A method and system for optimizing the configuration and layout of a low-utility redevelopment space

Abstract

The application discloses a low-efficiency land redevelopment space configuration and pattern optimization method, and belongs to the technical field of land resource management and urban planning. The method comprises the following steps: multi-source dynamic data fusion and low-efficiency land intelligent identification, modeling of spatial dependence relationship of land blocks based on a graph neural network; a dynamic evaluation index system covering four dimensions of economy, society, ecology and space is constructed, and a TOPSIS model and an obstacle degree model are used for comprehensive evaluation and obstacle factor diagnosis; a multi-objective optimization algorithm integrated with an explainable artificial intelligence module is used for space pattern optimization, and an interpretable candidate scheme is generated; multi-scenario dynamic simulation and comprehensive benefit evaluation are performed on the candidate scheme, and an optimal scheme is output; and full-life-cycle dynamic monitoring and early warning feedback are performed on the implementation scheme. The application realizes full-process closed-loop management of low-efficiency land redevelopment from identification, evaluation, optimization to monitoring, and has the advantages of high identification precision, comprehensive evaluation and interpretable optimization results.

Description

Technical Field

[0001] This invention relates to the field of land resource management and urban planning technology, specifically to a method and system for spatial configuration and pattern optimization of inefficient land redevelopment. Background Technology

[0002] Urban development is shifting from incremental expansion to stock optimization, making the redevelopment of inefficient land an important means of revitalizing existing space and improving land use efficiency. However, the current redevelopment of inefficient land faces the following technical challenges: (1) Insufficient accuracy in identifying inefficient land use. Existing methods for identifying inefficient land use include literature analysis, analytic hierarchy process, and Delphi method, which are mainly qualitative and difficult to accurately identify inefficient land use with complex and diverse characteristics. Some methods use a single data source or static indicators and fail to fully consider the dynamic changes and spatial relationships of land use.

[0003] (2) The evaluation index system is not perfect. The existing land use evaluation index system is not perfect, the weight ratio is too subjective, and there is a lack of verification methods. Existing evaluations focus more on economic benefits, and do not give enough consideration to social benefits, ecological benefits and spatial benefits, and lack a dynamic adjustment mechanism.

[0004] (3) Spatial optimization lacks interpretability. Most existing spatial decision-making models can only be used for updating schemes of small plots, and there are still few scientific quantitative schemes applicable to large-scale updating decisions. Most existing optimization algorithms are "black box" models, making it difficult for decision-makers to understand the generation logic of optimization results, which affects the acceptability and feasibility of the scheme.

[0005] (4) Lack of closed-loop management throughout the entire process. Most existing studies have focused on a single stage of identification or optimization, lacking a closed-loop approach covering the entire lifecycle from identification to evaluation to optimization to simulation to monitoring.

[0006] To address the aforementioned problems, this invention provides a spatial configuration and pattern optimization method for the redevelopment of inefficient land, aiming to achieve accurate identification, scientific assessment, interpretable optimization, and dynamic management of inefficient land throughout its entire life cycle. Summary of the Invention

[0007] This invention addresses the technical shortcomings of existing inefficient land redevelopment methods, such as insufficient identification accuracy, a single evaluation system, lack of interpretability in optimization, and lack of closed-loop management throughout the entire process. It provides a spatial configuration and pattern optimization method and system for inefficient land redevelopment.

[0008] In a first aspect, the present invention provides a method for spatial configuration and pattern optimization in the redevelopment of inefficient land, comprising: Step S1: Multi-source dynamic data fusion and intelligent identification of inefficient land use: Acquire multi-source data of the target area, including land survey data, satellite remote sensing image data, mobile phone signaling data, enterprise electricity consumption data, enterprise tax data, POI data, and building census data; perform spatiotemporal registration and standardization processing on the multi-source data to construct a feature matrix for multi-source data fusion; construct an inefficient land use identification model based on graph neural network, abstracting the land parcels in the target area as nodes in a graph structure, using the spatial adjacency and functional association relationships between land parcels as edges, and using graph neural network to model the spatial dependency relationship of the feature matrix to identify inefficient land use and output the spatial distribution, type, and degree of inefficiency of inefficient land use; Step S2: Multidimensional Dynamic Assessment and Obstacle Factor Diagnosis: Construct a dynamic multidimensional assessment index system, which includes economic benefit, social benefit, ecological benefit, and spatial benefit dimensions; calculate the weight of each index using the entropy weight method combined with the time series dynamic weight adjustment method; calculate the comprehensive performance score of each plot based on the TOPSIS model; and diagnose the main obstacle factors affecting the comprehensive performance of each plot using the obstacle degree model. Step S3: Interpretable Spatial Pattern Optimization: Based on the inefficient land use identified in Step S1 and the obstacle factors diagnosed in Step S2, multiple optimization objectives are set and constraints are established; a multi-objective optimization algorithm integrating an interpretable artificial intelligence module is used to optimize the spatial configuration of inefficient land use, generating at least two candidate spatial optimization schemes; the interpretable artificial intelligence module calculates the contribution of each optimization objective to the decision variables and outputs the optimization decision explanation information of each candidate scheme. Step S4: Multi-dimensional scenario simulation and scheme evaluation: Dynamic simulations are performed on each candidate spatial optimization scheme generated in step S3 under economic, social and ecological scenarios, respectively, and the comprehensive benefit index of each scheme under each scenario is calculated; the candidate schemes are ranked according to the comprehensive benefit index, and the optimal spatial configuration and pattern optimization scheme is output. Step S5: Dynamic monitoring and feedback throughout the entire life cycle: Dynamically monitor the implementation process of the optimal solution output in Step S4, regularly collect multi-source data of the implementation area, update the inefficient land use identification results and comprehensive performance evaluation results, and trigger an early warning when the monitoring indicators deviate from the preset threshold and automatically feed back to Step S3 for solution adjustment.

[0009] Optionally, the graph neural network mentioned in step S1 is a graph attention network, which adaptively learns the contribution weights of different neighboring nodes to the central node through an attention mechanism; the inefficient land use identification model adopts a multi-level identification strategy: the first level is preliminary identification, which determines whether the plot is inefficient land use; the second level is fine identification, which determines the type and degree of inefficiency of inefficient land use; the third level is micro-identification, which identifies inefficient sub-regions caused by the mismatch of the smallest unit structure within the inefficient land use.

[0010] Optionally, the dynamic multidimensional evaluation index system described in step S2 includes the following indicators: Economic efficiency dimensions: tax revenue per unit area, operating income per unit area, output per unit of electricity consumption, and employment density; Social benefits dimensions: accessibility of public service facilities, public transportation coverage, and per capita public space area; Ecological benefit dimensions: carbon sequestration per unit area, green space ratio, and connectivity of ecological patches; Spatial efficiency dimensions: plot ratio, building density, and land use mix; The land use mix degree is calculated according to the following formula: ; In the formula, Let K represent the land use mix degree of the i-th plot, and K be the total number of land use types. This represents the area proportion of the k-th land use type in the i-th plot; This was obtained by spatially overlaying and analyzing the land parcel vector plots and land use classification data on a GIS platform.

[0011] Optionally, the obstacle degree model described in step S2 calculates the obstacle degree of each index for each plot according to the following formula: ; In the formula, Let the obstacle degree of the j-th indicator of the i-th plot be denoted as . The index value of the j-th index for the i-th plot is the standardized value. Let be the weight of the j-th indicator, and m be the total number of indicators; The original index values ​​were obtained by processing them using the range standardization method. The obstacle degree is calculated using the entropy weight method; the greater the obstacle degree, the stronger the constraint effect of the indicator on the overall performance of the land parcel.

[0012] Optionally, the optimization objectives of the multi-objective optimization algorithm in step S3 include: maximizing the comprehensive benefits of land use, maximizing spatial clustering, and minimizing ecological disturbance; the constraints include: uniqueness constraint of land use type configuration, upper and lower limits constraint of plot ratio, minimum proportion constraint of ecological land, and zoning control constraint of development intensity.

[0013] Optionally, the interpretability AI module in step S3 uses the SHAP value method to calculate the contribution of each optimization objective to the decision variable. The SHAP value is based on the Shapley value in cooperative game theory and calculates the marginal contribution of the feature to the prediction result according to the following formula: ; in, Let f(S) be the SHAP value of the j-th feature, F be the set of all features, S be the feature subset that does not contain the j-th feature, and f(S) be the model prediction value when only the feature subset S is used. By calculating the SHAP values ​​of each index of each plot for the optimization objective, a visual decision interpretation report of each candidate spatial optimization scheme is generated.

[0014] Optionally, the comprehensive benefit index mentioned in step S4 is calculated according to the following formula: ; In the formula, E is the comprehensive benefit index. To score economic benefits, For social benefit evaluation For ecological benefit scoring, α, β, and γ are the weighting coefficients for economic, social, and ecological benefits, respectively, and α+β+γ=1. The weighting coefficients are dynamically adjusted according to the development stage and policy orientation of the target area. The dynamic simulation adopts a system dynamics model or a multi-agent model to simulate and extrapolate land use changes under different scenarios.

[0015] Optionally, the monitoring indicators for dynamic monitoring in step S5 include: the actual comprehensive performance score of the redeveloped land, the redevelopment progress completion rate, the plot ratio realization rate, the industry introduction rate, and the proportion of ecological land; the preset threshold is dynamically calculated based on the historical mean and standard deviation of each monitoring indicator, and an early warning is triggered when the monitoring indicator is lower than the mean minus two standard deviations or higher than the mean plus two standard deviations.

[0016] Secondly, the present invention provides a spatial configuration and pattern optimization system for the redevelopment of inefficient land, comprising: Multiple types of sensor units are arranged on the stator core of the hydro-generator unit to collect vibration signals, temperature signals, air gap signals and partial discharge signals of the stator core; The operating condition parameter acquisition interface is used to connect with the hydro-generator unit monitoring system to synchronously acquire the unit's operating condition parameters; The data acquisition and preprocessing unit is connected to the multi-type sensor unit and the working condition parameter acquisition interface. It is used to filter and denoise the acquired multi-source signals, remove outliers and synchronize and align them in time, and output standardized multi-source monitoring data. The multiphysics coupling analysis unit has a built-in three-dimensional finite element coupling model of electromagnetic field-temperature field-structural field based on the geometric dimensions and material properties of the stator core. It is connected to the data acquisition and preprocessing unit and is used to calculate the electromagnetic stress distribution, temperature gradient distribution and mechanical deformation distribution of the stator core using the excitation current and voltage data in the standardized multi-source monitoring data as excitation, and to extract the multiphysics coupling characteristic parameters. A comprehensive health index calculation unit is connected to the multi-physics field coupling analysis unit and is used to calculate the comprehensive health index HI of the stator core based on the multi-physics field coupling characteristic parameters. The deep learning prediction and diagnosis unit has a built-in hybrid model of graph attention network and long short-term memory network trained on the training sample set. It is connected to the data acquisition and preprocessing unit and the comprehensive health index calculation unit. It is used to predict the health status of the stator core in the future based on the standardized multi-source monitoring data and the comprehensive health index HI, and output fault warning signals and fault type identification results. The early warning output and display unit is connected to the deep learning prediction and diagnosis unit and is used to display the real-time monitoring data of the stator core, the comprehensive health index HI, the health status prediction trend, and the fault early warning information.

[0017] In summary, the present invention has at least one of the following beneficial technical effects: High recognition accuracy: This invention employs graph neural networks to fuse and analyze multi-source data. Through graph attention mechanisms, it adaptively learns the spatial dependencies and functional relationships between land parcels, overcoming the shortcomings of traditional methods that rely solely on a single data source or ignore spatial relationships. The multi-level recognition strategy achieves progressive and accurate recognition, progressing from "whether it is inefficient" to "what kind of inefficiency and how inefficient it is" and finally to "the internal structure of inefficiency."

[0018] The evaluation system is comprehensive and dynamic: This invention constructs a dynamic multidimensional evaluation index system covering four dimensions: economic, social, ecological, and spatial, overcoming the shortcomings of existing evaluation systems that are singular and subjective in weighting. By combining the entropy weight method with dynamic weight adjustment over time series, the evaluation results can reflect the dynamic changes in land use performance.

[0019] The optimization results are interpretable: This invention integrates an interpretability-based artificial intelligence module into a multi-objective optimization algorithm, using the SHAP value method to quantitatively explain the contribution of each indicator to the optimization results. Decision-makers can not only see "what the optimization should look like," but also understand "why this optimization was done," significantly improving the acceptability of the solution.

[0020] Full life cycle closed-loop management: This invention establishes a full-process closed-loop method of "identification-evaluation-optimization-simulation-monitoring". Through dynamic monitoring and automatic early warning mechanisms, it realizes full-process closed-loop management of inefficient land redevelopment from scheme formulation to implementation feedback, overcoming the shortcomings of existing methods that only focus on single-point optimization.

[0021] Multi-scenario simulation ensures robustness of the scheme: This invention comprehensively evaluates the robustness of each scheme through dynamic simulation under three scenarios: economic, social, and ecological, ensuring that the selected scheme has good adaptability and robustness under different development scenarios. Attached Figure Description

[0022] Figure 1 This is the schematic diagram of the method flow in Embodiment 1 of this application; Figure 2 This is a system diagram of Embodiment 2 of this application. Detailed Implementation

[0023] The following combination Figures 1-2 The present invention will be described in further detail below.

[0024] Example 1: Refer to Figure 1 This application provides a method for optimizing the spatial configuration and pattern of inefficient land redevelopment, including: Step S1: Multi-source dynamic data fusion and intelligent identification of inefficient land use: Acquire multi-source data for the target area, including land survey data, satellite remote sensing image data, mobile phone signaling data, enterprise electricity consumption data, enterprise tax payment data, POI data, and building census data.

[0025] The multi-source data undergoes spatiotemporal registration and standardization. Specifically, all data is unified to the same spatial coordinate system (such as the CGCS2000 National Geodetic Coordinate System) and time reference; default value processing, outlier removal, and normalization are performed on various types of data; a feature matrix of multi-source data fusion is constructed, and the feature vector of each plot includes economic benefit characteristics (tax revenue per unit area, revenue per unit area, output per unit of electricity consumption), spatial characteristics (plot ratio, building density, land area), location characteristics (transportation accessibility, coverage of public service facilities), and dynamic characteristics (spatiotemporal distribution of population, trend of enterprise electricity consumption), etc.

[0026] An inefficient land use identification model is constructed based on a graph neural network. Land parcels in the target area are abstracted as nodes V in a graph structure, and the spatial adjacency and functional relationships between parcels are represented as edges E, constructing a graph G=(V,E). The node feature matrix XX is composed of feature vectors fused from the aforementioned multi-source data. A graph attention network (GAT) is used to model the spatial dependencies of this feature matrix. GAT adaptively learns the contribution weights of different neighboring nodes to the central node through an attention mechanism; its core calculation process is as follows: For node i and its neighboring nodes j∈N(i), the attention coefficients are... for: ; In the formula, and Let be the feature vectors of nodes i and j, respectively; W be the learnable weight matrix; a be the parameter vector of the attention mechanism; ∥ denotes the vector concatenation operation; and LeakyReLU be the activation function. The attention coefficients are normalized using the softmax function to obtain the final attention weights. .

[0027] The inefficient land use identification model adopts a multi-level identification strategy: the first level is preliminary identification, which determines whether a plot of land is inefficient through binary classification; the second level is fine identification, which classifies the plots initially identified as inefficient land into multiple categories to determine the type of inefficient land use (such as inefficient industrial land, idle industrial and mining land, urban villages, shantytowns, etc.) and the degree of inefficiency (mild, moderate, severe); the third level is micro-identification, which identifies inefficient sub-regions within inefficient land use caused by the mismatch of the smallest unit structure.

[0028] Step S2: Multidimensional dynamic assessment and diagnosis of impairment factors: Construct a dynamic, multi-dimensional evaluation index system, which includes four dimensions: 1) Economic benefits dimension: tax revenue per unit area (ten thousand yuan / hectare), operating income per unit area (ten thousand yuan / hectare), electricity output per unit (ten thousand yuan / ten thousand kilowatt-hours), and employment density (persons / hectare).

[0029] 2) Social benefit dimension: accessibility of public service facilities (measured by the distance to the nearest public service facility), public transportation coverage (coverage within 500 meters of bus stops), and per capita public space area (square meters per person).

[0030] 3) Ecological benefit dimension: carbon sink per unit area (tons of carbon / hectare·year), green space ratio (%), and connectivity of ecological patches (connectivity index based on landscape ecology).

[0031] 4) Spatial efficiency dimensions: plot ratio, building density (%), and land use mix.

[0032] The land use mix degree is calculated according to the following formula: ; In the formula, Let K represent the land use mix degree of the i-th plot, and K be the total number of land use types. This represents the area proportion of the k-th land use type in the i-th plot; This was obtained by spatially overlaying and analyzing the land parcel vector plots and land use classification data on a GIS platform.

[0033] The entropy weight method combined with a time series dynamic weight adjustment method is used to calculate the weights of each indicator. Specifically, the initial weights of each indicator are first calculated using the entropy weight method; then, a time decay factor is introduced to give more weight to recent data, thus achieving dynamic adjustment of the weights.

[0034] The TOPSIS model is used to calculate the overall performance score of each parcel. The TOPSIS model obtains the relative proximity as the overall performance score by calculating the distance between each evaluation object and the positive and negative ideal solutions.

[0035] The obstacle degree model is used to diagnose the main obstacles affecting the overall performance of each plot of land. The obstacle degree is calculated according to the following formula: ; In the formula, Let the obstacle degree of the j-th indicator of the i-th plot be denoted as . The index value of the j-th index for the i-th plot is the standardized value. Let be the weight of the j-th indicator, and m be the total number of indicators; The original index values ​​were obtained by processing them using the range standardization method. The obstacle degree is calculated using the entropy weight method; the greater the obstacle degree, the stronger the constraint effect of the indicator on the overall performance of the land parcel.

[0036] Step S3: Optimization of Interpretable Spatial Pattern: Based on the inefficient land use identified in step S1 and the obstacle factors diagnosed in step S2, multiple optimization objectives are set and constraints are established.

[0037] The optimization objectives include: (1) maximizing the comprehensive benefits of land use; (2) maximizing spatial agglomeration (promoting the concentrated layout of similar land use and reducing fragmentation); and (3) minimizing ecological disturbance (reducing disturbance to ecologically sensitive areas).

[0038] The constraints include: (1) the uniqueness of land use type configuration (each plot can only be configured with one type of land use); (2) the upper and lower limits of plot ratio (meeting the requirements of the control detailed planning); (3) the minimum proportion of ecological land (ensuring that the proportion of ecological land is not lower than the prescribed proportion); and (4) the development intensity zoning control constraints (meeting the requirements of urban development intensity zoning control).

[0039] A multi-objective optimization algorithm integrating Explainable Artificial Intelligence (XAI) modules is used to optimize the spatial allocation of inefficient land use. The preferred multi-objective optimization algorithm is the Non-Dominated Ranking Genetic Algorithm with an elitist strategy (NSGA-II) or a multi-objective particle swarm optimization algorithm. The optimization variable is the land use type of each plot (e.g., industrial land, commercial land, residential land, green space, public service land, etc.).

[0040] The interpretability AI module uses the SHAP (SHapley Additive exPlanations) value method to calculate the contribution of each optimization objective to the decision variables. The SHAP value is based on the Shapley value in cooperative game theory, and calculates the marginal contribution of features to the prediction results according to the following formula: ; in, Let f(S) be the SHAP value of the j-th feature, F be the set of all features, S be the feature subset that does not contain the j-th feature, and f(S) be the model prediction value when only the feature subset S is used. By calculating the SHAP values ​​of each index of each plot for the optimization objective, a visual decision interpretation report of each candidate spatial optimization scheme is generated.

[0041] Using the above method, at least two candidate spatial optimization schemes are generated, and the optimization decision explanation information for each candidate scheme is output, including: (a) the achievement status of each optimization objective; (b) the basis for adjusting the land use type of each plot; (c) the ranking of the contribution of each indicator to the optimization result; and (d) the spatial layout visualization and adjustment explanation of the scheme.

[0042] Step S4: Multi-dimensional scenario simulation and solution evaluation: For each candidate spatial optimization scheme generated in step S3, dynamic simulations are performed under economic, social, and ecological scenarios.

[0043] Economic scenario assumptions: Economic development is the core objective, with a high proportion of industrial land and priority given to commercial and service land use; Social scenario assumptions: Improving people's livelihoods is the core objective, with a high proportion of residential and public service land use; Ecological scenario assumptions: Ecological protection is the core objective, with a high proportion of green space and water area, and development intensity is strictly limited.

[0044] The dynamic simulation employs either a system dynamics model or a multi-agent model to simulate and extrapolate land use changes under different scenarios. The system dynamics model simulates the dynamic process of land use change by constructing causal feedback loops; the multi-agent model simulates the evolution of land use patterns by simulating the decision-making behaviors of stakeholders such as the government, developers, and residents.

[0045] Calculate the comprehensive benefit index of each option under each scenario. The comprehensive benefit index is calculated according to the following formula: ; In the formula, E is the comprehensive benefit index. Economic efficiency is scored (based on indicators such as tax revenue per unit area and revenue per unit area). The social benefits are scored (based on indicators such as accessibility of public service facilities and per capita public space). Ecological benefit scoring (calculated based on indicators such as carbon sink and green space ratio) is achieved, with α, β, and γ representing the weighting coefficients for economic, social, and ecological benefits, respectively, and α+β+γ=1. These weighting coefficients are dynamically adjusted according to the development stage and policy orientation of the target area. For example, α can be appropriately increased in the early stage of industrialization, while the three coefficients can be balanced in the high-quality development stage. The dynamic simulation uses a system dynamics model or a multi-agent model to simulate and extrapolate land use changes under different scenarios.

[0046] The candidate schemes are ranked according to the comprehensive benefit index, and the optimal spatial configuration and pattern optimization scheme is output by comprehensively considering the performance stability under each scenario.

[0047] Step S5: Dynamic monitoring and feedback throughout the entire lifecycle: The implementation process of the optimal solution output in step S4 is dynamically monitored. Monitoring indicators include: the actual comprehensive performance score of the redeveloped land, the redevelopment progress completion rate, the plot ratio achievement rate, the industry introduction rate, and the proportion of ecological land.

[0048] Multi-source data of the implementation area are collected periodically (e.g., quarterly or semi-annually) to update the results of inefficient land use identification and comprehensive performance evaluation. The preset thresholds are dynamically calculated based on the historical mean and standard deviation of each monitoring indicator. Specifically, for each monitoring indicator, the mean μ and standard deviation of its historical data are calculated, and the threshold range is set to [μ-2σ, μ+2σ]. When a monitoring indicator is lower than μ-2σ or higher than μ+2σ, an early warning is triggered, indicating that an anomaly has occurred in the implementation process.

[0049] After the warning is triggered, the system automatically feeds back the abnormal information to step S3, and re-optimizes the spatial layout to realize the dynamic adjustment and iterative upgrade of the plan, forming a closed-loop management of "monitoring-early warning-feedback-optimization".

[0050] Example 2: This example uses the redevelopment of inefficient industrial land in the central urban area of ​​a prefecture-level city as an example to illustrate the method of the present invention in detail.

[0051] A high-tech industrial development zone in a prefecture-level city was selected as the study area. The zone covers approximately 45 square kilometers, of which about 12 square kilometers are industrial land. A large amount of inefficient industrial land exists within the zone, characterized by: some enterprises having a plot ratio below 0.8; tax revenue per mu below 50,000 yuan; and some factory buildings being idle or semi-idle. This area has been designated as a provincial-level pilot area for the redevelopment of inefficient land.

[0052] Data Acquisition and Preprocessing: Following step S1, collect the following multi-source data: 1) Land survey data: Obtain the third national land survey data from the local natural resources department, including the boundaries, area, and current land use type of each plot; 2) Satellite remote sensing image data: Acquire satellite remote sensing images of the study area at a resolution of 0.5 meters (two time phases in 2024 and 2025) for extracting building information and detecting changes.

[0053] 3) Mobile signaling data: Obtain mobile signaling data (after anonymization) of the study area from January 2024 to December 2025 from local telecommunications operators for analysis of population spatiotemporal distribution.

[0054] 4) Enterprise electricity consumption data: Monthly electricity consumption data of all industrial enterprises in the study area (January 2024 to December 2025) were obtained from the local power supply company.

[0055] 5) Enterprise tax data: Obtain annual tax data (2024 and 2025) for all industrial enterprises in the study area from the local tax authorities.

[0056] 6) POI data: Obtain POI data for the research area from internet map platforms, including points of interest such as businesses, shops, and public service facilities.

[0057] 7) Building survey data: Obtain building survey data for the study area from the local housing and construction department, including the number of building floors, building area, and building age.

[0058] All data were standardized to the CGCS2000 national geodetic coordinate system, and spatiotemporal registration was performed on various data types. Default values ​​were handled using multiple interpolation, and outliers were identified and removed using box plots. Each land parcel was used as the basic analysis unit, and a parcel feature matrix containing 47 feature dimensions was constructed.

[0059] Intelligent identification of inefficient land use: Following step S1, construct a graph attention network model. Use the 1256 plots in the study area as graph nodes, and construct the graph structure with the spatial adjacency relationship (shared boundary or distance less than 50 meters) and functional association relationship (such as enterprises in the same industrial chain) between plots.

[0060] The land parcel feature matrix was input into a graph attention network for training. The training set consisted of 236 inefficient land parcels and 412 efficient land parcels identified in the study area as labeled samples, while the validation set consisted of 20% of the randomly selected samples. The model used the Adam optimizer with a learning rate of 0.001 and 500 training epochs.

[0061] After training, all land parcels in the study area were identified. The identification results are as follows: a total of 387 inefficient industrial land parcels were identified, covering an area of ​​approximately 4.6 square kilometers, accounting for 38.3% of the total industrial land area. Among them, 189 parcels were slightly inefficient (floor area ratio 0.8-1.2, tax revenue per mu 50,000-150,000 yuan), 132 parcels were moderately inefficient (floor area ratio 0.5-0.8, tax revenue per mu 20,000-50,000 yuan), and 66 parcels were severely inefficient (floor area ratio below 0.5, tax revenue per mu below 20,000 yuan). Compared with traditional single-indicator identification methods, the identification accuracy of this invention increased from 72.6% to 91.3%, and the false positive rate decreased from 15.4% to 5.2%.

[0062] Multidimensional dynamic assessment and diagnosis of impairment factors: Following step S2, a dynamic multidimensional evaluation index system was constructed. Data for each index was collected from 387 inefficient industrial sites within the study area.

[0063] Economic benefit data: Tax payment data and operating revenue data for enterprises in each plot of land for 2024 and 2025 were obtained from the tax authorities; monthly electricity consumption data for enterprises in each plot of land were obtained from the power supply company. The average tax revenue per mu (10,000 yuan), average operating revenue per mu (10,000 yuan), electricity output per unit (10,000 yuan / 10,000 kWh), and employment density (persons / hectare) for each plot of land were calculated.

[0064] Social benefit data: Based on POI data and road network data, the distance from each plot to the nearest public service facilities (schools, hospitals, bus stops) is calculated in the GIS platform; the population density around each plot is calculated based on mobile phone signaling data.

[0065] Ecological benefit data: vegetation cover information is extracted from remote sensing images to calculate the green space ratio and carbon sink per unit area of ​​each plot; ecological patch connectivity is calculated based on landscape ecology methods.

[0066] Spatial efficiency data: Calculate the plot ratio and building density of each plot based on building survey data; calculate the plot ratio and building density based on land use classification data using the formula. Calculate the land use mix.

[0067] The entropy weight method was used to calculate the weights of each indicator. Taking 2025 data as an example, the calculated weights for each dimension were: economic benefit dimension 0.35, social benefit dimension 0.25, ecological benefit dimension 0.20, and spatial benefit dimension 0.20.

[0068] The overall performance score of each parcel was calculated based on the TOPSIS model. The overall performance score ranged from 0.12 to 0.78, with an average of 0.43.

[0069] Using the obstacle degree model according to the formula The obstacle degree of each indicator for each plot was calculated. Statistical analysis of the obstacle degree of 387 inefficient plots revealed the top five obstacle factors as follows: excessively low plot ratio (average obstacle degree 18.7%), low tax revenue per unit area (average obstacle degree 16.2%), insufficient building density (average obstacle degree 14.5%), low land use mix (average obstacle degree 12.3%), and poor accessibility of public service facilities (average obstacle degree 9.8%). These five factors account for 71.5% of the total obstacle degree and are the key obstacle factors restricting the redevelopment of inefficient land in this area.

[0070] Interpretable spatial pattern optimization: Follow step S3 to set the optimization objective and constraints.

[0071] The optimization objectives are set as follows: (1) maximize the comprehensive benefits of land use (based on the comprehensive performance score of step S2); (2) maximize spatial agglomeration (promote the concentrated layout of similar land use); (3) minimize ecological disturbance (reduce the disturbance to ecological corridors and ecological patches).

[0072] The constraints are set as follows: (1) Each plot can only be configured with one type of land use; (2) The plot ratio of industrial land shall not be less than 1.2 and not more than 3.0; (3) The proportion of ecological land shall not be less than 15%; (4) The development intensity shall meet the requirements of urban development intensity zoning control (plot ratio of 2.0~4.0 in core area, 1.0~2.5 in general area, and 0~0.8 in ecological area).

[0073] The NSGA-II multi-objective genetic algorithm was used for optimization. The algorithm parameters were set as follows: population size 200, number of iterations 500, crossover probability 0.8, and mutation probability 0.05. The optimization variable was the land use type of 387 inefficient plots, with selectable types including: Class I industrial land, Class II industrial land, commercial and service land, residential land, public management and public service land, and green space and plaza land.

[0074] The SHAP interpretability module is integrated into the optimization algorithm. For each candidate scheme, the SHAP value of each indicator under each optimization objective is calculated. Taking a certain candidate scheme as an example, the SHAP value analysis shows that: the contribution of the plot ratio indicator to the overall benefit objective is +0.32 (the largest positive contribution), the contribution of the tax revenue per unit area is +0.28, the contribution of the land use mix indicator is +0.18, and the contribution of the green space ratio indicator is -0.12 (a negative contribution). This indicates that the scheme mainly improves the overall benefit by increasing the plot ratio and tax revenue per unit area, but the decrease in the green space ratio has a negative impact on the overall benefit—this explanatory information provides decision-makers with a clear optimization logic.

[0075] After 500 iterations of optimization, 15 candidate schemes were generated on the Pareto frontier. Three representative candidate schemes were selected for further evaluation: Scheme A focuses on economic benefits (highest proportion of industrial land), Scheme B focuses on comprehensive balance (balanced proportion of various land uses), and Scheme C focuses on ecological benefits (highest proportion of green space).

[0076] Multi-dimensional scenario simulation and solution evaluation: Following step S4, dynamic simulations are performed on the three candidate schemes under economic, social, and ecological scenarios, respectively.

[0077] Economic scenario parameters: GDP annual growth rate target of 8%, and tax revenue per unit area of ​​industrial land increased to 300,000 yuan / mu; Social scenario parameters: per capita public space area target increased to 12 square meters, and public service facilities coverage within 500 meters target increased to 90%; Ecological scenario parameters: green space ratio target increased to 25%, and carbon sequestration target increased by 20%.

[0078] A multi-agent simulation method was adopted, setting up three types of agents: government, enterprises, and residents. The simulation period was 10 years (2026-2035), with a time step of 1 year. The decision-making rule of the government agent was to adjust land supply policies and planning control requirements according to the scenario objectives; the decision-making rule of the enterprise agent was to choose whether to move in or out based on land use costs and location conditions; and the decision-making rule of the resident agent was to choose a place to live based on the living environment and commuting costs.

[0079] After the simulation is completed, follow the formula. Calculate the comprehensive benefit index of each scheme under each scenario. In this embodiment, based on the characteristics of the city being in a high-quality development transformation stage, α=0.3, β=0.35, and γ=0.30 are set.

[0080] The simulation results are as follows: Option A (Economic Focus): Comprehensive benefit index 0.82 under economic scenario, 0.61 under social scenario, 0.55 under ecological scenario, and comprehensive score 0.66.

[0081] Option B (Comprehensive Balance): Comprehensive benefit index 0.73 under economic scenario, 0.78 under social scenario, 0.72 under ecological scenario, and comprehensive score 0.74.

[0082] Option C (Ecological Focus): Comprehensive benefit index 0.58 in economic scenario, 0.69 in social scenario, 0.85 in ecological scenario, and comprehensive score 0.71.

[0083] Option B demonstrated the most balanced performance across all scenarios, achieving the highest overall score (0.74). Therefore, it was selected as the optimal spatial configuration and layout optimization solution. The core content of Plan B includes: retaining or upgrading 156 (approximately 40%) of the 387 inefficient plots to industrial land (increasing the plot ratio to 1.5-2.5), adjusting 89 (approximately 23%) to commercial and service land, adjusting 72 (approximately 19%) to residential land, adjusting 42 (approximately 11%) to public management and public service land, and adjusting 28 (approximately 7%) to green space and plaza land. The overall plot ratio will increase from the current 0.76 to 1.85, the industrial land agglomeration index will increase from 0.43 to 0.71, and the proportion of ecological land will increase from 9.2% to 16.8%.

[0084] Full lifecycle dynamic monitoring and feedback: Following step S5, establish a dynamic monitoring system. Set the monitoring indicators and initial thresholds as follows: Starting from January 2026, monitoring data will be collected quarterly. Taking the first quarter of 2026 as an example, the collected data is as follows: 43 plots of land have been redeveloped (accounting for 11.1% of all inefficient plots), with an average comprehensive performance score of 0.52, a plot ratio achievement rate of 82%, an industry introduction rate of 71%, and an ecological land use ratio of 14.2%. All indicators are within the threshold range, and the system displays a "normal" status.

[0085] In the second quarter of 2026, monitoring revealed that the industry occupancy rate had dropped to 58%, below the warning threshold of 65%, triggering an automatic alert. Analysis indicated that the reason was the lagging infrastructure development on some plots, which affected businesses' willingness to move in. The system automatically relayed this information to step S3, re-optimizing the spatial layout—the adjusted plan increased the proportion of land for infrastructure and optimized the development sequence of industrial land. After the adjustments, the industry occupancy rate rebounded to 72% in the third quarter of 2026.

[0086] Through the above-mentioned dynamic monitoring and feedback throughout the entire life cycle, the scientific implementation and dynamic optimization of inefficient land redevelopment plans are ensured.

[0087] Example 3, referring to Figure 2This embodiment provides a spatial configuration and pattern optimization system for the redevelopment of inefficient land, including: The data fusion and standardization module is used to acquire multi-source data of the target area, perform spatiotemporal registration and standardization processing, and generate a fused land parcel feature matrix. The inefficient land use intelligent identification module, connected to the data fusion and standardization module, is used to abstract each plot as a graph structure node. Using the spatial adjacency and functional association between plots as edges, it uses a graph attention network to model the spatial dependency relationship of the feature matrix and outputs the distribution, type and inefficiency level of inefficient land use. This module includes a preliminary identification unit, a fine classification unit and a microstructure diagnosis unit, which are used to determine inefficiency attributes, subdivide inefficiency types and locate internal structural mismatch sub-regions, respectively. The multidimensional dynamic assessment and obstacle diagnosis module is connected to the inefficient land use intelligent identification module. It is used to construct an assessment indicator system from four dimensions: economic, social, ecological and spatial, dynamically calculate the indicator weights, comprehensively evaluate the performance of each plot, and identify the main obstacles that restrict performance improvement. An interpretable spatial pattern optimization module, connected to the multidimensional dynamic evaluation and obstacle diagnosis module, is used to set multiple optimization objectives and constraints based on the identification and diagnosis results, and to perform spatial configuration optimization using a multi-objective optimizer with integrated interpretable algorithms, generating at least two candidate schemes and their corresponding decision interpretation information; the optimization objectives include at least maximizing comprehensive benefits, increasing spatial clustering, and minimizing ecological disturbance, and the interpretation information is used to explain the contribution of each indicator to the optimization results; The multi-scenario simulation and scheme selection module is connected to the interpretable spatial pattern optimization module. It is used to perform dynamic simulations of candidate schemes under three scenarios: economic, social, and ecological. It calculates the comprehensive benefits of each scheme, ranks them, and outputs the optimal configuration scheme. The dynamic monitoring and feedback module is connected to the multi-scenario simulation and scheme selection module and the interpretable spatial pattern optimization module. It is used to periodically monitor the implementation scheme, trigger an early warning when key indicators deviate from the preset dynamic threshold, and automatically send the abnormal information back to the optimization module for scheme adjustment.

[0088] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for spatial configuration and pattern optimization in the redevelopment of inefficient land, characterized in that, include: Step S1: Multi-source dynamic data fusion and intelligent identification of inefficient land use: Acquire multi-source data of the target area, including land survey data, satellite remote sensing image data, mobile phone signaling data, enterprise electricity consumption data, enterprise tax data, POI data and building census data; Perform spatiotemporal registration and standardization processing on the multi-source data to construct a feature matrix for multi-source data fusion; A graph neural network-based inefficient land use identification model is constructed. The land parcels in the target area are abstracted as nodes in a graph structure. The spatial adjacency and functional association between land parcels are used as edges. The graph neural network is used to model the spatial dependency relationship of the feature matrix to identify inefficient land use and output the spatial distribution, type and degree of inefficiency of inefficient land use. Step S2: Multidimensional Dynamic Assessment and Obstacle Factor Diagnosis: Construct a dynamic multidimensional assessment index system, which includes economic benefit, social benefit, ecological benefit, and spatial benefit dimensions; calculate the weight of each index using the entropy weight method combined with the time series dynamic weight adjustment method; calculate the comprehensive performance score of each plot based on the TOPSIS model; and diagnose the main obstacle factors affecting the comprehensive performance of each plot using the obstacle degree model. Step S3: Interpretable spatial pattern optimization: Based on the inefficient land use identified in Step S1 and the obstacle factors diagnosed in Step S2, set multiple optimization objectives and establish constraints. A multi-objective optimization algorithm with an integrated interpretable artificial intelligence module is used to optimize the spatial configuration of inefficient land use and generate at least two candidate spatial optimization schemes. The interpretability AI module calculates the contribution of each optimization objective to the decision variables and outputs the optimization decision explanation information for each candidate solution. Step S4: Multi-dimensional scenario simulation and scheme evaluation: Dynamic simulations are performed on each candidate spatial optimization scheme generated in step S3 under economic, social and ecological scenarios, respectively, and the comprehensive benefit index of each scheme under each scenario is calculated; the candidate schemes are ranked according to the comprehensive benefit index, and the optimal spatial configuration and pattern optimization scheme is output. Step S5: Dynamic monitoring and feedback throughout the entire life cycle: Dynamically monitor the implementation process of the optimal solution output in Step S4, regularly collect multi-source data of the implementation area, update the inefficient land use identification results and comprehensive performance evaluation results, and trigger an early warning when the monitoring indicators deviate from the preset threshold and automatically feed back to Step S3 for solution adjustment.

2. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 1, characterized in that, The graph neural network mentioned in step S1 is a graph attention network, which adaptively learns the contribution weights of different neighboring nodes to the central node through an attention mechanism; the inefficient land use identification model adopts a multi-level identification strategy: the first level is preliminary identification, which determines whether the plot is inefficient land use; The second level is fine-grained identification, which determines the type and degree of inefficiency of inefficient land use; the third level is micro-level identification, which identifies inefficient sub-regions caused by the mismatch of the smallest unit structure within inefficient land use.

3. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 1, characterized in that, The dynamic multidimensional evaluation index system described in step S2 includes the following indicators: Economic efficiency dimensions: tax revenue per unit area, operating income per unit area, output per unit of electricity consumption, and employment density; Social benefits dimensions: accessibility of public service facilities, public transportation coverage, and per capita public space area; Ecological benefit dimensions: carbon sequestration per unit area, green space ratio, and connectivity of ecological patches; Spatial efficiency dimensions: plot ratio, building density, and land use mix; The land use mix degree is calculated according to the following formula: ； In the formula, Let K represent the land use mix degree of the i-th plot, and K be the total number of land use types. This represents the area proportion of the k-th land use type in the i-th plot; This was obtained by spatially overlaying and analyzing the land parcel vector plots and land use classification data on a GIS platform.

4. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 3, characterized in that, In step S2, the obstacle degree model calculates the obstacle degree of each index for each plot according to the following formula: ； In the formula, Let the obstacle degree of the j-th indicator of the i-th plot be denoted as . The index value of the j-th index for the i-th plot is the standardized value. Let be the weight of the j-th indicator, and m be the total number of indicators; The original index values ​​were obtained by processing them using the range standardization method. The obstacle degree is obtained by calculation using the entropy weight method; the greater the obstacle degree, the stronger the constraint effect of the indicator on the overall performance of the land parcel.

5. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 4, characterized in that, The optimization objectives of the multi-objective optimization algorithm in step S3 include: maximizing the comprehensive benefits of land use, maximizing spatial clustering, and minimizing ecological disturbance. The constraints include: uniqueness of land use type configuration, upper and lower limits of plot ratio, minimum proportion of ecological land, and zoning control of development intensity.

6. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 4, characterized in that, In step S3, the interpretability AI module uses the SHAP value method to calculate the contribution of each optimization objective to the decision variables. The SHAP value is based on the Shapley value in cooperative game theory and calculates the marginal contribution of features to the prediction results according to the following formula: ； in, Let f(S) be the SHAP value of the j-th feature, F be the set of all features, S be the feature subset that does not contain the j-th feature, and f(S) be the model prediction value when only the feature subset S is used. By calculating the SHAP values ​​of each index of each plot for the optimization objective, a visual decision interpretation report of each candidate spatial optimization scheme is generated.

7. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 1, characterized in that, The comprehensive benefit index mentioned in step S4 is calculated according to the following formula: ； In the formula, E is the comprehensive benefit index. To score economic benefits, For social benefit evaluation For ecological benefit scoring, α, β, and γ are the weighting coefficients for economic, social, and ecological benefits, respectively, and α+β+γ=1. The weighting coefficients are dynamically adjusted according to the development stage and policy orientation of the target area. The dynamic simulation adopts a system dynamics model or a multi-agent model to simulate and extrapolate land use changes under different scenarios.

8. The spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in claim 1, characterized in that, The monitoring indicators for dynamic monitoring in step S5 include: the actual comprehensive performance score of the redeveloped land, the redevelopment progress completion rate, the plot ratio realization rate, the industry introduction rate, and the proportion of ecological land. The preset threshold is dynamically calculated based on the historical mean and standard deviation of each monitoring indicator. When the monitoring indicator is lower than the mean minus twice the standard deviation or higher than the mean plus twice the standard deviation, an early warning is triggered.

9. A spatial configuration and pattern optimization system for the redevelopment of inefficient land, employing the spatial configuration and pattern optimization method for the redevelopment of inefficient land as described in any one of claims 1-8, characterized in that, include: The data fusion and standardization module is used to acquire multi-source data of the target area, perform spatiotemporal registration and standardization processing, and generate a fused land parcel feature matrix. The inefficient land use intelligent identification module, connected to the data fusion and standardization module, is used to abstract each plot as a graph structure node. Using the spatial adjacency and functional association between plots as edges, it uses a graph attention network to model the spatial dependency relationship of the feature matrix and outputs the distribution, type and inefficiency level of inefficient land use. This module includes a preliminary identification unit, a fine classification unit and a microstructure diagnosis unit, which are used to determine inefficiency attributes, subdivide inefficiency types and locate internal structural mismatch sub-regions, respectively. The multidimensional dynamic assessment and obstacle diagnosis module is connected to the inefficient land use intelligent identification module. It is used to construct an assessment indicator system from four dimensions: economic, social, ecological and spatial, dynamically calculate the indicator weights, comprehensively evaluate the performance of each plot, and identify the main obstacles that restrict performance improvement. An interpretable spatial pattern optimization module, connected to the multidimensional dynamic evaluation and obstacle diagnosis module, is used to set multiple optimization objectives and constraints based on the identification and diagnosis results, and to perform spatial configuration optimization using a multi-objective optimizer with integrated interpretable algorithms, generating at least two candidate schemes and their corresponding decision interpretation information; the optimization objectives include at least maximizing comprehensive benefits, increasing spatial clustering, and minimizing ecological disturbance, and the interpretation information is used to explain the contribution of each indicator to the optimization results; The multi-scenario simulation and scheme selection module is connected to the interpretable spatial pattern optimization module. It is used to perform dynamic simulations of candidate schemes under three scenarios: economic, social, and ecological. It calculates the comprehensive benefits of each scheme, ranks them, and outputs the optimal configuration scheme. The dynamic monitoring and feedback module is connected to the multi-scenario simulation and scheme selection module and the interpretable spatial pattern optimization module. It is used to periodically monitor the implementation scheme, trigger an early warning when key indicators deviate from the preset dynamic threshold, and automatically send the abnormal information back to the optimization module for scheme adjustment.

Images