Low-efficiency land use redevelopment function prediction method based on graph neural network

Abstract

The invention discloses a low-efficiency land use redevelopment function prediction method based on a graph neural network, and the method comprises the steps: S1, obtaining the land use benefit of each space unit in a learning region, dividing each space unit into a high land use benefit type, a middle land use benefit type and a low land use benefit type, selecting a space unit with a high land utilization benefit type as an urban land use configuration case library; s2, taking the urban land use configuration case library obtained in the step S1 as a sample set, and constructing and training a functional feature learning model based on a graph neural network according to the sample set; and S3, obtaining each space unit in the prediction area, inputting the space units into the functional feature learning model obtained in the step S2, and performing low-efficiency land use redevelopment function prediction. By simulating urban space function interaction and mechanism innovation, a traditional land utilization prediction normal form based on static rules and spatio-temporal evolution is broken through, and scientific support is provided for land benefit improvement and function replacement decision in urban updating.

Description

Technical Field

[0001] This invention relates to the field of urban land management technology, and in particular to a method for predicting the redevelopment function of inefficient land use based on graph neural networks. Background Technology

[0002] In the later stages of urbanization, urban construction land expands rapidly, accompanied by problems such as extensive use, irrational land use, and low output of developed land resources. One goal of redeveloping inefficient land is to determine the most suitable land use plan for the redevelopment plots based on the existing urban spatial structure and land use pattern. By rationally allocating the composition and proportion of residential, commercial, green space, public service, industrial, and transportation land, the overall efficiency of the area can be improved and urban functions can be continuously optimized to better meet the needs of future socio-economic activities.

[0003] Currently, urban land use simulation typically follows a technical process of "data acquisition—model construction—parameter calibration—scenario simulation—evaluation and verification." Based on remote sensing imagery, land use status maps, and socio-economic statistics, it selects macro-level variables such as social, economic, and environmental factors as the main driving factors. Models such as Markov chains, cellular automata (CA), and CLUE-S are constructed to simulate the spatial evolution of land use on homogeneous raster cells. Subsequently, historical data is used to calibrate and optimize the model parameters, and future land use patterns are predicted under different defined development scenarios.

[0004] However, in urban renewal, the aforementioned land use simulation has the following inherent limitations in supporting the redevelopment of inefficient land: (1) Insufficient spatial representation capability: Traditional models usually construct spatial units based on large-scale grids, which makes it difficult to accurately depict the shape, attributes and spatial relationships of urban plots and cannot meet the needs of refined analysis; (2) Incomplete consideration of evolutionary mechanisms: Existing models usually simulate the evolutionary process based on historical driving factors and static interaction rules, lacking consideration of the impact of residents' needs and spatial behavior on land use, which may lead to problems such as demand response disconnect; (3) Lack of clear goal orientation: Traditional methods focus on simulating the natural evolution trend of land use. When predicting inefficient land use, there is no optimization mechanism with "efficiency improvement" as the goal, which cannot support redevelopment decisions.

[0005] Therefore, there is an urgent need for an innovative technical framework that can overcome the above limitations and achieve refined, dynamic, and goal-oriented prediction of the redevelopment functions of inefficient land. Summary of the Invention

[0006] Purpose of the invention: To address the above-mentioned shortcomings, this invention proposes a method for predicting the redevelopment functions of inefficient land based on graph neural networks. By simulating the interaction of urban spatial functions and innovating mechanisms, it breaks through the traditional prediction paradigm based on static rules and spatiotemporal evolution, providing scientific support for land efficiency improvement and functional replacement decisions in urban renewal.

[0007] Technical solution: This invention provides a method for predicting the redevelopment function of inefficient land use based on graph neural networks, comprising: S1. Obtain the land use benefits of each spatial unit in the learning area, and divide each spatial unit into three types of land use benefits: high, medium and low. Select the spatial unit with high land use benefits as the urban land allocation case library. S2. Using the urban land use configuration case library obtained in S1 as the sample set, a functional feature learning model based on graph neural networks is constructed as follows: (1) Obtain the geometric morphological features, location features, population density features and functional activity attribute features of each spatial unit in the sample set, and fuse them to generate a node feature matrix of multidimensional data; (2) Construct a graph structure based on travel trajectory data, including multi-type weighted graphs and inductive weighted graphs. The inductive weighted graph is constructed using the efficient land use obtained from the multi-type weighted graphs as nodes. (3) Construct a neighborhood feature fusion layer for the graph neural network based on the multi-type weighted graph, and construct a functional semantic learning layer for the graph neural network based on the inductive weighted graph, thereby obtaining the graph neural network; Using the urban land use configuration case library obtained in S1 as a sample set, the graph neural network is input into the graph neural network for training to obtain a functional feature learning model. S3. Obtain each spatial unit in the prediction area and input it into the functional feature learning model obtained in S2 to perform functional prediction of inefficient land redevelopment.

[0008] Specifically, in S2, the urban land use configuration case library obtained in S1 is used as a sample set, which is divided into a training set, a validation set, and a test set according to a set ratio. The sample set is then input into the constructed graph neural network for training to obtain a functional feature learning model.

[0009] More specifically, the training of the graph neural network adopts a staged training mechanism, as follows: The training for the first set number of epochs uses binary cross-entropy loss for functional existence classification learning. The training for the subsequent set number of epochs introduces weighted MAE loss and uses the Adam optimizer for training. The initial learning rate is set to 1×10. -3 The weight decay rate is set to 1×10 -4 During training, a cosine annealing strategy is introduced to dynamically adjust the learning rate, and an early stopping mechanism is also introduced.

[0010] Specifically, the geometric features of each spatial unit are obtained by combining the area of ​​the spatial unit with its contour compactness index; wherein, the contour compactness index of the spatial unit is calculated as follows: C = 4πA / (P) 2 ; Where C is the contour compactness index of the spatial unit, A is the area of ​​the spatial unit, and P is the perimeter of the spatial unit.

[0011] Specifically, the latitude and longitude coordinates of the centroid of the spatial unit after range normalization are used as its location characteristics, and the area and air profile compactness index of the spatial unit after logarithmic transformation and range normalization are used as its geometric morphological characteristics.

[0012] Specifically, based on the grid population data provided by mobile phone signaling data within a spatial unit, the population data is mapped to the corresponding spatial unit using the land fragmentation method, and the population density of the spatial unit is calculated by combining the area of ​​the spatial unit, thus obtaining the population density characteristics of the spatial unit.

[0013] Specifically, by statistically analyzing the data of points of interest within a spatial unit according to major categories and calculating the frequency proportion of each category of points of interest, the functional activity attribute characteristics of the spatial unit are obtained.

[0014] Specifically, in S2, the construction of the multi-type weighted graph is as follows: Using spatial units as nodes, edges are constructed based on the start and destination nodes of user travel trajectories within the spatial units. These edges include adjacency edges representing geographical adjacency relationships between spatial units and connection edges representing human flow relationships, with corresponding weights assigned to each edge. The weight of an adjacent edge is determined by the cosine similarity of the population density and functional activity attributes of the two nodes it connects, while the weight of a connecting edge is represented by the number of travelers between the originating node and the destination node. The constructed inductive weighted graph uses efficient land use in the multi-type weighted graph as nodes, where nodes are considered to be of the same type, and their edges and corresponding weights are correlated with the multi-type weighted graph. Figure 1 To.

[0015] Specifically, in S2, the functional semantic learning layer of the graph neural network is constructed based on the inductive weighted graph, as follows: The functional semantic learning layer employs a two-layer GraphSAGE model. The GraphSAGE model generates feature embeddings by sampling and aggregating the local neighbor features of a node. The sampling and aggregation of the local neighbor features of a node are as follows: ; in, Represents a node In the current layer's aggregation features, an average pooling aggregator is used to process features from the first layer. Features of layer neighbor nodes We obtain N(i) as the node. The set of neighboring nodes; Aggregation results Using a splicing method combined with the characteristics of the nodes themselves Combine and update to obtain the first Layer features: ; in, For nodes Features obtained from aggregation For activation function, This is the self-connection weight matrix. Indicates splicing the first Features of layer neighbor nodes and nodes Features obtained by aggregating features in the current layer.

[0016] Specifically, in S1, by acquiring the benefit indicators of each plot in each spatial unit of the learning area in one or more dimensions of spatial vitality, spatial quality and output and services, the benefit indicators of each dimension are normalized by the range standardization method, and the weights corresponding to each dimension are determined by the entropy weight method. The comprehensive benefit score of each plot in each spatial unit is calculated. Then, according to the area proportion of each plot in its corresponding spatial unit, the corresponding scores are weighted and summed to obtain the land use benefit of each spatial unit.

[0017] More specifically, in S1, the land use characteristics corresponding to the plots within the spatial unit include residential, commercial service, green space, public service, industrial, or transportation land use. For residential land, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. Among them, the spatial vitality dimension data includes nighttime population heat map and the diversity of supporting facilities. The nighttime population heat map is calculated by the average daily nighttime population per unit area, and the data source is obtained from the mobile phone signaling data of telecommunications operators. The diversity of supporting facilities adopts the Shannon diversity index of supporting facilities points of interest, and the data source is obtained from the database of map software or real estate software. The spatial quality dimension data includes plot ratio and green space ratio, and the data sources for both are obtained from the corresponding residential community data in the database of map software or real estate software. The output and service dimension data can adopt the benchmark land price, that is, the benchmark land price for residential land, which is obtained through relevant assessment, planning and mapping. For commercial service land, the dimensions of its benefit indicators include spatial vitality and output and services; The spatial vitality dimension data includes average daily foot traffic and business diversity. Average daily foot traffic is calculated based on the average number of visitors per unit area per day, and the data source is mobile signaling data from telecommunications operators. Business diversity is measured using the Shannon Diversity Index of Points of Interest for Commercial Facilities, and the data source is the database of map software or real estate software. The output and service dimension data includes consumption level and service scope. Consumption level is calculated based on the average total consumption per merchant, and the data source is merchant review data from local life consumption platforms. Service scope is obtained by measuring the origin of the arriving foot traffic, and the data source is mobile signaling data from telecommunications operators. For green space and public service land, the dimensions of their benefit indicators include spatial vitality and output and services; The spatial vitality dimension data includes activity intensity, activity connection intensity, average dwell time, and dwell time balance. Activity intensity is determined by the population size concentrated in the area within a specified time period; activity connection intensity is determined by the improvement centrality of the land use; average dwell time is determined by the average dwell time of the users in the area; and dwell time balance is determined by the balance of the number of people in the area during different time periods each day. The output and service dimension data is its service range, obtained by measuring the origin range of the arriving population. All data for this type of land use are obtained through mobile phone signaling data from telecommunications operators. For industrial land use, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. The spatial vitality dimension data includes activity intensity and enterprise survival rate. Activity intensity is calculated based on the population size concentrated in the area within a specified time period, with data obtained from mobile phone signaling data from telecommunications operators. Enterprise survival rate is calculated based on the proportion of enterprises that have survived for more than three years, with data obtained from business registration data from enterprise information query agencies. The spatial quality dimension data includes public transportation convenience and green space accessibility. Public transportation convenience is calculated based on the density of subway and bus stops within a specified range, with data obtained from map software databases. Green space accessibility is calculated based on the density of parks within a specified range, with data obtained from map software databases. The output and service dimension data is the patent data within the area, calculated as the total number of patent applications per unit area, with data obtained from database queries. For transportation-related land use, the dimensions of its benefit indicators include spatial vitality and output and services; Among them, the spatial vitality dimension data is its mixed use and function, which adopts the Shannon diversity index of points of interest, and the data source is obtained from the database of map software or real estate software; output and services are its benchmark land price, which is obtained through relevant assessment, planning and mapping.

[0018] Beneficial Effects: This invention guides the selection of land use functions for urban renewal through case studies. By comprehensively considering the static environment surrounding the land parcel and the dynamic interactions generated by pedestrian flow, it predicts land use functions, aiming for high-quality redevelopment of inefficient land. It constructs a land efficiency evaluation index system integrating resident perception and activity information, selects efficient land use as learning samples using spatial clustering, and uses spatiotemporal trajectory data to construct a hierarchical structure diagram representing spatial and functional interactions with efficient land use as the core. A designed "environmental perception-functional semantics" graph neural network architecture is then used to learn the functional combination characteristics of efficient land use. This model is then transferred to inefficient land use to predict its redevelopment functional schemes. This invention, through simulating urban spatial functional interactions and mechanism innovation, breaks through the traditional prediction paradigm based on static rules and spatiotemporal evolution, providing scientific support for land efficiency improvement and functional replacement decisions in urban renewal. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the inefficient land redevelopment function prediction method based on graph neural networks of the present invention.

[0021] Figure 2 A schematic diagram illustrating the modeling of the corresponding geographical proximity relationships and human flow interaction relationships.

[0022] Figure 3 This is an example diagram of the graph neural network constructed in this invention.

[0023] Figure 4 This is a schematic diagram illustrating the working principle of the functional feature learning model constructed and trained in this invention.

[0024] Figure 5 This is an example diagram of Nanjing city, which serves as the learning area for verification in this invention.

[0025] Figure 6 This is an example diagram of an example area for predicting the redevelopment function of inefficient land use in this invention.

[0026] Figure 7This is a schematic diagram illustrating the current functional composition of five key redevelopment projects for inefficient land use in the example area of ​​this invention.

[0027] Figure 8 This is a schematic diagram illustrating the functional composition of five key redevelopment projects for inefficient land use in the example area of ​​this invention after redevelopment.

[0028] Figure 9 is an example diagram of constructing a pedestrian flow network based on the analysis of residents' travel trajectory data.

[0029] Figure 10 This is a performance comparison chart of the three models used in the examples of this invention.

[0030] Figure 11 This is an example of the prediction results obtained using the functional feature learning model of this invention. Detailed Implementation

[0031] To make the objectives, technical solutions and advantages of the present invention clearer, the present application will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0032] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of the present invention should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0033] The method for predicting the redevelopment function of inefficient land use based on graph neural networks of the present invention is as follows: Figure 1 As shown, it includes: S1. Obtain the land use benefits of each spatial unit in the learning area. Based on this, divide each spatial unit into three types of land use benefits: high, medium, and low. Select the spatial unit with high land use benefits as the urban land allocation case library.

[0034] In this invention, the learning area can be set as a specific area of ​​a city, such as the urban area of ​​a city.

[0035] In this invention, to adapt to the prediction of subsequent models, irregular continuous plots of land naturally enclosed by urban roads are used as basic spatial units, each containing one or more plots and possessing relatively independent land use characteristics and spatial attributes.

[0036] The land use characteristics corresponding to the plots within each spatial unit may include residential, commercial service, green space, public service, industrial, or transportation land use.

[0037] In order to ensure that the learning objectives of the subsequent model are consistent with the core of improving land use efficiency, this invention takes land use efficiency as the guide. By quantifying the economic and social benefits exhibited by plots in each spatial unit of the learning area in one or more dimensions of spatial vitality, spatial quality and output and services, a high-quality urban land use configuration case library is selected and constructed.

[0038] In this invention, by acquiring the benefit indicators of each plot in each spatial unit of the learning area in one or more dimensions of spatial vitality, spatial quality, and output and services, the benefit indicators of each dimension are normalized using the range standardization method, and the weights corresponding to each dimension are determined by the entropy weight method. The comprehensive benefit score of each plot in each spatial unit is calculated, and then the corresponding scores are weighted and summed according to the area proportion of each plot in its corresponding spatial unit to obtain the land use benefit of each spatial unit.

[0039] Specifically: For residential land, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. Specifically, spatial vitality data includes nighttime population density and the diversity of supporting facilities. Nighttime population density can be calculated by the average daily nighttime population per unit area, and the data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. The diversity of supporting facilities can be achieved using the Shannon Diversity Index of Points of Interest (POIs), and the data source can be obtained from the databases of map software or real estate application software. Spatial quality data includes plot ratio and green space ratio, and the data for both can be obtained from the databases of corresponding residential communities in map software or real estate application software. The output and services dimension data can use benchmark land prices, specifically benchmark land prices for residential land, which can be obtained through relevant assessment, planning, and surveying.

[0040] For commercial service land, the dimensions of its benefit indicators include spatial vitality and output and services. Spatial vitality data includes average daily foot traffic and business diversity. Average daily foot traffic can be calculated by the average number of visitors per unit area per day, and the data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. Business diversity can be achieved using the Shannon Diversity Index for Points of Interest (POIs), and the data source can be obtained from the databases of map software or real estate software. Output and services data includes consumption levels and service scope. Consumption levels can be calculated by the average total consumption per merchant, and the data source can be obtained from merchant reviews on local lifestyle consumption platforms such as Dianping. Service scope can be obtained by identifying the origin areas of the arriving foot traffic, and the data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom.

[0041] For green space and public service land, the dimensions of their benefit indicators include spatial vitality and output and services. The spatial vitality dimension data includes activity intensity, activity connection intensity, average dwell time, and dwell time balance. Activity intensity can be measured by the population size concentrated in the area within a specified time period, and the data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. Activity connection intensity can be measured by the improved degree centrality of the land use, where the number of edges directly connecting the land use plot to other plots through pedestrian flow is the value. The more edges a plot (node) is connected to, the higher its degree centrality. Improved degree centrality takes into account the different importance of different connections and assigns a weight to each edge based on the degree centrality calculation principle. The data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. Average dwell time can be measured by the average dwell time of the user population in the area, and the data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. Dwell time balance can be measured by the balance of the number of people in the area during different time periods each day, and can be calculated as follows: E u This is the index of the uniformity of residence time in the space u of the land parcel. The percentage of people residing in plot space u during time period v on that day is the ratio of the total number of people residing there on that day; w is the total number of hours spent residing there. In this embodiment, w is 17, and the preferred time period is 6:00-23:00. The data source can be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. The output and service dimension data represents its service range, which can be obtained from the source area range corresponding to the arrival flow. The data source can also be obtained from mobile signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom.

[0042] Specifically, the aforementioned improvement degree centrality is calculated as follows: ; ; ; in, Let be the improved degree centrality of node i at time z; Let be the activity outflow intensity of node i at time z, which is equivalent to the population outflow of node i at time z. Let be the number of neighboring nodes of node i, that is, the total number of nodes where there is population outflow from node i at time z; Let be the activity inflow intensity of node i at time z, which is the population inflow of node i at time z; Let be the number of neighboring nodes of node i, that is, the total number of nodes where population flows into node i at time z; The coefficient is 0.5 in this invention.

[0043] For industrial land use, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. Spatial vitality data includes activity intensity and enterprise survival rate. Activity intensity can be measured by the population size concentrated in the land area within a specified time period, and the data can be obtained from mobile phone signaling data from telecommunications operators such as China Unicom, China Mobile, or China Telecom. Enterprise survival rate can be calculated by the proportion of enterprises that have survived for more than three years, and the data can be obtained from business registration data from enterprise information query agencies such as Qichacha. Spatial quality data includes public transportation convenience and green space accessibility. Public transportation convenience can be measured by the density of subway and bus stops within a specified range, and the data can be obtained from map software databases. Green space accessibility can be measured by the density of parks within a specified range, and the data can be obtained from map software databases. The specified range can be within 1km of the land area. Output and services data is the patent data within the land area, which can be measured by the total number of patent applications per unit area, and the data can be obtained through database queries.

[0044] For transportation-related land uses, the dimensions of their benefit indicators include spatial vitality and output and services. Spatial vitality data represents the degree of mixed-use development, which can be achieved using the Shannon Diversity Index for Points of Interest (POIs). Data can be obtained from the databases of map software or real estate application software. Output and services represent the benchmark land price, which can be obtained through relevant assessments, planning, and surveying.

[0045] In this invention, after obtaining the land use benefits of each spatial unit in the learning area, the k-means clustering algorithm can be used to classify each spatial unit.

[0046] Specifically, by combining the elbow rule (the inflection point is significant when k=3), each spatial unit is divided into three types of land use efficiency: high, medium, and low. Based on this, the present invention selects spatial units with high efficiency as urban land use configuration case libraries for subsequent feature learning.

[0047] S2. Using the urban land use configuration case library obtained in S1 as a sample set, divide it into training set, validation set and test set according to a set ratio, and construct and train a functional feature learning model based on graph neural network.

[0048] In this invention, a functional feature learning model based on graph neural networks is constructed, as follows: (1) Obtain the geometric morphological features, location features, population density features and functional activity attribute features of each spatial unit in the sample set, and fuse them to generate a node feature matrix of multidimensional data;

[0049] In this invention, to ensure that the final functional feature learning model based on graph neural networks can accurately capture the impact of spatial morphology on urban land use, the invention combines the area and contour compactness index of spatial units to obtain their geometric morphological features, and uses them as one of the node features. The geometric morphological features are then fused with their location features, population density features, and functional activity attribute features to generate a node feature matrix of multi-dimensional data, thereby improving the adaptability and accuracy of the functional feature learning model in predicting the functions of complex and heterogeneous urban plots.

[0050] In this invention, the contour compactness index of the spatial unit is calculated as follows: C = 4πA / (P) 2 ; Where C is the contour compactness index of the spatial unit, A is the area of ​​the spatial unit, and P is the perimeter of the spatial unit.

[0051] In this embodiment, the latitude and longitude coordinates of the centroid of the spatial unit after range normalization can be used as its location characteristics, and the area and air profile compactness index of the spatial unit after logarithmic transformation and range normalization can be used as its geometric morphological characteristics.

[0052] In this invention, population density characteristics can be obtained by mapping population data to corresponding spatial units using a land fragmentation method based on the grid population data provided by mobile signaling data within a spatial unit, and then calculating the population density of the spatial unit by combining the area of ​​the spatial unit. Specifically, the land fragmentation method involves cutting the spatial unit into several spatial unit fragments according to the mobile signaling grid. For each spatial unit fragment, the proportion of its area to the total area of ​​all fragments within its respective mobile signaling grid is calculated and recorded as the population allocation weight. Based on the population allocation weight, the population within the mobile signaling grid is proportionally allocated to each spatial unit fragment. The total population of each spatial unit is then obtained by summing these weights, thus mapping the population data of each spatial unit to the corresponding spatial unit.

[0053] In this invention, the functional activity attributes of a spatial unit can be obtained by statistically analyzing point-of-interest (POI) data within a spatial unit according to major categories and calculating the frequency proportion of each category. The POI data includes name, category (e.g., catering, transportation, public services), latitude and longitude, etc. Categories are categorized into major, medium, and minor categories. Major categories include catering services, scenic spots, financial insurance, and companies, etc. The frequency of each category is calculated as the number of POIs within that category.

[0054] In this invention, the modeling of the functional feature learning model is as follows: Figure 2As shown, the model is affected by two geographical effects (corresponding to the constructed geographical proximity relationship and the interaction relationship of human flow). The first part is the influence of the built environment, the second part is the influence of socio-economic activities caused by population flow, and the third part is the combined effect of the above two influences on land use function prediction.

[0055] (2) Construct a graph structure based on travel trajectory data, including multi-type weighted graphs and inductive weighted graphs. The inductive weighted graph is constructed using the efficient land use obtained from the multi-type weighted graphs as nodes. Existing methods for processing urban mobile signaling data often directly construct a full graph containing all spatial units and travel connections. This approach may introduce redundant information and fails to fully focus on high-quality development models that are instructive for urban renewal. Therefore, this invention constructs a graph structure with efficient urban spatial units as core nodes and spatial units with geographical proximity or pedestrian connections as associated nodes, creating a multi-type weighted graph. This allows learning how functional interactions within efficient spatial units collectively contribute to efficient land use. Furthermore, for efficient spatial units, an inductively weighted graph based on travel connections is constructed, focusing on learning their land use functional characteristics from efficient interactions. In this invention, both graphs share the same loss function, and during backpropagation, the parameters of both graphs are jointly optimized based on the overall loss.

[0056] In this invention, the construction of multi-type weighted graphs is as follows: Using spatial units as nodes, this invention constructs edges based on the start and destination nodes of user travel trajectories within the spatial units. Two types of edges are constructed, representing the geographical adjacency relationship and the pedestrian flow connection relationship between spatial units, respectively defined as adjacency edges and connection edges, and weights are set for the corresponding edges. Specifically, the weight of an adjacent edge can be determined by the cosine similarity of the population density feature and the functional activity attribute feature of the two nodes it connects, as follows: ; in, The weight of the adjacent edge. and This represents the population density characteristics of two nodes connected by adjacent edges. and The functional activity attributes of the two nodes connected by adjacent edges; The weight of the connecting edge is represented by the number of travelers between the origin node and the destination node; This allows us to obtain various types of weighted graphs, as follows: ; in, Represents a set of nodes, each node It represents the spatial unit represented by the starting node or the destination node of the travel trajectory; Let the set of edges be denoted by the node. The edges between them are represented as ,in, The set of edge types representing the relationships between nodes. ; The weight of the edge; The inductive weighted graph is an undirected graph. It uses efficient land use from the multi-type weighted graph as nodes, treating nodes as belonging to the same type. It also includes adjacent edges and connecting edges. Weights are set as in the multi-type weighted graph, resulting in: ; in, This represents the set of nodes corresponding to an efficient spatial unit, each node... Represents a specific efficient spatial unit. Let the set of edges be represented as follows: , is the weight, representing the symmetric weight value of an undirected edge.

[0057] (3) Based on the multi-type weighted graph constructed in step (2), construct the neighborhood feature fusion layer of the graph neural network, and based on the inductive weighted graph constructed in step (2), construct the functional semantic learning layer of the graph neural network, thereby obtaining the graph neural network, such as Figure 3 As shown;

[0058] In this invention, existing urban land use simulation or land use feature learning methods based on graph neural networks typically only consider single factors in the surrounding geographical environment of a plot or in residents' travel activities, lacking a holistic consideration of the differentiated contextual information provided by these two types of spatial interactions. Therefore, this invention constructs a neighborhood feature fusion layer of a graph neural network based on a multi-type weighted graph constructed in step (2), which integrates the multi-dimensional spatial interactions constituted by geographical proximity and human flow connections, thereby enriching the contextual feature representation of core nodes for efficient land use. Based on this, this invention constructs a functional semantic learning layer of a graph neural network based on an inductively weighted graph constructed in step (2), which is used to extract the interactive relationships and functional features between efficient land uses. Through the above hierarchical feature learning architecture, the layer-by-layer modeling and representation learning of land use influencing factors is achieved from local to global, and from single relationships to multiple relationships.

[0059] In this invention, the neighborhood feature fusion layer and the functional semantic learning layer can be implemented using the commonly used and mature RGCN and GraphSAGE algorithms in graph neural networks, respectively. During the backpropagation of loss, the weights in the neighborhood feature fusion layer and the functional semantic learning layer are updated simultaneously.

[0060] In this embodiment, considering that environmental information from geographic space and pedestrian space will affect the land use patterns of urban spatial units in different ways, the neighborhood feature fusion layer adopts a single-layer RGCN to distinguish and capture the impact of different spatial connection methods on node information aggregation. Unlike the classic GCN model, RGCN performs feature transformation according to the type and direction of the edges, enhancing its ability to process static geographic environmental features and dynamic resident travel connection information. During the layer-by-layer propagation of node information, the first... The formula for calculating the node features of a convolutional layer is as follows: ; in, For the activation function, ReLU is used in this embodiment to introduce nonlinear factors and help the model learn more complex patterns; It is a node The set of neighbors for edge type r; This is a normalization constant used to normalize the contributions of different nodes; Let be the transformation weight matrix of the l-th layer for type r, used to map the input features to another space; It is a self-connected weight matrix used to preserve the feature information of the nodes themselves, without going through the aggregation process of neighboring nodes; is the feature vector of the neighbor node at layer l.

[0061] In this embodiment, the functional semantic learning layer adopts a two-layer GraphSAGE model to ensure that the final functional feature learning model meets the requirements of flexible and scalable structure and inductive reasoning ability. The GraphSAGE model generates feature embeddings by sampling and aggregating the local neighbor features of nodes, thereby capturing the network structure and human flow interaction features of high-efficiency urban land use, and applying them to the task prediction of location nodes.

[0062] Specifically, the local neighbor features of a node are sampled and aggregated as follows: ; in, Represents a node In the current layer's aggregation features, an average pooling aggregator is used to process features from the first layer. Features of layer neighbor nodes We obtain N(i) as the node. The set of neighboring nodes; Aggregation results Using a splicing method combined with the characteristics of the nodes themselves Combine and update to obtain the first Layer features: ; in, For nodes Features obtained from aggregation Indicates splicing the first Features of layer neighbor nodes and nodes Features obtained from aggregation at the current layer.

[0063] In this invention, in the aforementioned framework, all nodes are projected to 64 dimensions through a linear layer, and after passing through a neighborhood feature fusion layer, i.e., a single RGCN convolution, the feature dimension is transformed to 128 dimensions, obtaining a new embedding of efficient land use nodes containing environmental information. After constructing a directed graph composed of all efficient land uses, a functional semantic learning layer, i.e., two SAGE convolutions, is used, with the feature dimension remaining at 128 dimensions, and the output being 64 dimensions.

[0064] In this invention, due to the varying sizes and shapes of urban plots, there is a data imbalance problem with functional labels, meaning that the functional characteristics of plots show significant differences in data representation, which in turn affects the model's feature learning ability. To address this, the graph neural network of this invention divides the task of learning the functional characteristics of spatial units into functional existence classification and functional proportion regression subtasks, respectively processing information on the two dimensions of "presence" and "proportion" of functions, thereby enhancing the model's ability to express the land use composition of multifunctional plots. Corresponding to two six-dimensional labels, namely functional existence label and functional proportion label, the former uses one-hot encoding to indicate whether there are six types of functions in the spatial unit: residential, commercial, green space, public service, industrial, and transportation. 0 represents the absence of the function, and 1 represents the presence of the function. The latter consists of the ratio of the area of ​​each of the six functional land uses to the total area of ​​the spatial unit, with a sum of 1.

[0065] In this invention, the urban land use configuration case library obtained in S1 is used as a sample set, which is divided into a training set, a validation set and a test set according to a set ratio. The sample set is then input into the graph neural network constructed above for training to obtain a functional feature learning model.

[0066] In this invention, the urban land use configuration case library obtained in S1 is used as a sample set. 60% of the data in the sample set is used for training, and the remaining 40% is used for verification and testing.

[0067] In this invention, to adapt to the needs of the functional presence classification and functional proportion regression subtasks, a phased training mechanism is proposed. The first set number of training rounds, such as 300 rounds, uses binary cross-entropy loss for functional presence classification learning. The subsequent set number of training rounds, such as 200 rounds, introduces weighted MAE loss to simultaneously optimize the functional presence classification and proportion regression tasks. In this invention, the Adam optimizer is used for training, with an initial learning rate set to 1×10⁻⁶. -3 The weight decay rate is set to 1×10-4 During training, a cosine annealing strategy is introduced to dynamically adjust the learning rate, and an early stopping mechanism is introduced to avoid overfitting. Training ends after the performance on the validation set stabilizes. This invention improves the stability and prediction accuracy of the model through stage-by-stage optimization.

[0068] S3. Obtain each spatial unit in the prediction area and input it into the functional feature learning model obtained in S2 to perform functional prediction of inefficient land redevelopment.

[0069] In this invention, when predicting the redevelopment function of inefficient land, the features of inefficient land nodes retain coordinates, area, and outline compactness features, while other dimensions are set to 0. An embedding is generated through a neighborhood feature fusion layer, and then input into the functional feature learning model obtained in S2 for prediction. Finally, the functional composition and corresponding proportions of the spatial unit are obtained, such as... Figure 4 As shown.

[0070] This invention also supports large-scale, automated system development, teaching, and practical research. It can be embedded in urban land use dynamic assessment platforms, providing strong support for urban renewal, land use planning indicator adjustments, and policy formulation under the background of high-quality development.

[0071] This invention aims to improve land use efficiency by integrating residents' travel activity characteristics. Through case studies, it guides the selection of land use functions for urban renewal. By comprehensively considering the static environment surrounding the land parcel and the dynamic interactions generated by pedestrian flow, it predicts land use functions, thereby promoting the efficient and sustainable use of land resources. Oriented towards high-quality redevelopment of inefficient land, it constructs a land efficiency evaluation index system that integrates residents' perceptions and activity information, and selects efficient land as learning samples using spatial clustering. This invention utilizes spatiotemporal trajectory data to construct a hierarchical structure diagram representing spatial and functional interactions, with efficient land as the core. It employs a designed "environmental perception-functional semantics" graph neural network architecture to learn its functional combination characteristics and transfers this model to inefficient land to predict its redevelopment functional schemes. Through simulating urban spatial functional interactions and mechanism innovation, this invention breaks through the traditional prediction paradigm based on static rules and spatiotemporal evolution, providing scientific support for land efficiency improvement and functional replacement decisions in urban renewal.

[0072] This invention was verified in Nanjing, the capital of Jiangsu Province, one of the first pilot cities for the redevelopment of inefficient land in China.

[0073] The learning area is designated as the urban area of ​​Nanjing (excluding the two outlying suburbs of Gaochun and Lishui). Figure 5 As shown, spatial units with high land use efficiency are selected as urban land use allocation case libraries. In the Shanbei area, a pilot area for smart urban renewal, inefficient land redevelopment function prediction is conducted. Figure 6 As shown.

[0074] In this embodiment, the total planned area of ​​the Shanbei area is approximately 105.87 hectares. As of the end of 2021, the permanent resident population was 58,040, the floating population was 10,875, and the total population was 68,915. Figure 7 This refers to the five key redevelopment projects for inefficient land use identified in the preliminary smart assessment phase, along with their current functional components. Figure 8 This refers to the functional composition of five projects after redevelopment within the regulatory detailed plan and the current urban redevelopment scheme.

[0075] Specifically, 770 spatial units with high land use efficiency were selected from a total of 8,100 spatial units in Nanjing's urban area as a case study database for urban land use allocation. Based on the land use characteristics and area proportion of each spatial unit, a six-dimensional functional label was constructed, corresponding to six functional attributes: residential, commercial services, green space, public services, industry, and transportation. Among the selected spatial units, 243 spatial units exhibited a single function, with only one non-zero functional label. Classification and statistical analysis according to the dominant functional type of the spatial units showed that: 449 were residential-dominant, 66 were commercial-dominant, 9 were green space-dominant, 45 were public services-dominant, 139 were industry-dominant, and 62 were transportation-dominant. Further analysis of the non-zero sample proportions for different functional dimensions yielded 63%, 56%, 24%, 47%, 30%, and 21%, respectively.

[0076] Overall, the urban land use configuration case library is consistent with the distribution characteristics of spatial units throughout the city in terms of the diversity of dominant functional types and functional labels, and has good representativeness and research value. The constructed functional label data provides a clear and feature-rich training foundation for subsequent functional embedding learning and prediction of inefficient land redevelopment schemes.

[0077] Furthermore, by analyzing residents' travel trajectory data to construct a population flow network, as shown in Figure 9, it was found that nearly 80% of the city's population flow occurred in the highly efficient spatial units, which account for less than 15% of the total number of spatial units. Their degree centrality ranks high among all nodes, making them the most active core nodes in the population flow network, with significant spatial influence and network control. The population flow network constructed in this way is highly similar to the overall network of the city in terms of structural characteristics, and can accurately reflect the most core resident travel activity patterns in the city.

[0078] To systematically evaluate the scientific validity of this invention in model construction and scenario application, the following two baseline models were selected for performance comparison: (1) Analysis of the impact of environmental information on the functional inference of efficient land use: The SAGE model is used as the baseline. The SAGE model infers the functional characteristics of efficient land use nodes based on the flow of people and information transmission between them; while the RGCN-SAGE model proposed in this invention further integrates environmental information from geographically adjacent plots and plots with flow of people. By comparing the performance of the models, the impact of environmental information on the functional inference of efficient land use can be clearly defined. (2) Impact analysis of neighborhood connectivity on the functional inference of high-efficiency land use: The GCN-SAGE model was used as the baseline. This model does not distinguish the relationship between geographically adjacent plots, pedestrian-connected plots, and high-efficiency land use. GCN updates the environmental information of high-efficiency land use in the isomorphic map and combines it with SAGE to infer its functional performance. By comparing their performance differences, it can be determined whether different types of environmental information will affect the functional inference of high-efficiency land use.

[0079] Regarding the selection of performance evaluation indicators, the Jaccard index (IoU) is used to measure the similarity between predicted and true labels for functional category existence prediction. The weighted F1 score is used to evaluate the spatial distribution of different land use functions, and the weighted absolute percentage error (Weighted-MAE) is used to evaluate functional proportion prediction. The final model performance comparison is as follows: Figure 10 As shown.

[0080] The results show that the RGCN-SAGE model of this invention outperforms both the single SAGE model and the combined GCN-SAGE model in terms of Jaccard Index (IoU) and Weighted-F1. By fusing functional, morphological, and population data from geographically adjacent and pedestrian-connected plots using the RGCN model, valuable information can be provided for the SAGE model to infer the functional combinations of urban land use and the spatial distribution of different land use functions. Based on this, distinguishing the spatial relationships of urban land use can further improve the prediction accuracy. In comparison, the method proposed in this invention performs slightly worse than the other two models in predicting functional proportions. However, in practical applications, due to the irregular shapes of urban plots, the predicted functional proportions can only serve as a preliminary reference, and a small difference between the accuracy of the proposed model and the baseline performance is acceptable.

[0081] The final model prediction is as follows Figure 11 As shown, the results indicate that: (1) In the forecast of the Dadi Construction Group project, the residential function has decreased by 50% compared to the current situation, the green space function remains the same, the area of ​​public services has increased to 40%, and the industrial function has been relocated. The above trends are generally consistent with the control plan. The commercial function is predicted to increase by 50%, which is consistent with the current renewal plan. The forecast plan increases the proportion of transportation function, which is different from the control plan and the current renewal plan. (2) In the forecast for the East Long-Distance Railway Station, the proportion of green space will increase to 10%, public services will decrease by 50% compared to the current situation, and transportation functions will remain the same as the current situation. The above trends are generally the same as those in the control plan. The commercial function is predicted to increase by 10%, which is consistent with the trend of the current renewal plan. The residential function is predicted to decrease by 50% compared to the current situation, while the industrial function is increased, and the proportion of transportation remains unchanged. This is different from both the control plan and the current renewal plan. (3) In the forecast of the Zijin Fresh Fruit Wholesale Market project, the residential function will decrease compared to the current situation, the proportion of green space will increase to 30%, the public service function will increase by 50%, the industrial function will remain the same as the current situation, and the proportion of transportation will rise to 10%. The above trends are the same as those in the control plan. The commercial function is predicted to increase by 30%, which is different from the trends in the control plan and the current update plan. (4) In the forecast of the Phoenix Cultural Trade Project, the proportion of green space will increase to 10%, while the functions of transportation, industry, public services, and commerce will remain consistent with the current situation. The above trends are generally the same as those of the control plan and the current renovation plan. The predicted residential function is reduced by 10% compared to the current situation, which is different from the trends of the control plan and the current renovation plan. (5) In the forecast of the Tianguang Institute project, the residential and green space functions remain consistent with the current situation, which is the same as the trend of the current urban renewal plan. The commercial function remains unchanged, the public service function is reduced by 40% compared with the current situation, and the proportion of industrial and transportation functions is increased, which is different from the trend of the control plan and the current urban renewal plan.

[0082] By comparison, we can see that: (1) In terms of function type prediction, by comparing the current situation with the control plan, the model can predict the new function types proposed in the control plan and the update plan; (2) Regarding the prediction of functional change trends, the model's consistency with existing planning schemes shows functional differences. The predicted change trends of green space functions are highly consistent with other schemes. The consistency of the change trends of residential, commercial, and public service functions reaches 80%. The prediction of transportation and industrial functions reflects the independence of the model more, with a consistency of 60% with existing schemes.

[0083] Overall, the prediction results of this invention are relatively independent in terms of the proportion of industrial and transportation functions, and generally increase the proportion of industrial and transportation functions. The reasons for this difference will be analyzed from the aspects of higher-level planning, functional characteristics of efficient urban land use, and human flow patterns.

[0084] The "Nanjing Xuanwu District Territorial Spatial Zoning Plan (2021-2035)" proposes a new functional positioning for the Shanbei area, to which the urban renewal project is located, including "actively integrating into the Zijin Mountain Science and Technology Innovation Belt, strengthening university-local government integration and technology incubation; promoting the landing of major science and technology innovation projects and new R&D institutions; improving livable supporting facilities, and promoting a vibrant transformation in conjunction with the construction of urban "Silicon Alleys." The study area is located at one of the core nodes of the Zijin Mountain Science and Technology Innovation Belt, adjacent to Nanjing Forestry University. Its future development will align with this strategic positioning, further expanding the demand for industrial function enhancement and transportation infrastructure. Furthermore, the selected high-vitality, high-efficiency plots exhibit spatial characteristics and functional structures of a balanced work-life ratio and convenient transportation. The area's current industrial and transportation functions are underdeveloped; appropriate functional integration will help enhance development vitality. Finally, residents' spatiotemporal trajectory data show frequent commuting inflows and outflows during morning and evening rush hours, indicating cross-regional tidal traffic between the renewal area and surrounding areas, potentially creating a need to enhance industrial and transportation functions.

[0085] This invention innovates upon traditional prediction methods in terms of research perspective, methodology, and application scenarios. In practice, it supplements static adjacency with dynamic associations, reconstructing the fixed spatial relationships between urban land uses into a dynamic association network based on real-world pedestrian interactions. Irregular plots replace homogeneous grids, significantly improving the precision and realism of spatial representation. Methodologically, it replaces global analysis with hierarchical learning, employing a hierarchical strategy centered on efficient land use to extract and learn key features. In application, it achieves proactive optimization through pattern transfer. A neighbor sampling mechanism enables the model to be extensible, transferring functional combination patterns learned from efficient land use to inefficient land use, and providing a clear objective for the prediction process. Case studies show that the model achieves 80% consistency with urban planning and planners' experience-based solutions.

[0086] This invention breaks through the traditional land use prediction paradigm based on static rules and spatiotemporal evolution by learning case characteristics, simulating human-land interaction, and designing a goal-oriented mechanism.

[0087] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of the invention as described above, which are not provided in the details for the sake of brevity.

[0088] The embodiments of this invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this invention should be included within the protection scope of this invention.

Claims

1. A method for predicting the redevelopment function of inefficient land use based on graph neural networks, characterized in that, include: S1. Obtain the land use benefits of each spatial unit in the learning area, and divide each spatial unit into three types of land use benefits: high, medium and low. Select the spatial unit with high land use benefits as the urban land allocation case library. S2. Using the urban land use configuration case library obtained in S1 as the sample set, a functional feature learning model based on graph neural networks is constructed as follows: (1) Obtain the geometric morphological features, location features, population density features and functional activity attribute features of each spatial unit in the sample set, and fuse them to generate a node feature matrix of multidimensional data; (2) Construct a graph structure based on travel trajectory data, including multi-type weighted graphs and inductive weighted graphs. The inductive weighted graph is constructed using the efficient land use obtained from the multi-type weighted graphs as nodes. (3) Construct a neighborhood feature fusion layer for the graph neural network based on the multi-type weighted graph, and construct a functional semantic learning layer for the graph neural network based on the inductive weighted graph, thereby obtaining the graph neural network; Using the urban land use configuration case library obtained in S1 as a sample set, the graph neural network is input into the graph neural network for training to obtain a functional feature learning model. S3. Obtain each spatial unit in the prediction area and input it into the functional feature learning model obtained in S2 to perform functional prediction of inefficient land redevelopment.

2. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, In step S2, the urban land use configuration case library obtained in step S1 is used as a sample set. It is divided into a training set, a validation set, and a test set according to a set ratio. The sample set is then input into the constructed graph neural network for training to obtain a functional feature learning model.

3. The method for predicting the redevelopment function of inefficient land use according to claim 2, characterized in that, The training of the graph neural network adopts a phased training mechanism, as follows: The training for the first set number of epochs uses binary cross-entropy loss for functional existence classification learning. The training for the subsequent set number of epochs introduces weighted MAE loss and uses the Adam optimizer for training. The initial learning rate is set to 1×10. -3 The weight decay rate is set to 1×10 -4 During training, a cosine annealing strategy is introduced to dynamically adjust the learning rate, and an early stopping mechanism is also introduced.

4. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, The geometric features of each spatial unit are obtained by combining the area of ​​the spatial unit with its contour compactness index; wherein, the contour compactness index of the spatial unit is calculated as follows: C=4πA / (P) 2 ; Where C is the contour compactness index of the spatial unit, A is the area of ​​the spatial unit, and P is the perimeter of the spatial unit.

5. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, The latitude and longitude coordinates of the centroid of a spatial unit after range normalization are used as its locational characteristics. The area and air profile compactness index of the spatial unit after logarithmic transformation and range normalization are used as its geometric morphological characteristics.

6. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, Based on the grid population data provided by mobile phone signaling data within a spatial unit, the population data is mapped to the corresponding spatial unit using the land use fragmentation method, and the population density of the spatial unit is calculated by combining the area of ​​the spatial unit, thus obtaining the population density characteristics of the spatial unit.

7. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, In S2, the multi-type weighted graph and the inductive weighted graph are specifically as follows: Using spatial units as nodes, edges are constructed based on the start and destination nodes of user travel trajectories within the spatial units. These edges include adjacency edges representing geographical adjacency relationships between spatial units and connection edges representing human flow relationships, with corresponding weights assigned to each edge. The weight of an adjacent edge is determined by the cosine similarity of the population density and functional activity attributes of the two nodes it connects, while the weight of a connecting edge is represented by the number of travelers between the originating node and the destination node. The constructed inductive weighted graph uses efficient land use in the multi-type weighted graph as nodes, where nodes are considered to be of the same type, and their edges and corresponding weights are consistent with those in the multi-type weighted graph.

8. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, In step S2, the functional semantic learning layer of the graph neural network is constructed based on the inductive weighted graph, as follows: The functional semantic learning layer employs a two-layer GraphSAGE model. The GraphSAGE model generates feature embeddings by sampling and aggregating the local neighbor features of a node. The sampling and aggregation of the local neighbor features of a node are as follows: ； in, Represents a node In the current layer's aggregation features, an average pooling aggregator is used to process features from the first layer. Features of layer neighbor nodes We obtain N(i) as the node. The set of neighboring nodes; Aggregation results Using a splicing method combined with the characteristics of the nodes themselves Combine and update to obtain the first Layer features: ； in, For nodes Features obtained from aggregation For activation function, This is the self-connection weight matrix. Indicates splicing the first Features of layer neighbor nodes and nodes Features obtained by aggregating features in the current layer.

9. The method for predicting the redevelopment function of inefficient land use according to claim 1, characterized in that, In step S1, by acquiring one or more benefit indicators of each plot in each spatial unit of the learning area in terms of spatial vitality, spatial quality, and output and services, the benefit indicators of each dimension are normalized using the range standardization method, and the weights corresponding to each dimension are determined by the entropy weight method. The comprehensive benefit score of each plot in each spatial unit is calculated, and then the corresponding scores are weighted and summed according to the area proportion of each plot in its corresponding spatial unit to obtain the land use benefit of each spatial unit.

10. The method for predicting the redevelopment function of inefficient land use according to claim 9, characterized in that, In S1, the land use characteristics corresponding to the plots within the spatial unit include residential, commercial service, green space, public service, industrial, or transportation land use. For residential land, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. Among them, the spatial vitality dimension data includes nighttime population heat map and the diversity of supporting facilities. The nighttime population heat map is calculated by the average daily nighttime population per unit area, and the data source is obtained from the mobile phone signaling data of telecommunications operators. The diversity of supporting facilities adopts the Shannon diversity index of supporting facilities points of interest, and the data source is obtained from the database of map software or real estate software. The spatial quality dimension data includes plot ratio and green space ratio, and the data sources for both are obtained from the corresponding residential community data in the database of map software or real estate software. The output and service dimension data can adopt the benchmark land price, that is, the benchmark land price for residential land, which is obtained through relevant assessment, planning and mapping. For commercial service land, the dimensions of its benefit indicators include spatial vitality and output and services; The spatial vitality dimension data includes average daily foot traffic and business diversity. Average daily foot traffic is calculated based on the average number of visitors per unit area per day, and the data source is mobile signaling data from telecommunications operators. Business diversity is measured using the Shannon Diversity Index of Points of Interest for Commercial Facilities, and the data source is the database of map software or real estate software. The output and service dimension data includes consumption level and service scope. Consumption level is calculated based on the average total consumption per merchant, and the data source is merchant review data from local life consumption platforms. Service scope is obtained by measuring the origin of the arriving foot traffic, and the data source is mobile signaling data from telecommunications operators. For green space and public service land, the dimensions of their benefit indicators include spatial vitality and output and services; The spatial vitality dimension data includes activity intensity, activity connection intensity, average dwell time, and dwell time balance. Activity intensity is determined by the population size concentrated in the area within a specified time period; activity connection intensity is determined by the improvement centrality of the land use; average dwell time is determined by the average dwell time of the users in the area; and dwell time balance is determined by the balance of the number of people in the area during different time periods each day. The output and service dimension data is its service range, obtained by measuring the origin range of the arriving population. All data for this type of land use are obtained through mobile phone signaling data from telecommunications operators. For industrial land use, the dimensions of its benefit indicators include spatial vitality, spatial quality, and output and services. The spatial vitality dimension data includes activity intensity and enterprise survival rate. Activity intensity is calculated based on the population size concentrated in the area within a specified time period, with data obtained from mobile phone signaling data from telecommunications operators. Enterprise survival rate is calculated based on the proportion of enterprises that have survived for more than three years, with data obtained from business registration data from enterprise information query agencies. The spatial quality dimension data includes public transportation convenience and green space accessibility. Public transportation convenience is calculated based on the density of subway and bus stops within a specified range, with data obtained from map software databases. Green space accessibility is calculated based on the density of parks within a specified range, with data obtained from map software databases. The output and service dimension data is the patent data within the area, calculated as the total number of patent applications per unit area, with data obtained from database queries. For transportation-related land use, the dimensions of its benefit indicators include spatial vitality and output and services; Among them, the spatial vitality dimension data is its mixed use and function, which adopts the Shannon diversity index of points of interest, and the data source is obtained from the database of map software or real estate software; output and services are its benchmark land price, which is obtained through relevant assessment, planning and mapping.

Images