Simulation method and system for land utilization change

Abstract

The invention discloses a land utilization change simulation method and system, and relates to the field of spatio-temporal data modeling and land planning application, and the method comprises the steps: taking a grid unit of a to-be-tested research region as a cell through employing a cellular automaton (CA), inputting the neighborhood land utilization condition, the land utilization type and the land development suitability of each cell into the trained convolutional neural network CNN to obtain the transition probability of each cell; and taking the transition probability, the constraint condition and the random factor as a conversion rule of a cellular automaton (CA), obtaining the development probability of each cell through the conversion rule, and determining the land utilization type of each cell after change through the development probability. The method can simulate and predict land utilization and urban development trends.

Description

Technical Field

[0001] The present invention relates to the field of spatiotemporal data modeling and land planning applications, and in particular to a method and system for simulating land use changes. Background Art

[0002] Land use or land cover change is a basic process and important result that reflects the interaction between human activities and the land resources on which they depend for survival. Using simulation models to study the evolutionary process and pattern of land use or land cover change not only helps to reveal the mechanism of human-land interaction and alleviate the contradiction between man and land, but also provides a scientific decision-making basis for regional land use and urban planning, and plays an important role in promoting regional sustainable development.

[0003] In the prior art, the logistic regression method in the cellular automaton model is used to define land conversion rules, which reflect whether the land state will change, and then simulate the change of land use. However, the disadvantage of the above method is that it cannot determine which land use type the land will become after the change. Summary of the invention

[0004] The embodiments of the present invention provide a method and system for simulating land use change, which can solve the problem in the prior art that it is impossible to determine which land use type the land will become after a change occurs.

[0005] The embodiment of the present invention provides a method for simulating land use change, comprising the following steps: obtaining land-related data in a study area, the land-related data comprising: neighborhood land use conditions, land use types and land development suitability; the neighborhood land use conditions are used to indicate the proportion of different land types in a determined range to all land in the range, and the land development suitability is used to indicate the result of land use change being jointly affected by multiple spatial variables; training a convolutional neural network model CNN output transition probability according to the land-related data to obtain a transition probability model, wherein the transition probability indicates the probability of a current land use type being transformed into a different land use type; using a cellular automaton CA to regard a grid unit of a study area to be tested as a cell, inputting the land-related data of each cell into the transition probability model to obtain the transition probability of each cell; using the transition probability, the constraint conditions for indicating the influence on the land use change and the random factors for increasing the randomness as the conversion rules of the cellular automaton CA; predicting the development probability of each cell according to the conversion rule, and determining the land use type of each cell after the change through the development probability.

[0006] Furthermore, the convolutional neural network model CNN is trained according to the land-related data to output the transition probability, and the specific steps include: The land-related data including neighborhood land use conditions, land use types, and land development suitability are input into the convolutional layer of CNN for feature extraction to obtain feature data; Input the feature data into the pooling layer to change the spatial dimension of the feature, and map the feature data after the spatial dimension change to the corresponding transition probability through the fully connected layer; The transition probability of land use type is output through the classifier layer softmax.

[0007] Furthermore, the method of determining the land use type of each cell after the change through the development probability specifically includes the following steps: Predict the development probability of each cell according to the transformation rules P ij , the formula is: ; in, CP ij Represents a cell ( i,j ) of land use type transition probability, Z ij Represents a cell ( i,j ) constraints, R ij Represents a cell ( i,j ) of random factors; A probability threshold is set to determine whether the current cell is transferred. When the development probability is greater than the probability threshold, the current cell is determined to be converted into the main land use type.

[0008] Furthermore, the constraint condition indicates that the land use type of water areas, ecological protection areas, and high-quality farmland protection areas does not change; The constraints Z ij The formula for obtaining is: ; in, c ij represents the state of the cell, con (·) is a function if the cell ( i,j ) is within the restricted area, it returns 0, otherwise it returns 1.

[0009] Furthermore, the random factor introduces random variables into the cellular automaton, making the cellular automaton random; The random factor R ij The formula for obtaining is: ; in, ais a random number from 0 to 1, and u is a parameter that controls the intensity of random perturbations, with a value in the range of 1-10.

[0010] Furthermore, the convolutional neural network model CNN is composed of 3 convolutional layers, 2 pooling layers, 1 fully connected layer, 1 classifier layer softmax and activation function ReLU, and the first convolutional layer is composed of 8 convolution kernels, the second pooling layer is composed of 3×3 pooling layers, the third convolutional layer is composed of 16 5×5 convolution kernels, the fourth pooling layer is composed of 3×3 pooling layers, the fifth convolutional layer is composed of 16 5×5 convolution kernels, and the sixth fully connected layer is composed of 48 neurons.

[0011] Furthermore, the obtaining of land-related data within the study area specifically comprises the following steps: obtaining neighborhood land use conditions and land use types through an initial land use map, wherein the initial land use map is used to display the land use situation at a fixed time point within the study area; based on the initial land use map, using a logistic regression method to analyze the ground slope, distance to the urban area, and spatial distribution of GDP to obtain land development suitability.

[0012] An embodiment of the present invention provides a land use change simulation system, including: A condition acquisition module is used to acquire land-related data in the study area, wherein the land-related data includes: neighborhood land use conditions, land use types and land development suitability; The transition probability acquisition module is used to train the convolutional neural network model CNN according to the land-related data to output the transition probability, and obtain the transition probability model, wherein the transition probability represents the probability of the current land use type changing to a different land use type; a grid unit of a research area to be tested is regarded as a cell by using a cellular automaton CA, and the land-related data of each cell is input into the transition probability model to obtain the transition probability of each cell; The land use simulation module is used to use the transfer probability, the constraints that affect land use changes, and the random factors that increase randomness as the transformation rules of the cellular automaton CA; the development probability of each cell is predicted according to the transformation rules, and the land use type of each cell after the change is determined by the development probability.

[0013] The embodiment of the present invention provides a method and system for simulating land use change. Compared with the prior art, the beneficial effects thereof are as follows: Using cellular automata CA, a grid unit of the research area to be tested is regarded as a cell, and the land-related data of each cell is input into the transition probability model to obtain the transition probability of each cell. The transition probability represents the probability of the current land use type changing to a different land use type. The transfer probability, constraints and random factors are used as the conversion rules of the cellular automaton CA; the development probability of each cell is predicted according to the conversion rules, and finally the land use type of each cell after the change can be determined by the development probability, and the land use type of the study area to be tested after the land use type changes can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A model flow chart provided for an embodiment of the present invention; Figure 2 A convolutional neural network model framework provided for an embodiment of the present invention; Figure 3 Classified land use maps of 2005 and 2015 provided in an embodiment of the present invention, wherein (a) is a classified land use map of 2005, and (b) is a classified land use map of 2015; Figure 4 A comparison diagram of simulation results in 2015 generated by different models in the study area provided by an embodiment of the present invention, wherein (a) is an actual land use map in 2015, (b) is a simulation result diagram of the LR-CA model, and (c) is a simulation result diagram of the LR-CNN-CA model. DETAILED DESCRIPTION

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0016] The present invention proposes a cellular automaton CA (combining logistic regression LR (Logistic Regression) and convolutional neural network CNN) CellularAutomata ) model, and named the model as LR-CNN-CA model to simulate land use change. The model framework is as follows Figure 1As shown. In the LR-CNN-CA model framework, the initial land use map is used to obtain the neighborhood land use conditions and the land use type of the current cell, and the logistic regression method is used to analyze spatial variables such as ground slope, distance to the urban area, and spatial distribution of GDP (Gross Domestic Product) to obtain land development suitability. Then, the neighborhood land use conditions, the land use type of the current cell, and the land development suitability are connected in series as the input of CNN. CNN is used to capture the deep characteristics of the land unit, accurately obtain the transition probability of each cell changing state, determine which land use type the cell that has changed state will be transformed into at the next moment, and perform the final land use type classification.

[0017] See also Figure 1 The embodiment of the present invention provides a method for simulating land use change, comprising the following steps: Step 1: Obtain land-related data in the study area, which include: neighborhood land use conditions, land use types and land development suitability; neighborhood land use conditions are used to indicate the proportion of different land types within a certain range to all land within the range, and land development suitability is used to indicate the result of land use changes being affected by multiple spatial variables.

[0018] Step 2: Train the convolutional neural network model CNN according to the land-related data to output the transition probability and obtain the transition probability model. The transition probability represents the probability of the current land use type changing to a different land use type. Use the cellular automaton CA (Cellular Automaton) to regard a grid unit of the study area to be tested as a cell, and input the land-related data of each cell into the transition probability model to obtain the transition probability of each cell.

[0019] Step 3: Use the transfer probability, the constraints that affect land use changes, and the random factors that increase randomness as the transformation rules of the cellular automaton CA; predict the development probability of each cell based on the transformation rules, and determine the land use type of each cell after the change through the development probability.

[0020] The detailed content of the invention is as follows: 1. Development Probability In cellular automata, the transformation rules for land use change simulation usually consist of four parts: land development suitability, neighborhood land use conditions, constraints and random factors.

[0021] The development probability of cells in cellular automata can be expressed as follows: (1) in,( i,j ) represents the row and column number of the cell,P ij Represents a cell ( i,j ) development probability, S ij is a cell ( i, j ) land development suitability, N ij is a cell ( i,j )’s neighborhood land use conditions, Z ij is a cell ( i,j ) constraints, R ij is a cell ( i,j ) is a random variable.

[0022] 1. Land development suitability specifically includes: Land use change is affected by many spatial variables, such as distance from the city center, distance from the highway, terrain elevation, slope, population and GDP. Land development suitability is calculated by a function composed of a series of independent influencing factors. This paper uses the logistic regression method to analyze spatial variables to obtain these influencing factors. It can be summarized by the following formula: (2) in S ij For cells ( i,j ) land development suitability, a is a constant, b k are the coefficients of the logistic regression model, x k is a set of spatial variables.

[0023] 2. Neighborhood land use conditions include: Neighborhood land use conditions refer to the land use pattern within a certain range, which is usually determined by the number and distribution of cells of various land use types within the neighborhood. n x n In the neighborhood of , the more cells of a certain land use type there are, the greater the possibility that the state of the central cell will be converted to this land use type. This can be summarized by the following formula: (3) in N ij is a cell ( i,j )’s neighborhood land use conditions, c ij Represents the state of the cell. The neighborhood window size used in this paper is 5x5.

[0024] 3. The specific constraints include: Generally speaking, land use changes also need to objectively consider external constraints. For example, land use types such as water areas, ecological protection areas, and high-quality farmland protection areas generally do not change. The formula is as follows: (4) in, con (·) is a function if the cell ( i,j ) is within the restricted area, it returns 0, otherwise it returns 1.

[0025] 4. Random factors include: The random factor is a random variable. Introducing the random variable into the cellular automation model makes the model uncertain and random, which can make the simulation results closer to the actual land use changes. It can be expressed by the following formula: (5) in R ij is a cell ( i,j ), a It is a random number from 0 to 1, and u is a parameter that controls the intensity of random disturbance. Its value ranges from 1 to 10. The larger the value, the stronger the random disturbance and the greater the calculated probability value. This paper uses the value of u as 1.

[0026] 2. Convolutional Neural Networks The convolutional neural network is further developed to calculate the transition probability required by the underlying cellular automation CA model, which is used to determine which land use type the land unit with a state change will transform into at the next moment, determine the land use type to which the land unit with a state change will transform, and perform the final land use type classification in order to carry out the next step of land use change simulation. The structure of the CNN model consists of seven parts, including 3 convolutional layers, 2 pooling layers, 1 fully connected layer, and 1 softmax classifier layer (usually located in the last few layers of the network, used to convert the original output of the neural network into a probability distribution). The activation function used by the convolutional layer and the fully connected layer is ReLU (The Rectified Linear Units). CNN uses a 5x5 convolution kernel and a 3x3 pooling window. The first convolution layer consists of 8 convolution kernels, the second layer consists of a 3x3 pooling layer, the third layer consists of 16 5x5 convolution kernels, the fourth layer is also a 3x3 pooling layer, and the fifth layer is also composed of 16 convolution kernels. The next layer is a fully connected layer with 48 neurons. Finally, the softmax layer is used to calculate the transition probability of each land use type of each cell, determine which land use type the cell will be converted to at the next moment, and determine the land use type that the cell will be converted to. The structure of the CNN model is as follows: Figure 2As shown. The variables contained in the CNN input vector represent the neighborhood land use conditions, the land use type of the current cell, and the suitability of land development. The input sequence is shown in Table 1. The entire model is trained using supervised learning. Cross entropy is used as the loss function. Specific behaviors are learned by minimizing the loss of the target mode. Reconstruction loss is used to drive the goal. Specifically, the error between the simulation result and the target mode is reestablished. The error is reduced by back propagation, and the error is minimized in a convergence state, so that the simulation result is closer to the target mode.

[0027] Table 1 Description of input sequences.

[0028]

[0029] Will N ij and obtained by logistic regression method S ij Combined with the transition probability derived by CNN CP ij , which can represent the suitability of land development and the land use conditions of the neighborhood at the same time. Taking into account various influences, the development probability of each cell can be described as: (6) In equation (6), the constraints are determined by policies and the natural characteristics of land use, and the random variables are set randomly.

[0030] (7) In each cycle, land use changes often account for only a small proportion, so a threshold is introduced to control the scale of change. The conversion threshold is used to compare with the calculated development probability value to determine whether the conversion can be carried out. The value is in the range of 0-1. The larger the value, the more difficult it is to convert. The fewer cells are converted in each cycle, and the longer the iteration takes. In the experiment, multiple iterative simulations are needed to adjust the threshold to find the most appropriate threshold.

[0031] An embodiment of the present invention provides a land use change simulation system, including: The condition acquisition module is used to obtain land-related data in the study area. The land-related data include: neighborhood land use conditions, land use types and land development suitability. The transition probability acquisition module is used to train the convolutional neural network model CNN based on the land-related data to output the transition probability and obtain the transition probability model. The transition probability represents the probability of the current land use type changing to a different land use type. The cellular automaton CA is used to treat a grid unit of the study area to be tested as a cell, and the land-related data of each cell is input into the transition probability model to obtain the transition probability of each cell. The land use simulation module is used to use the transition probability, the constraints that represent the changes in land use, and the random factors that increase randomness as the conversion rules of the cellular automaton CA; the development probability of each cell is predicted according to the conversion rules, and the land use type of each cell after the change is determined by the development probability.

[0032] A specific embodiment is as follows: The research area is a city in a certain province. The land use raster data of the city in 2005 and 2015 and various spatial variable data are from the resource and environment data cloud platform of a certain academy of sciences, with a resolution of 90 meters. Figure 3 The land use map of a city in 2005 and 2015 is shown. Land use change is jointly affected by many spatial variables, especially land use change in urban areas, which is not only affected by natural factors such as soil and environment, but also closely related to human activities, socio-economic development and government decision-making to a large extent. Therefore, considering many factors, in this study, information such as traffic conditions, economic conditions, hydrological conditions, and topography are selected as spatial variables for simulating land use change in a city. Specifically, it includes (a) distance to urban area, (b) distance to construction land, (c) distance to rural settlements, (d) distance to national highways, (e) distance to provincial highways, (f) distance to county roads, (g) distance to expressways, (h) distance to railways, (i) distance to waters, (j) spatial distribution of GDP, (k) spatial distribution of population, and (l) ground slope. The description of spatial variables is shown in Table 2.

[0033] Table 2 Description of spatial variables

[0034] Experimental results and discussion: In order to quantitatively evaluate the land use change simulation results, the total accuracy, Kappa Coefficient, producer accuracy, user accuracy and quality factor FOM (Figure of Merit) are used to evaluate the accuracy of the model. These formulas are as follows: (8) in, TP represents a true positive, FP Indicates a false positive. TN represents a true negative, FN Indicates a false negative. Overallaccuracy Represents the overall accuracy.

[0035] (9) Kappa The coefficient is used for consistency testing, focusing on the consistency of spatial position, and is often used to measure the classification accuracy of remote sensing data. p 0 represents the observation consistency rate, p e K represents the expected consistency rate. appa The index value is generally in the range of [0,1], and is usually fuzzily classified into five levels: [0.0,0.2] is extremely low similarity, indicating that the current map and the simulation map have extremely low consistency and the simulation effect is extremely poor; (0.2,0.4] is relatively low similarity, indicating that the current map and the simulation map have low consistency and the simulation effect is poor; (0.4,0.6] is average similarity, indicating that the current map and the simulation map have average consistency and the simulation effect is average; (0.6,0.8] is relatively high similarity, indicating that the current map and the simulation map have high consistency and the simulation effect is good; (0.8,0.1] is extremely high similarity, indicating that the current map and the simulation map have extremely high consistency and the simulation effect is excellent.

[0036] (10) (11) (12) FOM This method only evaluates the cells that have changed, which can better identify the accuracy of the simulation, especially for large-scale regional simulations where the amount of land use change accounts for a small proportion of the total area of ​​the study area, and can more objectively and truly evaluate the accuracy of the simulation. PA represents the producer accuracy, UA Indicates the user accuracy, A Indicates the area where changes are observed, which does not change during simulation. B Indicates the area where a change is observed, which changes to the correct category during simulation. C Indicates the region where a change was observed, which changed to the wrong category during simulation. D Indicates an area where no change is observed, but changes occur during simulation.

[0037] For the discussion of the experimental results, during the simulation process, the model receives the network parameters obtained through training and uses them as the conversion rules for model simulation to simulate the land use changes in the study area. During the simulation process, the land use status observation data and related spatial variable data in 2005 were used to simulate the land use in 2015. After the simulation, the obtained land use simulation data and observation data in 2015 were compared and analyzed to verify the simulation ability of the improved model.

[0038] The constructed LR-CA model and the LR-CNN-CA model proposed in this paper are used to simulate the land use changes of a city during 2005-2015, and the simulation results of the models are compared. The experimental program in this paper is implemented in Python language and based on the Pytorch framework. In order to evaluate the effect of the cellular automaton model combining LR and CNN in the dynamic simulation of land use change, we compared the LR-CNN-CA model and the LR-CA model to verify that compared with the LR-CA model, the LR-CNN-CA model is used in the simulation of land use change. CNN is used to capture the deep spatial characteristics of the land unit, reflect the complex relationship between spatial variables, and perform the final land use type classification. It is determined which land use type the land unit will be transformed into at the next moment, and the land use type to which the land unit will be transformed can be determined. This can solve the shortcomings of the LR-CA model in the simulation of land use change, improve the simulation accuracy of land use change, and reflect the real changes in urban land use forms. Table 3 lists the performance evaluation results of different models.

[0039] Table 3 Evaluation indicators of different models.

[0040]

[0041] The differences between the simulated land use maps of 2015 obtained by different models and the actual land use situation were compared, and the actual simulation accuracy of different models was analyzed and compared. The results show that the LR-CNN-CA model we proposed has a high simulation ability for the land use changes in a certain city during 2005-2015. The performance of the model is much better than the LR-CA model, with an overall accuracy of 0.93597, producer accuracy and user accuracy of 0.55449 and 0.61000, which can well simulate the land use changes in a certain city and significantly improve the simulation accuracy of land use changes. By superimposing the real observation map and simulation result map of 2015, the change confusion matrix generated by the number of matches of each land use type in spatial position is counted point by point, and its Kappa index value is calculated. The K between the simulation results of this model and the real map appaThe index value is 0.84561, indicating that it has a very high consistency in spatial distribution, reflecting that the model has good simulation effect and credibility. FOM The higher the precision value, the higher the consistency between the simulation and observation. FOM The precision value is 0.40937, which shows that the land use change area simulated by the model has a high spatial consistency with the change area obtained by observation, which reflects that the simulation effect of the model is good and credible.

[0042] In order to evaluate the effect of the cellular automaton model combining LR and CNN in the dynamic simulation of land use change, we compared the LR-CNN-CA model and the LR-CA model. The LR-CA model uses the transformation rules defined by logistic regression to simulate complex land use changes. Compared with the LR-CA model, the LR-CNN-CA model adds a CNN component on the basis of the LR-CA model to capture the deep characteristics of the land unit, reflect the complex relationship between spatial variables, perform the final land use type classification, and determine which land use type the land unit with a changed state will be transformed into at the next moment, which is used to optimize the LR-CA model and simulate land use changes. Compared with the LR-CA model, the overall accuracy of the LR-CNN-CA model increased from 0.92256 to 0.93597, an increase of about 1.45%, and K appaThe index value increased from 0.81329 to 0.84561, an increase of about 3.97%, and the FOM accuracy value increased from 0.32729 to 0.40937, an increase of about 25.08%. This means that the LR-CNN-CA model proposed in this study has better simulation results than the LR-CNN model, which verifies the use of the LR-CNN-CA model in land use change simulation, using CNN to capture the deep spatial characteristics of the land unit, reflect the complex relationship between spatial variables, and perform the final land use type classification, determine which land use type the land unit will be transformed into at the next moment, and determine the land use type to which the land unit will be transformed. This can solve the shortcomings of the LR-CA model in land use change simulation, improve the simulation accuracy of land use change, and reflect the real changes in urban land use morphology. The contribution of the CNN component in land use change simulation is proved, indicating that adding the CNN component to simulate land use change on the basis of the LR-CNN model helps to improve the performance of the model and can effectively reflect the complex relationship between spatial variables. The effect of simulating land use changes by using the cellular automaton model combining LR and CNN was verified, which strongly proved the effectiveness of the cellular automaton model combining LR and CNN proposed by us in simulating land use changes. It shows that the simulation accuracy can be improved by using the cellular automaton model combining LR and CNN to simulate land use changes. Compared with the LR-CA model, the simulation results are closer to the actual land use situation.

[0043] We randomly select a part of the study area to illustrate the difference in simulation results between the LR-CNN model and the cellular automaton model combining LR and CNN, such as Figure 4 The actual land use in 2015 and the land use simulated by different CA models are shown. Figure 4It can be seen that the model we proposed has excellent ability in land use change simulation. It can well simulate the spatial change trend of land use in a certain city from 2005 to 2015, better match the actual land use map of a certain city in 2015, and the simulation results are closer to the actual land use change situation. The land use results of a certain city in 2015 simulated by the LR-CNN-CA model proposed by us are close to the actual observation results in spatial distribution, and the performance is very good compared with other models. The simulation value of the model can well simulate the spatial distribution characteristics of urban land, other construction land, and water areas in a certain city. It can be determined that the neuron cellular automaton model combining LR and CNN is used in the simulation of land use change, and CNN is used to capture the deep spatial characteristics of the land unit, reflect the complex relationship between spatial variables, and perform the final land use type classification, judge which land use type the state of the land unit will be transformed into at the next moment, and determine the land use type to which the land unit will be transformed. This is in line with the actual situation of land use change, and is conducive to improving the accuracy of land use change simulation, and can also solve the shortcomings of the LR-CA model in land use change simulation.

[0044] This paper proposes a new model that uses a cellular automaton model that combines LR and CNN to accurately simulate land use changes, which can achieve higher simulation accuracy and conform to the actual situation of land use changes. The model we proposed integrates CNN into the LR-CA model to simulate complex land use changes. It uses CNN's excellent feature extraction and powerful feature representation capabilities to capture the deep characteristics of land units, reflect the complex relationship between spatial variables, determine which land use type the land unit will be transformed into at the next moment, determine the land use type to which the land unit will be transformed, and perform the final land use type classification, solving the shortcomings of the LR-CA model in land use change simulation. A city in China was selected as a research case to verify the performance of the LR-CNN-CA model we proposed. The experimental results show that the LR-CNN-CA model significantly improves the simulation accuracy of land use changes, has a high simulation capability for land use changes in a city from 2005 to 2015, and can well simulate the spatial change trend of land use in a city from 2005 to 2015, with an overall accuracy of 0.93597 and K appa The precision value is 0.84561, FOM The index is 0.40937, and the simulation accuracy is improved by 1.45% compared with the LR-CA model. This verifies the effect of simulating land use changes by using the cellular automaton model combining LR and CNN, which can effectively reflect the real changes in urban land use patterns. It proves that the use of the cellular automaton model combining LR and CNN in land use change simulation is conducive to improving the simulation accuracy of land use changes.

[0045] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for simulating land use change, characterized in that: The following steps are involved: Obtaining land-related data in the study area, the land-related data includes: neighborhood land use conditions, land use types and land development suitability; the neighborhood land use conditions are used to indicate the proportion of different land types within a certain range to all land within the range, and the land development suitability is used to indicate the result of land use change being jointly affected by multiple spatial variables; The convolutional neural network model CNN is trained according to the land-related data to output the transition probability, thereby obtaining a transition probability model, wherein the transition probability represents the probability of the current land use type changing to a different land use type; Using cellular automata CA, a grid unit of the research area to be measured is regarded as a cell, and the land-related data of each cell is input into the transition probability model to obtain the transition probability of each cell; The transition probability, the constraints that affect land use change, and the random factors that increase randomness are used as the transformation rules of the cellular automaton CA. The development probability of each cell is predicted according to the transformation rules, and the land use type of each cell after the change is determined by the development probability.

2. A method for simulating land use change according to claim 1, characterized in that: The convolutional neural network model CNN is trained according to the land-related data to output the transfer probability, and the specific steps include: The land-related data including neighborhood land use conditions, land use types, and land development suitability are input into the convolutional layer of CNN for feature extraction to obtain feature data; Input the feature data into the pooling layer to change the spatial dimension of the feature, and map the feature data after the spatial dimension change to the corresponding transition probability through the fully connected layer; The transition probability of land use type is output through the classifier layer softmax.

3. A method for simulating land use change according to claim 2, characterized in that: The specific steps of determining the land use type after each cell changes by development probability include: Predict the development probability of each cell according to the transformation rules P ij , the formula is: ； in, CP ij Represents a cell ( i,j ) of land use type transition probability, Z ij Represents a cell ( i,j ) constraints, R ij Represents a cell ( i,j ) of random factors; A probability threshold is set to determine whether the current cell is transferred. When the development probability is greater than the probability threshold, the current cell is determined to be converted into the main land use type.

4. A method for simulating land use change according to claim 1, characterized in that: The constraint condition indicates that the land use type of water areas, ecological protection areas, and high-quality farmland protection areas shall not change; The constraints Z ij The formula for obtaining is: ； in, c ij represents the state of the cell, con (·) is a function if the cell ( i,j ) is within the restricted area, it returns 0, otherwise it returns 1.

5. A method for simulating land use change according to claim 1, characterized in that: The random factor introduces random variables into the cellular automaton, making the cellular automaton random; The random factor R ij The formula for obtaining is: ； in, a is a random number from 0 to 1, and u is a parameter that controls the intensity of random perturbations, with a value in the range of 1-10.

6. A method for simulating land use change according to claim 1, characterized in that: The convolutional neural network model CNN is composed of 3 convolutional layers, 2 pooling layers, 1 fully connected layer, 1 classifier layer softmax and activation function ReLU, and the first convolutional layer is composed of 8 convolution kernels, the second pooling layer is composed of 3×3 pooling layers, the third convolutional layer is composed of 16 5×5 convolution kernels, the fourth pooling layer is composed of 3×3 pooling layers, the fifth convolutional layer is composed of 16 5×5 convolution kernels, and the sixth fully connected layer is composed of 48 neurons.

7. A method for simulating land use change according to claim 1, characterized in that: The specific steps of obtaining land-related data in the study area include: Obtaining neighborhood land use conditions and land use types through an initial land use map, wherein the initial land use map is used to display the land use situation at a fixed time point in the study area; Based on the initial land use map, the spatial distribution of ground slope, distance to urban area and GDP was analyzed using logistic regression method to obtain the land development suitability.

8. A land use change simulation system, characterized in that: include: A condition acquisition module is used to acquire land-related data in the study area, wherein the land-related data includes: neighborhood land use conditions, land use types and land development suitability; The transition probability acquisition module is used to train the convolutional neural network model CNN according to the land-related data to output the transition probability, and obtain the transition probability model, wherein the transition probability represents the probability of the current land use type changing to a different land use type; a grid unit of a research area to be tested is regarded as a cell by using a cellular automaton CA, and the land-related data of each cell is input into the transition probability model to obtain the transition probability of each cell; The land use simulation module is used to use the transfer probability, the constraints that affect land use changes, and the random factors that increase randomness as the transformation rules of the cellular automaton CA; the development probability of each cell is predicted according to the transformation rules, and the land use type of each cell after the change is determined by the development probability.

Images