Urban and rural space collaborative layout optimization method and system

Abstract

The invention relates to the technical field of territorial space planning, and discloses an urban and rural space collaborative layout optimization method and system.The urban and rural space collaborative layout optimization method comprises the steps that multi-source heterogeneous space data is obtained and preprocessed; urban and rural space elements are intelligently identified and accurately extracted; quantitative evaluation is carried out on the collaborative development level of the urban and rural space; carrying out space unit division and carrying out preliminary classification on each unit; solving an optimal configuration scheme of urban and rural spatial layout; development demand changes are predicted, different policy scenes are designed for scene simulation, and layout optimization is executed again on each scene; carrying out comprehensive evaluation and intelligent screening on the multi-scene dynamic optimization scheme; according to the invention, deep fusion of multi-source heterogeneous data and high-precision intelligent identification of urban and rural space elements are realized.

Description

Technical Field

[0001] This invention relates to the field of land and space planning technology, and more specifically, to a method and system for optimizing the coordinated layout of urban and rural spaces. Background Technology

[0002] Traditional urban and rural planning methods employ a fragmented layout strategy, resulting in a lack of effective coordination between urban and rural areas in terms of spatial functions, resource allocation, and industrial development. This leads to prominent contradictions in the urban-rural dual structure, inefficient resource allocation, and continuously increasing pressure on the ecological environment. While the rapid development of big data, artificial intelligence, and remote sensing technologies has provided new technical means for optimizing the coordinated layout of urban and rural spaces, existing technologies still have significant shortcomings in data integration, intelligent analysis, and dynamic optimization.

[0003] Existing urban and rural spatial layout optimization technologies mainly suffer from the following problems: Multi-source heterogeneous data are difficult to integrate effectively; remote sensing images, geographic information, statistical data, and real-time population flow data differ significantly in spatiotemporal scale and semantic expression, making it difficult for existing methods to achieve deep integration and collaborative analysis; There is a lack of quantitative evaluation methods for urban-rural spatial synergy, with qualitative descriptions being more prevalent and failing to scientifically measure urban-rural relationships; The accuracy of identifying boundaries in urban-rural transition zones is low, with traditional remote sensing classification methods struggling to accurately identify blurred transitional areas; Multi-objective, multi-constraint optimization is difficult to converge quickly, with traditional optimization algorithms exhibiting slow convergence speeds and a tendency to get trapped in local optima when facing high-dimensional and complex urban and rural spatial layout problems; There is a lack of dynamic optimization and scenario prediction capabilities, with existing methods primarily being static planning and unable to adapt to the dynamic changes in urban and rural development.

[0004] Therefore, there is an urgent need to propose a method for optimizing the coordinated layout of urban and rural spaces that can effectively integrate multi-source heterogeneous data, accurately identify urban and rural spatial elements, scientifically evaluate the degree of urban-rural synergy, quickly solve multi-objective optimization problems, and support dynamic prediction and scenario simulation, so as to solve the above-mentioned technical problems and provide technical support for the integrated development of urban and rural areas. Summary of the Invention

[0005] This invention provides a method and system for optimizing the coordinated layout of urban and rural spaces, which solves the technical problems in related technologies such as difficulty in integrating multi-source data, low accuracy in identifying urban-rural transition zones, slow convergence of multi-objective optimization, and lack of dynamic prediction capabilities.

[0006] This invention provides a method for optimizing the coordinated layout of urban and rural spaces, comprising: Acquire multi-source heterogeneous spatial data and preprocess them to obtain a standardized spatiotemporal data set; Based on standardized spatiotemporal datasets, urban and rural spatial elements are identified and extracted to obtain urban and rural spatial distribution maps and urban and rural boundary vector data. Based on urban and rural spatial distribution maps and urban and rural boundary vector data, the level of coordinated development of urban and rural spaces is quantitatively evaluated, and the evaluation results of urban and rural spatial coordination degree are obtained. Based on the evaluation results of urban-rural spatial coordination, spatial units are divided and each unit is initially classified to obtain optimized basic units and preliminary spatial classification results; Based on the evaluation results of urban-rural spatial coordination, the optimization of basic units, and the preliminary spatial classification results, the optimal configuration scheme of urban-rural spatial layout is solved to obtain the non-inferior layout optimization scheme. Based on the non-dominated layout optimization scheme, we predict changes in development needs and conduct scenario simulations. We then re-execute the layout optimization for each scenario to obtain a multi-scenario dynamic optimization scheme. The dynamic optimization schemes for multiple scenarios are evaluated and screened to obtain intelligent recommendation scheme documents.

[0007] In a preferred embodiment, the acquisition and preprocessing of multi-source heterogeneous spatial data includes: Collect satellite remote sensing image data and geographic information system data, acquire multispectral image data, and obtain administrative boundary vector data, road traffic network data, river system data, and topographic elevation data from the National Basic Geographic Information Database; Collect socioeconomic statistics and real-time population flow data. To address the issue of personal privacy in real-time population flow data, a differential privacy protection mechanism is adopted, and personal location information is protected by adding Laplace noise. Spatiotemporal registration, semantic alignment, and quality control are performed on multi-source heterogeneous spatial data. A time reference table is established, ontology mapping technology is used to establish conceptual associations across data sources, and a multidimensional spatiotemporal data cube is constructed.

[0008] In a preferred embodiment, the intelligent identification and accurate extraction of urban and rural spatial elements includes: We constructed a weakly supervised training sample set, extracted land use annotation data of the study area from open street maps as coarse-grained supervision signals, and invited urban and rural planning professionals to perform fine annotation; Design a boundary-enhanced semantic segmentation network model, using an encoder-decoder structure as the backbone network and adding a boundary enhancement module; Self-supervised pre-training was performed using publicly available remote sensing image datasets, with image inpainting task as the pre-training objective. Domain adaptation techniques were employed to reduce the distribution differences between the source and target domains. The semantic segmentation network model is trained and urban and rural spatial elements are identified, and the boundary is optimized using a conditional random field model.

[0009] In a preferred embodiment, the quantitative evaluation of the level of coordinated development of urban and rural spaces includes: Construct an evaluation index system for urban-rural spatial coordination, including spatial structure coordination, functional configuration coordination, factor flow coordination, and ecological protection coordination. Calculate the scores of each indicator in the urban-rural spatial coordination evaluation index system; The comprehensive score of urban-rural spatial coordination was calculated using a combined weighting method, and the weights of each indicator were determined by a combination of the analytic hierarchy process and the entropy weighting method.

[0010] In a preferred embodiment, the step of dividing the space into units and performing preliminary classification of each unit includes: Identify ecological red lines and ecological security patterns. Based on the vector data of ecological protection red lines determined by the national land spatial planning, identify ecological source areas using the ecosystem service importance assessment method and identify ecological corridors using the minimum cumulative resistance model. The basic units were divided and optimized. Natural geographical elements were used as unit boundaries. Voronoi diagram method was used for preliminary unit division. Urban centers, town centers and important transportation nodes were selected as generation points. The unit boundaries were adjusted in combination with topography, land use status and road network. The overall suitability of spatial units is evaluated, including suitability for urban development, suitability for agricultural production, and importance for ecological protection. A preliminary spatial classification was conducted, and the dominant function discrimination method was used to compare the relative sizes of the three suitability scores of each unit. The optimized basic units were then classified into four categories: ecological space, living space, production space, and composite space.

[0011] In a preferred embodiment, the process of finding the optimal configuration scheme for urban and rural spatial layout includes: Establish a multi-objective optimization mathematical model, define decision variables to assign functional types to each basic optimization unit, establish multi-objective functions to evaluate layout schemes respectively, and establish constraints. We designed a knowledge-guided particle initialization strategy, established an expert knowledge rule base in the field of urban and rural planning, and used a heuristic construction method to generate some high-quality initial solutions. An adaptive inertia weight and dynamic crowding distance mechanism are designed to dynamically adjust the inertia weight based on the algorithm's iterative progress and the population state. The optimization algorithm is executed to solve and output the optimization scheme. An improved multi-objective particle swarm optimization algorithm is adopted, and the particles are classified by a fast non-dominated sorting method.

[0012] In a preferred embodiment, the step of predicting changes in development demand and designing different policy scenarios for scenario simulation includes: Construct a model for predicting urban and rural development trends; Design multiple scenario parameters and conduct scenario simulations to predict the development parameters of each scenario at future time points; Execute scenario optimization and generate dynamic solutions. Re-execute the improved multi-objective particle swarm optimization algorithm for each parameter setting and perform incremental optimization based on the non-dominated layout optimization scheme. Analyze the spatial evolution trajectory and generate a scenario comparison report. Use the spatial transition matrix method to statistically analyze the mutual transformation of various spaces at different time points and generate a scenario comparison analysis report.

[0013] In a preferred embodiment, the comprehensive evaluation and intelligent selection of multi-scenario dynamic optimization schemes includes: A comprehensive evaluation index system for the proposed solution is established, comprising five dimensions: technical feasibility, economic rationality, social acceptance, ecological friendliness, and ease of implementation. The TOPSIS method was used to comprehensively rank the candidate schemes. The scores of the candidate schemes were calculated according to the comprehensive evaluation index system. Experts in urban and rural planning were invited to score and evaluate the indicators. A weighted normalized decision matrix was constructed to determine the ideal optimal solution and the ideal worst solution. The relative closeness of each scheme was calculated. A preference screening mechanism is used to generate a set of recommended solutions. An interactive questionnaire is used to collect decision-makers’ preferences for different goals and scenarios. A preference-based solution screening method is used, and a multi-attribute utility theory is used to construct the decision-makers’ utility function. Robustness and sensitivity analyses of the proposed solution were conducted, and scenario analysis was used to simulate the impact of external shocks on the implementation effectiveness of the solution.

[0014] In a preferred embodiment, the generation of the intelligent recommendation scheme document includes: Develop a visual decision support platform and generate decision documents. The main functional modules include a data display module, a scheme comparison module, a 3D visualization module, an interactive adjustment module, and a report generation module. The 3D visualization module uses 3D Earth Engine technology to construct a 3D scene of the study area and displays the optimization scheme in the form of 3D building models and terrain rendering. The interactive adjustment module allows decision-makers to fine-tune the recommended scheme. By clicking on the map and selecting a specific spatial unit, the functional type assignment can be modified. The system recalculates the adjusted objective function value and evaluation score in real time based on a multi-objective constraint optimization mathematical model. The report generation module automatically generates the final intelligent recommendation solution document based on the decision-maker's needs.

[0015] In a preferred embodiment, an urban-rural spatial collaborative layout optimization system is used to execute the above-described urban-rural spatial collaborative layout optimization method, including: The data preprocessing module is used to acquire multi-source heterogeneous spatial data and preprocess it to obtain a standardized spatiotemporal data set; The spatial element identification module, based on a standardized spatiotemporal dataset, identifies and extracts urban and rural spatial elements to obtain urban and rural spatial distribution maps and urban and rural boundary vector data. The coordination evaluation module, based on the urban-rural spatial distribution map and urban-rural boundary vector data, quantifies the level of coordinated development of urban and rural spaces and obtains the evaluation results of urban-rural spatial coordination. The spatial unit division module, based on the evaluation results of urban-rural spatial coordination, divides spatial units and performs preliminary classification of each unit, resulting in optimized basic units and preliminary spatial classification results; The layout optimization module, based on the evaluation results of urban and rural spatial coordination, the optimization of basic units, and the preliminary spatial classification results, solves the optimal configuration scheme of urban and rural spatial layout and obtains the non-inferior layout optimization scheme. The scenario simulation module, based on the non-dominated layout optimization scheme, predicts changes in development needs and performs scenario simulations. For each scenario, the layout optimization is re-executed to obtain a multi-scenario dynamic optimization scheme. The intelligent filtering module is used to evaluate and filter dynamic optimization solutions for multiple scenarios, and obtain intelligent recommendation solution documents.

[0016] The beneficial effects of this invention are as follows: It employs privacy-preserving data fusion technology, addressing the challenge of acquiring sensitive data through differential privacy and federated learning methods; it utilizes spatiotemporal registration and semantic alignment techniques to achieve collaborative analysis across data sources, constructing a multidimensional spatiotemporal data cube to provide a high-quality data foundation for subsequent analysis; and it employs weakly supervised deep learning methods, combined with transfer learning and active learning strategies, to achieve high-precision recognition even with insufficient labeled samples. Furthermore, it designs a boundary-enhanced semantic segmentation network to specifically address the problem of blurred boundaries in urban-rural transition zones, improving recognition accuracy and providing an accurate spatial information foundation for layout optimization.

[0017] This invention proposes an evaluation index system for urban-rural synergy encompassing four dimensions: spatial structure, functional configuration, factor flow, and ecological protection. It employs a combined weighting method to achieve scientific quantitative evaluation, providing targeted support for identifying weaknesses and optimizing improvements. Furthermore, it proposes a knowledge-guided improved multi-objective particle swarm optimization algorithm. By incorporating expert rules, it enhances the quality of initial solutions and utilizes adaptive inertia weights and dynamic crowding distance mechanisms to accelerate convergence while maintaining solution set diversity, thus shortening the optimization solution time.

[0018] By employing long short-term memory networks to establish predictive models that support dynamic optimization across multiple scenarios, and by using multi-criteria decision analysis and visualization technologies to provide intelligent solution recommendations and interactive decision support, the scientific, forward-looking, and practical nature of urban and rural spatial layout optimization has been comprehensively improved, providing important technical support for promoting integrated urban and rural development and optimizing the national land spatial pattern. Attached Figure Description

[0019] Figure 1 This is a flowchart of the main process of a method for optimizing the coordinated layout of urban and rural spaces in this invention; Figure 2 This is a detailed flowchart of a method for optimizing the coordinated layout of urban and rural spaces according to the present invention; Figure 3 This is a module diagram of an urban-rural spatial collaborative layout optimization system according to the present invention. Detailed Implementation

[0020] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0021] At least one embodiment of the present invention discloses a method for optimizing the coordinated layout of urban and rural spaces, such as... Figures 1 to 2 As shown, it includes: S1. Acquire multi-source heterogeneous spatial data and preprocess it to obtain a standardized spatiotemporal data set; S11 involves collecting satellite remote sensing imagery and geographic information system (GIS) data. Based on the geographic coordinates of the study area, multispectral imagery with a spatial resolution of one meter, obtained from Gaofen-2 satellite remote sensing images, reflects spatial characteristics such as land cover type, vegetation health, and building distribution. The time span covers the four seasons (spring, summer, autumn, and winter) over the past five years, resulting in a remote sensing imagery dataset. Simultaneously, GIS data acquisition tools were used to obtain administrative boundary vector data, road network data, river system data, and topographic elevation data from the National Basic Geographic Information Database, resulting in a basic geographic information dataset.

[0022] S12. Collect socio-economic statistics and real-time population flow data. Based on statistical yearbooks and government open data platforms, collect population statistics, GDP data, industrial structure data, fixed asset investment data, and public service facility distribution data for the study area over the past five years to obtain a socio-economic statistics dataset. To address the privacy concerns associated with real-time population flow data, a differential privacy protection mechanism is employed. This mechanism uses Laplace noise addition to protect individual location information while ensuring the overall distribution characteristics of the statistical data remain undistorted. This method is used to obtain time-specific population heat map data for weekdays and holidays within the study area over a year, data on major population flow channels, and urban-rural population exchange intensity data, resulting in a privacy-preserving population flow dataset.

[0023] S13. Collect industry association network data and point-of-interest (POI) data. Based on enterprise registration information and industry chain maps, web crawling and natural language processing technologies are used to extract industry classifications, supply chain relationships, and investment relationships of enterprises within the region, constructing an industry association network and obtaining industry association map data. The industry association map uses enterprises as nodes and supply, investment, and cooperation relationships as edges, reflecting the spatial distribution characteristics and association strength of urban and rural industries. Simultaneously, using a map application programming interface (API), POI data for the study area is obtained, including spatial location and attribute information for fifteen major categories and one hundred and twenty subcategories such as catering services, shopping facilities, educational institutions, medical institutions, cultural and entertainment venues, and government agencies, totaling over 100,000 records, resulting in a POI dataset.

[0024] S14. Spatiotemporal registration, semantic alignment, and quality control are performed on multi-source data. Based on the acquired multi-source heterogeneous data, coordinate transformation technology is used to unify the data from different coordinate systems to the national geodetic coordinate system, ensuring spatial consistency. For remote sensing image data, geometric correction and image registration techniques are used, with precise registration using ground control points, ensuring pixel-level alignment of images from different time phases, with registration accuracy controlled within half a pixel. A time reference table is established, standardizing the timestamps of the multi-source heterogeneous data to Beijing time, and establishing a time hierarchy structure according to year, season, month, day, and hour.

[0025] To address the semantic differences between different data sources, ontology mapping technology is employed to establish cross-data source conceptual associations. An ontology library for the urban and rural spatial domain is constructed, defining core concepts such as land type, building type, facility type, and activity type, along with their hierarchical relationships. Semantic association mapping is performed between construction land in remote sensing image classification results and buildings in point-of-interest data, as well as built-up area in statistical data, establishing a unified conceptual system. A deep semantic embedding network is used to map features from different data sources to a unified high-dimensional semantic space, calculate semantic similarity, and establish an automated semantic alignment mechanism.

[0026] To address the issues of missing data and outliers, statistical analysis methods are employed to identify outliers. The quartile range for each attribute is calculated, and values ​​exceeding 1.5 times the interquartile range are marked as suspected outliers. Real outliers are then confirmed using a combination of domain knowledge and manual judgment. For confirmed outliers and missing values, spatiotemporal kriging interpolation is used for completion. Spatial autocorrelation is modeled using a variogram, and the optimal unbiased estimate is calculated for imputation. For data that cannot be interpolated, a generative adversarial network (GAN) is used for intelligent completion. The generator is trained to learn the underlying distribution patterns of the data, generating completed values ​​consistent with the statistical characteristics of the real data, resulting in semantically aligned, high-quality multi-source data.

[0027] S15. Constructing a multi-dimensional spatiotemporal data cube. Based on multi-source data after spatiotemporal registration and semantic alignment, a spatiotemporal data cube construction technique is employed, organizing the data according to three dimensions: longitude, latitude, and time. The study area is spatially divided into grid units of appropriate scale and temporally layered according to monthly intervals. For each spatiotemporal unit, multi-dimensional attributes such as remote sensing image features, number and type of points of interest, population flow intensity, and industrial distribution density within the unit are integrated to form a feature vector for the spatiotemporal unit. After quality control, the data integrity rate reaches over 97%, resulting in a standardized spatiotemporal dataset.

[0028] Furthermore, given the varying data availability across different regions, federated learning techniques can be employed to achieve cross-regional collaborative data fusion in data-rich areas. The aim is to achieve joint modeling of multi-regional data without sharing original sensitive data, thereby enhancing the model's generalization ability. Specifically, local data processing nodes are deployed in each region. Each node trains local model parameters using local data and uploads the homomorphically encrypted model parameter gradients to a central server. The central server aggregates the parameter updates from each node, forming global model parameters which are then distributed to all nodes. Each node updates its local model using these global parameters, achieving joint learning through multiple iterations. A secure multi-party computation protocol is used to protect privacy during parameter transmission, ensuring that no single node can reverse-engineer the original data of other nodes. This method is particularly suitable for cross-administrative regional urban-rural spatial collaborative layout optimization, enabling the joint mining of data value while protecting local data sovereignty.

[0029] S2, based on a standardized spatiotemporal dataset, identifies and extracts urban and rural spatial elements to obtain urban and rural spatial distribution maps and urban and rural boundary vector data; S21. Constructing a Weakly Supervised Training Sample Set. Based on remote sensing image data from a standardized spatiotemporal dataset, samples for urban and rural spatial element identification are constructed. Addressing the issue of the high cost of manual annotation in identifying urban and rural spatial elements, a weakly supervised learning strategy is employed to construct training samples. Land use annotation data for the study area is extracted from open street maps and used as coarse-grained supervision signals. Specifically, vector elements such as building outlines, road networks, water body boundaries, and green space extents are extracted from the open street maps. These vector elements are rasterized and overlaid with remote sensing images to generate pixel-level weakly labeled samples. A preliminary segmentation model is trained using these weakly labeled samples to predict all remote sensing images. The prediction confidence of each pixel is calculated, and the entropy method is used to measure prediction uncertainty. The one thousand image patches with the highest uncertainty are selected as samples to be labeled. Urban and rural planning professionals are invited to perform detailed annotation on these one thousand image patches, covering eleven categories: urban built-up areas, rural settlements, industrial land, commercial land, public service facilities land, road and transportation land, farmland, forest land, grassland, water bodies, and unused land. Weakly labeled samples and finely labeled samples are mixed in a ratio of nine to one to obtain a weakly supervised training sample set.

[0030] S22, Design a boundary-enhanced semantic segmentation network architecture. Addressing the issue of blurred boundaries in urban-rural transition zones, a boundary-enhanced semantic segmentation network architecture is designed. An encoder-decoder structure is used as the backbone network. The encoder uses a residual network to extract multi-scale features, and the decoder restores spatial resolution through upsampling and feature fusion. Based on this, a boundary enhancement module is added, which includes an edge detection branch and an attention mechanism branch.

[0031] The edge detection branch uses the Laplacian operator and the Canni edge detection algorithm to extract boundary features; the attention mechanism branch uses a cascaded structure of spatial attention and channel attention to highlight the feature response of the boundary region.

[0032] In the loss function design of semantic segmentation networks, a weighted combination of cross-entropy loss and boundary loss is adopted. Cross-entropy loss evaluates the accuracy of pixel category prediction, while boundary loss is specifically calculated for boundary pixels, and a distance transformation map is used to assign higher weights to boundary pixels.

[0033] S23 employs transfer learning and self-supervised pre-training to improve model performance. To enhance the model's generalization ability under data scarcity, transfer learning and self-supervised pre-training strategies are adopted. Self-supervised pre-training is performed using a large-scale public remote sensing image dataset. Image inpainting is used as the pre-training objective, randomly occluding parts of the remote sensing image. The semantic segmentation network is trained to predict the content of the occluded area based on the visible area, allowing the semantic segmentation network to learn general feature representations such as texture, structure, and spectrum of the remote sensing image.

[0034] Based on self-supervised pre-training, supervised fine-tuning was performed using other existing urban remote sensing image segmentation datasets to learn the discriminative features of urban building, road, and vegetation categories. The pre-trained and fine-tuned model parameters were then used as initialization parameters for transfer learning fine-tuning on a weakly supervised training sample set in the study area.

[0035] To address the differences in geographic features across different urban and rural areas, a domain adaptation technique is employed to reduce the distributional discrepancies between the source and target domains. Specifically, an adversarial domain adaptation method is used, adding a domain discriminator to the semantic segmentation network. The domain discriminator distinguishes feature sources, while the feature extractor aims to extract domain-invariant features to deceive the domain discriminator. Through adversarial training, the feature extractor learns cross-domain general features, improving the model's generalization performance in the target domain.

[0036] S24, Model Training and Urban-Rural Spatial Feature Recognition. Based on a weakly supervised training sample set and a designed boundary-enhanced semantic segmentation network, the model is trained using a stochastic gradient descent optimization algorithm. Appropriate batch size and learning rate are set, and a cosine annealing strategy is used to dynamically adjust the learning rate, with appropriate training iteration cycles. Data augmentation techniques are employed during training, including random flipping, random rotation, random cropping, brightness adjustment, and contrast adjustment, to increase the diversity of training samples and prevent overfitting. Model performance is evaluated on a validation set every ten training cycles, and the model parameters with the highest accuracy on the validation set are saved as the final model. After training, the final model is used to perform inference predictions on all remote sensing images of the study area. A sliding window strategy is used to input the remote sensing images into the model in blocks, and the prediction results for overlapping areas are fused using a weighted average to obtain pixel-by-pixel classification results for the entire area. The classification results are post-processed, using morphological opening and closing operations to remove isolated noise points and fill small holes. A conditional random field (CRF) model is used to optimize the boundary. The CRF establishes spatial consistency constraints between pixels and uses the class correlation of adjacent pixels to smooth the boundary, eliminating jagged edges and obtaining a distribution map of urban-rural spatial features.

[0037] S25: Extract urban-rural boundaries and perform multi-task function recognition. Based on the urban-rural spatial element distribution map obtained in S24, a boundary extraction algorithm is used to identify the boundaries of the urban-rural transition zone. A morphological gradient operator is used to calculate the boundary line between urban built-up areas and rural areas, and buffer analysis is performed on the boundary line to determine the range of the urban-rural transition zone.

[0038] Based on spatial element identification, this study integrates point-of-interest (POI) data and population flow data, employing a multi-task learning framework to identify spatial function types. Specifically, it statistically analyzes the number and density of different types of POIs within each spatial grid unit, calculates the temporal variation characteristics of population flow, and uses clustering analysis to classify spatial units into six major function types: residential-dominated, employment-dominated, commercial service-dominated, industrial production-dominated, ecological conservation-dominated, and mixed-function-dominated. The spatial element category and spatial function type are used as the two labels for multi-task learning, jointly training a multi-task network. By sharing a feature extraction layer, the two tasks mutually reinforce each other, improving identification accuracy. The result is a dual-labeled urban-rural spatial distribution map containing both spatial element category and function type data, as well as urban-rural boundary vector data.

[0039] Furthermore, due to the high computational complexity and slow inference speed of traditional semantic segmentation networks, lightweight neural network architectures can be adopted for applications requiring real-time or near-real-time processing. The aim is to significantly reduce the number of model parameters and computational load while maintaining high recognition accuracy, thereby improving processing efficiency. Specifically, depthwise separable convolution is used to replace traditional convolution operations. Depthwise separable convolution decomposes standard convolution into two steps: channel-wise convolution and pointwise convolution, reducing the number of parameters and computational load to one-eighth to one-ninth of the original. Channel pruning techniques are used to remove redundant feature channels. By calculating the contribution of each channel to the final output, channels with high contributions are retained, while those with low contributions are pruned. Knowledge distillation techniques are employed, using a large, complex model as the teacher model and a lightweight model as the student model. The student model not only learns the real labels but also the output distribution of the teacher model, absorbing the teacher model's knowledge. Through these lightweight techniques, the number of model parameters can be reduced to one-tenth of the original, inference speed is increased by more than five times, while recognition accuracy only decreases by two to three percent, allowing for efficient operation even on resource-constrained edge computing devices.

[0040] S3, based on the urban and rural spatial distribution map and urban and rural boundary vector data, quantitatively evaluates the level of coordinated development of urban and rural space, and obtains the evaluation results of urban and rural spatial coordination degree; S31. Construct an evaluation index system for urban-rural spatial synergy. This system encompasses four dimensions: spatial structure synergy, functional configuration synergy, factor mobility synergy, and ecological protection synergy. The spatial structure synergy dimension includes three secondary indicators: urban-rural spatial connectivity, urban-rural spatial compactness, and urban-rural accessibility. The functional configuration synergy dimension includes three secondary indicators: equalization of public services, infrastructure integration, and industrial division of labor and cooperation. The factor mobility synergy dimension includes four secondary indicators: population mobility intensity, capital mobility efficiency, technology diffusion speed, and information interaction frequency. The ecological protection synergy dimension includes three secondary indicators: ecological spatial integrity, ecological corridor connectivity, and ecosystem service value. These thirteen secondary indicators constitute the complete evaluation system.

[0041] S32, Calculating the sub-indices of spatial structure synergy. Based on urban and rural spatial distribution maps, the urban and rural spatial connectivity index is calculated using the landscape connectivity index from landscape ecology. Urban built-up areas and rural settlements are considered as patches, and road traffic networks are considered as connecting corridors. The integral connectivity index measures the overall connectivity level of urban and rural spaces. Specifically, the integral connectivity index measures overall connectivity by calculating the ratio of the area product of all patch pairs to their connection distance, then dividing by the square of the total landscape area. For the urban and rural spatial compactness index, a combination of compactness index and fractal dimension is used based on urban and rural boundary vector data. The compactness index reflects the regularity of spatial morphology by calculating the ratio of the perimeter to the area of ​​urban and rural construction land; higher compactness indicates a more intensive spatial layout. The fractal dimension calculation method is based on urban and rural boundary vector data, using box counting. The urban and rural boundary lines are covered with squares of different scales, the number of squares containing the boundary lines is counted, and the slope of the logarithmic relationship between the number of squares and the square scale is calculated to obtain the fractal dimension value. A higher value indicates a more complex boundary, reflecting the degree of fragmentation and sprawl in urban and rural spaces. For urban-rural accessibility indicators, spatial syntax theory is used for calculation based on road traffic network and point-of-interest (POI) data. Specifically, the road network is abstracted as an axis diagram, and the integration and connectivity of each road axis are calculated. The integration degree is calculated by counting the number of other axes that can be reached from a given road axis with the fewest turns. The fewer the turns and the more axes that can be reached, the higher the integration degree of that axis, reflecting its accessibility level in the overall network. The connectivity degree is calculated by counting the number of other axes that directly intersect with a given road axis. The more intersecting axes, the higher the connectivity degree, reflecting the strength of the direct connection between that road and adjacent roads. Further, the shortest path time cost from the city center to various rural points is calculated. The driving speed is determined based on factors such as road grade, traffic congestion, and road surface conditions. The shortest path is selected, and the overall accessibility index is obtained by weighting the rural points by population size.

[0042] After normalizing the three indicators—urban-rural spatial connectivity, urban-rural spatial compactness, and urban-rural accessibility—and weighting them in a 4:3:3 ratio, the scores for the spatial structure synergy sub-item are obtained.

[0043] S33, Calculation of the functional configuration synergy sub-indicator. For the public service equalization indicator, three basic public services—education, healthcare, and culture—are selected for evaluation. Based on point-of-interest data, the number and spatial distribution of facilities such as primary schools, middle schools, hospitals, clinics, and cultural activity centers in urban and rural areas are statistically analyzed. The per capita availability is calculated by separately counting the total number of various facilities in urban and rural areas and dividing it by the corresponding regional population. The service radius coverage rate is calculated by setting standard service radii for various facilities: one kilometer for primary schools, three kilometers for middle schools, and five kilometers for hospitals. Service circles are drawn with the facilities as centers, and the population covered within the circles is counted and divided by the total population to obtain the coverage rate. The Gini coefficient is used to evaluate the level of urban-rural public service equalization; a smaller coefficient indicates a smaller urban-rural disparity. Simultaneously, a two-step mobility search method is used to calculate the spatial accessibility of public services, comprehensively considering factors such as facility supply capacity, population demand, and transportation distance to obtain a comprehensive evaluation score for urban-rural public service equalization.

[0044] For infrastructure integration indicators, the focus is on evaluating the level of urban-rural integration of infrastructure such as transportation, water supply, drainage, power supply, and communication. Based on geographic information system data, data such as urban and rural road network density, water supply network coverage, sewage treatment facility service area, power grid reliability, and broadband network penetration rate are extracted. The difference coefficient is calculated by averaging the values ​​of various infrastructure indicators in urban and rural areas separately, dividing the absolute value of the difference between the two by the average value, and then averaging the difference coefficients of all indicators. The smaller the difference coefficient, the higher the level of integration.

[0045] For the industrial division of labor and cooperation index, network analysis is used to calculate the correlation strength between urban and rural industries based on industrial association map data. Specifically, the number of supply chain links, investment relationship strength, and frequency of technological cooperation between urban and rural enterprises are calculated. Network density is calculated by dividing the actual number of connections in the industrial network by the theoretical maximum number of connections. The clustering coefficient is calculated by dividing the actual number of connections between adjacent enterprises of each enterprise node by the theoretical maximum number of connections, and averaging over all nodes. These two indicators together measure the tightness of the industrial network. Furthermore, an industrial complementarity index is used to assess the functional complementarity level between urban high-end manufacturing and modern service industries and rural characteristic agriculture and rural tourism. The three secondary indicators are normalized and weighted, and then summed to obtain the sub-scores for functional configuration synergy.

[0046] S34 calculates the sub-indicators of factor mobility synergy. For the population mobility intensity indicator, based on the privacy-protected population mobility data obtained in S1, the total number and intensity of two-way population flow are calculated by statistically analyzing the daily commuting population, weekend leisure population, and holiday return-to-hometown population between urban and rural areas. The population mobility index is calculated by dividing the total number of urban and rural two-way migrants by the total urban and rural population, then multiplying by a mobility frequency weighting coefficient. Daily commuting has a weight of 1, weekend leisure has a weight of 0.5, and holiday return-to-hometown has a weight of 0.3. The weighted sum is then used to obtain the population mobility index. A higher index indicates more frequent urban-rural personnel exchanges and smoother factor mobility.

[0047] For the capital flow efficiency index, based on fixed asset investment data and financial institution credit data, the scale of urban capital investment in rural areas and the transaction amount of rural assets in cities are statistically analyzed to calculate the total amount and velocity of capital flow between urban and rural areas. The capital allocation efficiency index is calculated by dividing the actual rate of return generated by urban and rural capital flows by the theoretical optimal rate of return, and simultaneously calculating the actual risk level by dividing the theoretical minimum risk level. The ratio of return efficiency to risk efficiency is used as the capital allocation efficiency index to assess the degree of optimization of capital allocation between urban and rural areas.

[0048] For the technology diffusion speed indicator, based on patent data and industry-academia-research cooperation data, the number and frequency of technology transfers from urban research institutions and high-tech enterprises to rural enterprises are statistically analyzed. The technology diffusion lag is calculated by calculating the time interval from the first application of each technology in the city to its application in the countryside, and then weighting the lags of all technologies. The weights are determined according to the importance of the technologies; the shorter the lag, the faster the technology diffusion speed. For the information interaction frequency indicator, based on social media and e-commerce data, the information interaction frequency is calculated by statistically analyzing the number of social media interactions and e-commerce transactions by urban and rural residents, then standardizing on a monthly basis and weighting the sum. The weight for social interaction is 0.6, and the weight for e-commerce transactions is 0.4, resulting in the information interaction frequency index, which reflects the activity level of information exchange between urban and rural areas. After normalizing the above four secondary indicators and weighting the sum, the sub-scores of the factor flow synergy are obtained.

[0049] S35, Calculate the sub-indices of ecological protection synergy. For the ecological space integrity index, based on the land use classification results in the urban and rural spatial distribution map, extract ecological land such as forest land, grassland, wetland, and water bodies, and calculate the area proportion and patch integrity of ecological land. The landscape fragmentation index is calculated by dividing the total number of ecological space patches by the total area of ​​ecological space, then multiplying by a standardization coefficient, and simultaneously calculating the ratio of the average patch area to the largest patch area. The two indicators are weighted and averaged to obtain the fragmentation index. The lower the fragmentation, the more complete the ecological space. Simultaneously, the encroachment of construction land within the ecological red line area is identified. The ecological red line protection effectiveness is calculated by dividing the area within the ecological red line area that maintains its original ecological function by the total area of ​​the ecological red line, then subtracting the proportion of encroached or degraded areas to obtain the protection effectiveness index.

[0050] For the ecological corridor connectivity index, ecological source areas are identified based on the assessment of the importance of ecosystem services, and ecological corridors are identified using the minimum cumulative resistance model. The total length, average width, and number of connected ecological source areas of the corridors are statistically analyzed, and a weighted average is used to calculate the ecological corridor connectivity index score.

[0051] For the ecosystem service value index, based on ecosystem service assessment methods, the value of ecosystem services such as water conservation, soil retention, carbon sequestration and oxygen release, and biodiversity maintenance in the study area is calculated. Using a unit area ecosystem service value equivalent table, combined with land use type and area, the total service value of various ecosystems is calculated to evaluate the effectiveness of integrated urban-rural ecological protection. The three secondary indicators are normalized and then weighted and summed to obtain the sub-scores for the degree of synergy in ecological protection.

[0052] S36. A combined weighting method is used to calculate the comprehensive score of urban-rural spatial coordination. Based on the scores of the thirteen secondary indicators across the four dimensions mentioned above, the weights of each indicator are determined using a combination of the analytic hierarchy process (AHP) and the entropy weighting method. Subjective weights are obtained by constructing a judgment matrix, calculating eigenvectors, and performing consistency checks. The entropy weighting method determines weights based on the objective distribution of the data; the greater the dispersion of the indicator values, the greater the information entropy and the higher the weight of that indicator. The subjective and objective weights are combined using a multiplicative synthesis method to calculate the combined weight.

[0053] Specifically, subjective and objective weight vectors are calculated for each of the thirteen secondary indicators. The corresponding elements of the two weight vectors are multiplied and normalized to obtain a combined weight vector. A weighted summation method is used to multiply the scores of the thirteen secondary indicators by their combined weights and then sum them to obtain a comprehensive score for urban-rural spatial coordination. The score ranges from zero to one; a higher score indicates a higher level of coordinated urban-rural spatial development. Simultaneously, a spatial distribution map of urban-rural spatial coordination is generated, and the comprehensive score is visualized according to spatial grid units. A heat map is used to intuitively reflect the level of coordination in different regions, identifying weak links with low coordination levels. This provides a targeted basis for subsequent optimization, resulting in an evaluation result and a diagnostic report on weak links for urban-rural spatial coordination.

[0054] Furthermore, given the differences in development stages and resource endowments across regions, and the varying focuses on coordinated urban-rural development, a fuzzy comprehensive evaluation method can be used instead of the weighted summation method. The aim is to better address the fuzziness and uncertainty in the evaluation process and provide more robust evaluation results. Specifically, a fuzzy evaluation set is established, dividing the evaluation results into five levels: excellent, good, average, poor, and very poor. For each secondary indicator, a membership function is established from the quantitative score to the fuzzy evaluation level. Commonly used membership functions include the trapezoidal distribution function and the normal distribution function. The membership degree of each indicator to its respective evaluation level is calculated based on its actual score. Fuzzy operators are used for multi-level fuzzy synthesis operations. First-level fuzzy synthesis within each dimension yields a four-dimensional fuzzy evaluation vector, and second-level fuzzy synthesis yields a comprehensive fuzzy evaluation vector. Finally, the maximum membership principle or the weighted average principle is used to determine the final evaluation level. This method can more comprehensively reflect the uncertainty of the evaluation and provide evaluation conclusions that are more consistent with human cognitive habits.

[0055] S4. Based on the evaluation results of urban-rural spatial coordination, spatial units are divided and each unit is initially classified to obtain optimized basic units and preliminary spatial classification results. S41. Identify ecological red lines and ecological security patterns. Based on the vector data of ecological protection red lines determined by the national land spatial planning, overlay them onto the study area to extract the scope of the ecological protection red lines. The space within this scope serves as a rigid constraint and is not involved in optimization adjustments. Ecological protection red lines include the core areas and buffer zones of nature reserves, primary protection zones for drinking water sources, important wetlands, and areas prone to geological disasters. These areas must be strictly protected, and any urban construction and industrial development activities are prohibited.

[0056] Based on the ecological protection red line, further identify the regional ecological security pattern. Employ ecosystem service importance assessment methods to evaluate the importance of ecological functions such as water conservation, soil and water conservation, biodiversity maintenance, and windbreak and sand fixation, identifying areas of extremely important and important ecological functions as ecological source areas. Ecological source areas are the core supporting areas for regional ecological security and require key protection and appropriate restoration.

[0057] Ecological corridors were identified using a minimum cumulative resistance model. Ecological resistance surfaces were constructed, and resistance values ​​were assigned based on the degree of obstruction to species migration and ecological processes caused by different land use types: forest and grassland resistance was set to 1, farmland to 10, rural settlements to 50, urban construction land to 100, and roads to 20-80 depending on their classification. A cost-path algorithm was used to calculate the minimum cumulative resistance for outward diffusion from each ecological source area, identifying the paths with the minimum cumulative resistance connecting different ecological source areas as potential ecological corridors.

[0058] The identified potential ecological corridors will be screened and optimized, retaining those that connect important ecological sources, are of moderate length and sufficient width, while eliminating those that are too narrow or fragmented. Ultimately, a regional ecological security pattern will be formed, consisting of ecological sources and ecological corridors.

[0059] S42, Delineation and Optimization of Basic Units. Based on the evaluation results of urban-rural spatial synergy, weak links with low synergy are identified. Under the constraints of the ecological security pattern, the optimizable space, excluding the ecological protection red line, is divided into units. Natural geographical elements are used as unit boundaries, and major rivers, ridgelines, administrative boundaries, etc., are extracted as natural boundaries for unit division to ensure that the unit division conforms to the natural geographical pattern.

[0060] The Voronoi diagram method was used for initial unit division, selecting city centers, town centers, and important transportation nodes as generation points. Voronoi polygons were generated based on Euclidean distance, where the distance from a point within a polygon to its generation point is less than its distance to other generation points. This polygon constitutes a preliminary spatial unit, reflecting the radiating influence of the central location on the surrounding space. Based on the Voronoi units, unit boundaries were adjusted by considering topography, land use, and road networks. The constrained Delaunay triangulation method was used, with linear elements such as rivers and roads as constraint edges, to subdivide and merge the Voronoi polygons, aligning unit boundaries with real-world boundaries. Units with control areas between 5 and 20 square kilometers were merged, and excessively large units were subdivided to ensure a suitable unit scale for optimization. Finally, the study area was divided into several basic optimization units, each with relatively independent geographical characteristics and functional attributes, serving as the basic spatial operation unit for layout optimization.

[0061] S43, Evaluate the comprehensive suitability of spatial units. For each optimized basic unit, evaluate its comprehensive suitability score in three aspects: suitability for urban construction, suitability for agricultural production, and importance for ecological protection.

[0062] The suitability assessment for urban development selects five factors: topographic slope, geological disaster risk, transportation location conditions, current development intensity, and water resource carrying capacity. Units with a slope of less than 15 degrees, low geological disaster risk, proximity to major roads, moderate current development intensity, and relatively abundant water resources are rated as highly suitable, while those with higher slopes are rated as low suitable. The comprehensive score for urban development suitability is calculated by assigning weights to the five factors. In this embodiment, the weights for topographic slope, geological disaster risk, transportation location conditions, current development intensity, and water resource carrying capacity are 0.3, 0.2, 0.25, 0.15, and 0.1, respectively. The comprehensive score is obtained by multiplying the score of each factor by its weight and then summing the results.

[0063] Agricultural production suitability assessment selects five factors: soil quality, irrigation conditions, plot flatness, climate suitability, and farmland contiguousness. Units with fertile soil, convenient irrigation, flat terrain, suitable climate, and concentrated farmland contiguousness are rated as highly suitable. Permanent basic farmland protection zones are specifically identified; units within these zones receive the highest level of agricultural production suitability assessment and may not be converted into construction land.

[0064] The assessment of the importance of ecological protection selected five factors: ecosystem service value, biodiversity richness, ecological sensitivity, distance from the ecological source area, and vegetation cover. Units with high ecosystem service value, rich species diversity, high ecological sensitivity, proximity to the ecological source area, and high vegetation cover were rated as highly important.

[0065] The three suitability scores for each unit are normalized. The higher the score, the higher the suitability of that type of use, thus obtaining the comprehensive suitability evaluation result of the space unit.

[0066] S44. Preliminary spatial classification. Based on the comprehensive suitability evaluation results of spatial units, combined with the current land use types and urban-rural spatial coordination evaluation results, each unit is preliminarily classified into production space, living space, and ecological space. Production space primarily undertakes agricultural and industrial production functions, living space primarily undertakes residential and public service functions, and ecological space primarily undertakes ecological protection and restoration functions. The dominant function discrimination method is used to compare the relative suitability scores of the three categories for each unit.

[0067] Units with the highest score for ecological protection importance, exceeding the scores of the other two categories, are classified as ecological spaces. These units are primarily focused on ecological protection and restoration, with strict control over development and construction activities. Units with the highest score for urban suitability are further assessed to determine if they are currently urban built-up areas or planned development zones. If so, they are classified as living spaces. These units are primarily focused on residential, public service, and commercial services.

[0068] For units with the highest agricultural production suitability scores, it is determined whether they are located in permanent basic farmland protection zones or important agricultural product production protection zones. If so, they are classified as production spaces, which are primarily engaged in agricultural and industrial production. For units with similar suitability scores across the three categories and no significant differences, they are classified as composite spaces. These units can perform multiple functions and have greater flexibility for optimization and adjustment.

[0069] Based on the above classification rules, each basic optimization unit is classified into four categories: ecological space, living space, production space, and composite space, generating a preliminary classification vector map of the three-life space. This classification result serves as the initial state for layout optimization, resulting in the preliminary spatial classification result.

[0070] S5. Based on the evaluation results of urban and rural spatial coordination, the optimization of basic units and the preliminary spatial classification results, the optimal configuration scheme of urban and rural spatial layout is solved to obtain the non-inferior layout optimization scheme. S51. Establish a multi-objective optimization mathematical model. Based on the basic optimization unit, define the decision variable as the functional type allocation for each basic optimization unit, represented by integer variables: one represents ecological space, two represents production space, and three represents living space. Under the constraints, find the optimal functional type allocation scheme so that multiple objective functions simultaneously reach or are close to the optimal.

[0071] The first objective function evaluates the economic benefits generated by the layout scheme. Based on the spatial attributes of the basic units, it specifically calculates the economic output of each unit under different functional types. Living space is calculated based on the output per unit area of ​​urban construction land, production space based on the output per unit area of ​​industrial and agricultural land, and ecological space based on the ecosystem service value. The total economic benefit of the scheme is obtained by summing the economic output of all units, and the optimization objective is to maximize this value. The second objective function evaluates the degree to which the layout scheme promotes urban-rural integration. Based on the synergy evaluation model established from the urban-rural spatial synergy evaluation results, the optimized spatial layout scheme is substituted into the evaluation model to recalculate the spatial structure synergy, functional configuration synergy, factor flow synergy, and ecological protection. The scores of the four dimensions of synergy are weighted and summed to obtain the optimized comprehensive score of urban-rural spatial synergy, with the optimization objective being to maximize this value. The third objective function assesses the negative impact of the layout scheme on the ecological environment, specifically calculating the area of ​​ecological space encroached upon by the development of living and production spaces, the length of ecological corridors cut, and the degree of enclosure of ecological source areas. An ecological impact assessment model is used to calculate the ecological loss value, with the optimization objective being to minimize this value. The fourth objective function assesses the spatial morphological intensification of the layout scheme, calculating the fractal dimension and compactness index of living and production spaces. An excessively high fractal dimension indicates that the space is too fragmented and sprawling, while an excessively low compactness indicates that the space is not used intensively. A reasonable compactness target range is set, with the optimization objective being to control the compactness index within the target range.

[0072] Establish constraints, including constraints related to ecological protection red lines, permanent basic farmland protection, urban development boundaries, total area of ​​various land uses, and spatial adjacency coordination. Within ecological protection red lines, the functional type of the units must be ecological space that cannot be altered; within permanent basic farmland protection zones, the functional type of the units must be production space that cannot be altered; the total area of ​​living space must not exceed the upper limit of the urban development boundary area determined by the national land space plan; and the total area of ​​ecological space must not be lower than the lower limit of the ecological protection target.

[0073] By establishing the above objective function and constraints, a multi-objective constrained optimization mathematical model is formed.

[0074] S52 introduces a knowledge-guided particle initialization strategy. A multi-objective particle swarm optimization (MPS) algorithm is employed to simulate bird flock foraging behavior to find the optimal solution. In the MPS algorithm, each particle represents a candidate layout scheme, and the particle's position corresponds to the value of the decision variable established in S51. The particle swarm gradually approaches the optimal solution of the multi-objective constrained optimization mathematical model through information sharing and collaborative search.

[0075] The design employs a knowledge-guided particle initialization strategy to establish an expert knowledge rule base in the field of urban and rural planning, extracting planning standards and expert experience to form rules. For example, living spaces should be prioritized for placement in units with convenient transportation, flat terrain, and high current development intensity; industrial land in production spaces should be far away from ecologically sensitive areas and residential areas; and ecological spaces should maintain connectivity to avoid fragmentation.

[0076] Based on these expert rules, a heuristic construction method is used to generate some high-quality initial solutions. Based on the preliminary spatial classification results, the functional types of units corresponding to ecological protection red lines and permanent basic farmland are fixed. Based on the comprehensive suitability scores of the optimized basic units, units are sorted from high to low according to their comprehensive suitability scores, and the most suitable functional type is assigned to units with high suitability scores first. During the assignment process, it is checked whether the constraints and expert rules established in S51 are met. If they are met, the assignment is confirmed; otherwise, a suboptimal selection is attempted. This method generates 30% of the high-quality initial particles in the population. The remaining 70% of particles are still generated using a random initialization method to maintain population diversity and avoid premature convergence caused by overly similar initial solutions. The knowledge-guided particles and randomly generated particles are mixed to form the initial particle swarm.

[0077] S53 employs an adaptive inertia weighting and dynamic crowding distance mechanism. A particle swarm optimization algorithm is used, where particle velocity consists of three parts: inertia, individual cognition, and social cognition. Inertia weighting controls the degree to which particles maintain their original direction of motion; higher weights are beneficial for global search, while lower weights are beneficial for finer local search.

[0078] An adaptive inertia weight adjustment strategy is designed to dynamically adjust the inertia weight based on the algorithm's iterative progress and the population state. Specifically, a relatively large inertia weight of 0.9 is set in the early stages of iteration to encourage particles to explore the search space extensively. In the middle stages of iteration, the inertia weight is adaptively adjusted based on the population's convergence rate. If the population diversity decreases rapidly, it indicates that the algorithm is converging quickly, so the inertia weight is reduced to accelerate the convergence speed. If the population diversity remains high, it indicates that further exploration is needed, so a relatively large inertia weight is maintained. In the later stages of iteration, a smaller inertia weight of 0.4 is set to encourage particles to perform local fine-tuning searches near the current optimal solution.

[0079] Population variance is used as a diversity measure, and inertial weights are dynamically adjusted based on the variance change rate to achieve adaptive balance between global and local search.

[0080] To address the maintenance problem of the Pareto optimal solution set in multi-objective optimization, a dynamic crowding distance mechanism is designed. During algorithm iteration, an external archive is maintained, storing all currently searched non-dominated solutions, i.e., Pareto optimal solutions (in multi-objective optimization, the solution for which improvement in any one objective cannot worsen other objectives). When the number of solutions in the external archive exceeds a preset size, some solutions need to be deleted to control the archive size. Traditional methods use a fixed crowding distance calculation method, which may lead to uneven distribution of the solution set.

[0081] A dynamic crowding distance calculation method is designed, which dynamically adjusts the distance calculation weights based on the distribution of solutions in the target space. For regions with sparse solution distribution, the crowding distance of solutions in that region is increased, reducing the probability of deletion and encouraging the algorithm to search in that region, thereby increasing the breadth of the solution set. For regions with dense solution distribution, the crowding distance of solutions in that region is decreased, increasing the probability of deletion and preventing the solution set from becoming overly concentrated.

[0082] S54 executes the optimization algorithm to solve the problem and output the optimized solution. Based on the initial particle swarm, adaptive inertia weight strategy, and dynamic crowding distance mechanism, an improved multi-objective particle swarm optimization algorithm is executed. Algorithm parameters are set, including population size, external archive size, maximum number of iterations, learning factor, etc., and the constraint violation penalty coefficient is adaptively set according to the degree of violation.

[0083] In each iteration, the four objective function values ​​for each particle are calculated based on the multi-objective constraint optimization mathematical model, and it is checked whether the constraints defined in S51 are satisfied. For particles that violate the constraints, a constraint repair strategy is used for correction. If the rigid constraints are still violated after correction, a large penalty value is applied to reduce the fitness of the particle. A fast non-dominated sorting method is used to classify the particles, dividing the population into multiple non-dominated layers.

[0084] Update the individual optimal position and global optimal position of each particle. The individual optimal position is the best solution found during the particle's historical search. The global optimal position is selected from the external archive using a crowding distance selection mechanism, prioritizing solutions with larger crowding distances as global leaders to encourage the population to explore different regions of the solution space. Calculate the new velocity and position of each particle according to the velocity and position update formulas. Since the decision variables defined in S51 are integers, rounding is used to convert continuous values ​​to discrete integer values ​​after the position update.

[0085] Add the non-dominated solutions generated in the current iteration to the external archive and update the optimal solution front in the archive. If the number of solutions in the archive exceeds the size limit, use a dynamic crowding distance mechanism to delete solutions with small crowding distances. Determine the algorithm's termination condition: if the maximum number of iterations is reached or the optimal solution front has not improved for twenty consecutive iterations, then terminate the iteration.

[0086] After the algorithm terminates, the solutions saved in the external archive are the final optimal solution set, yielding the non-dominated layout optimization schemes. Each scheme corresponds to a spatial unit function type allocation result, and the values ​​of the scheme on the four objective functions established in S51 are given. A scheme comparison table is generated to show the performance differences of different schemes, resulting in a comparison data of non-dominated layout optimization schemes and objective function values.

[0087] Furthermore, since different decision-makers have different preferences for various optimization objectives, preference information can be introduced into the optimization process for preference-guided optimization. The aim is to focus the optimization search on the target regions favored by decision-makers, reduce the waste of computational resources in non-favored regions, accelerate the algorithm's convergence speed, and provide solutions that better meet decision-making needs. Specifically, before the algorithm starts, the decision-makers' weight preferences for each objective are collected through an interactive interface. For example, if a decision-maker indicates a greater emphasis on ecological environmental protection, then ecological objectives are assigned higher weights. During particle evaluation and selection, a weighted aggregation function is used to transform the multi-objective problem into a single-objective problem with preferences. The values ​​of each objective function are weighted and summed according to the weights given by the decision-makers, so that solutions that satisfy preferences have higher fitness. At the same time, a certain proportion of the population is still retained for unbiased global search to avoid premature convergence to local optima. Through this preference-guided mechanism, the algorithm can generate more high-quality solutions concentrated in the decision-maker's preferred regions while ensuring the diversity of the solution set, thereby improving the practicality of the solutions and the efficiency of decision-making.

[0088] S6. Based on the non-dominated layout optimization scheme, predict changes in development needs and conduct scenario simulations. Re-execute layout optimization for each scenario to obtain a multi-scenario dynamic optimization scheme. S61. Constructing a Predictive Model for Urban and Rural Development Trends. Based on non-dominated layout optimization schemes and specific layout scheme data, combined with time series data from the standardized spatiotemporal dataset obtained in S1, including population statistics, economic development data, and land use change data for the past five years, a Long Short-Term Memory (LSTM) network is used to construct a predictive model for urban and rural development trends. LSTM is a special type of recurrent neural network capable of learning long-term dependencies in time series data, making it suitable for predicting urban and rural development processes with complex temporal dynamics.

[0089] A population growth prediction model is constructed, taking into account historical total population, natural population growth rate, mechanical population growth rate, urbanization rate, number of jobs, and public service level. The output is the total population and urban-rural population distribution for each future year. A multi-layered Long Short-Term Memory (LSTM) network is designed with a decreasing number of neurons, and the final result is output via a fully connected layer. Monthly data from the past five years are used as training samples, with training and validation sets divided according to requirements. Mean squared error is used as the loss function, and the Adam optimizer is used for training, with appropriate learning rate and batch size set.

[0090] The same method was used to construct industrial development forecasting models and land demand forecasting models. The industrial development forecasting model takes into account historical GDP, industrial structure ratios, fixed asset investment, and technological innovation input data, and outputs the future growth rate and output scale of each industry. The land demand forecasting model takes into account historical construction land area, land development intensity, and per capita construction land area data, and outputs the future demand area for urban construction land and agricultural land. After model training, the prediction errors on the validation set were all controlled within 8%, resulting in the urban and rural development trend forecasting model.

[0091] S62, Design multi-scenario parameters and conduct scenario simulations. Based on the prediction model, design three different policy scenarios for scenario simulation.

[0092] Scenario 1 is a rapid urbanization scenario, assuming that the urbanization rate will continue to be relatively fast in the future, with an average annual growth rate of 3% for the urban population, a continuous increase in the demand for urban construction land, economic development led by urban industry and service industries, and relatively relaxed ecological protection constraints.

[0093] Scenario 2 is a scenario of balanced urban and rural development. This scenario emphasizes the integration and coordinated development of urban and rural areas, controls the speed of urbanization to coordinate with rural development, sets the average annual growth rate of urban population at 1.5%, strictly controls the growth of urban construction land, vigorously develops rural industries and characteristic towns, and promotes the two-way flow of factors between urban and rural areas.

[0094] Scenario 3 is the ecological priority scenario. This scenario prioritizes ecological environmental protection, strictly controls development intensity, and ensures that urban population growth is mainly supported by increasing the level of intensive land use and building density. The average annual growth rate of urban population is one percent, the area of ​​ecological space is increased, and ecological restoration projects are implemented.

[0095] For each scenario, a predictive model is used to forecast development parameters such as population, industry, and land demand at three time points over the next five, ten, and fifteen years. These forecast parameters are then updated into a multi-objective optimization mathematical model, with adjustments made to constraints and objective function weights to reflect the policy orientation of different scenarios, resulting in nine sets of parameter settings for each of the three scenarios and three time points.

[0096] S63, Execute scenario optimization and generate dynamic solutions. For nine sets of parameter settings, re-execute the improved multi-objective particle swarm optimization algorithm to solve for the optimal layout scheme for each scenario at each time point. To accelerate computation, based on the non-dominated layout optimization scheme, the non-dominated scheme is used as the initial solution for the current time step. Incremental optimization is then performed on this basis, adjusting the functional types of some spatial units to adapt to new development needs and constraints.

[0097] Taking Scenario 1, rapid urbanization, as an example, the urban population is projected to increase by 200,000 over the next five years, requiring an additional 30 square kilometers of urban construction land. In the optimization process, priority is given to converting production and mixed-use spaces in the surrounding areas of towns, with convenient transportation and good development conditions, into living spaces to meet the needs of urban expansion. Simultaneously, the location of ecological corridors is adjusted to avoid them being fragmented by urban construction. After optimization, the resulting five-year layout scheme for this scenario shows a 12% increase in living space area, a 0.05% improvement in urban-rural synergy score, and an increased ecological and environmental impact that remains within acceptable limits.

[0098] For the ten-year and fifteen-year periods, incremental optimization will continue, with further adjustments based on the five-year plan. The ten-year plan aims to increase the urban population by another 200,000 and construction land by another 30 square kilometers, with spatial expansion extending along major transportation axes. In the fifteen-year period, the pace of urbanization will begin to slow, the incremental increase will decrease, and the focus of optimization will shift to functional optimization and efficiency improvement of existing spaces.

[0099] For Scenario 2 and Scenario 3, optimization was performed at three time points to generate their respective dynamic layout scheme sequences. Scenario 2's scheme resulted in a relatively balanced expansion of urban and rural space, moderate urban growth, and effective protection and rational utilization of rural space, achieving the highest urban-rural synergy score among the three scenarios. Scenario 3's scheme resulted in highly intensive urban space, with a large amount of production space being converted into ecological space, improving the quality of the ecological environment, but with slower economic growth.

[0100] S64 analyzes spatial evolution trajectories and generates scenario comparison reports. Based on nine sets of dynamic optimization schemes, it analyzes the evolution trajectory and change patterns of urban and rural spaces under different scenarios. Using the spatial transition matrix method, it statistically analyzes the mutual transformation of various spatial types between different time points. A three-dimensional spatial transition matrix is ​​constructed, where rows represent the spatial type at the initial time, columns represent the spatial type at the target time, and matrix elements represent the area or number of units transformed from one type to another.

[0101] The spatial transformation rate and direction under different scenarios were calculated to reveal the dynamic mechanism of spatial evolution. Scenario 1 is characterized by the continuous transformation of production space and ecological space into living space, rapid urban expansion, and a clear trend of spatial sprawl. Scenario 2 is characterized by the relative stability of various types of space, a low transformation rate, and a relatively balanced spatial pattern. Scenario 3 is characterized by the transformation of some production space and degraded living space into ecological space.

[0102] A comparative analysis report of three scenarios is generated, comparing the performance of different scenario solutions from four dimensions: economic benefits, urban-rural coordination, ecological and environmental impact, and spatial compactness. Radar charts are created to visually demonstrate the comprehensive performance of each scenario across multiple indicators, and time-series curves are generated to show the evolution trend of each indicator over time. Textual descriptions are written to explain the advantages and limitations of each scenario, providing a scientific basis for decision-makers to choose development paths, and resulting in a report analyzing dynamic optimization solutions and future evolution trends across multiple scenarios.

[0103] Furthermore, given the numerous uncertainties surrounding future development, a single deterministic prediction may deviate from reality. Monte Carlo simulation can be used to assess the robustness of alternative plans. The aim is to quantify the impact of predictive uncertainty on the optimization plan and identify robust plans that perform well under multiple possible scenarios. Specifically, a probability distribution is set for the input parameters of the prediction model; for example, the population growth rate follows a normal distribution, and the industry growth rate follows a log-normal distribution, reflecting the uncertainty of the parameters. Using Monte Carlo random sampling, a set of parameter values ​​is randomly selected from the parameter distribution and substituted into the prediction model to obtain a set of prediction results. This sampling is repeated one thousand times to obtain one thousand possible future development scenarios. Layout optimization is performed on each scenario to obtain one thousand optimized plans. The distribution characteristics of these one thousand plans are statistically analyzed, and the frequency of occurrence of each spatial unit's functional type is calculated. Allocation plans with high frequencies indicate that they are better choices under multiple scenarios and have strong robustness. Spatial units with high robustness and sensitive units with low robustness are identified. Units with high robustness are given priority for protection or development, while flexible adjustment plans are developed for sensitive units. A robustness assessment report is generated to provide decision-makers with a basis for risk management.

[0104] S7 evaluates and filters dynamic optimization schemes for multiple scenarios to obtain intelligent recommendation scheme documents; S71. Construct a comprehensive evaluation index system for the proposed solutions. Based on the actual decision-making needs for optimizing urban and rural spatial layout, a comprehensive evaluation index system for the proposed solutions is constructed. This index system includes five dimensions: technical feasibility, economic rationality, social acceptability, ecological friendliness, and ease of implementation. The technical feasibility dimension evaluates whether the proposed solutions comply with planning technical specifications, meet engineering construction conditions, and have the technical basis for implementation, selecting three secondary indicators: compliance with specifications, engineering feasibility, and technological maturity.

[0105] The economic rationality dimension evaluates the economic benefits and cost efficiency of the proposed plan, selecting four secondary indicators: direct economic benefits, investment payback period, cost-benefit ratio, and job creation capacity. The social acceptance dimension evaluates the impact of the plan on residents' lives and public recognition, selecting three secondary indicators: resident satisfaction, degree of conflict of interest, and public participation. Data is collected through questionnaires and public hearings.

[0106] The ecological and environmental performance of the eco-friendly dimension evaluation scheme is assessed using four secondary indicators: ecological space protection rate, biodiversity impact, carbon emission intensity, and resource conservation degree. The ease of implementation dimension evaluates the operational complexity of implementing the scheme, using three secondary indicators: the number of stakeholders involved, the required policy adjustments, and the length of the implementation period. A total of seventeen secondary indicators constitute the complete evaluation system.

[0107] S72 uses the TOPSIS method for comprehensive ranking of schemes. Based on multi-scenario dynamic optimization schemes, for the nine groups of dynamic optimization schemes output from S6 and the non-dominated layout optimization schemes, a total of fifty-seven candidate schemes are calculated according to the comprehensive evaluation index system on seventeen secondary indicators. The Delphi method is used to invite fifteen experts in the field of urban and rural planning to score and evaluate the indicators. The values ​​of quantitative indicators are calculated directly, while the expert scoring method is used for qualitative indicators. The average of the expert scores is taken as the final score.

[0108] The TOPSIS method was used to comprehensively rank the schemes, and the score matrix of the seventeen indicators was normalized to eliminate the influence of different indicator dimensions. Weight vectors were determined based on indicator importance, and subjective weights were calculated using the analytic hierarchy process (AHP).

[0109] Construct a weighted normalized decision matrix by multiplying the normalized index values ​​by their corresponding weights. Determine the ideal optimal solution, where each index value is the best value among all solutions for that index, taking the maximum value for positive indices and the minimum value for negative indices. Determine the ideal worst solution, where each index value is the worst value among all solutions for that index.

[0110] Calculate the Euclidean distance from each solution to the ideal optimal solution and the ideal worst solution, taking into account the differences across seventeen index dimensions. Calculate the relative closeness of each solution, which is equal to the distance to the worst solution divided by the sum of the distances to the optimal and worst solutions. The closeness value ranges from zero to one; a higher closeness value indicates that the solution is closer to the ideal optimal solution. Sort the fifty-seven solutions from highest to lowest relative closeness to obtain the overall ranking result.

[0111] S73 employs a preference-based screening mechanism to generate a set of recommended solutions. Based on the comprehensive ranking of the solutions, decision-makers' preference information is incorporated for intelligent screening. An interactive questionnaire is used to collect decision-makers' preferences for different goals and scenarios, asking them whether they prioritize economic development or ecological protection, whether they prefer rapid urbanization or balanced development, and their tolerance for uncertainty and risk. Based on the decision-makers' responses, fuzzy inference methods are used to infer their preference functions.

[0112] A preference-based solution selection method is employed to calculate the degree of matching between each solution and the decision-maker's preferences. Specifically, the decision-maker's emphasis on each objective is converted into a weighting coefficient. A weighted score is calculated based on the objective function values ​​of the four solutions, with solutions having higher weighted scores indicating a greater alignment with the decision-maker's preferences. The ranking of the solutions in the comprehensive evaluation is also considered, with solutions ranking higher overall being given priority for recommendation.

[0113] Using multi-attribute utility theory, a utility function for decision-makers is constructed, which comprehensively reflects their preferences and risk attitudes towards various indicators. The expected utility value of each option is calculated, and the option with the highest utility value is the optimal recommended option. From fifty-seven candidate options, the top five options with the highest utility values ​​are selected as key recommended options. These five options perform optimally under the decision-maker's preferences and have high overall evaluation scores, resulting in an intelligent recommendation solution set.

[0114] S74. Conduct robustness and sensitivity analyses of the proposed solutions. For the five solutions in the recommended solution set, robustness and sensitivity analyses are performed to assess their stability under parameter disturbances and environmental changes. Based on the future evolution trend analysis report, the robustness analysis employs scenario analysis to simulate the impact of external shocks such as macroeconomic fluctuations, policy adjustments, and natural disasters on the implementation effectiveness of the solutions. Three external environments—optimistic, baseline, and pessimistic—are set up to evaluate the magnitude of performance changes for each solution under different environments. Solutions with smaller magnitudes of change are considered more robust.

[0115] Sensitivity analysis employed a single-factor sensitivity analysis method to identify the key parameters that have the greatest impact on the effectiveness of the proposed solution. Ten key parameters were selected, including population growth rate, economic growth rate, land price, ecological protection standards, and infrastructure investment. The value of each parameter was changed one by one, and the changes in the objective function value and comprehensive evaluation score of the proposed solution were observed. Sensitivity coefficients were calculated; a large absolute value of the coefficient indicates that the proposed solution is highly sensitive to that parameter, requiring close attention and risk control.

[0116] Generate robustness analysis reports and sensitivity analysis charts, identifying the robustness level and key sensitivities of each recommended solution, providing decision-makers with risk warning information. Simultaneously, propose risk response suggestions, develop monitoring and early warning mechanisms and emergency adjustment plans for sensitive parameters, and enhance the security of solution implementation.

[0117] S75: Develop a visual decision support platform and generate decision documents. Based on multi-scenario dynamic optimization schemes and future evolution trend analysis reports, as well as recommended scheme sets and analysis reports, develop a Web GIS-based visual decision support platform. Employing a front-end / back-end separation architecture and modern Web GIS technology, it provides service interfaces for data management, algorithm invocation, and result analysis, enabling the visual display of geographic information.

[0118] The platform's main functional modules include a data display module, a solution comparison module, a 3D visualization module, an interactive adjustment module, and a report generation module, providing intuitive map display, solution comparison, and interactive adjustment functions.

[0119] The 3D visualization module utilizes 3D Earth Engine technology to construct a 3D scene of the study area, showcasing optimized solutions in the form of 3D architectural models and terrain renderings, providing an immersive visual experience and a more intuitive spatial perception. Decision-makers can freely control the viewing angle and observe the effects of the solutions from different perspectives.

[0120] The interactive adjustment module allows decision-makers to fine-tune recommended solutions. By clicking on a specific spatial unit on the map, they can modify its functional type assignment. The system recalculates the adjusted objective function value and evaluation score in real time based on a multi-objective constraint optimization mathematical model, determines whether the adjustment violates the constraints, and provides prompts. Decision-makers can save the adjusted solution as a new candidate solution.

[0121] The report generation module automatically generates decision support documents based on decision-makers' needs, comprising three levels: a technical report, a decision summary, and a public version of the explanation. The technical report, aimed at professional planners, includes a complete data analysis process, model building details, algorithm parameter settings, optimization solutions, and result verification. The decision summary, aimed at government decision-makers, extracts core conclusions and key recommendations, concisely explaining the scheme's features, expected effects, and implementation path using a combination of text and graphics. The public version of the explanation, aimed at the general public, uses easily understandable language to explain the main content of the planning scheme and its impact on residents' lives, accompanied by diagrams and renderings to enhance public understanding and participation.

[0122] The platform is deployed on a cloud server, supporting multi-user online access and collaborative decision-making. The system employs a permission management mechanism, with different user roles having different operational permissions. Administrators can manage data and algorithm parameters, planners can execute optimization and adjustment plans, decision-makers can view plans and generate reports, and the public can browse public information and submit feedback.

[0123] Generate the final intelligent recommendation solution document, which includes detailed descriptions of the five recommendation solutions, spatial layout diagrams, indicator comparison tables, implementation suggestions, risk warnings, and other content.

[0124] A system for optimizing the coordinated layout of urban and rural spaces, such as Figure 3 As shown, a method for optimizing the coordinated layout of urban and rural spaces, as described above, includes: The data preprocessing module is used to acquire multi-source heterogeneous spatial data and preprocess it to obtain a standardized spatiotemporal data set; The spatial element identification module, based on a standardized spatiotemporal dataset, identifies and extracts urban and rural spatial elements to obtain urban and rural spatial distribution maps and urban and rural boundary vector data. The coordination evaluation module, based on the urban-rural spatial distribution map and urban-rural boundary vector data, quantifies the level of coordinated development of urban and rural spaces and obtains the evaluation results of urban-rural spatial coordination. The spatial unit division module, based on the evaluation results of urban-rural spatial coordination, divides spatial units and performs preliminary classification of each unit, resulting in optimized basic units and preliminary spatial classification results; The layout optimization module, based on the evaluation results of urban and rural spatial coordination, the optimization of basic units, and the preliminary spatial classification results, solves the optimal configuration scheme of urban and rural spatial layout and obtains the non-inferior layout optimization scheme. The scenario simulation module, based on the non-dominated layout optimization scheme, predicts changes in development needs and performs scenario simulations. For each scenario, the layout optimization is re-executed to obtain a multi-scenario dynamic optimization scheme. The intelligent filtering module is used to evaluate and filter dynamic optimization solutions for multiple scenarios, and obtain intelligent recommendation solution documents.

[0125] In one embodiment of the present invention, a specific example is provided: This invention focuses on a project for optimizing the coordinated urban-rural spatial layout in a prefecture-level city in the Yangtze River Delta region, within the application field of urban-rural integration development. The city covers an area of ​​8,500 square kilometers, comprising a central urban area, three county-level cities, and five counties, where the imbalance between urban and rural development is a prominent issue.

[0126] Examples of multi-source heterogeneous data acquisition using the method of this invention are shown in Table 1: Table 1: Examples of multi-source heterogeneous data acquisition; Table 2 shows the results of urban and rural spatial element identification and coordination evaluation using the method of this invention: Table 2: Results of Identification and Coordination Evaluation of Urban and Rural Spatial Elements; After optimization using the method of this invention, the generated recommended scheme has achieved significant improvements in urban-rural coordination, equalization of public services, and ecological space protection, providing a scientific basis and technical support for the city's territorial spatial planning.

[0127] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for optimizing the coordinated layout of urban and rural spaces, characterized by, The method comprises the following steps: acquiring multi-source heterogeneous spatial data and preprocessing to obtain a standardized spatio-temporal data set; based on the standardized spatio-temporal data set, identifying and extracting urban and rural spatial elements to obtain an urban and rural spatial distribution map and urban and rural boundary vector data; based on the urban and rural spatial distribution map and the urban and rural boundary vector data, quantitatively evaluating the collaborative development level of urban and rural space to obtain an urban and rural space collaboration degree evaluation result; based on the urban and rural space collaboration degree evaluation result, dividing the space into units and preliminarily classifying each unit to obtain an optimized basic unit and a spatial preliminary classification result; based on the urban and rural space collaboration degree evaluation result, the optimized basic unit and the spatial preliminary classification result, solving the optimal configuration scheme of the urban and rural space layout to obtain a non-inferior layout optimization scheme; based on the non-inferior layout optimization scheme, predicting development demand changes and performing scenario simulation, re-executing layout optimization for each scenario to obtain a multi-scenario dynamic optimization scheme; evaluating and screening the multi-scenario dynamic optimization scheme to obtain an intelligent recommendation scheme document. 2.The method of claim 1, wherein, The acquisition of multi-source heterogeneous spatial data and preprocessing comprises: collecting satellite remote sensing image data and geographic information system data, acquiring multi-spectral image data, obtaining administrative boundary vector data, road traffic network data, river system data and terrain elevation data from the national basic geographic information database; collecting social and economic statistical data and real-time population flow data, for the problem of personal privacy involved in real-time population flow data, adopting a differential privacy protection mechanism to protect personal location information through a Laplace noise addition method; performing spatio-temporal registration, semantic alignment and quality control on the multi-source heterogeneous spatial data, establishing a time reference table, using ontology mapping technology to establish concept association across data sources, and constructing a multi-dimensional spatio-temporal data cube. 3.The method of claim 1, wherein, The intelligent identification and accurate extraction of urban and rural spatial elements comprises: constructing a weakly supervised training sample set, extracting land use annotation data of the research area from an open street map as a coarse-grained supervision signal, and inviting urban and rural planning professionals for fine annotation; designing a boundary-enhanced semantic segmentation network model, using an encoder-decoder structure as the backbone network, and adding a boundary-enhanced module; using public remote sensing image data sets for self-supervised pre-training, using an image inpainting task as the pre-training target, and using domain adaptation technology to reduce the distribution difference between the source domain and the target domain; performing semantic segmentation network model training and urban and rural space element identification, and using a conditional random field model to optimize the boundary.

4. The method of claim 1, wherein, The quantitative evaluation of the collaborative development level of urban and rural space comprises: constructing an urban and rural space collaboration degree evaluation index system, including spatial structure collaboration degree, functional configuration collaboration degree, element flow collaboration degree and ecological protection collaboration degree; calculating the scores of each index in the urban and rural space collaboration degree evaluation index system respectively; using a combination weighting method to calculate the comprehensive score of the urban and rural space collaboration degree, and using a combination weighting method of the analytic hierarchy process and the entropy weight method to determine the weights of each index.

5. The method of claim 1, wherein, The spatial unit division and preliminary classification of each unit comprise: Identify ecological red line and ecological security pattern, based on the ecological protection red line vector data determined by national spatial planning, identify ecological source based on ecosystem service importance evaluation method, and identify ecological corridor based on minimum cumulative resistance model; Divide the optimization basic unit, use natural geographical elements as the unit boundary, use Voronoi diagram method for preliminary unit division, select city center, township center and important traffic node as generation point, and combine terrain, land use status and road network to adjust the unit boundary; Evaluate the comprehensive suitability of the space unit, including urban construction suitability, agricultural production suitability and ecological protection importance; Preliminary classification of space, using the dominant function discrimination method, comparing the size relationship of the three types of suitability scores of each unit, and dividing each optimization basic unit into ecological space, living space, production space and composite space.

6. The method of claim 1, wherein, The optimal configuration scheme for urban and rural spatial layout includes: Establish a multi-objective optimization mathematical model, define the decision variable as the function type allocation of each optimization basic unit, establish a multi-objective function to evaluate the layout scheme, and establish the constraint condition; Design a knowledge-guided particle initialization strategy, establish an expert knowledge rule base in the field of urban planning, and use heuristic construction method to generate part of high-quality initial solution; Design adaptive inertia weight and dynamic crowding distance mechanism, dynamically adjust the inertia weight according to the iteration progress and population state of the algorithm; Execute the optimization algorithm to solve and output the optimization scheme, use improved multi-objective particle swarm optimization algorithm, and use fast non-dominated sorting method to classify particles.

7. The method of claim 1, wherein, The prediction of development demand change and the design of different policy scenarios for scenario simulation include: Construct an urban and rural development trend prediction model; Design multiple scenario parameters and perform scenario simulation, predict the development parameters of each scenario at future time nodes; Execute scenario optimization and generate dynamic scheme, re-execute improved multi-objective particle swarm optimization algorithm for each parameter setting, and perform incremental optimization based on non-inferior layout optimization scheme; Analyze the spatial evolution trajectory and generate scenario comparison report, use spatial transfer matrix method to calculate the mutual conversion of each type of space between different time nodes, and generate scenario comparison report. 8.The method of claim 1, wherein, The comprehensive evaluation and intelligent screening of multi-scenario dynamic optimization scheme includes: Construct a scheme comprehensive evaluation index system, including technical feasibility, economic rationality, social acceptance, ecological friendliness and implementation difficulty; Use TOPSIS method for scheme comprehensive sorting, calculate the score of candidate scheme according to the comprehensive evaluation index system, invite experts in the field of urban planning to score and evaluate the index, construct weighted and normalized decision matrix, determine ideal optimal solution and ideal worst solution, and calculate the relative closeness of each scheme; Use preference screening mechanism to generate recommended scheme set, collect decision makers' preferences for different targets and scenarios through interactive questionnaire, use solution screening method based on preference, use multi-attribute utility theory to construct decision makers' utility function; Perform scheme robustness analysis and sensitivity analysis, use scenario analysis method to simulate the influence of external shock on the implementation effect of the scheme. 9.The method of claim 1, wherein, The generation of the intelligent recommendation scheme document includes: Develop a visual decision support platform and generate a decision document, the main functional modules include data display module, scheme comparison module, three-dimensional visualization module, interactive adjustment module, report generation module; The three-dimensional visualization module uses three-dimensional earth engine technology to build a three-dimensional scene of the research area, and displays the optimization scheme in the form of three-dimensional building models and terrain rendering; The interactive adjustment module allows decision makers to fine-tune the recommended scheme, modify the functional type allocation of specific spatial units by clicking on the map, and the system recalculates the adjusted objective function value and evaluation score based on the multi-objective constraint optimization mathematical model in real time; The report generation module automatically generates the final intelligent recommendation scheme document according to the decision maker's requirements.

10. A system for optimizing the coordinated layout of urban and rural space, characterized in that it comprises: A method for performing the urban and rural space collaborative layout optimization of any one of claims 1-9, comprising: A data preprocessing module for obtaining and preprocessing multi-source heterogeneous spatial data to obtain a standardized spatio-temporal data set; A spatial feature identification module based on the standardized spatio-temporal data set, identifies and extracts urban and rural spatial features, and obtains urban and rural spatial distribution maps and urban and rural boundary vector data; A collaborative degree evaluation module based on the urban and rural spatial distribution map and the urban and rural boundary vector data, quantitatively evaluates the collaborative development level of urban and rural space, and obtains the evaluation result of the urban and rural space collaborative degree; A spatial unit division module based on the urban and rural space collaborative degree evaluation result, divides the spatial unit and preliminarily classifies each unit to obtain the optimization basic unit and the spatial preliminary classification result; A layout optimization module based on the urban and rural space collaborative degree evaluation result, the optimization basic unit and the spatial preliminary classification result, solves the optimal configuration scheme of urban and rural space layout, and obtains the non-inferior layout optimization scheme; A scenario simulation module based on the non-inferior layout optimization scheme, predicts the change of development demand and performs scenario simulation, re-executes the layout optimization for each scenario, and obtains the multi-scenario dynamic optimization scheme; An intelligent screening module for evaluating and screening the multi-scenario dynamic optimization scheme to obtain an intelligent recommendation scheme document.

Images