A rural space planning method and related devices

Abstract

The present application relates to the field of rural space governance and intelligent decision-making technology, in particular to a rural space planning method and related equipment, the method is: acquiring multi-source data of the target village, the multi-source data contains the land data, population data and rural economic data of the target village; the multi-source data is carried out spatio-temporal alignment fusion processing, and the fusion data is obtained; constructing the spatio-temporal knowledge graph based on the fusion data, extracting entity attributes and spatial relationships; the entity attributes and spatial relationships are input into the pre-trained planning optimization model, the rural space matching strategy is iteratively optimized according to the rigid constraint, and the optimal planning result is output, which effectively solves the technical problems that the rural planning in the prior art faces, such as data fragmentation, factor analysis isolation, planning scheme static and actual development demand disconnection, decision-making process relying on subjective experience and difficult to quantify economic-ecological benefit trade-off.

Description

Technical Field

[0001] This invention relates to the field of rural spatial governance and intelligent decision-making technology, specifically to a rural spatial planning method and related equipment. Background Technology

[0002] The "Three Zones and Three Lines" (urban space, agricultural space, ecological space, and permanent basic farmland, ecological protection red line, and urban development boundary) is the core control framework in the national land spatial planning system. The "Three Zones" are the functional classifications of national land space, and the "Three Lines" are the rigid control boundaries corresponding to the "Three Zones." The two are mutually matched and coordinated in their management. Currently, rural "people-land-industry" matching decision-making technology mainly relies on two methods: one is GIS-based static spatial analysis technology, which delineates functional zones by overlaying relevant data; the other is single-element optimization models, which use mathematical models to independently optimize the layout of land, population, or industry.

[0003] However, existing technologies for matching rural populations, land, and industries have four shortcomings: First, static data is disconnected from dynamic elements, relying on low-frequency data updates and failing to integrate real-time dynamic data, leading to planning delays; second, elements are fragmented and lack coordination, focusing on optimizing single elements without establishing a coupled analysis framework, resulting in unbalanced resource allocation; third, policy constraints are too rigid, treating the "three zones and three lines" as absolute forbidden zones, lacking flexible adjustment mechanisms, and making planning schemes lack resilience; fourth, the decision-making process is fragmented, with various tools operating independently, lacking collaborative extrapolation capabilities, and resulting in low decision-making efficiency. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a rural spatial planning method and related equipment to address the shortcomings of the prior art, so as to solve the technical problem that it is impossible to dynamically plan and optimize the rural "people-land-industry" system under the constraints of "three zones and three lines".

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

[0006] In a first aspect, the present invention provides a rural spatial planning method, comprising:

[0007] Acquire multi-source data for the target village, including land data, population data, and rural economic data of the target village;

[0008] After performing spatiotemporal alignment and fusion processing on the multi-source data, fused data is obtained;

[0009] A spatiotemporal knowledge graph is constructed based on the fused data. The spatiotemporal knowledge graph includes several entities, corresponding entity attributes, and spatial relationships between entities.

[0010] The entity attributes and spatial relationships are input into a pre-trained planning optimization model. Based on rigid constraints, the rural space matching strategy is iteratively optimized to obtain the optimal planning result.

[0011] As a further improvement of the present invention, it also includes data cleaning processing of the multi-source data before fusion processing, the data cleaning processing including:

[0012] Abnormal land parcels were identified by using the GPS confidence field in the land data, and the boundaries of these abnormal land parcels were simplified using the Douglas-Peucker algorithm.

[0013] The 3σ principle is used to identify abnormal fluctuations in rural economic data, and the abnormal fluctuations are smoothed using a moving average method.

[0014] As a further improvement of the present invention, the method for spatiotemporal alignment and fusion processing of multi-source data includes:

[0015] The population heatmap coordinate system used for population data is converted to the coordinate system corresponding to the land data by using the offset parameters calibrated by the regional control points;

[0016] After interpolating rural economic data using the spatiotemporal kriging method, we obtain economic data that is spatially aligned with national land data.

[0017] As a further improvement of the present invention, a spatiotemporal knowledge graph is constructed based on the fused data, including:

[0018] Data related to land data, population data, and rural economic data are extracted from the fused data, and a skill density index and development suitability score are calculated.

[0019] Population, land, and industry are considered as three types of entities. The skill density index is considered as the entity attribute of the target rural population, the development suitability score is considered as the entity attribute of land, and the correlation between industry and population and the distribution of industry and land are considered as the entity attributes of industry.

[0020] By using the Dempster-Shafer evidence theory algorithm and entity attributes to calculate the spatial relationships and conflict probabilities between various entities, a spatiotemporal knowledge graph about the target village is obtained.

[0021] As a further improvement of the present invention, the planning optimization model employs a constraint policy optimization deep reinforcement learning algorithm, which includes:

[0022] Based on the physical attributes and spatial relationships of the target village, the planning of the target village is divided according to the set size to obtain several grid units. The grid unit uses four-dimensional features to describe the current state vector. The four-dimensional features are land use type code, population density value, dominant industry classification code and ecological value.

[0023] The policy network architecture is used to extract and fuse features based on the state vectors of grid cells. Then, it is mapped to the action space through a fully connected layer to output land use adjustment suggestions for each grid cell.

[0024] As a further improvement of the present invention, the policy network architecture has a built-in dual-channel feature extraction structure, which includes a spatial feature channel and a temporal feature channel:

[0025] The spatial feature channel uses a convolutional neural network to extract features from the state vector of the grid cell, thereby obtaining the spatial distribution features associated with the land use network and the four-dimensional features.

[0026] The time-series feature channel utilizes a long short-term memory network to capture the time-series features of economic trends associated with network units based on four-dimensional features;

[0027] After concatenating the spatial distribution features and temporal features, the fused features are input into the fully connected layer of the policy network architecture.

[0028] As a further improvement of the present invention, during the training process of the planning optimization model, Monte Carlo tree search is used to update the planning policy space, and the parameters of the policy network architecture are updated in combination with the near-end policy optimization algorithm to achieve convergence of constraints.

[0029] As a further improvement of the present invention, the optimal planning results are visualized through at least one of the following methods: spatial configuration map, dynamic simulation report, and conflict early warning list.

[0030] Secondly, the present invention provides a rural spatial planning system for implementing the above-mentioned rural spatial planning method, comprising:

[0031] The data input layer includes a data interface for acquiring multi-source data of the target village, including land data, population data, and rural economic data of the target village.

[0032] The intelligent processing layer, communicatively connected to the data input layer, sequentially includes a data fusion module, a knowledge graph construction module, and a decision optimization module. The data fusion module processes the fusion logic of multi-source data to obtain fused data. The knowledge graph construction module constructs a spatiotemporal knowledge graph based on the output of the data fusion module. The spatiotemporal knowledge graph includes several entities, corresponding entity attributes, and spatial relationships between entities. The decision optimization module includes a planning optimization model for optimizing rural spatial matching strategies.

[0033] The decision output layer communicates with the intelligent processing layer and is used to output the optimal planning results.

[0034] Thirdly, the present invention provides a computer-readable storage medium for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the above-described rural spatial planning method.

[0035] Fourthly, the present invention provides a computing device, comprising:

[0036] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including steps for performing the above-described rural spatial planning method.

[0037] The beneficial effects of this invention are as follows: This invention provides a rural spatial planning method that, through the spatiotemporal fusion of multi-source data, greatly improves the completeness and accuracy of decision-making information, overcoming the limitations of a single data source. By considering low-frequency land data, high-frequency population data, and medium-frequency economic data, it solves the problem of the disconnect between static data and dynamic actual development in traditional planning, providing comprehensive data support for the precise matching of "people-land-industry" elements. The fusion of multi-source data breaks down the format barriers and spatial misalignment problems of heterogeneous multi-source data, achieving precise spatiotemporal correlation of land, population, and economic data, and solving the problem of resource allocation imbalance caused by data silos in traditional planning. Through the dynamic optimization capability of the planning optimization model, it solves the problem of the lack of resilience in solutions caused by excessively rigid policy constraints in traditional planning, achieving a dynamic balance between rigid constraints and actual needs through flexible adjustment mechanisms (such as fine-tuning of development boundaries and ecological compensation linkage). This invention effectively resolves the contradiction between element allocation and policy constraints in rural planning, achieving intelligent and refined matching of the rural "people-land-industry" system while adhering to the "three zones and three lines" policy. Ultimately, the optimal planning results not only promote the sustainable growth of the rural economy, but also effectively protect the ecological environment, take into account social interests, and promote the in-depth implementation of rural revitalization, demonstrating the enormous potential and value of technical solutions in practice. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the rural space planning method in an embodiment of the present invention.

[0040] Figure 2 This is a flowchart of the data processing in the rural spatial planning method in an embodiment of the present invention.

[0041] Figure 3 This is a schematic diagram of the constraint policy optimization deep reinforcement learning algorithm in an embodiment of the present invention.

[0042] Figure 4 This is a flowchart of the CPO-DRL training steps and constraint verification training used in the embodiments of the present invention.

[0043] Figure 5 This is a comparison diagram of the technical effects of the embodiments of the present invention and the technical effects of the prior art.

[0044] Figure 6 This is a schematic diagram of the structure of a computer device in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives and technical solutions of this invention clearer and easier to understand, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0046] Terminology Explanation:

[0047] Douglas-Peucker algorithm: used for boundary correction in land data preprocessing;

[0048] GPS (Global Positioning System)

[0049] WGS84 coordinate system (World Geodetic System 1984): World Geodetic System 1984;

[0050] CGCS2000 Coordinate System (China Geodetic Coordinate System 2000): The 2000 National Geodetic Coordinate System;

[0051] Dempster-Shafer evidence theory algorithm: used for quantitative modeling of spatial relationships between entities in spatiotemporal knowledge graphs;

[0052] UCB: Upper Confidence Bound;

[0053] CPO-DRL (Constrained Policy Optimization-Deep Reinforcement Learning): A combination of constrained policy optimization and deep reinforcement learning.

[0054] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment discloses a rural spatial planning method that achieves precise allocation of rural planning elements through the synergistic effect of multi-source data fusion, dynamic knowledge modeling, and intelligent optimization algorithms. The method mainly includes: acquiring multi-source data of the target village; performing spatiotemporal alignment fusion processing on the multi-source data to obtain fused data; constructing a spatiotemporal knowledge graph based on the fused data, and obtaining the entity attributes corresponding to entities in the spatiotemporal knowledge graph and the spatial relationships between entities; inputting the entity attributes and spatial relationships into a pre-trained planning optimization model, and iteratively optimizing the rural spatial matching strategy based on rigid constraints to obtain the optimal planning result.

[0057] The multi-source data includes land data, population data, and rural economic data of the target village. In this embodiment, the land data is low-frequency updated data in vector format, such as land use status maps and ecological red line ranges. The population data is a raster-formatted population heat map, which is high-frequency updated data. The economic data is in tabular format and is medium-frequency updated data, such as industrial output value reports.

[0058] The rural spatial planning method in this embodiment acquires land use data, population data, and rural economic data of the target village through multiple channels. This data covers the current land use status, population distribution dynamics, and industrial economic characteristics of the target village, providing detailed foundational information for subsequent intelligent decision-making. Next, advanced spatiotemporal alignment and fusion technology is used to process the aforementioned multi-source data, ensuring a high degree of consistency between land use data, population heatmaps, and industrial economic data in both spatial and temporal dimensions, forming integrated fused data and laying a solid data foundation for rural planning decisions. Subsequently, a refined spatiotemporal knowledge graph is constructed based on the fused data. This graph not only includes entity attribute information, such as arable land, population density, dominant industry types, and their ecological value, but also quantifies the spatial competition and interdependence relationships between entities, providing comprehensive decision-making references for the intelligent agent. The constructed entity attributes and their spatial relationships are used as input to a pre-trained planning optimization model, and the rigid constraints of the "three zones and three lines" are strictly considered during model operation. The planning optimization model employs a deep learning algorithm, which continuously learns and optimizes rural spatial matching strategies through interaction with the environment until the optimal planning result is achieved. This optimization process fully respects the restrictive rules on rural land use, such as the upper limit on the conversion ratio of permanent basic farmland and the inviolability of ecological protection red lines, ensuring the compliance and feasibility of the planning scheme.

[0059] Furthermore, to ensure data quality, after acquiring multi-source data, data cleaning is also performed. The data cleaning process includes: identifying anomalous plots in the land data using the GPS confidence field; simplifying the polygonal boundaries of these anomalous plots using the Douglas-Peucker algorithm; and retaining key nodes to correct the geometry. For anomalous fluctuations in rural economic data, outliers are identified using the 3σ principle, and smoothed using a moving average method to eliminate noise and anomalous fluctuations in the data. Data cleaning improves the accuracy and reliability of the data, providing a solid foundation for subsequent spatiotemporal knowledge graph construction and planning optimization.

[0060] Furthermore, the multi-source data undergoes spatiotemporal alignment and fusion processing, including: converting the population heatmap coordinate system used for population data to the coordinate system corresponding to the land data using offset parameters calibrated by regional control points; and interpolating rural economic data using spatiotemporal kriging to obtain economic data spatially aligned with the land data. The spatiotemporal alignment of multi-source data is achieved through a data fusion module, specifically by converting raster-format population heatmaps (such as the WGS84 coordinate system) generated by mobile operators to vector-format land data (CGCS2000 coordinate system) provided by the natural resources department, and using spatiotemporal kriging to fill in missing values ​​in the economic data, ensuring the uniformity and integrity of the data in both spatial and temporal dimensions. Data fusion overcomes the problem of disconnect between static and dynamic elements in existing technologies, achieving full-chain data analysis and improving the scientific and comprehensive nature of planning decisions. In other embodiments, more advanced data processing technologies, such as deep learning image recognition technology, can be introduced to automatically correct abnormal plots in the land data, or more accurate spatiotemporal interpolation algorithms can be used to improve the predictive accuracy of economic data, further optimizing the data fusion effect.

[0061] Furthermore, a spatiotemporal knowledge graph is constructed based on the fused data. This includes: extracting data related to land, population, and rural economic data from the fused data, and calculating the skill density index and development suitability score; treating population, land, and industry as three types of entities, with the skill density index (based on the target village) as the entity attribute of population, and the development suitability score as the entity attribute of land; and considering the correlation between industry and population, and the distribution attributes of industry and land as the entity attributes of industry. The spatial relationships and conflict probabilities among various entities are calculated using the Dempster-Shafer evidence theory algorithm, thereby obtaining a spatiotemporal knowledge graph about the target village. Entity attributes are transformed into basic probability allocation parameters in the Dempster-Shafer evidence theory algorithm, making them necessary support for deriving spatial relationships between entities and calculating conflict probabilities. By constructing the spatiotemporal knowledge graph, deep correlation and dynamic analysis of the "people-land-industry" elements are achieved, improving the scientific and comprehensive nature of planning decisions.

[0062] Labor resources are quantified using the skill density index, which is calculated using the following formula:

[0063]

[0064] in, This is a skill density index. The area of ​​the target village. For the first Skill level For the first The number of employees at the highest skill level. The total number of skill level categories included in the calculation is determined by the level classification of the national vocational qualification standards. Skill levels are divided into 1 to 5 levels according to the national vocational qualification standards (e.g., level 1 is entry-level worker, level 5 is senior technician), so... .

[0065] Furthermore, the planning optimization model in this embodiment employs a constraint-based strategy optimization deep reinforcement learning algorithm. This learning algorithm uses rigid constraints as action constraints, dividing the planning of the target village into several grid cells according to a set size based on the entity attributes and spatial relationships of the target village. Each grid cell uses a four-dimensional feature to describe its current state vector; a policy network architecture is used to adjust the state vectors of the grid cells, mapping them to the action space through fully connected layers, and outputting land use adjustment suggestions for each grid cell. The four-dimensional features include land use type code, population density value, dominant industry classification code, and ecological value.

[0066] The strategy network architecture incorporates a dual-channel feature extraction structure to capture the spatial and temporal features of grid cells. This dual-channel structure includes a spatial feature channel and a temporal feature channel. The spatial feature channel utilizes a convolutional neural network to extract features from the state vectors of grid cells, obtaining the spatial distribution features associated with the land use network and four-dimensional features. The temporal feature channel uses a long short-term memory network to capture the temporal features of economic trends associated with network cells based on the four-dimensional features. The data from the spatial distribution features and temporal features are concatenated, and the resulting fused feature is input into the fully connected layer of the strategy network architecture. This enables dynamic adjustment and optimization of rural spatial configuration, overcoming the problems of fragmented elements and lack of coordination in existing technologies, and improving the compliance and resource conservation of planning schemes.

[0067] Furthermore, during the training of the planning optimization model, Monte Carlo tree search is used to update the planning policy space, combined with a proximal policy optimization algorithm to update the policy network parameters, thereby achieving convergence of constraints. The Monte Carlo tree explores the policy space by simulating the decision-making process, selecting policies with high potential rewards; the proximal policy optimization algorithm optimizes the policy network parameters using gradient descent, maximizing the reward function while satisfying the constraints. This combination of Monte Carlo tree search and proximal policy optimization enables dynamic response and optimization to constraints, improving the compliance and economic efficiency of planning decisions.

[0068] Ultimately, the optimal planning results are visualized through spatial configuration maps, dynamic simulation reports, and conflict warning lists. This intuitive visualization interface improves the transparency and interactivity of the decision-making process, enhancing the acceptability of planning decisions and the efficiency of implementation.

[0069] Example 2

[0070] Based on the rural spatial planning method in Example 1, this example provides a specific implementation method. The data processing flow in this example mainly includes four core steps: data cleaning, spatiotemporal alignment, knowledge graph construction, and dynamic optimization.

[0071] During the data cleaning phase, outlier data is filtered out from the land spatial vector data (such as land use status maps). Plots with positioning errors exceeding 5 meters are identified using the GPS confidence field (e.g., farmland outlines shifted to the center of a river). The Douglas-Peucker algorithm is used to simplify the polygon boundaries of these plots, preserving key nodes to correct their geometry. For abnormal fluctuations in industrial economic data (e.g., a sudden 300% increase in monthly output), outliers are identified using the 3σ principle and smoothed using a moving average method.

[0072] It's worth noting that the Douglas-Peucker algorithm is a classic algorithm for simplifying curves or polygons. It preserves the key feature points of the curve and removes redundant points, reducing the amount of data while maintaining the original shape as much as possible. The 3σ rule, also known as the three-standard-deviation rule, is a data judgment criterion in statistics based on the characteristics of a normal (Gaussian) distribution. It is used to identify outliers in a dataset or determine the normal range of data fluctuations.

[0073] During the spatiotemporal alignment phase, multi-source data fusion is achieved through coordinate transformation and interpolation. Specifically, the coordinate transformation steps include converting the WGS84 coordinate system of the population flow heatmap to the CGCS2000 coordinate system used for land data. The transformation formula is as follows:

[0074]

[0075] in, In CGCS2000 coordinate system axis coordinate values, In CGCS2000 coordinate system axis coordinate values, In WGS84 coordinate system axis coordinate values, In WGS84 coordinate system axis coordinate values, For the offset parameter axis coordinate values, For the offset parameter Axis coordinate values.

[0076] Offset parameter (for example, in this embodiment, the offset parameter is set to) =12.5m, =-8.3m) is determined by regional control points. For example Figure 2 As shown, the data processing in this embodiment is divided into three stages: data cleaning, spatiotemporal alignment, and dynamic optimization, with conflict detection and verification embedded throughout the entire process. In the preprocessing stage of data cleaning, outlier filtering (e.g., removing samples with positioning errors greater than 5m), 3σ outlier detection, and moving average smoothing are performed on the observed data to stabilize the temporal and spatial sample quality. Subsequently, in the spatiotemporal alignment stage, the system estimates and applies coordinate transformation models. , Equal offset parameters are used, and missing data items are appropriately imputed using spatiotemporal kriging interpolation, weighted accordingly. λ Pixel-level conflict detection and labeling are performed on cultivated land, construction land, and ecological land. Based on the spatial competition relationship between entities, the conflict probability between plots is calculated using Dempster-Shafer evidence theory. After alignment, a dynamic optimization phase begins, using path and layout as decision variables and comparing several candidate solutions based on a cost-benefit evaluation function. All solutions must undergo conflict detection verification again at the end (after evaluation and adjustment), and only solutions that simultaneously meet the threshold and control constraints are output.

[0077] Raster-formatted population data (population heatmap) is resampled to 100-meter resolution using bilinear interpolation to ensure spatial consistency with vector land data. For missing values ​​in tabular data (such as missing tea production value for a certain quarter in a specific village), spatiotemporal kriging interpolation is used to calculate weighting coefficients based on contemporaneous data from neighboring areas and historical data from the village. Impute missing values. Weighting coefficients. The formula:

[0078]

[0079] in, These are the weighting coefficients. For spatial distance, For time difference.

[0080] In the knowledge graph construction phase, entity attributes and dynamic relationships are defined. The "population" entity, based on administrative villages, quantifies labor resources using a skill density index, calculated as follows:

[0081]

[0082] in, This is a skill density index. The area of ​​the target village. For the first Skill level For the first The number of employees corresponding to each skill level The total number of skill level categories included in the calculation is determined by the level classification of the national vocational qualification standards. Skill levels are divided into 1-5 levels according to the national vocational qualification standards (e.g., Level 1 is entry-level worker, Level 2 is intermediate worker, Level 3 is advanced worker, Level 4 is technician, and Level 5 is senior technician). The development suitability score of the "land" entity is based on 10 indicators, including terrain slope (weight 0.3), soil fertility (weight 0.2), and road proximity (weight 0.5). Each indicator is normalized and then weighted and summed. Spatial competition between entities is calculated using Dempster-Shafer evidence theory to determine conflict probabilities. For example, this embodiment depicts a spatiotemporal knowledge graph mechanism, with the corresponding logic being object definition, relational reasoning, and dynamic updating. The object layer covers population (such as the Skill Density Index (SDI), land (a suitability index weighted by slope, soil, and road proximity), and thematic graphs that can serve as spatiotemporal nodes. The relational layer distinguishes between two mechanisms: competition and dependence. The former describes the mutual exclusion strength between adjacent plots or schemes, while the latter characterizes the correlation between elements such as industry-land use or agriculture-market. Taking "the correlation coefficient between the tea garden and other entity elements is 0.91" as an example, as events occur and data is updated, the graph weights edges through a transmission link of "association identification - weight decay - event triggering." W Perform rolling updates; an example update formula is: W new = 0.7× W old + 0.3× ΔW ;in, W new For the updated edge weights, W old As the historical marginal weight, ΔW This increments the edge weights triggered by new events. When key nodes such as the processing plant experience state changes, the association degree is updated immediately (in the example, it is updated to 0.4), providing the latest and interpretable structured prior parameters for subsequent optimization searches.

[0083] During the dynamic optimization phase, the scheme is iteratively optimized using deep reinforcement learning algorithms. When spatial conflicts are detected (such as the proposed project overlapping basic farmland by more than 5 mu), the system triggers an automatic adjustment mechanism: it calls a path planning algorithm to calculate the optimal migration path (such as shifting 280 meters southwest to unused land) and reassesses the economic and ecological impacts after the adjustment (such as a predicted decrease in output value of 2.1%, but a reduction in the soil erosion risk level by 1 level). The optimized scheme is then simulated and extrapolated through multiple rounds of Monte Carlo tree search to select the strategy with the highest comprehensive reward value.

[0084] In this embodiment, the deep reinforcement learning algorithm uses a constraint policy optimization method, such as... Figure 3 As shown, the goal of the constrained policy optimization deep reinforcement learning algorithm is to generate the optimal rural planning scheme through dynamic interaction between the agent and the environment under the rigid constraints of "three zones and three lines". The implementation of the constrained policy optimization deep reinforcement learning algorithm includes four operational processes: environment modeling, policy network design, reward mechanism, and training optimization. Figure 3 As shown, this embodiment constructs an integrated solution framework of environmental dynamics, policy network, and reward function. The environmental dynamics model uses a grid as the smallest unit, and the attribute vector of each grid can be divided into four dimensions: land use type, population density, dominant industry type, and ecological value, and its state is updated at set time intervals. For example, Figure 3 The table in the environmental dynamics modeling section. On the policy network, a 3-layer Convolutional Neural Network (CNN) extracts the spatial channel structure, and then a Long Short-Term Memory (LSTM) temporal subnetwork is used to characterize the trend, outputting a 256*256 dimensional spatiotemporal feature vector lip. The network directly outputs the probability distribution of each grid cell in terms of residential / agricultural / other land use actions, and incorporates hard constraints such as "basic farmland conversion ratio ≤ 5%" to ensure that the optimal policy meets the hard key red line (the permanent basic farmland conversion ratio does not exceed 5%, and there are no construction projects within the ecological protection red line). During the model training and optimization process, a combination of Monte Carlo tree search and proximal policy optimization is used, with policy priors used to expand the search tree (e.g., ...). Figure 3 The training optimization section of the algorithm includes a dendrogram (which also includes pruning regions), which is used to simultaneously update the policy and value assessment during simulation and backpropagation. Specifically, by defining pruning regions to clarify the pruning path (such as node sequences), new states are explored, and then the policy network is updated using a pruning-based gradient descent method. This reduces computational redundancy and, combined with the feedback from the reward function, allows the policy network to better balance economic, ecological, and cost considerations while meeting constraints such as the basic farmland conversion ratio ≤ 5%, thereby achieving better land use decisions for residential and agricultural land. The reward function uses a comprehensive reward value (…). With the objective of [missing information], the system simultaneously outputs visualized results of three-dimensional indicators (economic, ecological, and cost) and a comprehensive score, which serve as reasonable criteria for optimal result selection, training termination, and backtracking optimization. The formula for calculating the comprehensive reward value is:

[0085]

[0086] Based on the strategy with the highest overall reward value, an executable planning scheme is finally generated.

[0087] The environmental modeling operation divides the planning area into 1km×1km grid units. The state vector of each grid contains four dimensions: land use type, population density, dominant industry type, and ecological value.

[0088] In this embodiment, land use types include the coded values ​​for cultivated land, forest land, and construction land; population density is derived from heat map interpolation results, with units of people / km². 2 The dominant industry type refers to the code coded according to the national economic industry classification; ecological value is calculated based on indicators such as vegetation coverage and soil erosion rate, with a score of 0-100. For example, a certain grid state can be represented as (2, 150, 0132, 85), corresponding to "forest land, population density 150 people / km²". 2 Tea processing industry, ecological value 85 points.

[0089] The policy network adopts a dual-channel hybrid structure, which is used to capture spatial feature channels and temporal feature channels respectively, to obtain dual-channel feature vectors.

[0090] In this embodiment, the spatial feature channel extracts the land use spatial distribution pattern through a 3-layer convolutional neural network. The input is a gridded state matrix (e.g., a 10×10 grid area). The first layer has a 3×3 kernel size (stride 1) and outputs 32 feature maps; the second layer has a 2×2 kernel size (e.g., stride 1) and outputs 64 feature maps; the third layer generates a 256-dimensional spatial feature vector through a fully connected layer.

[0091] In this embodiment, the time-series feature channel uses a Long Short-Term Memory (LSTM) network to capture industrial economic trends. The LSM network takes five years of historical industrial output data (sliced ​​by quarter) as input, has 128 hidden layer nodes, and outputs a 256-dimensional time-series feature vector.

[0092] After concatenating the dual-channel feature vectors, they are mapped to the action space through a fully connected layer, outputting land use adjustment suggestions for each grid cell. These suggestions include maintaining the current status quo, converting to industrial use, or converting to residential use. Action selection must meet the following constraints:

[0093]

[0094] In the formula, For the first The amount of arable land converted refers to the area of ​​arable land (i.e., land used for agricultural production) that is converted to other uses (such as construction land or forest land). The total cultivated land area of ​​the target village. The total number of entities participating in farmland conversion and ecological red line constraints. The formula is used to determine the amount of damage to the ecological protection red line. The conclusion is that the conversion rate of permanent basic farmland does not exceed 5%, and there are no construction projects within the ecological protection red line.

[0095] The reward function considers economic, ecological, and social benefits, and its calculation formula is as follows:

[0096]

[0097] in, For the reward function, For the implementation of policies / projects The value reflects the total economic output after implementation. Before implementing policies / projects value, For ecological benefits, Relocation costs are used to reflect the economic costs of implementation. This represents the maximum budget limit, i.e., the highest expenditure allowed by this policy / project. The economic gain weight is 0.6 (determined through training with 200 sets of historical data), and the ecological benefit... Based on vegetation cover (weight 0.4), soil erosion rate (weight 0.3), and biodiversity (weight 0.3), the demolition cost is normalized to a proportion of the budget ceiling. When the reward function... The higher the value, the better the economic benefits of the policy / project.

[0098] In this embodiment, the training process of the model adopts a combination of Monte Carlo tree search and proximal policy optimization. First, Monte Carlo tree search includes a selection phase, an expansion and simulation phase, and a backtracking update phase.

[0099] During the selection phase, starting from the root node, the selection... The child node with the largest value The formula is:

[0100]

[0101] in, The upper limit confidence estimate for the action options. Accumulated rewards for nodes, The number of visits is 1.5, and the exploration coefficient is 1.5. The total number of attempts for all action options.

[0102] During the expansion and simulation phase, when an unexplored leaf node is reached, a new node is expanded and the process is deduced through a random strategy until the termination state (e.g., the planning period is 5 years), and the path reward value is recorded.

[0103] During the backtracking update phase, the values ​​of all nodes on the path are updated based on the simulation results. Value and value.

[0104] The network parameters are updated using a near-end strategy optimization algorithm, with the loss function being:

[0105]

[0106] in, For loss function, The probability ratio between the old and new strategies. For the dominant function, the shearing parameter =0.2 (to prevent policy mutations) For expectation operator, For editing functions, For input parameters.

[0107] like Figure 4 As shown, the policy learning process uses deep reinforcement learning under constrained policy optimization (such as...). Figure 4 The system is structured around the CPO-DRL framework. First, in environmental interactions, the grid state is used as input, and trajectories and reports are collected in a rolling manner. Simultaneously, it supports action space decisions such as "maintain the status quo" and "convert to industrial use," constructing corresponding expected states. PPO and other algorithms are used as optimization strategies for updates, and value networks and advantage estimation are combined to improve sample efficiency. Monte Carlo tree search is introduced in the search phase, efficiently traversing the decision space under the guidance of policy priors, following the sequence of "selection-expansion-simulation-backpropagation." Both training and evaluation are subject to hard, real-time constraints. Once violations such as "basic farmland conversion rate > 5%" are detected, the corresponding action is immediately deemed invalid and penalized in the backpropagation, ensuring that the final converged strategy meets the requirements of the control red line.

[0108] Example 3

[0109] This embodiment provides a rural spatial planning system, which includes a three-layer architecture: a data input layer, an intelligent processing layer, and a decision output layer. Each layer module forms a closed loop through data flow and feedback mechanisms.

[0110] The data input layer obtains multi-source data from three independent interfaces. The multi-source data mainly includes vector-format land data (such as land use status maps and ecological red line ranges) provided by the natural resources department, raster-format population heat maps generated by mobile operators (updated every 15 minutes), and tabular-format economic data (such as industrial output reports) from the statistics department.

[0111] Before entering the intelligent processing layer, multi-source data needs to undergo format parsing and preliminary cleaning. For example, land parcels with a positioning error of more than 5 meters in the land data will be automatically filtered out, and population heat maps will be resampled to 100-meter resolution using bilinear interpolation to ensure spatial alignment with the land data.

[0112] The core modules of the intelligent processing layer include a data fusion module, a knowledge graph construction module, and a decision optimization module. In the data fusion module, data from different coordinate systems are transformed using offset parameters determined by regional control points (e.g., converting the WGS84 coordinate system to the CGCS2000 coordinate system). The transformation formula is:

[0113]

[0114] in, In CGCS2000 coordinate system axis coordinate values, In CGCS2000 coordinate system axis coordinate values, In WGS84 coordinate system axis coordinate values, In WGS84 coordinate system axis coordinate values, For the offset parameter axis coordinate values, For the offset parameter Axis coordinate values. Missing values ​​in economic data are filled in using spatiotemporal kriging interpolation.

[0115] The knowledge graph module is used to construct a network of "people-land-industry" entities. The "people" entity quantifies labor resources using the Skill Density Index (SDI) at the administrative village level. The "land" entity calculates development suitability scores using indicators such as slope and soil fertility. The spatial competition relationship between entities is dynamically calculated using the Dempster-Shafer evidence theory to determine the probability of conflict.

[0116] The decision optimization module employs a deep reinforcement learning algorithm (CPO-DRL), dividing the planning area into 1-kilometer grid cells. The state of each grid cell includes attributes such as land use type and population density. The policy network consists of a dual-channel structure: a spatial channel extracts land use distribution features using a convolutional neural network, while a temporal channel captures economic trends using an LSTM network. During training, the algorithm must meet rigid constraints (e.g., the conversion rate of permanent basic farmland ≤ 5%) and explore the optimal strategy through Monte Carlo Tree Search (MCTS).

[0117] Ultimately, the decision output layer generates the optimal planning results: a spatial configuration map (marking proposed land use adjustments), a dynamic simulation report (showing economic and ecological impact predictions), and a conflict warning list (listing proposed projects requiring adjustments and alternative solutions), such as... Figure 5 As shown. Figure 5This paper presents a comparison of the method in this embodiment with existing technologies in terms of process and effectiveness. Traditional approaches often involve isolated analysis, statically displaying land, population, and economic entities separately, relying on GIS visualization technology, and lacking capabilities for cross-factor collaboration and temporal extrapolation. This embodiment, through a collaborative system of multi-source data fusion (people-land-industry), intelligent spatial configuration optimization, and real-time simulation and dynamic extrapolation, tightly couples the data foundation, spatial constraint control, interpretive reasoning, and online iteration. It can directly generate feasible configuration schemes that meet the objective function and hard constraints, achieving precise allocation of rural planning elements and continuously improving comprehensive benefits through feedback, thus promoting the transformation of territorial spatial planning from static display to dynamic optimization oriented towards decision-making.

[0118] Example 4

[0119] In another embodiment of the present invention, a computer-readable storage medium is provided as a storage component within a terminal device, the function of which is to store programs and data. It should be noted that the computer-readable storage medium here encompasses not only the built-in storage components of the terminal device but also extended storage components supported by the device. Essentially, it is a tangible medium capable of containing or storing programs that can be invoked by or in conjunction with an instruction execution system, device, or apparatus. This storage medium provides storage areas for the terminal's operating system and stores one or more instructions suitable for processor loading and execution, which can constitute one or more computer programs containing program code.

[0120] Specifically, examples of computer-readable storage media (a non-exclusive list) include: electrical connections with one or more wires, portable disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable optical disc read-only memory, optical storage devices, magnetic storage devices, or any reasonable combination of the above types.

[0121] The storage medium may also include data signals propagated as part of a baseband portion or a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any reasonable combination of both. Furthermore, computer-readable storage medium may also refer to other readable media besides conventional readable storage media, capable of sending, propagating, or transmitting programs for use or operation by an instruction execution system, apparatus, or device. Program code on the storage medium can be transmitted via any suitable medium, including but not limited to wireless, wired, optical fiber, or any reasonable combination thereof.

[0122] The program code used to implement the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C. The execution modes of the program code include: running entirely on the user's computing device, running partially on the user's device as a standalone software package, running partially in a distributed manner on both the user's device and a remote computing device, or running entirely on a remote computing device or server. When a remote computing device is involved, the device can be connected to the user's computing device via any type of network such as a local area network (LAN) or a wide area network (WAN), or connected to an external computing device via the Internet through an Internet service provider.

[0123] The processor is capable of loading and executing one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the rural spatial planning method described in Example 1.

[0124] Example 5

[0125] Figure 6 This is a schematic diagram of a computer device provided according to an embodiment of the present invention.

[0126] Please see Figure 6 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the processor 61 executes the computer program 63, it implements the rural spatial planning method of this embodiment. To avoid repetition, these details are not elaborated here. Alternatively, when the processor 61 executes the computer program 63, it implements the functions of each model / unit in the computational system constituting the rural spatial planning method of this embodiment. To avoid repetition, these details are not elaborated here.

[0127] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 6 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0128] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, CPUs, graphics processing units (GPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, quantum computing-based data processing logic units, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0129] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or RAM of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0130] Furthermore, the memory 62 may include both internal storage units of the computer device 60 and external storage devices. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0131] Any references to memory, databases, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0132] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

Claims

1. A method for rural space planning, characterized by, The method comprises the following steps: acquiring multi-source data of a target village, wherein the multi-source data comprises territorial data, population data and village economic data of the target village; performing spatio-temporal alignment and fusion processing on the multi-source data to obtain fused data; constructing a spatio-temporal knowledge graph according to the fused data, wherein the spatio-temporal knowledge graph comprises a plurality of entities, corresponding entity attributes and spatial relationships between the entities; inputting the entity attributes and spatial relationships into a pre-trained planning optimization model, and iteratively optimizing a village space matching strategy based on rigid constraints to obtain an optimal planning result; constructing a spatio-temporal knowledge graph according to the fused data, comprising: extracting data related to territorial data, population data and village economic data from the fused data, and calculating a skill density index and a development suitability score; taking population, land and industry as three types of entities, taking the skill density index as an entity attribute of the population of the target village, taking the development suitability score as an entity attribute of the land, and taking the association attributes between the industry and the population and the distribution attributes of the industry and the land as entity attributes of the industry; calculating spatial relationships and conflict probabilities between entities of various types by using a Dempster-Shafer evidence theory algorithm and entity attributes, and then obtaining a spatio-temporal knowledge graph about the target village; the planning optimization model adopts a constraint strategy optimization deep reinforcement learning algorithm, which comprises: dividing the planning of the target village according to a set size according to entity attributes and spatial relationships of the target village to obtain a plurality of grid cells, wherein the grid cells use four-dimensional features to describe a current state vector, and the four-dimensional features are land use type codes, population density values, dominant industry classification codes and ecological values; using a strategy network architecture to perform feature extraction and feature fusion on the state vector of the grid cell, and then mapping to an action space through a fully connected layer to output land use adjustment suggestions for each grid cell.

2. The rural space planning method according to claim 1, wherein, The method further comprises data cleaning processing on the multi-source data before fusion processing, which comprises: identifying abnormal land blocks from the territorial data through a GPS confidence field, and simplifying the boundaries of the abnormal land blocks by using a Douglas-Peucker algorithm; identifying abnormal fluctuation values in the village economic data by using a 3σ principle, and smoothing the abnormal fluctuation values by using a moving average method.

3. The rural space planning method according to claim 1, wherein, The method of spatio-temporal alignment and fusion processing on the multi-source data comprises: converting a population heat map coordinate system used by the population data into a coordinate system corresponding to the territorial data by using offset parameters of regional control points; performing interpolation processing on the village economic data by using a spatio-temporal Kriging interpolation method to obtain economic data that is spatially aligned with the territorial data.

4. The rural space planning method according to claim 1, wherein, The strategy network architecture has a built-in dual-channel feature extraction structure, which comprises a spatial feature channel and a time sequence feature channel: the spatial feature channel uses a convolutional neural network to perform feature extraction on the state vector of the grid cell to obtain spatial distribution features associated with the land use network and the four-dimensional features; The timing feature channel utilizes a long short-term memory network to capture the timing features of the economic trends associated with the network units based on four-dimensional features; After the spatial distribution features and the timing features are spliced, the spliced fusion features are input into a fully connected layer of the strategy network architecture.

5. The rural space planning method according to claim 1, wherein, In the training process of the planning optimization model, a Monte Carlo tree search is used to update the planning strategy space, and a proximal policy optimization algorithm is used to update the parameters of the strategy network architecture, so as to realize the convergence of the constraint conditions.

6. The rural space planning method according to claim 1, wherein, The optimal planning result is visually displayed in at least one of a spatial configuration diagram, a dynamic deduction report and a conflict early warning list.

7. A rural space planning system for implementing the rural space planning method according to any one of claims 1 to 6, characterized in that The method comprises the following steps: a data input layer comprising a data interface for acquiring multi-source data of a target village, wherein the multi-source data comprises territorial data, population data and rural economic data of the target village; an intelligent processing layer in communication connection with the data input layer and comprising, in sequence, a data fusion module, a knowledge graph construction module and a decision optimization module; the data fusion module is used to process the fusion logic of the multi-source data to obtain fused data; the knowledge graph construction module is used to construct a spatio-temporal knowledge graph according to the output of the data fusion module; the spatio-temporal knowledge graph comprises a plurality of entities, corresponding entity attributes and spatial relationships between the entities; the decision optimization module comprises a planning optimization model for optimizing a rural space matching strategy; a decision output layer in communication connection with the intelligent processing layer and used to output an optimal planning result.

8. A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that when executed by a computer cause the computer to perform a method of any of claims 1-7. The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform the method of any one of claims 1 to 6.

Images