National land space resource planning system based on big data

Abstract

The present invention relates to the field of land and resources planning technology, specifically to a land and space resource planning system based on big data, the system comprising: a spatiotemporal graph modeling module that aggregates land use data, extracts the land use type of each geographical unit, embeds the type data into the spatiotemporal graph node, calculates the spatiotemporal dependency between nodes, and generates a spatiotemporal graph adjacency matrix. In the present invention, the spatiotemporal graph processing models the spatiotemporal dependency between nodes, accurately captures the complexity of spatial and temporal changes, combines differentiated time windows and small and large window convolution kernels, the system can effectively identify short-term and long-term change characteristics, makes the prediction of land use change more flexible, multi-scale spatial heterogeneity modeling further adjusts the spatial adjacency matrix, accurately depicts the mutual relationship between regions, improves the accuracy of spatial dependency calculation, weightedly fuses data trends at different time steps, effectively predicts land use changes, and provides a more accurate reference for decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of land resources planning, and in particular to a land space resources planning system based on big data. Background Art

[0002] The technical field of land and resources planning mainly involves the rational use, protection and management of natural resources. Its core goal is to optimize the spatial layout of land, improve resource utilization efficiency and achieve sustainable development of resources through scientific planning and management methods. This field includes land use planning, ecological and environmental protection, infrastructure construction, urban and rural development, etc., and uses various technical means such as remote sensing, geographic information system (GIS), big data analysis, etc. to support decision-making and policy implementation. By comprehensively considering multiple factors such as economy, society and environment, land and resources planning technology ensures the rational allocation and long-term utilization of resources.

[0003] Among them, the national land space resources planning system based on big data refers to a system that uses big data technology to collect, store, analyze and mine various information related to land resources, and combines geographic information systems (GIS), remote sensing technology and other means to scientifically plan and manage land resources, mineral resources, water resources, etc.

[0004] Existing technologies lack the ability to dynamically predict land use changes, usually relying only on static data models and ignoring spatiotemporal dependencies. The short-term and long-term characteristics of resource changes are not effectively distinguished, resulting in inaccurate predictions of long-term development trends. The relationship between spatial data and historical data has not been fully explored, resulting in resource management being unable to respond to rapidly changing environments in a timely manner. Existing methods do not adequately consider spatial heterogeneity and are unable to handle diverse spatial dependencies in complex environments, resulting in deviations in land resource development and protection decisions. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and propose a national land space resource planning system based on big data.

[0006] In order to achieve the above purpose, the present invention adopts the following technical scheme: The national land space resource planning system based on big data includes:

[0007] The spatiotemporal graph modeling module aggregates land use data, extracts the land use type of each geographic unit, embeds the type data into the spatiotemporal graph nodes, calculates the spatiotemporal dependencies between nodes, and generates the spatiotemporal graph adjacency matrix;

[0008] The spatiotemporal convolution processing module performs graph convolution operation on each node according to the spatiotemporal graph adjacency matrix, transfers the spatial information of adjacent nodes, performs convolution processing on the node data in combination with the time information, extracts the spatial dependency features at each moment, and generates spatiotemporal convolution output;

[0009] The multi-scale change learning module is based on the spatiotemporal convolution output, extracts short-term change features and long-term change features by setting differentiated time windows, applies small window convolution kernels and large window convolution kernels respectively, transfers node features layer by layer and updates them, and obtains multi-scale change features;

[0010] The spatial heterogeneity modeling module uses the multi-scale variation characteristics to adjust the spatial adjacency matrix in the space-time graph according to the regional spatial characteristics, calculates the mutual relationship between multiple spatial units, adjusts the spatial dependence intensity, and generates spatial heterogeneity characteristics;

[0011] The prediction result fusion module combines the spatial heterogeneity characteristics, extracts multi-time step information, combines it with the long-term dependency characteristics, calculates the data change trend of multiple time steps, and performs weighted fusion to generate land use change prediction values.

[0012] The adjacency matrix of the spatiotemporal graph includes land use types, geographic units, adjacent relationships, and spatiotemporal dependencies. The spatiotemporal convolution output includes spatial dependency features, temporal features, and node data changes. The multi-scale change features include short-term change features, long-term change features, small window convolution kernels, and large window convolution kernels. The spatial heterogeneity features include regional spatial features, spatial unit relationships, and spatial dependency intensity. The land use change prediction value includes multi-time step information, long-term dependency features, and weighted fusion results.

[0013] As a further solution of the present invention, the step of obtaining the spatiotemporal graph adjacency matrix is ​​specifically as follows:

[0014] Summarize land use data, extract geographic location parameters and time series information corresponding to the land use type of each geographic unit, combine proximity and time change rate, calculate the basic correlation value between geographic units, and generate the initial spatiotemporal correlation matrix;

[0015] Normalizing the correlation values ​​in the initial spatiotemporal correlation matrix, setting a normalization factor, adjusting the range of the correlation value according to the difference between the maximum value and the minimum value, and performing a binarization operation on the normalized matrix according to a specified threshold to obtain a preliminary adjacency binary matrix;

[0016] Based on the preliminary adjacency binary matrix, combined with the spatiotemporal correlation strength of adjacent geographic units and their own correlation weights, the formula is adopted:

[0017] ;

[0018] Calculate the association values ​​between multiple cells in the adjacency matrix to generate the spatiotemporal graph adjacency matrix;

[0019] in, Represents the geographic unit in the adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Representing geographical units The self-correlation strength of Representing geographical units The self-correlation strength of Representing geographical units With other geographical units The association weight between Representing geographical units With other geographical units The association weight between represents the intensity difference adjustment coefficient, which is used to control the impact of the difference in self-association intensity. Represents the weight adjustment coefficient, which is used to control the impact of the weight of neighboring geographic units on the results.

[0020] As a further solution of the present invention, the step of obtaining the spatiotemporal convolution output is specifically as follows:

[0021] Based on the spatiotemporal graph adjacency matrix, spatial feature transfer is performed on the data of each node and the adjacent nodes, and node features are weighted based on the spatial correlation matrix, and the node itself and neighbor node information are integrated to generate a spatial feature weighted matrix;

[0022] The spatial feature weighted matrix is ​​combined with the time series data of the node, and for each time window, the current spatial feature of the node is integrated with the previous time data, and the node feature is adjusted by time weighting to obtain an integrated spatiotemporal feature matrix;

[0023] The graph convolution operation is applied to the integrated spatiotemporal feature matrix using the formula:

[0024] ;

[0025] Perform spatial convolution on the node features at each time point to capture the dynamic spatial dependencies of the nodes and generate spatiotemporal convolution outputs;

[0026] in, Indicates time point The convolution output of is the time attenuation coefficient, which represents the weakening of the influence of time span on convolution. Is a node In time The characteristic time label, Is a node and The spatial connection weights between Is a node In time Features, represents the target time point, Representative Node With Node The spatial connection weights between Represents the number of neighbor nodes associated with the target node, Represents the total number of all associated nodes referenced in the calculation.

[0027] As a further solution of the present invention, the step of acquiring the multi-scale variation feature is specifically as follows:

[0028] Based on the spatiotemporal convolution output, the time series data is divided according to the set short-term and long-term time windows, the node feature change range in multiple time windows is extracted, and the short-term change feature matrix and the long-term change feature matrix are respectively constructed according to the time window type to generate the spatiotemporal feature matrix of the node;

[0029] For the short-term and long-term change feature matrices in the spatiotemporal feature matrix of the node, a small window convolution kernel and a large window convolution kernel are applied respectively, and a weighted operation is performed on the feature matrix based on the convolution kernel weight and the node adjacency relationship, the node features are updated and the information in the window is integrated to obtain a convolution update matrix;

[0030] Based on the convolution update matrix, by fusing the short-term and long-term convolution results, the formula is adopted:

[0031] ;

[0032] Combining short-term and long-term change characteristics, extracting change patterns within multi-scale time ranges, and generating multi-scale change characteristics;

[0033] in, Represents the multi-scale variation characteristics of nodes, Represents the convolution kernel weight under the short-term time window, which is weighted for each node feature in the short-term feature matrix. The points represent the convolution kernel weights under the long-term time window, and weight each node feature in the long-term feature matrix. represents the node feature value after convolution of the short-term time window, represents the node feature value after long-term time window convolution, is the fusion adjustment coefficient, which is used to balance the proportion of short-term and long-term features in the results. is the time distance factor between nodes, is the time decay coefficient, which controls the weight of the influence of time distance on long-term features.

[0034] As a further solution of the present invention, the step of acquiring the spatial heterogeneity feature is specifically:

[0035] Based on the multi-scale variation characteristics, extract the local features and adjacency weights of each spatial unit, calculate the initial influence coefficients between spatial units, and generate a preliminary spatial influence matrix by performing matrix operations in combination with the local features and adjacency weights;

[0036] The preliminary spatial influence matrix is ​​called, and the adjacency strength of each spatial unit is normalized and adjusted in combination with the regional spatial characteristics, and the influence range of the spatial unit is recalculated according to the normalized strength to generate an adjusted spatial adjacency matrix;

[0037] Based on the adjusted spatial adjacency matrix, the weights and comprehensive relationship strengths between superimposed units are calculated using the formula:

[0038] ;

[0039] Update the dependency strength between spatial units to generate spatial heterogeneity features;

[0040] in, Representation unit and The spatial heterogeneity eigenvalue of is the spatial relationship value between cells in the adjusted adjacency matrix, is the weight value between units, , Unit and The regional characteristic value of is the regional characteristic adjustment coefficient, which is used to balance the impact of regional characteristic value differences between differentiated units. is the weight interaction adjustment coefficient, which controls the influence of the sum of squares of weight interactions between adjacent units. , Respectively represent units and With unit The weight value of .

[0041] As a further solution of the present invention, the step of obtaining the predicted value of land use change is specifically as follows:

[0042] Combining the spatial heterogeneity characteristics and multi-time step information, extracting the feature variation range of each spatial unit in multiple time steps, calculating the local trend matrix of the time step according to the time step length and the feature variation, and generating an initial time step variation matrix;

[0043] Using the initial time step change matrix, the long-term dependent features are fused with the time step change information, and matrix weighted calculation is performed based on the interaction weights of the local trend of each time step and the long-term dependent features to generate a fused time step trend matrix;

[0044] Based on the fused time step trend matrix, the characteristic change values ​​of multiple time steps are weighted and the formula is adopted:

[0045] ;

[0046] Comprehensively calculate the data change trends of multiple time steps to generate land use change prediction values;

[0047] in, represents the predicted value of land use change, Represents the time step The weight parameter is used to measure the influence of the current time step. Represents the time step The characteristic value of reflects the characteristic change in the time step. is the time step The characteristic value of Represents the coefficient for adjusting short-term trends, controlling the impact of short-term feature differences on predictions, Represents the adjustment dependency coefficient, which adjusts the contribution of long-term feature fusion. is the temporal distance factor between spatial units, is the time attenuation coefficient, Represents space unit The weight parameter measures the impact of the spatial unit on the overall prediction. Represents space unit The eigenvalues ​​of describe the variation within the unit, represents the total number of time steps and is used to determine the overall length of the time series. Represents the total number of spatial units and is used to define the range of all units in the space.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are:

[0049] In the present invention, the prediction ability of land resources is improved by extracting land use types and their spatiotemporal dependencies. Spatiotemporal graph processing models the spatiotemporal dependencies between nodes and accurately captures the complexity of spatial and temporal changes. Combined with differentiated time windows and small and large window convolution kernels, the system can effectively identify short-term and long-term change characteristics, making the prediction of land use changes more flexible. Multi-scale spatial heterogeneity modeling further adjusts the spatial adjacency matrix, accurately depicts the relationship between regions, and improves the accuracy of spatial dependency calculations. Weighted fusion of data trends at different time steps effectively predicts land use changes and provides a more accurate reference for decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a system flow chart of the present invention;

[0051] Figure 2 A flow chart of the steps for obtaining the adjacency matrix of the space-time graph of the present invention;

[0052] Figure 3 Flow chart of the steps for obtaining the spatiotemporal convolution output of the present invention;

[0053] Figure 4 A flowchart of the steps for obtaining multi-scale variation features of the present invention;

[0054] Figure 5 Flow chart of the steps for obtaining the spatial heterogeneity characteristics of the present invention;

[0055] Figure 6 This is a flow chart of the steps for obtaining the predicted value of land use change in the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0058] Embodiment 1

[0059] See also Figure 1 , the national land space resource planning system based on big data includes:

[0060] The spatiotemporal graph modeling module aggregates land use data, extracts the land use type of each geographic unit, embeds the type data into the spatiotemporal graph nodes, calculates the spatiotemporal dependencies between nodes, and generates the spatiotemporal graph adjacency matrix;

[0061] The spatiotemporal convolution processing module performs graph convolution operations on each node based on the spatiotemporal graph adjacency matrix, transfers the spatial information of adjacent nodes, performs convolution processing on the node data in combination with the time information, extracts the spatial dependency features at each moment, and generates spatiotemporal convolution output;

[0062] The multi-scale change learning module is based on the spatiotemporal convolution output. By setting differentiated time windows, it extracts short-term change features and long-term change features, applies small window convolution kernels and large window convolution kernels respectively, transfers node features layer by layer and updates them, and obtains multi-scale change features.

[0063] The spatial heterogeneity modeling module uses multi-scale variation characteristics to adjust the spatial adjacency matrix in the space-time graph according to the regional spatial characteristics, calculates the relationship between multiple spatial units, adjusts the spatial dependence intensity, and generates spatial heterogeneity characteristics;

[0064] The prediction result fusion module combines spatial heterogeneity features, extracts multi-time step information, combines it with long-term dependency features, calculates the data change trends of multiple time steps, and performs weighted fusion to generate land use change prediction values.

[0065] The adjacency matrix of the spatiotemporal graph includes land use type, geographic unit, adjacent relationship, and spatiotemporal dependency. The spatiotemporal convolution output includes spatial dependency features, temporal features, and node data changes. Multi-scale change features include short-term change features, long-term change features, small window convolution kernels, and large window convolution kernels. Spatial heterogeneity features include regional spatial features, spatial unit relationships, and spatial dependency intensity. The predicted value of land use change includes multi-time step information, long-term dependency features, and weighted fusion results.

[0066] See also Figure 2 , the specific steps for obtaining the adjacency matrix of the space-time graph are:

[0067] Summarize land use data, extract geographic location parameters and time series information corresponding to the land use type of each geographic unit, combine proximity and time change rate, calculate the basic correlation value between geographic units, and generate the initial spatiotemporal correlation matrix;

[0068] Based on the extraction of associated data based on geographic unit types, by analyzing the association between geographic unit types and location parameters, further integrating time series information, evaluating the spatial proximity and temporal correlation between geographic units, and obtaining the calculation basis of basic spatiotemporal correlation values, this process involves a comprehensive evaluation of the geographical location of each unit type and its time mark. By calculating the spatial and temporal differences between these geographic units, a preliminary spatiotemporal correlation matrix is ​​constructed, which can reflect the intensity of interaction between different units within a specific time, and provide the necessary basic data for subsequent normalization and threshold binarization processing. These data are obtained and calculated through geographic information systems and time series analysis tools, ensuring the accuracy of data processing and the reliability of correlation analysis.

[0069] Normalize the correlation values ​​in the initial spatiotemporal correlation matrix, set the normalization factor, adjust the range of the correlation value according to the difference between the maximum value and the minimum value, perform a binarization operation on the normalized matrix according to the specified threshold, and obtain a preliminary adjacency binary matrix;

[0070] Starting from the initial spatiotemporal correlation matrix, the proportional factor of the normalization processing is determined by analyzing the maximum and minimum values ​​of the correlation values ​​in the matrix. This factor is set based on the distribution characteristics of the correlation values ​​to ensure that all values ​​are evenly distributed in the range of 0 to 1, thereby improving the consistency of data processing and the accuracy of subsequent operations. The binary processing converts the normalized correlation values ​​to 0 or 1 according to the set threshold. This process involves threshold judgment of the correlation values ​​and logical operations on the matrix to ensure that the final adjacent binary matrix can accurately reflect the spatiotemporal proximity between geographical units. These operations are completed through numerical analysis and logical judgment algorithms, ensuring the rigor of the processing process and the availability of the results.

[0071] Based on the preliminary adjacency binary matrix, combined with the spatiotemporal correlation strength of adjacent geographic units and their own correlation weights, the formula is adopted:

[0072] ;

[0073] Calculate the association values ​​between multiple cells in the adjacency matrix to generate the spatiotemporal graph adjacency matrix;

[0074] in, Represents the geographic unit in the adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Representing geographical units The self-correlation strength of Representing geographical units The self-correlation strength of Representing geographical units With other geographical units The association weight between Representing geographical units With other geographical units The association weight between represents the intensity difference adjustment coefficient, which is used to control the impact of the difference in self-association intensity. Represents the weight adjustment coefficient, which is used to control the impact of the weight of neighboring geographic units on the results.

[0075] formula:

[0076] ;

[0077] The benefit of the formula is that by combining the direct correlation between geographic units with the impact of each unit on other units, it can more carefully characterize the complex interactions between geographic units, thereby providing a richer data basis for establishing accurate adjacency relationships.

[0078] Detailed explanation of the formula and the process of formula calculation and derivation:

[0079] Assume that it is known , , , , total number of geographical units , and for each , known and The value of is 0.1, and the adjustment coefficient , :

[0080] ;

[0081] ;

[0082] This result shows that geographical units and The correlation value between them is about 0.353, indicating that the two geographical units have a moderate degree of interaction when considering the influence of time and space. This value will be used to construct the adjacency matrix of the spatiotemporal graph, which intuitively reflects the intensity of interaction between geographical units under specific time and space conditions.

[0083] See also Figure 3 , the steps to obtain the spatiotemporal convolution output are as follows:

[0084] Based on the spatiotemporal graph adjacency matrix, the spatial features of each node and its adjacent nodes are transferred. At the same time, the node features are weighted based on the spatial correlation matrix, and the node information and neighbor node information are integrated to generate a spatial feature weighted matrix.

[0085] In the spatial feature transmission stage, the data of the node and its adjacent nodes are weightedly transmitted through the spatiotemporal graph adjacency matrix to fuse the information of each node. The weighted average method of each node's own features and its neighboring node features is adopted. Each eigenvalue of the node is obtained by weighted summation of the relevant features of its direct neighbors. The weight is provided by the elements in the adjacency matrix. The processed spatial feature weighted matrix not only contains the information of the node itself, but also integrates the spatial information of the surrounding environment. The implementation of this step effectively prepares the spatial comprehensive feature data for the subsequent time series integration, and provides a basis for capturing the trend of time changes.

[0086] Combine the spatial feature weighted matrix with the time series data of the node. For each time window, integrate the current spatial feature of the node with the previous time data, adjust the node features by time weighting, and obtain the integrated spatiotemporal feature matrix.

[0087] On the basis of the spatial feature weighted matrix, the time series data of the node is integrated through a specific time window. In this process, the data at each time point must not only consider the current spatial features, but also the changes in the time dimension. The node features at different time points are adjusted through the time weighted method to reflect the impact of time changes. In this process, the features of each node at different time points are adjusted to reflect the impact of time flow. The integrated spatiotemporal feature matrix obtained provides a feature set that contains both spatial and temporal dimension information for the final convolution operation.

[0088] To integrate the spatiotemporal feature matrix, apply graph convolution operation, using the formula:

[0089] ;

[0090] Perform spatial convolution on the node features at each time point to capture the dynamic spatial dependencies of the nodes and generate spatiotemporal convolution outputs;

[0091] in, Indicates time point The convolution output of is the time attenuation coefficient, which represents the weakening of the influence of time span on convolution. Is a node In time The characteristic time label, Is a node and The spatial connection weights between Is a node In time Features, represents the target time point, Representative Node With Node The spatial connection weights between Represents the number of neighbor nodes associated with the target node, Represents the total number of all associated nodes referenced in the calculation.

[0092] formula:

[0093] ;

[0094] The benefit of the formula is that it enables the model to consider both the influence of time and space by combining the time decay factor and the space weight, thereby more accurately simulating the state changes of nodes at different time points.

[0095] Detailed explanation of the formula and the process of formula calculation and derivation:

[0096] Assume that at time point Next, node The characteristic time label is , the time decay coefficient is , the spatial connection weights between nodes is 0.5, node In time Features is 20. Calculate each part of the formula:

[0097] ;

[0098] ;

[0099] ;

[0100] Combining the above values:

[0101] ;

[0102] Assumptions nodes, .

[0103] The results showed that at the time point , after considering time decay and spatial weight, the node The comprehensive eigenvalue of is 163.74, which reflects the state of the node in a specific time and space and is used for further spatiotemporal data analysis and pattern recognition.

[0104] See also Figure 4 , the specific steps for obtaining multi-scale change features are:

[0105] Based on the spatiotemporal convolution output, the time series data is divided according to the set short-term and long-term time windows, the node feature change range in multiple time windows is extracted, and the short-term change feature matrix and the long-term change feature matrix are constructed according to the time window type to generate the spatiotemporal feature matrix of the node;

[0106] Based on the data output by spatiotemporal convolution, the time series of node feature data is divided into short-term and long-term through the set time window method. Through this division, the changes in node behavior on different time scales can be observed more carefully. The capture of short-term change characteristics mainly focuses on rapidly changing data points, while long-term change characteristics include a wider time range, so that more macro data trends can be grasped. This method not only improves the flexibility of data processing, but also makes data analysis more accurate. Therefore, in the specific implementation, it is necessary to comprehensively consider the size of the time window and the characteristics of the data to ensure that the obtained feature matrix can accurately reflect the actual situation of the node. The short-term and long-term time window settings are adjusted according to the change rate of the actual monitoring data. The node feature matrix will be used in subsequent more complex data processing steps.

[0107] For the short-term and long-term change feature matrices in the spatiotemporal feature matrix of the node, a small window convolution kernel and a large window convolution kernel are applied respectively. Based on the convolution kernel weight and the node adjacency relationship, a weighted operation is performed on the feature matrix, the node features are updated and the information in the window is integrated to obtain the convolution update matrix.

[0108] After obtaining the node feature matrix based on the time window, convolution kernels of different sizes will be applied to these matrices. Small window convolution kernels are suitable for capturing subtle changes in the short term, while large window convolution kernels are more suitable for capturing long-term trend changes. This method can ensure the comprehensive update of node features and integrate short-term and long-term data analysis results. The weight setting and selection of the convolution kernel are the optimal parameters obtained based on historical data analysis, which requires continuous adjustment through experiments and iterative tests to achieve the best data processing effect. The updated node features will better reflect the dynamic changes of time series data, providing a solid foundation for the next step of data analysis and application.

[0109] Based on the convolution update matrix, by fusing the short-term and long-term convolution results, the formula is adopted:

[0110] ;

[0111] Combining short-term and long-term change characteristics, extracting change patterns within multi-scale time ranges, and generating multi-scale change characteristics;

[0112] in, Represents the multi-scale variation characteristics of nodes, Represents the convolution kernel weight under the short-term time window, which is weighted for each node feature in the short-term feature matrix. The points represent the convolution kernel weights under the long-term time window, and weight each node feature in the long-term feature matrix. represents the node feature value after convolution of the short-term time window, represents the node feature value after long-term time window convolution, is the fusion adjustment coefficient, which is used to balance the proportion of short-term and long-term features in the results. is the time distance factor between nodes, is the time decay coefficient, which controls the weight of the influence of time distance on long-term features.

[0113] formula:

[0114] ;

[0115] The benefit of the formula is that it can effectively integrate short-term and long-term data features, adjust the weight of long-term data features through the time decay coefficient, and increase the adaptability and flexibility of the model.

[0116] Detailed explanation of the formula and the process of formula calculation and derivation:

[0117] Set specific values: , , , , , , , , .

[0118] ;

[0119] The result shows that the comprehensive value of the multi-scale change characteristics is 112, which indicates that the change characteristics of the nodes are integrated into a comprehensive indicator after short-term and long-term convolution processing, providing a basis for subsequent data analysis.

[0120] See also Figure 5 , the specific steps for obtaining spatial heterogeneity features are:

[0121] Based on the multi-scale variation characteristics, the local features and adjacency weights of each spatial unit are extracted, the initial influence coefficients between spatial units are calculated, and the preliminary spatial influence matrix is ​​generated by performing matrix operations by combining local features and adjacency weights.

[0122] Firstly, the feature weight of each unit is calculated according to the regional spatial distribution of multi-scale change characteristics, and the feature weight is combined with the adjacency weight in the adjacency matrix. The initial influence coefficient between each unit and its neighbors is calculated by traversing each unit one by one. Then, each initial influence coefficient is normalized by weight normalization to ensure that the mutual influence values ​​between all units are distributed within a specific interval. After completing the influence calculation of all units, the preliminary spatial influence matrix is ​​obtained by integration.

[0123] Call the preliminary spatial influence matrix, normalize and adjust the adjacency strength of each spatial unit in combination with the regional spatial characteristics, recalculate the influence range of the spatial unit according to the normalized strength, and generate the adjusted spatial adjacency matrix;

[0124] The preliminary spatial influence matrix is ​​called to analyze the key parameters contained in the regional spatial characteristics item by item. For the spatial adjacency relationship, all non-zero elements in the matrix are normalized by combining the weight distribution of local unit features to generate an adjacency matrix with a higher correlation. Then, the normalized matrix is ​​used to reconstruct the spatial influence range of the unit, and the dependency strength of each spatial unit is gradually adjusted. After superimposing the adjusted results of all units, an adjusted spatial adjacency matrix is ​​formed.

[0125] Based on the adjusted spatial adjacency matrix, the weights and comprehensive relationship strengths between superimposed units are calculated using the formula:

[0126] ;

[0127] Update the dependency strength between spatial units to generate spatial heterogeneity features;

[0128] in, Representation unit and The spatial heterogeneity eigenvalue of is the spatial relationship value between cells in the adjusted adjacency matrix, is the weight value between units, , Unit and The regional characteristic value of is the regional characteristic adjustment coefficient, which is used to balance the impact of regional characteristic value differences between differentiated units. is the weight interaction adjustment coefficient, which controls the influence of the sum of squares of weight interactions between adjacent units. , Respectively represent units and With unit The weight value of .

[0129] formula:

[0130] ;

[0131] The benefit of the formula is that by introducing the interactive operation of the regional feature difference factor and the sum of squares of the adjacency relationship weights, it can express the complex dependencies between spatial units in a more detailed manner, thereby improving the accuracy and relevance of feature capture.

[0132] Detailed explanation of the formula and the process of formula calculation and derivation:

[0133] Obtained through monitoring (the spatial relationship value between cells in the adjacency matrix), (inter-unit weight value), , , (Regional characteristics adjustment coefficient), (weight interaction adjustment coefficient), and the weights between neighboring cells in the adjacency matrix and They are and .

[0134] Calculation formula for the first term:

[0135] ;

[0136] Calculate the second term of the formula:

[0137] ;

[0138] Comprehensive formula:

[0139] ;

[0140] The results show that the unit and unit The spatial heterogeneity characteristic value of , indicating that the two units have a medium level of mutual correlation after adjustment of the spatial characteristics of the current region.

[0141] See also Figure 6 , the specific steps for obtaining the predicted value of land use change are:

[0142] Combining spatial heterogeneity characteristics and multi-time step information, extract the feature variation range of each spatial unit in multiple time steps, calculate the local trend matrix of the time step according to the time step length and feature variation, and generate the initial time step variation matrix;

[0143] Based on the preliminary fusion of multi-time step information and spatial heterogeneity characteristics, the local trend analysis of spatial units at each time step becomes a key factor in the data-driven model. The local characteristics construct the initial time step change matrix by comprehensively analyzing the data changes at multiple time points. This process involves a large amount of data normalization processing and time series analysis to ensure that the data at each time point reflects the true dynamics of the spatial unit. Through the precise calculation of the time step and the amount of change, the changing trend of land use in the short term can be revealed, thus providing a basis for the long-term trend. The accuracy of these data processing and analysis directly affects the reliability of the model prediction. In the data normalization processing, each data point needs to be standardized according to the statistical characteristics of its time step to ensure the consistency of the model input and the effectiveness of the comparison.

[0144] Using the initial time step change matrix, the long-term dependent features are fused with the time step change information, and matrix weighted calculation is performed based on the interaction weights of the local trend and long-term dependent features of each time step to generate a fused time step trend matrix.

[0145] In the construction of the fusion time step trend matrix, the weight adjustment of long-term dependent features and time step change information is a complex data fusion process. Every calculation step involved must be executed accurately to ensure that the final trend analysis can accurately reflect the change pattern within the time span. Through the fine adjustment of the interaction weights, data fusion not only needs to deal with the feature differences of each time step, but also consider the influence of long-term data. In this process, the allocation of weights determines the fusion effect of short-term and long-term data features, thereby affecting the model's prediction accuracy of future change trends. Comprehensive consideration of the weight configuration of various factors is a key step to improve the model's prediction performance.

[0146] Based on the fusion time step trend matrix, the characteristic change values ​​of multiple time steps are weighted and the formula is adopted:

[0147]

[0148] Comprehensively calculate the data change trends of multiple time steps to generate land use change prediction values;

[0149] in, represents the predicted value of land use change, Represents the time step The weight parameter is used to measure the influence of the current time step. Represents the time step The characteristic value of reflects the characteristic change in the time step. is the time step The characteristic value of Represents the coefficient for adjusting short-term trends, controlling the impact of short-term feature differences on predictions, Represents the adjustment dependency coefficient, which adjusts the contribution of long-term feature fusion. is the temporal distance factor between spatial units, is the time attenuation coefficient, Represents space unit The weight parameter measures the impact of the spatial unit on the overall prediction. Represents space unit The eigenvalues ​​of describe the variation within the unit, represents the total number of time steps and is used to determine the overall length of the time series. Represents the total number of spatial units and is used to define the range of all units in the space.

[0150] formula:

[0151] ;

[0152] The benefit of the formula is that it accurately calculates the predicted values ​​of land use change by dynamically adjusting the weights and feature differences at each time step, while taking into account the complex interactions in time and space.

[0153] Detailed explanation of the formula and the process of formula calculation and derivation:

[0154] Set the specific scene parameters and assume that the total time step , each time step Weight is 1.2, the characteristic value of the current step is 30, the eigenvalue of the previous step is 28, adjusting the coefficient of short-term trend is 0.05, the distance attenuation coefficient is 0.1, the distance factor is 2, and there are three spatial units, the weight of each unit is 1.1 and the eigenvalue =25. In the formula, the change trend of each time step is first calculated, and then the attenuation effect caused by the spatial distance is added to obtain the predicted value of land use change:

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] The results show that the predicted value of land use change after integrating time and space factors is 195.52, which shows that the model is able to predict future change trends based on multi-dimensional factors. Furthermore, this result will help to carry out more accurate land planning and management.

[0162] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The national land space resource planning system based on big data is characterized by: The system comprises: The spatiotemporal graph modeling module aggregates land use data, extracts the land use type of each geographic unit, embeds the type data into the spatiotemporal graph nodes, calculates the spatiotemporal dependencies between nodes, and generates the spatiotemporal graph adjacency matrix; The spatiotemporal convolution processing module performs graph convolution operation on each node according to the spatiotemporal graph adjacency matrix, transfers the spatial information of adjacent nodes, performs convolution processing on the node data in combination with the time information, extracts the spatial dependency features at each moment, and generates spatiotemporal convolution output; The multi-scale change learning module is based on the spatiotemporal convolution output, extracts short-term change features and long-term change features by setting differentiated time windows, applies small window convolution kernels and large window convolution kernels respectively, transfers node features layer by layer and updates them, and obtains multi-scale change features; The spatial heterogeneity modeling module uses the multi-scale variation characteristics to adjust the spatial adjacency matrix in the space-time graph according to the regional spatial characteristics, calculates the mutual relationship between multiple spatial units, adjusts the spatial dependence intensity, and generates spatial heterogeneity characteristics; The prediction result fusion module combines the spatial heterogeneity characteristics, extracts multi-time step information, combines it with the long-term dependency characteristics, calculates the data change trend of multiple time steps, and performs weighted fusion to generate land use change prediction values; The steps of acquiring the multi-scale variation features are specifically as follows: Based on the spatiotemporal convolution output, the time series data is divided according to the set short-term and long-term time windows, the node feature change range in multiple time windows is extracted, and the short-term change feature matrix and the long-term change feature matrix are respectively constructed according to the time window type to generate the spatiotemporal feature matrix of the node; For the short-term and long-term change feature matrices in the spatiotemporal feature matrix of the node, a small window convolution kernel and a large window convolution kernel are applied respectively, a weighted operation is performed on the feature matrix based on the convolution kernel weight and the node adjacency relationship, the node features are updated and the information in the window is integrated to obtain a convolution update matrix; Based on the convolution update matrix, by fusing the short-term and long-term convolution results, the formula is adopted: ； Combining short-term and long-term change characteristics, extracting change patterns within multi-scale time ranges, and generating multi-scale change characteristics; in, Represents the multi-scale variation characteristics of nodes, Represents the convolution kernel weight under the short-term time window, which is weighted for each node feature in the short-term feature matrix. The points represent the convolution kernel weights under the long-term time window, and weight each node feature in the long-term feature matrix. represents the node feature value after convolution of the short-term time window, represents the node feature value after long-term time window convolution, is the fusion adjustment coefficient, which is used to balance the proportion of short-term and long-term features in the results. is the time distance factor between nodes, is the time decay coefficient, which controls the weight of the influence of time distance on long-term characteristics; The steps for obtaining the spatial heterogeneity feature are specifically as follows: Based on the multi-scale variation characteristics, extract the local features and adjacency weights of each spatial unit, calculate the initial influence coefficients between spatial units, and generate a preliminary spatial influence matrix by performing matrix operations in combination with the local features and adjacency weights; The preliminary spatial influence matrix is ​​called, and the adjacency strength of each spatial unit is normalized and adjusted in combination with the regional spatial characteristics, and the influence range of the spatial unit is recalculated according to the normalized strength to generate an adjusted spatial adjacency matrix; Based on the adjusted spatial adjacency matrix, the weights and comprehensive relationship strengths between superimposed units are calculated using the formula: ； Update the dependency strength between spatial units to generate spatial heterogeneity features; in, Representation unit and The spatial heterogeneity eigenvalue of is the spatial relationship value between cells in the adjusted adjacency matrix, is the weight value between units, , Unit and The regional characteristic value of is the regional characteristic adjustment coefficient, which is used to balance the impact of regional characteristic value differences between differentiated units. is the weight interaction adjustment coefficient, which controls the influence of the sum of squares of weight interactions between adjacent units. , Respectively represent units and With unit The weight value of .

2. The national land space resource planning system based on big data according to claim 1 is characterized in that: The adjacency matrix of the spatiotemporal graph includes land use types, geographic units, adjacent relationships, and spatiotemporal dependencies. The spatiotemporal convolution output includes spatial dependency features, temporal features, and node data changes. The multi-scale change features include short-term change features, long-term change features, small window convolution kernels, and large window convolution kernels. The spatial heterogeneity features include regional spatial features, spatial unit relationships, and spatial dependency intensity. The land use change prediction value includes multi-time step information, long-term dependency features, and weighted fusion results.

3. The national land space resource planning system based on big data according to claim 2 is characterized in that: The steps for obtaining the spatiotemporal graph adjacency matrix are specifically as follows: Summarize land use data, extract geographic location parameters and time series information corresponding to the land use type of each geographic unit, combine proximity and time change rate, calculate the basic correlation value between geographic units, and generate the initial spatiotemporal correlation matrix; Normalizing the correlation values ​​in the initial spatiotemporal correlation matrix, setting a normalization factor, adjusting the range of the correlation value according to the difference between the maximum value and the minimum value, and performing a binarization operation on the normalized matrix according to a specified threshold to obtain a preliminary adjacency binary matrix; Based on the preliminary adjacency binary matrix, combined with the spatiotemporal correlation strength of adjacent geographic units and their own correlation weights, the formula is adopted: ； Calculate the association values ​​between multiple cells in the adjacency matrix to generate the spatiotemporal graph adjacency matrix; in, Represents the geographic unit in the adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Represents the geographic unit in the normalized preliminary adjacency matrix and The correlation value between Representing geographical units The self-correlation strength of Representing geographical units The self-correlation strength of Representing geographical units With other geographical units The association weight between Representing geographical units With other geographical units The association weight between represents the intensity difference adjustment coefficient, which is used to control the impact of the difference in self-association intensity. Represents the weight adjustment coefficient, which is used to control the impact of the weight of neighboring geographic units on the results.

4. The national land space resource planning system based on big data according to claim 3 is characterized in that: The steps for obtaining the spatiotemporal convolution output are specifically as follows: Based on the spatiotemporal graph adjacency matrix, spatial feature transfer is performed on the data of each node and the adjacent nodes, and node features are weighted based on the spatial correlation matrix, and the node itself and neighbor node information are integrated to generate a spatial feature weighted matrix; The spatial feature weighted matrix is ​​combined with the time series data of the node, and for each time window, the current spatial feature of the node is integrated with the previous time data, and the node feature is adjusted by time weighting to obtain an integrated spatiotemporal feature matrix; The graph convolution operation is applied to the integrated spatiotemporal feature matrix using the formula: ； Perform spatial convolution on the node features at each time point to capture the dynamic spatial dependencies of the nodes and generate spatiotemporal convolution outputs; in, Indicates time point The convolution output of is the time attenuation coefficient, which represents the weakening of the influence of time span on convolution. Is a node In time The characteristic time label, Is a node and The spatial connection weights between Is a node In time Features, represents the target time point, Representative Node With Node The spatial connection weights between Represents the number of neighbor nodes associated with the target node, Represents the total number of all associated nodes referenced in the calculation.

5. The national land space resource planning system based on big data according to claim 4 is characterized in that: The steps for obtaining the predicted value of land use change are specifically as follows: Combining the spatial heterogeneity characteristics and multi-time step information, extracting the feature variation range of each spatial unit in multiple time steps, calculating the local trend matrix of the time step according to the time step length and the feature variation, and generating an initial time step variation matrix; Using the initial time step change matrix, the long-term dependent features are fused with the time step change information, and matrix weighted calculation is performed based on the interaction weights of the local trend of each time step and the long-term dependent features to generate a fused time step trend matrix; Based on the fused time step trend matrix, the characteristic change values ​​of multiple time steps are weighted and the formula is adopted: ； Comprehensively calculate the data change trends of multiple time steps to generate land use change prediction values; in, represents the predicted value of land use change, Represents the time step The weight parameter is used to measure the influence of the current time step. Represents the time step The characteristic value of reflects the characteristic change in the time step. is the time step The characteristic value of Represents the coefficient for adjusting short-term trends, controlling the impact of short-term feature differences on predictions, Represents the adjustment dependency coefficient, which adjusts the contribution of long-term feature fusion. is the temporal distance factor between spatial units, is the time attenuation coefficient, Represents space unit The weight parameter measures the impact of the spatial unit on the overall prediction. Represents space unit The eigenvalues ​​of describe the variation within the unit, represents the total number of time steps and is used to determine the overall length of the time series. Represents the total number of spatial units and is used to define the range of all units in the space.

Images