and iteration will cease. ， ， Where R is the reward value; It is the matching score between the graph network G and the goal-oriented approach; These are represented as the current and next steps in the graph network structure, respectively.

5. The method for identifying and delineating potential urban renewal areas driven by artificial intelligence according to claim 4, characterized in that, Step S4 automatically corrects the boundaries of potential urban renewal areas using an edge morphology optimization algorithm. Specifically, it includes: first, merging road buffer zones by generating a dynamic width buffer zone based on the centerline of the city's main and secondary roads; then, performing spatial intersection calculations between the potential renewal area polygons and the administrative boundary through administrative division topology verification; for renewal units crossing administrative boundaries, if the cross-area ratio is <15%, merging them according to the administrative region with the largest area; if the cross-area ratio is ≥15%, dividing them into independent units along the administrative boundary; finally, morphological optimization correction of the map features; and for micro-void areas with areas between 500-5000㎡ resulting from the correction, automatic filling based on neighborhood analysis and generation of a topology check report.

6. The method for identifying and delineating potential urban renewal areas driven by artificial intelligence according to claim 5, characterized in that, Step S5 establishes a holographic projection sand table to interactively display the urban renewal area identification and delineation scheme driven by artificial intelligence. The required equipment includes a 3D physical sand table, a digital holographic projector, and a gesture recognizer. Specifically, this includes: constructing a 3D physical sand table using ArcGIS software to realistically reproduce the urban topography, building layout, and spatial structure as a basic display platform; using Unity or Unreal Engine, combined with a holographic projection plugin, accurately projecting the identification and delineation results of the urban renewal area onto the 3D sand table using digital holographic projection. The projected content includes the boundaries of the renewal units, type guidance information, and dynamically changing optimization schemes; installing a gesture recognizer to capture user gestures, enabling interactive operations with the holographic sand table. Users can zoom and rotate the sand table view, select renewal units to view detailed attribute information, switch between different delineation schemes, and dynamically simulate the evolution process of the renewal area using gestures; finally, the system automatically extracts the delineation results from the database and generates display content, including spatial range, type labels, and attribute data. It also supports exporting as image, video, or spatial data files for use in urban planning meetings, public participation, and educational research scenarios.

Images