Urban land utilization change simulation method, medium and equipment

Abstract

The invention discloses an urban land utilization change simulation method, medium and equipment, and relates to the technical field of land utilization change simulation, the method comprises the following steps: constructing a graph structure according to vector plot data, plots are nodes, each node corresponds to a vector cellular unit, edges exist if the plots have common points, and constructing an adjacent matrix; acquiring a feature matrix of nodes by using time sequence data, DEM data, POI data and OSM road network data of historical land utilization; constructing a graph attention neural network based on the graph structure, processing features output by the graph attention neural network, and inputting the processed features into a gating circulation unit; and based on the final hidden state of the gating circulation unit, calculating the conversion probability of the land parcel by using a vector cellular automaton, and carrying out land utilization change simulation based on the conversion probability of the land parcel. According to the method provided by the invention, the land utilization change of the plot-level data can be simulated from two scales of space and time, and the land utilization simulation precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of land use change simulation, and in particular to a method, medium and equipment for simulating urban land use change. Background Art

[0002] Urban land use change simulation is an important tool for analyzing and studying the trend and effect of urban land use change, and has always been the core content of land use / cover change research. Vector Cellular Automata (VCA) model, as a bottom-up dynamic modeling method, has been widely used in urban expansion and urban land use simulation. However, the existing urban land use change simulation method based on VCA lacks the consideration of the important factor of spatiotemporal interaction, and has certain limitations in solving the spatial heterogeneity problem and the temporal dependence problem at the plot scale.

[0003] Prior art, such as the patent application with publication number CN118691171A, combines graph convolutional neural network (GCN) and vector cellular automaton to achieve more accurate simulation of urban land use change. However, GCN inevitably inherits some shortcomings of convolutional neural network because it applies the convolution operation in convolutional neural network to graph structure. For example, due to the existence of the global shared parameter mechanism, GCN treats all neighbor nodes equally when performing convolution operation, and cannot assign different weights to each neighbor node according to the importance of neighbor nodes. This feature can capture local structural information well when classifying images because image information is locally related. However, in land use change simulation, land use types and change patterns in different regions may be affected by different factors, and this global parameter sharing may not be able to capture the specific characteristics of these regions. From the above, it can be found that the prior art will ignore the different effects of different neighbors on the target entity, resulting in deviations from neighbors and thus affecting the final simulation results.

[0004] In addition, the time and space of land use change have a strong interaction. The temporal dependence of land use change refers to the trend and pattern of changes in land use types and patterns in different periods. Both current and historical states and conditions will affect land use change. For example, suburban land used for urban expansion is undergoing rapid and continuous land use transformation, while the land use pattern of ecological protection areas remains unchanged for a long time. Therefore, in the simulation process, it is necessary not only to model land use changes at the spatial scale, but also to simulate the dynamic process of land use change at the temporal scale to achieve a more comprehensive and accurate prediction of land use change. However, existing technologies lack the mining of temporal driving factors and simulation at a finer scale, and cannot effectively explore the temporal dependence mechanism of land use change. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the existing land use change simulation method lacks time land use simulation, and proposes a method for simulating urban land use change, comprising the following steps:

[0006] S1. Obtain urban land data, including: vector plot data, time series data of historical land use, DEM data, POI data and OSM road network data;

[0007] S2. Build a graph structure based on the vector plot data, treat the plots as nodes, and construct an adjacency matrix if there are common points between two plots; use the time series data of historical land use, DEM data, POI data, and OSM road network data to obtain the feature matrix of the nodes, and each node of the graph structure corresponds to a vector cell unit;

[0008] S3. Build a graph attention neural network based on the graph structure, process the features output by the graph attention neural network and input them into the gated recurrent unit;

[0009] S4. Based on the final hidden state of the gated recurrent unit, a vector cellular automaton is used to calculate the conversion probability of the plot, and land use change simulation is performed based on the conversion probability of the plot.

[0010] Further,

[0011] POI data includes: school areas, store areas, residential areas, park areas, bus station areas, and industrial areas;

[0012] OSM road network data includes expressways, trunk roads, main roads, secondary roads, tertiary roads and their connecting roads, branch roads, and unclassified roads.

[0013] Further,

[0014] Get slope data from DEM data;

[0015] Calculate the Euclidean distance data from the pixels in the study area to the POI data and OSM road network data, convert the Euclidean distance data into raster data, and reproject the raster data;

[0016] Slope data, reprojected raster data, and time series data of historical land use are used as node features, and the reprojected raster data are the spatial distribution of driving factor values ​​in the study area.

[0017] Furthermore, the attention coefficient between a node and its neighbor nodes is expressed as:

[0018]

[0019] α ij =softmax(e ij )

[0020] Among them, e ij represents the attention coefficient between the i-th node and the j-th neighbor node, LeakyReLU represents the ReLU linear activation function, represents a trainable parameter vector, W represents a learnable weight matrix, represents the feature matrix of the l-th layer input of the i-th node, represents the feature matrix of the jth neighbor node’s lth layer input, || represents concatenation, α ij represents the normalized e ij .

[0021] Furthermore, the output of each layer of the graph neural network is expressed as:

[0022]

[0023] in, represents the feature matrix of the output of the l+1th layer of the i-th node, P represents the number of attention heads, represents the concatenation of the features of P attention heads, σ represents the sigmoid activation function, VN(i) represents the set of neighbor nodes of the i-th node, represents the normalized attention coefficients of the i-th and j-th nodes of the p-th attention head, W p represents the learnable weight matrix of the p-th attention head, Represents the feature matrix of the j-th neighbor node's l-th layer output.

[0024] Furthermore, the output data of the last layer of the graph neural network is converted into three-dimensional data and input into the gated recurrent unit.

[0025] Furthermore, the principle of the gated recurrent unit is expressed as:

[0026] r t =σ(W r x t +U r h t-1 +b r )

[0027]

[0028] Among them, r t represents the reset gate output at time step t; σ represents the sigmoid function; x t represents the input data at time step t; h t-1 represents the hidden state at time step t-1; represents the candidate hidden state at time step t, W, W r , U r ,U,b r , b are learnable parameters, where W, W r , U r , U represents the weight matrix, b r , b represents the bias term, h t represents the hidden state at time step t, z t represents the update gate output at time step t, ⊙ represents the Hadamard product;

[0029] Among them, when t = 0, x 0 The three-dimensional data converted from the output data of the last layer of the graph neural network.

[0030] Furthermore, based on the final hidden state of the gated recurrent unit, the conversion probability of the plot is calculated using a vector cellular automaton and expressed as:

[0031]

[0032] in, represents the constraint condition for the kth land use change of the ith plot at time step t; represents the random factor of the kth land use change occurring in the ith plot at time step t; It represents the final conversion probability of the kth land use change of the i-th plot at the t+1 time step; It represents the conversion suitability of the kth land use change of the ith plot at time step t; h t It is obtained through a fully connected layer and SoftMax.

[0033] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the urban land use change simulation method is implemented.

[0034] The present invention also proposes an electronic device, comprising a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned urban land use change simulation method.

[0035] The beneficial effects brought by the technical solution provided by the present invention are:

[0036] The present invention first constructs a graph structure and a graph attention neural network, uses the graph attention network to quantify the intensity of the influence of neighboring plots on the central plot, learns the heterogeneity between different nodes, and realizes the differential aggregation of the central plot, thereby effectively extracting potential spatial neighborhood features. Then, a gated recurrent unit is used to capture historical land use change information, and VCA technology is used to simulate land use changes in combination with the spatial neighborhood features extracted by the graph attention network. The method proposed in the present invention can well simulate land use changes of plot-level data from both spatial and temporal scales, and can significantly improve the accuracy of land use simulation while ensuring high computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of a method for simulating urban land use change according to an embodiment of the present invention;

[0038] Figure 2 is a block diagram of an electronic device in an exemplary embodiment of the present invention;

[0039] Figure 3 Detailed comparison diagram of urban land use change simulation of the method of the embodiment of the present invention and different methods; wherein Figure 3 (a) is a simulation display diagram of urban land use change by GU-VCA (method of the present invention), Figure 3 (b) is a simulation diagram of urban land use change using GAT-VCA (Graph Attention Network-Vector Cellular Automaton). Figure 3 (c) is a simulation diagram of urban land use change using ANN-VCA (artificial neural network-vector cellular automaton). Figure 3 (d) is a simulation diagram of urban land use change using RF-VCA (Random Forest-Vector Cellular Automaton). Figure 3 (e) is a simulation diagram of urban land use change using LR-VCA (logistic regression-vector cellular automaton). DETAILED DESCRIPTION

[0040] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0041] The flowchart of the urban land use change simulation method according to the embodiment of the present invention is as follows: Figure 1 , specifically including the following steps:

[0042] S1. Obtain urban land data, including: vector plot data, time series data of historical land use, DEM data, POI data and OSM road network data.

[0043] Vector parcel data are vector polygons that describe the location and shape of parcels.

[0044] POI (point of interest) data is point data. POI data includes: school areas, store areas, residential areas, park areas, bus station areas and industrial areas. POI data serves as a driving factor for VCA.

[0045] OSM road network data includes expressways, trunk roads, main roads, secondary roads, tertiary roads and their connecting roads, branch roads, and unclassified roads.

[0046] S2. Build a graph structure based on vector plot data, treat the plots as nodes, and if there are common points between two plots, there are edges, and build an adjacency matrix; use the time series data of historical land use, DEM data, POI data and OSM road network data to obtain the feature matrix of the node. Each node of the graph structure corresponds to a vector cell unit.

[0047] The graph structure can be expressed as G = (V, E, X, A), V = {v 1 ,v 2 ,...,v N} represents a set of nodes, where N is the number of nodes. And E represents the edge defined by the adjacency relationship between plots. The feature matrix X∈N×M represents the M attributes of the corresponding N plots. In the adjacency matrix A∈N×N, each element A ij Represents the connectivity between plots i and j. The construction process of the adjacency matrix can be expressed by the following equation:

[0048]

[0049] Where VN(i) is the neighbor set of the ith node, which is determined by the adjacency relationship between the plots. If the ith plot has a common edge or vertex with the jth plot, then A i,j will be set to 1; otherwise A i,jwill be set to 0. Since the adjacency between plots is mutual, the derived graph is an undirected graph.

[0050] Obtain slope data from DEM data; calculate the Euclidean distance data from pixels in the study area to POI data and OSM road network data, convert the Euclidean distance data into raster data; and reproject the raster data.

[0051] Slope data, reprojected raster data, and time series data of historical land use are used as node features, and the reprojected raster data are the spatial distribution of driving factor values ​​in the study area.

[0052] S3. Build a graph attention neural network based on the graph structure, and process the features output by the graph attention neural network and input them into the gated recurrent unit. The graph attention neural network is responsible for extracting the potential spatial features of the neighborhood, and the gated recurrent unit is responsible for capturing the time-dependent information of land use change data.

[0053] The graph structure constructed above is input into the graph attention neural network. In the graph attention network, the state of each node (plot) is not only affected by its own characteristics, but also by the state of its neighboring nodes (plots). The connection between each node and its neighboring nodes has a corresponding attention coefficient (denoted as e ij ), this coefficient measures the strength of the influence of neighbor nodes on the central node. The attention coefficient between a node and its neighbor nodes is expressed as:

[0054]

[0055] Among them, e ij represents the attention coefficient between the i-th node and the j-th neighbor node, LeakyReLU represents the ReLU linear activation function, represents a trainable parameter vector, W represents a learnable weight matrix, represents the feature matrix of the i-th node's l-th layer input, represents the feature matrix of the jth neighbor node’s lth layer input, || represents concatenation, α ij represents the normalized e ij .

[0056] Next, neighborhood aggregation is performed to assign a normalized attention coefficient a to each neighbor node. ij To update the characteristics of the central node. The result is obtained according to the following formula:

[0057]

[0058] The above is the result of a single attention head. A single operation may lead to insufficient detection of different interactions and ignore some important neighbors. Therefore, the graph attention will be repeated multiple times in parallel, combining all these similar graph attentions together to produce the final attention coefficient, called multi-head attention. The output of each layer of the graph neural network under multi-head attention is expressed as:

[0059]

[0060] in, represents the feature matrix of the output of the l+1th layer of the i-th node, P represents the number of attention heads, represents the concatenation of the features of P attention heads, σ represents the sigmoid activation function, VN(i) represents the set of neighbor nodes of the i-th node, represents the normalized attention coefficients of the i-th and j-th nodes of the p-th attention head, W p represents the learnable weight matrix of the p-th attention head, Represents the feature matrix of the j-th neighbor node's l-th layer output.

[0061] The new state of the central block is the weighted sum of its own state and the state of its neighboring nodes, and the weight is the normalized attention coefficient a calculated above. ij . This means that if a neighbor node has a greater influence on the central node, then its attention coefficient will be higher, and thus it will account for a larger proportion in the weighted sum. Based on the result of this weighted sum, the state of the central node will be updated to reflect its current state and the influence of the neighbor nodes. This process can be viewed as an adaptive and dynamic way to integrate local neighborhood information, where the influence of each neighbor is not fixed, but dynamically adjusted according to their characteristics and the structure of the network. Such a mechanism enables the network to more flexibly capture the complex interactions between different plots, thereby more accurately predicting the state evolution of the plots.

[0062] The output data of the last layer of the graph neural network is converted into three-dimensional data and input into the gated recurrent unit for learning long time series information.

[0063] The principle of the gated recurrent unit is expressed as:

[0064] r t =σ(W r x t +U r h t-1 +b r )

[0065]

[0066] Among them, rt represents the reset gate output at time step t; σ represents the sigmoid function; x t represents the input data at time step t; h t-1 represents the hidden state at time step t-1; represents the candidate hidden state at time step t, W, W r , U r ,U,b r , b are learnable parameters, where W, W r , U r , U represents the weight matrix, b r , b represents the bias term, h t represents the hidden state at time step t, z t represents the update gate output at time step t, and ⊙ represents the Hadamard product.

[0067] The reset gate determines how new input information is combined with previous memory. t When it is close to 0, almost all previous memories are discarded; when r t When it is close to 1, more previous memories are retained. The update gate controls the extent to which the state information of the previous moment is brought into the current state. t The closer it is to 1, the more old information is retained; the closer it is to 0, the more is forgotten.

[0068] Among them, when t = 0, x 0 The three-dimensional data converted from the output data of the last layer of the graph neural network.

[0069] S4. Based on the final hidden state of the gated recurrent unit, a vector cellular automaton is used to calculate the conversion probability of the plot, and land use change simulation is performed based on the conversion probability of the plot.

[0070] In the vector cellular automaton model, the final conversion probability of a plot consists of four parts: environmental driving effect P g , neighborhood effect Ω, constraint condition P c , random factor RA. Environmental driving effect refers to the impact of environmental driving factors (such as terrain characteristics, location conditions and socio-economic conditions) on land use conversion within a plot. Neighborhood effect refers to the impact of surrounding plots on land use change of the central plot. Constraints represent the prohibition of certain land use conversions on specific plots. Random factors are used to reflect the randomness in the development process of complex urban systems.

[0071] Therefore, the probability that plot i experiences the kth land use change at time step t is expressed as

[0072]

[0073] in, represents the conversion probability of the kth land use change occurring in the ith plot at time step t+1; represents the environmental driving effect of the kth land use change in the ith plot at time step t; represents the neighborhood effect of the kth land use change occurring in the ith plot at time step t; represents the constraint condition for the kth land use change of the ith plot at time step t. If the plot is allowed to be developed, it is assigned a value of 1, otherwise it is assigned a value of 0; The random factor representing the kth land use change of the ith plot at time step t represents the unknown disturbance, RA = 1 + (-lnγ) α , where γ is set to a random value between 0 and 1, and α is a parameter between 1 and 10.

[0074] However, in the embodiment of the present invention, a graph attention neural network is constructed based on the graph structure, and the features output by the graph attention neural network are processed and input into the gated recurrent unit. As a result, the dependency relationship between adjacent plots is learned, so that the central plot can integrate the environmental driving effect and the neighborhood effect. Indicates that Environmental driving effects have been incorporated and neighborhood effects Therefore, in the embodiment of the present invention, the probability that the plot i experiences the kth land use change at time step t is expressed as:

[0075]

[0076] in, It represents the final conversion probability of the kth land use change of the i-th plot at the t+1 time step; It represents the conversion suitability of the kth land use change of the ith plot at time step t; h t It is obtained through a fully connected layer and SoftMax.

[0077] The model output result h t Perform feature dimensionality reduction through a fully connected layer. For example, if the model output h t The dimension of is 16, and the present invention needs to classify 5 different land use types, so the role of the fully connected layer is to compress the 16-dimensional feature space into 5 dimensions to match the requirements of the classification task. After passing through the fully connected layer, the output dimension changes from 16 to 5, and each dimension represents the score of a category.

[0078] Next, these scores are processed through the SoftMax layer. The main function of the SoftMax layer is to normalize the scores on these five dimensions so that their sum is equal to 1. In this way, each score can be interpreted as the probability of the category. For example, an output vector [0.5, 0.3, 0.1, 0.05, 0.05], where each value represents the probability of the corresponding category, and their sum is 1.

[0079] Finally, the cellular automaton will simulate the probability of land use type conversion at the next time step (denoted as t+1) based on these probabilities.

[0080] Calculate the final Finally, the highest probability is selected and compared with the threshold to ensure that the simulation results match the actual distribution. If the probability is higher than the threshold, the plot will be changed to the corresponding land use category; otherwise, the cell remains unchanged. The threshold is determined during the simulation based on the total changed area from time t to t+1 according to the historical data of urban land use.

[0081] In order to quantitatively evaluate the simulation results of the model in the present invention, a variety of indicators are used for model evaluation, including overall accuracy (OA), quality factor (FoM), producer accuracy (PA) and user accuracy (UA). OA measures the global accuracy derived from the confusion matrix of each plot, which can be calculated by the following formula.

[0082]

[0083] Where m represents the number of land use categories. uu Area represents the total area of ​​the plots whose true category is u in the ground truth and is correctly predicted in category u. uv Represents the total area of ​​the plots whose true category is u in the real situation and is predicted in category v.

[0084] Since urban growth is a long-term process and most urban land uses remain unchanged over a short period of time, using only the confusion matrix for model evaluation may lead to overestimation of the simulation results. For example, even if the input dataset is used as the simulation result, high accuracy can still be obtained. Therefore, this study adopts FOM as an additional evaluation metric that can specifically consider the changing part. In addition, PA and UA are also derived together with FOM to evaluate the model performance. The relevant equations are as follows:

[0085]

[0086] Where A represents the area of ​​the plot where the observed change is predicted as persistence, B represents the area of ​​the plot where the observed change is simulated as change, C represents the area of ​​the plot where the observed change is predicted as the error gain class, and D represents the number of plots where the area observed persistence is predicted as change. Higher FOM values ​​indicate higher agreement at the plot level.

[0087] In order to solve the problem that existing technologies cannot simultaneously solve the spatial heterogeneity and time dependence problems at the plot scale, this case improves the calculation method of land use conversion suitability. The improved calculation method is more reasonable, conforms to the actual situation of land use changes, and can more accurately predict land use changes.

[0088] In an exemplary embodiment, a computer-readable storage medium is included, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned urban land use change simulation method is implemented.

[0089] See also Figure 2 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0090] Wherein, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned urban land use change simulation method.

[0091] In order to verify the effectiveness of the method of the present invention, in an embodiment of the present invention, the graph attention neural network consists of two graph attention layers. The first graph attention layer contains 8 attention heads, and 8-dimensional features are calculated for each head, while the second graph attention layer contains 1 attention head, which collects 64-dimensional features of the nodes and reshapes them into the input shape required by GRU and inputs them into the GRU layer, which outputs the output of the last time step. During model training, the learning rate is set to 0.001 and the training epochs is 1000. The cross entropy loss function is introduced into the neural network to select the best performance model. The present invention uses land use data from 2017 to 2022 as training data and land use data from 2023 as accuracy verification data. The ratio of training set to test set data is 7:3.

[0092] The proposed method is compared with several classical methods, including logistic regression, random forest, artificial neural network and basic graph attention network. In this case, the neighborhood distance is set to 600 meters. In the ANN-based model, the number of input nodes is consistent with the number of spatial variables, the number of output nodes is consistent with the number of land use conversion types, the number of hidden layers is set to 3, and 10-fold cross validation is used to obtain the optimal urban land use change simulation model. In the random forest method, the number of trees is set to 80, the best fitting model is obtained by OOB estimation, and the OOB (Out-of-Bag) rate is set to 0.3.

[0093] Table 1 lists the performance results of different methods (LR-VCA (logistic regression-vector cellular automaton), ANN-VCA (artificial neural network-vector cellular automaton), RF-VCA (random forest-vector cellular automaton) and GAT-VCA (graph attention network-vector cellular automaton) as well as the method of the present invention)

[0094] Table 1

[0095] Evaluation indicators LR-VCA ANN-VCA RF-VCA GAT-VCA Method of the present invention PA 0.1923 0.3625 0.3121 0.4369 0.4652 OA 0.8278 0.8788 0.8643 0.8851 0.8961 UA 0.3194 0.6429 0.4160 0.6112 0.6740 FoM 0.1423 0.3076 0.2227 0.3587 0.3990

[0096] Table 1 shows that the proposed method performs best compared with these four models. Compared with GAT-VCA, the overall accuracy is improved by 1.2%, UA is improved by 10.2%, PA is improved by 6.4%, and the simulation accuracy FOM is improved by 11.2%. Representing the plot as a graph can express spatial information more efficiently, which is helpful for classification and prediction tasks.

[0097] In order to more intuitively demonstrate the differences between these simulation results and the ground truth land use data, three small parts of the study area were selected for visual comparison. Figure 3 As shown, Figure 3 Detailed comparison diagram of urban land use change simulation of the method of the embodiment of the present invention and different methods; wherein Figure 3 (a) is a simulation display diagram of urban land use change by GU-VCA (method of the present invention), Figure 3 (b) is a simulation diagram of urban land use change using GAT-VCA (Graph Attention Network-Vector Cellular Automaton). Figure 3 (c) is a simulation diagram of urban land use change using ANN-VCA (artificial neural network-vector cellular automaton). Figure 3 (d) is a simulation diagram of urban land use change using RF-VCA (Random Forest-Vector Cellular Automaton). Figure 3 (e) is a simulation diagram of urban land use change using LR-VCA (logistic regression-vector cellular automaton).

[0098] The differences between the simulation results are mainly concentrated near the construction land. Except for the method proposed in this paper and GAT-VCA, the simulation results of other methods are very different from the actual ground conditions. Among these methods, the simulation results of the method proposed in this paper have the highest matching degree with the actual ground conditions, which shows that the method proposed in this paper can better extract the domain effect and use rich time series information to capture clearer time dependencies, thereby more effectively exploring the driving mechanism of land use change.

[0099] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for simulating urban land use change, characterized in that: The following steps are involved: S1. Obtain urban land data, including: vector plot data, time series data of historical land use, DEM data, POI data and OSM road network data; S2. Build a graph structure based on the vector plot data, treat the plots as nodes, and construct an adjacency matrix if there are common points between two plots; use the time series data of historical land use, DEM data, POI data, and OSM road network data to obtain the feature matrix of the nodes, and each node of the graph structure corresponds to a vector cell unit; S3. Build a graph attention neural network based on the graph structure, process the features output by the graph attention neural network and input them into the gated recurrent unit; S4. Based on the final hidden state of the gated recurrent unit, a vector cellular automaton is used to calculate the conversion probability of the plot, and land use change simulation is performed based on the conversion probability of the plot.

2. The method for simulating urban land use change according to claim 1, characterized in that: POI data includes: school areas, store areas, residential areas, park areas, bus station areas, and industrial areas; OSM road network data includes: expressways, trunk roads, main roads, secondary roads, tertiary roads and their connecting roads, branch roads, and unclassified roads.

3. The method for simulating urban land use change according to claim 1, characterized in that: Get slope data from DEM data; Calculate the Euclidean distance data from the pixels in the study area to the POI data and OSM road network data, convert the Euclidean distance data into raster data, and reproject the raster data; Slope data, reprojected raster data, and time series data of historical land use are used as node features, and the reprojected raster data are the spatial distribution of driving factor values ​​in the study area.

4. The method for simulating urban land use change according to claim 1, characterized in that: The attention coefficient between a node and its neighbor nodes is expressed as: a ij =softmax(e ij ) Among them, e ij represents the attention coefficient between the i-th node and the j-th neighbor node, LeakyReLU represents the ReLU linear activation function, represents a trainable parameter vector, W represents a learnable weight matrix, represents the feature matrix of the i-th node's l-th layer input, represents the feature matrix of the jth neighbor node’s lth layer input, || represents concatenation, α ij represents the normalized e ij .

5. The method for simulating urban land use change according to claim 1, characterized in that: The output of each layer of the graph neural network is expressed as: in, represents the feature matrix of the output of the l+1th layer of the i-th node, P represents the number of attention heads, represents the concatenation of the features of P attention heads, σ represents the sigmoid activation function, VN(i) represents the set of neighbor nodes of the i-th node, represents the normalized attention coefficients of the i-th node and the j-th node of the p-th attention head, W p represents the learnable weight matrix of the p-th attention head, Represents the feature matrix of the jth neighbor node's lth layer output.

6. The method for simulating urban land use change according to claim 5, characterized in that: The output data of the last layer of the graph neural network is converted into three-dimensional data and input into the gated recurrent unit.

7. The method for simulating urban land use change according to claim 6, characterized in that: The principle of the gated recurrent unit is expressed as: r t =σ(W r x t +U r h t-1 +b r ) Among them, r r represents the reset gate output at time step t; σ represents the sigmoid function; x t represents the input data at time step t; h t-1 represents the hidden state at time step t-1; represents the candidate hidden state at time step t, W, W r , U r ,U,b r , b are learnable parameters, where W, W r , U r , U represents the weight matrix, b r , b represents the bias term, h t represents the hidden state at time step t, z t represents the update gate output at time step t, ⊙ represents the Hadamard product; Among them, when t=0, x0 is the three-dimensional data converted from the output data of the last layer of the graph neural network.

8. The method for simulating urban land use change according to claim 7, characterized in that: Based on the final hidden state of the gated recurrent unit, the conversion probability of the plot is calculated using a vector cellular automaton as: in, represents the constraint condition for the kth land use change of the ith plot at time step t. If the plot is allowed to be developed, it is assigned a value of 1, otherwise it is assigned a value of 0; The random factor representing the kth land use change of the ith plot at time step t represents the unknown disturbance, RA = 1 + (-lnγ) α , where γ is set to a random value between 0 and 1, and α is a parameter between 1 and 10; It represents the final conversion probability of the kth land use change of the i-th plot at the t+1 time step; It represents the conversion suitability of the kth land use change of the ith plot at time step t; h t It is obtained through a fully connected layer and SoftMax.

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

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 8.

Images