A land use driven pattern mining method and related equipment

Abstract

The present invention provides a land use driving pattern mining method and related equipment, which processes the original raster data of the study area to obtain processed raster data; calculates the processed raster data to obtain land use change characteristics, neighborhood land use characteristics and land use scene characteristics of the study area; inputs the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scene characteristics into the generated minimum spanning tree to partition, and obtain partition results; and uses the partition results to iterate the initial Bayesian network to obtain an iterated Bayesian network. The partition results are optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining result of the study area; and accurate mining of the land use driving pattern can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of spatial data mining, and in particular to a land use driving pattern mining method and related equipment. Background Art

[0002] With the progress of urbanization and social and economic development, the accurate identification of land use driving mechanisms has become a key and difficult issue in the scientific research of land change. Land use change is a complex spatial process, which is jointly influenced by multiple factors such as natural environment, social economy and policy system. Its driving mechanisms vary significantly in different regions and at different times. Accurately revealing the driving mechanisms of land use change is of great theoretical and practical significance for formulating scientific national land space planning and promoting regional sustainable development.

[0003] However, the spatial heterogeneity of land use driving forces makes it difficult for global modeling methods to accurately capture local variation characteristics, affecting the accuracy of land use change prediction and the effectiveness of policy making. In order to address the spatial heterogeneity of land use driving forces, there are currently three main types of spatial partitioning methods: partitioning methods based on spatial clustering, partitioning methods based on spatial optimization, and partitioning methods based on networks.

[0004] Spatial clustering-based zoning methods mainly include the K-Means clustering method, the hierarchical clustering algorithm, and the density-based spatial data clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN). These methods cluster spatial units with similar attribute characteristics into the same area. These methods are computationally efficient but usually ignore spatial continuity constraints, resulting in spatial dispersion of areas of the same category, making them difficult to directly apply to regional planning and management practices.

[0005] Partitioning methods based on spatial optimization mainly include the Automated Zoning Procedure (AZP) and the Regionalization with dynamically constrained agglomerative clustering and partitioning (REDCAP). By setting spatial continuity constraints, they optimize regional division to maximize intra-region homogeneity. This type of method can maintain spatial continuity, but has low computational efficiency when processing large-scale data and is prone to falling into local optimal solutions. At the same time, the optimization function with fixed parameters is difficult to capture nonlinear change characteristics, resulting in a decrease in boundary recognition accuracy.

[0006] Network-based partitioning methods utilize graph-theoretic network structures to represent spatial relationships and achieve spatial partitioning through methods such as network pruning or community discovery. Widely used network partitioning methods include the Minimum Spanning Tree (MST) and the Spatial 'K' Cluster Analysis by Tree Edge Removal (SKATER). While these methods achieve a balance between spatial continuity and attribute similarity by constructing and pruning spatial partitions based on the MST, SKATER's solution space is constrained by the initial MST structure, resulting in a lack of flexibility in the shape and size of partitions. Consequently, in highly heterogeneous regions such as urban-rural transition zones, SKATER may generate suboptimal partitions due to its fixed tree structure, failing to capture the complex spatial variation of driving mechanisms.

[0007] In summary, the defects of the existing technology are as follows:

[0008] (1) Spatial continuity is ignored and the partitioning is too rigid. Traditional clustering methods such as K-Means ignore spatial proximity constraints, resulting in spatial dispersion of regions of the same category, which is not conducive to regional management and policy implementation. Although MST-based methods such as SKATER maintain spatial continuity, the partitioning structure is too rigid and limited by the initial tree structure, making it difficult to adapt to the spatial variation of driving mechanisms in complex geographical environments. In particular, it performs poorly in areas such as urban-rural transition zones.

[0009] (2) Lack of uncertainty quantification mechanism: Uncertainty in land use driving force analysis is a key consideration in policy making. Existing zoning methods focus more on extracting deterministic patterns and lack effective uncertainty quantification mechanisms. Regression models assume linear relationships and independence between variables and are difficult to capture complex nonlinear interactions. Association rule mining focuses mainly on the co-occurrence frequency of events and cannot express causal relationships and uncertainty.

[0010] (3) The separation of driving force analysis and partitioning process: Existing methods usually regard spatial partitioning and driving force analysis as two independent processes. First, spatial partitioning is performed based on certain preset criteria, and then driving force analysis is performed separately in each partition. This separation may result in the partitioning results not being able to optimally serve the excavation of driving mechanisms, affecting the accuracy of subsequent analysis. Summary of the Invention

[0011] The present invention provides a land use driving pattern mining method and related equipment, the purpose of which is to achieve accurate mining of land use driving patterns.

[0012] In order to achieve the above object, the present invention provides a land use driving pattern mining method, comprising:

[0013] Step 1: Obtain land use classification data, socioeconomic data, natural data, and transportation data of the study area as original raster data;

[0014] Step 2, processing the original raster data to obtain processed raster data;

[0015] Step 3: Calculate the processed raster data to obtain the land use change characteristics, neighborhood land use characteristics, and land use scene characteristics of the study area;

[0016] Step 4: Partition the processed raster data, land use change characteristics, neighborhood land use characteristics, and land use scenario characteristics into the generated minimum spanning tree to obtain the partitioning results, and use the partitioning results to iterate the initial Bayesian network to obtain the iterated Bayesian network. The partitioning results are optimized according to the iterative Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area.

[0017] More specifically, step 2 includes:

[0018] Resample the land use classification data in the original raster data to obtain multiple grid cells;

[0019] Align the socioeconomic data, natural data, and traffic data in the original raster data to each grid cell according to the temporal and spatial locations to obtain aligned grid cells;

[0020] Discretize the socioeconomic data, natural data and traffic data in each grid unit to obtain discretized socioeconomic data, natural data and traffic data;

[0021] The aligned grid cells and discretized socioeconomic data, natural data, and traffic data are used as processed raster data.

[0022] More specifically, step 3 includes:

[0023] The mode of land use classification in each aligned grid cell is counted as the land use characteristics of the study area;

[0024] The land use characteristics are used to calculate the land use characteristic change vectors for multiple years to obtain the land use change characteristics of the study area;

[0025] The neighborhood of each aligned grid cell is calculated using the nearest neighbor algorithm, and the mode of land use characteristics in each neighborhood is counted to obtain the neighborhood land use characteristics of each grid cell in the study area.

[0026] The land use classification data in each aligned grid cell is converted into vector data to generate a buffer zone, which is then remapped and assigned to each aligned grid cell to obtain the land use scene characteristics of the study area.

[0027] Furthermore, before the processed raster data, land use change characteristics, neighborhood land use characteristics, and land use scenario characteristics are input into the generated minimum spanning tree for partitioning, the following steps are also included:

[0028] Based on each aligned grid cell, an initial spatial adjacency graph is established through a triangulated network. The edges in the initial spatial adjacency graph are randomly assigned values ​​to generate a weight matrix for representing the undirected graph. Based on the weight matrix, a spatial adjacency graph is obtained.

[0029] The Kruskal algorithm is used to generate the initial minimum spanning tree on the spatial adjacency graph. The initial minimum spanning tree includes Edge, Indicates the number of aligned grid cells.

[0030] Furthermore, based on the update operator, the partitioning result is used to iterate the initial Bayesian network to obtain the iterated Bayesian network, and the partitioning result is optimized according to the iterated Bayesian network.

[0031] More specifically, the update operators include:

[0032] The first update operator is used to randomly select a partition in the partition result and split it into two sub-partitions;

[0033] The second update operator is used to randomly select two adjacent partitions from the partition results and merge them into one partition;

[0034] The third update operator is used to randomly select a partition in the partition result and split it into two sub-partitions, and then select two adjacent partitions to merge into one partition;

[0035] The fourth update operator is used to randomly assign edge weights between 0 and 1 in the same partition and regenerate the minimum spanning tree.

[0036] Furthermore, the iteration termination conditions include:

[0037] The number of iterations reaches the preset total number of iterations;

[0038] There is no partitioning result and Bayesian network that is better than the previous iteration in several consecutive iterations.

[0039] The present invention also provides a land-utilizing driving mode excavation device, comprising:

[0040] The acquisition module is used to obtain land use classification data, socioeconomic data, natural data and transportation data of the study area as raw raster data;

[0041] A processing module is used to process the original raster data to obtain processed raster data;

[0042] The calculation module is used to calculate the processed raster data to obtain the land use change characteristics of the study area, the neighborhood land use characteristics and the land use scene characteristics;

[0043] The iterative module is used to partition the minimum spanning tree generated by inputting the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scenario characteristics to obtain the partitioning results, and use the partitioning results to iterate the initial Bayesian network to obtain the iterated Bayesian network. The partitioning results are optimized according to the iterative Bayesian network until the iteration termination conditions are met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area.

[0044] The present invention also provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a land use driving pattern mining method is implemented.

[0045] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the land use driving pattern mining method is implemented.

[0046] The above solution of the present invention has the following beneficial effects:

[0047] The present invention processes the acquired land use classification data, socio-economic data, natural data and traffic data of the study area as raw raster data to obtain processed raster data; calculates the processed raster data to obtain land use change characteristics, neighborhood land use characteristics and land use scene characteristics of the study area; inputs the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scene characteristics into the generated minimum spanning tree for partitioning to obtain partitioning results, and uses the partitioning results to iterate the initial Bayesian network to obtain an iterated Bayesian network. The partitioning results are optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area; compared with the existing technology, the present invention combines the minimum spanning tree and the Bayesian network. After each partitioning of the minimum spanning tree, the partition quality is immediately evaluated by the Bayesian network, and the minimum spanning tree is guided to optimize the previous partition. This tightly coupled design can achieve accurate mining of land use driving patterns.

[0048] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of the structure of the Bayesian network in an embodiment of the present invention;

[0051] Figure 3 Schematic diagram of the structure of the land use driving mode excavation device in an embodiment of the present invention;

[0052] Figure 4 Schematic diagram of the structure of the terminal device in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] To make the technical problems, technical solutions, and advantages to be solved by the present invention more clear, the following is a detailed description with reference to the accompanying drawings and specific embodiments. It is obvious that the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0054] In the description of the present invention, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and should not be understood as indicating or implying relative importance.

[0055] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0056] In view of the existing problems, the present invention provides a land use driving pattern excavation method and related equipment.

[0057] like Figure 1 As shown, an embodiment of the present invention provides a land use driving pattern mining method, comprising:

[0058] Step 1: Obtain land use classification data, socioeconomic data, natural data, and transportation data of the study area as original raster data;

[0059] Step 2, processing the original raster data to obtain processed raster data;

[0060] Step 3: Calculate the processed raster data to obtain the land use change characteristics, neighborhood land use characteristics, and land use scene characteristics of the study area;

[0061] Step 4, partitioning the minimum spanning tree generated by inputting the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scenario characteristics to obtain partitioning results, and using the partitioning results to iterate the initial Bayesian network to obtain the iterated Bayesian network. The partitioning results are optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area; in an embodiment of the present invention, the optimal partition structure may include urban areas, agricultural areas, and mountainous areas, and the Bayesian network of each partition is used to display the driving mechanism of land use change within the partition.

[0062] In an embodiment of the present invention, land use classification data uses existing Landsat image interpretation products. Landsat image interpretation products refer to surface information and features extracted by processing and analyzing remote sensing image data obtained by Landsat satellites, usually including land use status, vegetation cover, water distribution, urban expansion data, etc. Socioeconomic data are characterized by three indicators: gross domestic product, population density data, and night light data. The gross domestic product uses existing, rasterized gross domestic product data corresponding to the study area, and the night light data uses NPP-VIIRS image data. NPP-VIIRS image data is a type of night light remote sensing image data with higher spatial resolution and more accurate night light radiation intensity records. Natural data is mainly terrain data, represented by digital elevation models and their derived slope data. These natural data are based on the public SRTM DEM product. The SRTM DEM product is a digital elevation model (DEM) produced based on data obtained by the Shuttle Radar Topography Mission. Traffic data was calculated using OpenStreetMap road network data and Baidu Maps point of interest (POI) data to calculate the target area's distance to the city center (citydis), railway (CZT_D2R), major roads, and key facilities. All data was ultimately generated as raw raster data with a resolution of 300 meters.

[0063] Specifically, step 2 includes:

[0064] Resample the land use classification data in the original raster data to obtain multiple grid cells;

[0065] Align the socioeconomic data, natural data, and traffic data in the original raster data to each grid cell according to the temporal and spatial locations to obtain aligned grid cells;

[0066] Discretize the socioeconomic data, natural data and traffic data in each grid unit to obtain discretized socioeconomic data, natural data and traffic data;

[0067] The aligned grid cells and discretized socioeconomic data, natural data, and traffic data are used as processed raster data.

[0068] In the embodiment of the present invention, the land use classification data in the original raster data is resampled into 300-meter grid cells. The resampling is based on the mode of the land use classification in the 10×10 grid cells. The resampling calculation formula is:

[0069] ;

[0070] in, Represents the resampling result, represents land use classification data, represents the majority operation, represents the rows and columns of the grid cells, Indicates the offset in the row direction of the raster data. Indicates the offset in the column direction of the raster data.

[0071] Specifically, the socioeconomic data, natural data, and traffic data in the original raster data are aligned to each grid cell according to their temporal and spatial locations. The aligned grid cells are obtained, including:

[0072] ArcGIS was used to perform spatial calculations on the socio-economic data, natural data, and traffic data in the original raster data to obtain continuous values;

[0073] Then discretize based on the natural breakpoint method to obtain the eigenvalue;

[0074] Then, the eigenvalues ​​are assigned to the grid cells at the same spatial and temporal positions according to the spatial positions, and the grid cells whose eigenvalue quantiles are above the top 5% are deleted to obtain the aligned grid cells. The representation of the aligned grid cells is:

[0075] ;

[0076] in, Indicates the Aligned grid cells, Indicates the The center coordinates of the aligned grid cells, Indicates the Land use classification data in the aligned grid cells, Indicates the The first of the aligned grid cells eigenvalues.

[0077] Specifically, the embodiment of the present invention discretizes the socioeconomic data, natural data, and traffic data in each grid cell through a decision tree. The decision tree uses information gain as the splitting indicator. The discretization process is as follows:

[0078] Given a continuous feature and target variables (e.g. land use change type) , the expression for calculating information gain is:

[0079] ;

[0080] in, represents information gain, Threshold The discretized features, is the target variable entropy, is the conditional entropy;

[0081] Based on the optimal split point , select the threshold that maximizes the information gain:

[0082] ;

[0083] When selecting the optimal splitting point, the effect of the splitting is evaluated based on the reduction in entropy. At each splitting step, an attempt is made to maximize the information gain to automatically find the "optimal" threshold for converting continuous data into discrete intervals, thus completing the discretization process.

[0084] Specifically, step 3 includes:

[0085] The mode of land use classification in each aligned grid cell is counted as the land use characteristics of the study area;

[0086] The land use characteristics are used to calculate the land use characteristic change vectors for multiple years to obtain the land use change characteristics of the study area. The calculation expression is:

[0087] ;

[0088] in, Indicates the Land use change characteristics in the aligned grid cells, Indicates the The aligned grid cells are in Land use characteristics during the year;

[0089] The neighborhood of each aligned grid cell is calculated by the nearest neighbor algorithm, and the mode of land use characteristics in each neighborhood is counted to obtain the neighborhood land use characteristics of each grid cell in the study area. The expression is:

[0090] ;

[0091] in, Indicates the Neighborhood land use characteristics of the aligned grid cells;

[0092] The land use classification data in each aligned grid cell is converted into vector data to generate a buffer zone, which is then remapped and assigned to each aligned grid cell to obtain the land use scene characteristics of the study area.

[0093] In the embodiment of the present invention, the buffer zone is remapped into features 1 to 3, which respectively represent the cultivated land core area, the urban core area, the forest core area, and the land use edge area.

[0094] Specifically, before the processed raster data, land use change characteristics, neighborhood land use characteristics, and land use scenario characteristics are input into the generated minimum spanning tree for partitioning, the following steps are also included:

[0095] Based on each aligned grid cell, an initial spatial adjacency graph is established through a triangulated network. The edges in the initial spatial adjacency graph are randomly assigned values ​​to generate a weight matrix for representing the undirected graph. Based on the weight matrix, a spatial adjacency graph is obtained.

[0096] The Kruskal algorithm is used to generate the initial minimum spanning tree on the spatial adjacency graph. The initial minimum spanning tree includes Edge, Indicates the number of aligned grid cells.

[0097] In an embodiment of the present invention, an initial spatial adjacency graph is established based on each aligned grid cell through the Delaunay triangulation, which is expressed as:

[0098] ;

[0099] in, represents the initial spatial adjacency graph, represents a collection of grid cells, represents an edge set;

[0100] Then for The edges in the graph are randomly assigned values ​​0 to 1 to generate a weight matrix for representing the undirected graph. :

[0101] ;

[0102] Finally, the number of edges generated by Kruskal algorithm is The initial minimum spanning tree of ,in For grid cells the number of

[0103] Generated Include Edge, yes The spanning tree subgraph of is represented as:

[0104] ;

[0105] in, Represents a connected node and The edge of , and satisfies Minimum.

[0106] In the embodiment of the present invention, the Bayesian network is a probabilistic graphical model that represents the conditional independence between variables through a directed acyclic graph, which is formally expressed as ,in is a graph structure, For the parameter set, the Bayesian network score is calculated as follows:

[0107] ;

[0108] in, is the number of variables, For nodes The number of parent node configurations, For nodes The number of values, Represents a node in the dataset The value is And the parent node is configured as The number of samples, , is a priori hyperparameter, , is the gamma function.

[0109] To improve computational efficiency, the present invention utilizes the decomposability of the scoring function during Bayesian network training and optimizes the network structure search process through a dynamic programming algorithm. Specifically, the algorithm does not directly search in the vast directed acyclic graph space, but decomposes the learning problem of the global optimal structure into a series of local optimization sub-problems: for each variable (node) (in ), find the parent node set The optimal parent node configuration and its corresponding optimal local score when limited to a subset of variables By systematically computing, storing, and reusing these local optimal solutions for different subsets of variables, dynamic programming can avoid repeated scoring calculations for a large number of possible parent node combinations, thereby significantly reducing the complexity of finding the global optimal structure from super-exponential to single-exponential (with respect to the number of variables). ), which effectively improves the computational efficiency of structure learning.

[0110] The embodiment of the present invention also introduces a complexity penalty term to limit network complexity:

[0111] ;

[0112] in, represents the complexity penalty term, is the penalty coefficient, It is a measure of network complexity, expressed as the ratio of the number of times the parent and child transaction categories appear in the dataset to the number of times the parent transaction category appears in the dataset, multiplied by the product of the parent and child transaction categories. During the optimization process, the closer the network score is to zero, the more significant the learning effect of the Bayesian network.

[0113] Specifically, based on the update operator, the partitioning result is used to iterate the initial Bayesian network to obtain the iterated Bayesian network, and the partitioning result is optimized according to the iterated Bayesian network.

[0114] Specifically, the update operators include:

[0115] The first update operator is used to randomly select a partition from the partition results and split it into two sub-partitions. The acceptance probability of this update operator is equal to the operation probability ratio × the Bayesian network fitness ratio × (1-penalty term c);

[0116] The second update operator is used to randomly select two adjacent partitions from the partition results and merge them into one partition. The acceptance probability of this update operator is equal to the operation probability ratio × the Bayesian network fitness ratio / (1-penalty term c);

[0117] The third update operator is used to randomly select a partition in the partition result to split it into two sub-partitions, and then select two adjacent partitions to merge into one partition. The acceptance probability of this update operator is equal to the operation probability ratio × the Bayesian network fitness ratio × (1-penalty term c);

[0118] The fourth update operator is used to randomly assign edge weights between 0 and 1 in the same partition and regenerate the minimum spanning tree.

[0119] During the iteration process, based on the current partition results, the above four update operators are used to generate candidate partitions, and then the acceptance probability is calculated. And compare it with a random number between 0 and 1. If the random number is greater than or equal to the acceptance probability, the candidate partition is used as the new partition result.

[0120] It should be noted that in order to avoid the optimization process falling into a local optimal solution, the embodiment of the present invention adopts an optimization mechanism similar to simulated annealing, the core of which is the use of a non-deterministic state acceptance criterion. This criterion allows the system to accept not only the state that can improve the fitness score (usually means Better than or equal to ) of the candidate partition states, and according to a specific probability Acceptance may result in a temporary decrease in fitness (i.e. Inferior ), this ability to randomly accept "inferior" solutions gives the optimization algorithm the possibility to escape from the local optimal basin of attraction and explore a wider solution space, thereby increasing the chance of ultimately finding the global optimal or high-quality partitioning solution.

[0121] Specific acceptance probability The calculation is as follows:

[0122] ;

[0123] This formula shows that when the fitness score of the candidate partition No worse than the current score When , the acceptance probability is 1 (definite acceptance); when Inferior When , the acceptance probability is the ratio of the absolute value of the current score to the absolute value of the new score ( )of The absolute value is used here to handle the situation where the score may be negative (such as the BIC score). What is compared is the "goodness" of the score rather than the simple numerical size.

[0124] Among them, the key parameters affecting the calculation of acceptance probability are (Its role is similar to the temperature in the simulated annealing algorithm), its baseline value (or directly affect Other relevant parameters calculated) will vary with the number of iterations Gradually decrease and decay according to the following formula:

[0125] ;

[0126] in, is the initial reference value, is the decay rate (usually set to 0.0005), is the current iteration number;

[0127] As the iteration proceeds, (or related parameters) decreases, resulting in the probability of accepting an inferior solution The overall decrease makes the algorithm tend to explore widely in the early stage of search (accept more inferior solutions), while it focuses more on utilization in the later stage (concentrating near high-quality solutions), and eventually converges.

[0128] The probability of selecting the operation corresponding to the operator is dynamically adjusted according to the current number of partitions: when the number of partitions is 1, it is set to , and normalized; the number of partitions is When set to , and normalized; otherwise, set to , and normalized, if given an unnormalized probability vector , and its normalized calculation formula is:

[0129] ;

[0130] That is, vector Divide each element in by The sum of all elements in .

[0131] It should be noted that the embodiment of the present invention is for each partition Dataset , using Bayesian network to train a single partition model, where "single partition model" refers to the partition A subset of data within , using the precise structure learning method based on BIC scoring and dynamic programming and the corresponding parameter learning, to build a local Bayesian network model that can best fit the partition data Then compare all partition sets after the operation The partition set before the operation , and get the overall fitness score .

[0132] The core part of the fitness calculation method is corresponding to a single partition The contribution of is calculated based on the following formula:

[0133] ;

[0134] in, Represents a partition Internal Node The value is And the parent node is configured as The number of samples, . The formula calculates the partition The log-likelihood of the internal data, which reflects the learned local Bayesian network model For this partition data In actual evaluation, the contribution of all partitions (or more accurately, the complete local score including the model complexity penalty, such as ) is accumulated and may be combined with other global penalty items (such as the penalty for partition size) to obtain the overall fitness score This fitness score essentially measures the degree to which the entire dataset can be explained by a series of local Bayesian network models under the current partitioning scheme, or the certainty of the rules. A better score (for example, a smaller absolute value of the BIC score) indicates a higher overall quality of the partitioning scheme and the corresponding network structure.

[0135] Specifically, the iteration termination conditions include:

[0136] The number of iterations reaches the preset total number of iterations;

[0137] There is no partitioning result and Bayesian network that is better than the previous iteration in several consecutive iterations.

[0138] The present invention uses the land use classification data of a certain urban area from 2015 to 2020 to illustrate the provided method:

[0139] First, select the original raster data within a certain urban area. The original raster data includes land use classification data, socioeconomic data, natural data, and traffic data. The resolution of the land use classification data in the original raster data is 30 meters.

[0140] The 30-meter-resolution land use classification data was resampled into 300-meter grid cells using a 10×10 window. The socioeconomic data, natural data, and transportation data in the original raster data were aligned to each grid cell, and abnormal grid cells were deleted.

[0141] The mode of land use classification data in the aligned grid cells is counted as the land use feature, and the annual change series of land use features from 2015 to 2020 is calculated as the land use change feature. The neighboring cells of each aligned grid cell are calculated using the K8 nearest neighbor algorithm, and the mode of land use features in the neighborhood is counted as the neighborhood land use feature;

[0142] All aligned raster cells were converted into vector data, and a 200m internal buffer was generated to identify the cultivated land core area, urban core area, forest core area, and land use edge area, which were remapped to land use scenario features of 0-3.

[0143] For continuous natural and transportation data (such as altitude, slope, and distance), a decision tree-based discretization method is applied, using information gain as the splitting metric. For example, altitude values ​​are discretized into low altitude zones (0-200m), medium altitude zones (200-500m), and high altitude zones (>500m). For discretized categorical factors (such as land use type), their original category codes are retained.

[0144] Based on the aligned grid cells, a spatial adjacency graph is constructed using the Delaunay triangulation. , randomly assign weights of 0-1 to the edges in the graph and generate a weight matrix , using Kruskal algorithm in Generate the initial minimum spanning tree , including Edge ( is the number of grid cells), for the 15,241 grid cells in the study area, the initial minimum spanning tree contains 15,240 edges;

[0145] Based on the initial partition data, a Bayesian network model was trained and the BDeu score was calculated. A dynamic programming algorithm was used to optimize the network structure search. The Bayesian network obtained from the initial training showed that land use changes in the study area were mainly affected by GDP, transportation data, and neighborhood land use data. The BDeu score of the initial Bayesian network was -368,742.5.

[0146] Partition iterative optimization; set initial benchmark value , attenuation rate The number of iterations was preset to 15,000. During the optimization process, the selection probabilities of the four update operators (Birth, Death, Change, and Hyper) were dynamically adjusted according to the current number of partitions. After 5,217 iterations, the optimal partition structure was obtained, consisting of five spatially continuous partitions. The BDeu score of the Bayesian network was improved to -256,389.2, a 30.5% improvement compared to the initial Bayesian network.

[0147] Results evaluation: The optimal partition structure contains multiple spatially continuous partitions, namely urban areas where land use changes are mainly affected by population density data and GDP, plain agricultural areas where land use changes are mainly affected by water distribution in natural data, and mountainous areas where land use changes are mainly affected by slope and altitude. The Bayesian network structure of each partition is as follows Figure 2 As shown, the driving mechanisms of land use change in different regions are clearly different;

[0148] To further optimize the connection method of the partition structure, the edge weights are resampled according to the existing partition boundary constraints. The edge weights in the same partition are randomly assigned in the range of 0-0.5, and the edge weights in different partitions are randomly assigned in the range of 0.5-1. The Kruskal algorithm is used to generate a new minimum spanning tree. The newly generated tree structure maintains the original partition boundaries unchanged, while optimizing the connection relationship within the partition, improving the interpretability and visualization of the model.

[0149] The embodiment of the present invention processes the acquired land use classification data, socioeconomic data, natural data and traffic data of the study area as raw raster data to obtain processed raster data; calculates the processed raster data to obtain land use change characteristics, neighborhood land use characteristics and land use scene characteristics of the study area; inputs the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scene characteristics into the generated minimum spanning tree for partitioning to obtain partitioning results, and uses the partitioning results to iterate the initial Bayesian network to obtain an iterated Bayesian network. The partitioning results are optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area; compared with the prior art, the embodiment of the present invention combines the minimum spanning tree and the Bayesian network. After each partition of the minimum spanning tree is completed, the partition quality is immediately evaluated by the Bayesian network, and the minimum spanning tree is guided to optimize the previous partition. This tightly coupled design can achieve accurate mining of land use driving patterns.

[0150] Corresponding to the land use driving pattern mining method described in the above embodiment, Figure 3 As shown, an embodiment of the present invention further provides a land use driving mode mining device 100, the land use driving mode mining device 100 comprising:

[0151] An acquisition module 101 is used to acquire land use classification data, socioeconomic data, natural data and traffic data of a study area as original raster data;

[0152] The processing module 102 is used to process the original raster data to obtain processed raster data;

[0153] The calculation module 103 is used to calculate the processed raster data to obtain the land use change characteristics of the study area, the neighborhood land use characteristics and the land use scene characteristics;

[0154] The iterative module 104 is used to partition the minimum spanning tree generated by inputting the processed raster data, land use change characteristics, neighborhood land use characteristics and land use scenario characteristics to obtain partition results, and use the partition results to iterate the initial Bayesian network to obtain an iterated Bayesian network. The partition results are optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining results of the study area.

[0155] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0156] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0157] The embodiment of the present invention further provides a terminal device, such as Figure 4 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the above-mentioned land use driving pattern mining method when executing the computer program D102.

[0158] The terminal device D10 can be a computing device such as a desktop computer, a notebook, a PDA, a server, a server cluster, a cloud server, etc. The terminal device may include, but is not limited to, a processor D100 and a memory D101. It will be understood by those skilled in the art that Figure 4 This is merely an example of the terminal device D10 and does not constitute a limitation on the terminal device D10 . The terminal device D10 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device D10 may also include input and output devices, network access devices, etc.

[0159] The processor D100 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0160] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device D10. Furthermore, the memory D101 may include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory D101 may also be used to temporarily store data that has been output or is about to be output.

[0161] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0162] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0163] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the land use driving pattern mining method is implemented.

[0164] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a construction device / terminal device, a recording medium, computer memory, read-only memory (ROM), random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium. Examples include a USB flash drive, a removable hard drive, a magnetic disk, or an optical disk.

[0165] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A land use driving pattern mining method, characterized in that: include: Step 1: Obtain land use classification data, socioeconomic data, natural data, and transportation data of the study area as original raster data; Step 2, processing the original raster data to obtain processed raster data; Step 3, calculating the processed raster data to obtain the land use change characteristics, neighborhood land use characteristics and land use scene characteristics of the study area; Step 4, partitioning the minimum spanning tree generated by inputting the processed raster data, the land use change characteristics, the neighborhood land use characteristics, and the land use scene characteristics to obtain a partitioning result, and using the partitioning result to iterate the initial Bayesian network to obtain an iterated Bayesian network, the partitioning result is optimized according to the iterated Bayesian network until the iteration termination condition is met, and the optimal partition structure of the study area and the Bayesian network of each partition in the optimal partition structure are output as the land use driving pattern mining result of the study area; The step 2 includes: Resampling the land use classification data in the original raster data to obtain a plurality of grid cells; Aligning the socioeconomic data, natural data, and traffic data in the original raster data to each grid cell according to spatiotemporal locations to obtain aligned grid cells; Discretize the socioeconomic data, natural data and traffic data in each grid unit to obtain discretized socioeconomic data, natural data and traffic data; The aligned grid cells and discretized socioeconomic data, natural data and traffic data are used as processed raster data; The step 3 includes: Counting the mode of land use classification in each aligned grid cell as the land use characteristics of the study area; Calculating land use characteristic change vectors for multiple years using the land use characteristics to obtain land use change characteristics of the study area; The neighborhood of each aligned grid cell is calculated by the nearest neighbor algorithm, and the mode of the land use characteristics in each neighborhood is counted to obtain the neighborhood land use characteristics of each grid cell in the study area; The land use classification data in each aligned grid cell is converted into vector data to generate a buffer zone, and the buffer zone is remapped and assigned to each aligned grid cell to obtain the land use scene characteristics of the study area.

2. The land use driving pattern mining method according to claim 1, characterized in that: Before inputting the processed raster data, the land use change characteristics, the neighborhood land use characteristics, and the land use scene characteristics into the generated minimum spanning tree for partitioning, the method further includes: Based on each aligned grid cell, an initial spatial adjacency graph is established through a triangulated network, and edges in the initial spatial adjacency graph are randomly assigned values ​​to generate a weight matrix for representing an undirected graph, and a spatial adjacency graph is obtained based on the weight matrix; An initial minimum spanning tree is generated on the spatial adjacency graph using the Kruskal algorithm, wherein the initial minimum spanning tree includes Edge, Indicates the number of aligned grid cells.

3. The land use driving pattern mining method according to claim 2, characterized in that: Based on the update operator, the initial Bayesian network is iterated using the partitioning result to obtain an iterated Bayesian network, and the partitioning result is optimized according to the iterated Bayesian network.

4. The land use driving pattern mining method according to claim 3, characterized in that: The update operator includes: A first update operator is used to randomly select a partition from the partition results and split it into two sub-partitions; A second update operator is used to randomly select two adjacent partitions from the partition results and merge them into one partition; A third update operator is used to randomly select a partition from the partition results and split it into two sub-partitions, and then select two adjacent partitions to merge into one partition; The fourth update operator is used to randomly assign edge weights between 0 and 1 in the same partition and regenerate the minimum spanning tree.

5. The land use driving pattern mining method according to claim 1, characterized in that: The iteration termination conditions include: The number of iterations reaches the preset total number of iterations; There is no partitioning result and Bayesian network that is better than the previous iteration in several consecutive iterations.

6. A land use driving mode excavation device, characterized in that: include: The acquisition module is used to obtain land use classification data, socioeconomic data, natural data and transportation data of the study area as raw raster data; A processing module, configured to process the original raster data to obtain processed raster data; A calculation module is used to calculate the processed raster data to obtain the land use change characteristics, neighborhood land use characteristics and land use scene characteristics of the study area; an iterative module, configured to partition the minimum spanning tree generated by inputting the processed raster data, the land use change characteristics, the neighborhood land use characteristics, and the land use scenario characteristics to obtain partition results, and to iterate the initial Bayesian network using the partition results to obtain an iterated Bayesian network, wherein the partition results are optimized according to the iterated Bayesian network until an iteration termination condition is satisfied, and an optimal partition structure of the study area and a Bayesian network of each partition in the optimal partition structure are output as land use driving pattern mining results of the study area; The acquisition module includes: Resampling the land use classification data in the original raster data to obtain a plurality of grid cells; Aligning the socioeconomic data, natural data, and traffic data in the original raster data to each grid cell according to spatiotemporal locations to obtain aligned grid cells; Discretize the socioeconomic data, natural data and traffic data in each grid unit to obtain discretized socioeconomic data, natural data and traffic data; The aligned grid cells and discretized socioeconomic data, natural data and traffic data are used as processed raster data; The processing module includes: Counting the mode of land use classification in each aligned grid cell as the land use characteristics of the study area; Calculating land use characteristic change vectors for multiple years using the land use characteristics to obtain land use change characteristics of the study area; The neighborhood of each aligned grid cell is calculated by the nearest neighbor algorithm, and the mode of the land use characteristics in each neighborhood is counted to obtain the neighborhood land use characteristics of each grid cell in the study area; The land use classification data in each aligned grid cell is converted into vector data to generate a buffer zone, and the buffer zone is remapped and assigned to each aligned grid cell to obtain the land use scene characteristics of the study area.

7. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the land use driving pattern mining method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the land use driving pattern mining method according to any one of claims 1 to 5 is implemented.

Images