A method for predicting land use change in mountainous cities based on grey model and deep learning

Abstract

This invention discloses a method for predicting land use change in mountainous cities based on grey models and deep learning. The method includes the following steps: S1, collecting data and constructing a dataset; S2, data preprocessing; S3, constructing a state transition matrix; S4, establishing a grey system model; S5, simulating and predicting land use type change trends using a Markov model; S6, training and evaluating the suitability probability of land use types using an ANN module in deep learning; S7, simulating the interrelationships between different land use types; S8, constructing a land use change prediction model based on land use transition changes and driving factors; and S9, using the trained model to predict future land use changes under multiple scenarios. This invention achieves efficient land use prediction by integrating multi-source data, introducing an adaptive mechanism, combining decision support tools, and providing an ecological and environmental impact assessment.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method for predicting land use change in mountainous cities based on grey models and deep learning. Background Technology

[0002] With the acceleration of urbanization, land use change has had a profound impact on the ecological environment, economic development, and social life. Compared with plain cities, land use patterns in mountainous cities are particularly constrained by topography, facing challenges such as complex terrain, low land use efficiency, and insufficient land use control in small-scale areas on the city's periphery. Therefore, the rational use of land resources is crucial. Predicting land use change can help assess the potential environmental impacts of different projects and developments. By simulating future land use patterns, we can better understand potential environmental risks and challenges. However, under different development scenarios, land use in mountainous cities will exhibit different trends. Therefore, conducting multi-scenario predictions of land use in mountainous cities is helpful in understanding and mastering their changing trends. Land use simulation and prediction models mainly include CA (Cellular Automata) models, FLUS (Future Land Use Simulation) models, Markov models, and CLUE-S (Conversion of Land Use) models. Using a single model to predict future land use change often has significant limitations, cannot adapt to complex and changing geographical environments, and results in poor prediction performance. Summary of the Invention

[0003] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0004] In view of the problems existing in the above and / or existing land use change prediction technologies for mountainous cities, this invention is proposed.

[0005] Therefore, the purpose of this invention is to provide a land use change prediction method for mountainous cities based on grey models and deep learning. By combining the spatial analysis advantages of the FLUS model and the temporal analysis advantages of the grey Markov model, a new prediction model is established to reduce the difficulty of formulating land conversion rules and reduce excessive interference from human factors.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for predicting land use change in mountainous cities based on grey models and deep learning, comprising the following steps:

[0007] S1. Collect data and build a dataset;

[0008] S2, Data Preprocessing;

[0009] S3. Construct the state transition matrix;

[0010] S4. Establish a grey system model;

[0011] S5. Combine Markov model to simulate and predict the changing trends of land use types;

[0012] S6. Using the ANN module of deep learning, train and evaluate the suitability probability of land use types, and use the CA module to simulate the relationship between various land use types.

[0013] S7. Based on land use transfer and change and the driving factors of land use change, construct a land use change prediction model;

[0014] S8. Use the trained model to make multi-scenario predictions of future land use changes.

[0015] As a preferred embodiment of the mountain city land use change prediction method based on grey model and deep learning in this invention, in step S1, the collected data includes land use image data of the prediction area and land use change driving factor data in four aspects: natural environment, socio-economic, accessibility and policy restrictions.

[0016] As a preferred embodiment of the mountain city land use change prediction method based on grey model and deep learning in this invention, step S2 specifically includes the following steps:

[0017] S201. Determine the classification standards for land use types based on the attributes of the data source, land classification standards, the actual situation of the prediction area, and the requirements of the prediction objectives;

[0018] S202. Unify the data coordinate system of the raster data and transform it into the same projected coordinate system. Trim the data range from overlapping with the study area so that the number of raster cells in all raster data is equal.

[0019] S203. Determine the resolution of the raster data based on the size of the prediction area, the model operation time, and the calculation accuracy.

[0020] As a preferred embodiment of the land use change prediction method for mountainous cities based on grey models and deep learning in this invention, step S3 specifically includes the following steps:

[0021] S301. Using the Markov model, calculate the transition probability matrix between different land use types based on historical data. The expression is as follows:

[0022] (1);

[0023] (2);

[0024] Where, p ij (m) represents the probability of transitioning from land class i to land class j after m time units, p ij p is the probability of changing from type i to type j from the initial to the final stage, 0 ≤ p ij ≤1, S ij This represents the area that changes from type i to type j during the study period. Let represent the total area of ​​land use type i, and n be the number of land use data categories.

[0025] As a preferred embodiment of the mountain city land use change prediction method based on grey model and deep learning in this invention, step S4 specifically includes the following steps:

[0026] S401. Input data and construct a 7×5 matrix X, with units of km. 2 Its expression is,

[0027] (3);

[0028] 7 represents 7 types of land, 5 represents 5 years, and 5 years constitute 1 year. (i) (t) represents the area of ​​the i-th type of land in year t, where i=1,…,7 correspond to paddy fields, dry land, forest land, grassland, water area, construction land, and unused land, respectively.

[0029] S402. Data standardization to eliminate dimensional differences.

[0030] (4);

[0031] z (i) (t) represents the standardized area with a mean of 1;

[0032] S403, Accumulate and generate the sequence.

[0033] (5);

[0034] Among them, z (1,i) (t k ) represents the land of type i up to the year. The cumulative standardized area;

[0035] S404. Establish a grey differential equation to predict the area of ​​one type of land use, with the remaining six types as related variables;

[0036] (6);

[0037] Where a is the development gray number, reflecting its own decay rate; b i Z represents the association weight between the i-th type of land and changes in paddy fields. (1,1) z represents the standardized area of ​​the first land use type. (1,i) The area of ​​the i-th type of land after standardization;

[0038] S405. Use the least squares method to estimate the model parameters, parameter vector. ,

[0039] (7);

[0040] (8);

[0041] S406. Set up a sliding window. The initial window contains data from 1990-1995-2000, indicating that the land use status in 2005 is predicted using real data from 1995 and 2000. Slide to the next window: data from 1995-2000-2005, to predict the land use status in 2010, and so on. The time response function is obtained by accumulating the predicted values ​​of the generated sequence. ,

[0042]

[0043] in Let be the standardized area of ​​land of type i at time k.

[0044] As a preferred embodiment of the mountain city land use change prediction method based on grey model and deep learning in this invention, step S5 specifically includes the following steps:

[0045] S501. Area residual calculation and parameter update: Fit GM(1,N) to the data within the window, predict the next time point, and obtain the area residual.

[0046] (9);

[0047] e (i) (t k+1 ) represents the i-th type of land in t k+1 Annual area residual;

[0048] S502, Parameter Update

[0049] (10);

[0050] (11);

[0051] Where η is the learning rate, controlling the update magnitude; e (1) For the predicted area residual of the model for data points of land use type 1, sign(e (1) The direction (positive / negative) of the area residual determines whether the parameter is increased or decreased. (1) >0, sign(e (1) )=+1;e (1) =0, sign(e (1) )=0, e (1) <0, sign(e (1) )=-1; a new Here, 'a' represents the updated gray number of development, and 'b' represents the gray number of development before the update. i,new Let b be the updated association weight for the i-th land use type. i The association weight of the i-th land use type before the update;

[0052] S503. Based on autoregressive correction of area residuals, the historical area residuals of each land type are fitted.

[0053] (12);

[0054] These are the autoregressive coefficients, estimated using the least squares method; This represents the mean of the area residuals; For the next time step (t) of the i-th type of land in the time series k+1 Predicted values ​​of area residuals;

[0055] S504. Based on the current state and the transition probability matrix, correct the predicted values ​​of the original sequence. Its formula is,

[0056] (13).

[0057] S505. Set the transfer probability between different land types according to different development scenarios, and then calculate the number of pixels for future land use by combining the transfer probability matrix and the corrected prediction value.

[0058] As a preferred embodiment of the mountain city land use change prediction method based on grey model and deep learning in this invention, step S6 specifically involves:

[0059] S601. Select the driving factors affecting land use change, normalize the driving factor data using fuzzy membership, and obtain the suitability probability map of land use in the base year in the ANN module of the FLUS model. Its expression is:

[0060] (12);

[0061] (13);

[0062] Where i represents land use type; s represents hidden layer; r represents raster; and t represents time. For suitability probability; As weight, ent s (r,t) represents the feedback received by the grid r of the s-th hidden layer at training time t;

[0063] S602. Using the land use data for the forecast year and the constructed suitability probability atlas for the base year, the spatial pattern of land use for the forecast year is predicted in the CA module of the FLUS model. The expression is as follows: (14);

[0064] Among them, TP t r,k Let be the total probability of converting to land use type p in the t-th iteration. This represents the inertia coefficient of land use type k at time t. Indicates the cost of space type conversion. denoted by the number of grid cells generated by land use type i after the iteration, N is the Moore neighborhood in CA, and w is the variable neighborhood weight for each land use type;

[0065] S603. Select the initial year of land use, the land use change transition matrix, the land use transition suitability image set, and the prediction period to conduct land use change simulation. Compare the simulation results with actual land use data, and use the Kappa coefficient to evaluate the accuracy of the simulation results. The formula for calculating the Kappa coefficient is:

[0066] (15);

[0067] (16);

[0068] (17);

[0069] Where n is the total number of rasters; u1 is the number of rasters that are consistent with the simulation; U is the number of land types; and P0 represents the proportion of rasters that are consistent with the simulation.

[0070] S604. Perform error analysis on the simulation results, adjust the land use transfer rules according to the analysis results, re-run the model simulation and evaluate the simulation results until simulation results that meet the accuracy requirements are obtained.

[0071] S605. Analyze and predict the results, including changes in the quantity and spatial shifts of various land use types.

[0072] Compared with the prior art, the present invention has the following technical effects: the initial probability of the transition matrix is ​​calculated by the Markov model, and then the results of each land use type in the predicted year are corrected by the grey model. Then, the number of land use pixels in different future scenarios is calculated based on the corrected results and different land use transition probabilities. Finally, the corrected results are used as the input of the CA module, which improves the accuracy of the model prediction and makes the results more consistent with the actual land evolution process. Attached Figure Description

[0073] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0074] Figure 1 This is a schematic diagram of the process of the present invention.

[0075] Figure 2 This is a land use status image map for the present invention.

[0076] Figure 3 This is a comparison diagram of the actual land use simulation of the present invention.

[0077] Figure 4 This is a map showing the predicted future land use changes according to the present invention. Detailed Implementation

[0078] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0079] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0080] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0081] Example 1

[0082] Reference Figure 1 and Figure 4This invention provides a method for predicting land use change in mountainous cities based on grey models and deep learning, comprising the following steps:

[0083] S1. Collect data and build a dataset;

[0084] The collected data includes land use imagery data of the prediction area and data on four aspects of land use change drivers: natural environment, socio-economic factors, accessibility, and policy restrictions.

[0085] S2. Data preprocessing, specifically...

[0086] S201. Determine the classification standards for land use types based on the attributes of the data source, land classification standards, the actual situation of the prediction area, and the requirements of the prediction objectives;

[0087] S202. Unify the data coordinate system of the raster data and transform it into the same projected coordinate system. Trim the data range from overlapping with the study area so that the number of raster cells in all raster data is equal.

[0088] S203. Determine the resolution of the raster data based on the size of the prediction area, the model computation time, and the computational accuracy.

[0089] S3. Construct the state transition matrix, specifically as follows:

[0090] S301. Using the Markov model, calculate the transition probability matrix between different land use types based on historical data. The expression is as follows:

[0091] (1);

[0092] (2);

[0093] Where, p ij (m) represents the probability of transitioning from land class i to land class j after m time units, p ij p is the probability of changing from type i to type j from the initial to the final stage, 0 ≤ p ij ≤1, S ij This represents the area that changes from type i to type j during the study period. represents the total area of ​​land use type i, and n is the number of land use data categories;

[0094] S4. Establish a grey system model to predict the number of land pixels, specifically...

[0095] S401. Input data and construct a 7×5 matrix X, with units of km. 2 Its expression is,

[0096] (3);

[0097] 7 represents 7 types of land, 5 represents 5 years, and 5 years constitute 1 year. (i) (t) represents the area of ​​the i-th type of land in year t, where i=1,…,7 correspond to paddy fields, dry land, forest land, grassland, water area, construction land, and unused land, respectively.

[0098] S402. Data standardization to eliminate dimensional differences.

[0099] (4);

[0100] z (i) (t) represents the standardized area with a mean of 1;

[0101] S403, Accumulate and generate the sequence.

[0102] (5);

[0103] Among them, z (1,i) (t k ) represents the land of type i up to the year. The cumulative standardized area;

[0104] S404. Establish a grey differential equation to predict the area of ​​one type of land use, with the remaining six types as related variables;

[0105] (6);

[0106] Where a is the development gray number, reflecting its own decay rate; b i Z represents the association weight between the i-th type of land and changes in paddy fields. (1,1) z represents the standardized area of ​​the first land use type. (1,i) The area of ​​the i-th type of land after standardization;

[0107] S405. Use the least squares method to estimate the model parameters, parameter vector. ,

[0108] (7);

[0109] (8);

[0110] S406. Set up a sliding window. The initial window contains data from 1990-1995-2000, indicating that the land use status in 2005 is predicted using real data from 1995 and 2000. Slide to the next window: data from 1995-2000-2005, to predict the land use status in 2010, and so on. The time response function is obtained by accumulating the predicted values ​​of the generated sequence. ,

[0111]

[0112] in Let be the standardized area of ​​land of type i at time k.

[0113] S5. Using the transition probability matrix obtained from the Markov model and the results from the grey model, predict the future land use status, specifically as follows:

[0114] S501. Area residual calculation and parameter update: Fit GM(1,N) to the data within the window, predict the next time point, and obtain the area residual.

[0115] (9);

[0116] e (i) (t k+1 ) represents the i-th type of land in t k+1 Annual area residual;

[0117] S502, Parameter Update

[0118] (10);

[0119] (11);

[0120] Where η is the learning rate, controlling the update magnitude; e (1) For the predicted area residual of the model for data points of land use type 1, sign(e (1) The direction (positive / negative) of the area residual determines whether the parameter is increased or decreased. (1) >0, sign(e (1) )=+1;e (1) =0, sign(e (1) )=0, e (1) <0, sign(e (1) )=-1; a new Here, 'a' represents the updated gray number of development, and 'b' represents the gray number of development before the update. i,new Let b be the updated association weight for the i-th land use type. i The association weight of the i-th land use type before the update;

[0121] S503. Based on autoregressive correction of area residuals, the historical area residuals of each land type are fitted.

[0122] (12);

[0123] These are the autoregressive coefficients, estimated using the least squares method; This represents the mean of the area residuals; For the next time step (t) of the i-th type of land in the time series k+1 Predicted values ​​of area residuals;

[0124] S504. Based on the current state and the transition probability matrix, correct the predicted values ​​of the original sequence. Its formula is,

[0125] (13).

[0126] S505. Adjust the transition probability matrix according to the transition probabilities of different scenarios so that the probabilities of each land use type are added together to 1. Then, calculate the number of land use pixels under different scenarios based on the predicted value of each land use type after correction, and use this as the input of the CA module of the FLUS model.

[0127] S6. Using deep learning algorithms, based on land use transfer and change and the driving factors of land use change, a land use change prediction model is constructed based on the FLUS-Markov model to predict the land use situation under different future development scenarios. The relationship between various land use types is simulated through an adaptive inertial cell mechanism.

[0128] Specifically,

[0129] S601. Select the driving factors affecting land use change, normalize the driving factor data using fuzzy membership, and obtain the suitability probability map of land use in the base year in the ANN module of the FLUS model. Its expression is:

[0130] (12);

[0131] (13);

[0132] Where i represents land use type; s represents hidden layer; r represents raster; and t represents time. For suitability probability; w s,i As weight, ent s (r,t) represents the feedback received by the grid r of the s-th hidden layer at training time t;

[0133] S602. Using the land use data for the predicted year and the suitability probability atlas constructed for the base year, the spatial pattern of land use for the predicted year is predicted in the CA (Adaptive Inertial Cell Mechanism) module of the FLUS model. The expression is as follows:

[0134] (14);

[0135] Among them, TP t r,k Let r be the total probability that a grid r is converted to land use type k at time t. This represents the inertia coefficient of land use type k at time t. Indicates the cost of space type conversion. This represents the number of rasters generated after land use type i completes the iteration, where N is the Moore neighborhood in CA, and w i For land use type i, the variable weights of the area;

[0136] S603. Select the initial land use year, land use change transition matrix, land use transition suitability image set, number of pixels in the prediction year, and prediction period to conduct land use change simulation. Compare the simulation results with actual land use data and use the Kappa coefficient to evaluate the accuracy of the simulation results. The formula for calculating the Kappa coefficient is:

[0137] (15);

[0138] (16);

[0139] (17);

[0140] Where n is the total number of rasters; u1 is the number of rasters that are consistent with the simulation; U is the number of land types; and P0 represents the proportion of rasters that are consistent with the simulation.

[0141] S604. Perform error analysis on the simulation results, adjust the land use transfer rules according to the analysis results, re-run the model simulation and evaluate the simulation results until simulation results that meet the accuracy requirements are obtained.

[0142] S605. Analyze and predict the results, including changes in the quantity and spatial shifts of various land use types.

[0143] S7. Obtain model input parameters that meet the required accuracy through deep learning algorithms. Then, based on land use data, land use change drivers, the number of land use pixels in the predicted year, and the suitability atlas in the predicted year, construct a land use change prediction model based on the FLUS-Markov model (this is a conventional technique and not an improvement point of this application; its specific implementation steps are not described in detail in this application).

[0144] S8. Use the trained model to make multi-scenario predictions of future land use changes.

[0145] This invention calculates the initial probability of the transition matrix using a Markov model, then uses a grey model to correct the prediction results for each land use type in the predicted year. Next, based on the corrected results and different land use transition probabilities, it calculates the number of land use pixels for different future scenarios. Finally, the corrected results are used as input to the CA module, which improves the accuracy of the model prediction and makes the results more consistent with the actual land evolution process.

[0146] Example 2

[0147] like Figure 2 and Figure 3 This is the second embodiment of the present invention. This embodiment verifies the technical effect of using the present invention for land use change prediction through simulation experiments.

[0148] This embodiment uses land use change simulation and prediction in Chongqing as a practical application.

[0149] Regional Overview:

[0150] Chongqing Municipality is located in southwestern China, in the upper reaches of the Yangtze River. It covers a total area of ​​82,400 km², and currently administers 26 districts and 12 counties. Mountains and hills account for approximately 98% of its total area, making it a typical mountainous city. Chongqing is bordered by the Daba Mountains to the north, the Wu Mountains to the east, the Wuling Mountains to the southeast, and the Dalou Mountains to the south. The main urban area's elevation ranges from 168 to 400 meters, almost entirely built on mountains or hillsides. However, rapid economic development and urbanization have led to increasingly prominent problems with Chongqing's land ecological environment and land use security.

[0151] Five periods of land use imagery data (1km resolution) from 1995 to 2015 were collected in Chongqing. Spatial analysis tools were used for data preprocessing (cropping, correction, etc.). Land use classification standards were determined, and a land use transfer matrix was calculated. Factors influencing land use transfer and change were identified. This model considered factors such as mountain city DEM, slope, aspect, GDP, population density, nighttime light data, vegetation cover, distance from rivers, distance from railways, and distance from towns as driving factors. A land use transfer suitability atlas was created using an ANN model. A land use change prediction model was constructed based on the FLUS-Markov model as input, and land use simulation and prediction were conducted.

[0152] This embodiment classifies Chongqing's land use types into two categories: paddy fields and dry land, according to the secondary classification standards in Table 1. Other land use types are then classified into five major categories, totaling seven categories, according to the primary classification standards. Land use change simulation and prediction are then conducted based on this classification standard. A land use transition suitability image generated using the ANN model is shown below. Figure 2As shown, the land use change in 2015 was simulated based on the 2005-2015 Land Use Transfer Change Suitability Atlas. Figure 3 The Kappa coefficient of the model, compared with actual measurements, reached 0.9092, which is higher than the accuracy of traditional models such as CA (Cellular Automata), FLUS (Future Land Use Simulation), Markov, and CLUE-S (Conversion of Land Use) (a Kappa coefficient of 0.75 is generally considered to indicate a good simulation effect). This demonstrates that the land use change prediction model constructed based on this method has high accuracy. Furthermore, based on the 2015 land use pattern, it predicts the land use change pattern of Chongqing under different development scenarios in 2025. Figure 4 As shown.

[0153] Table 1 Land Type Classification Standards

[0154]

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting land use change in mountainous cities based on grey model and deep learning, characterized in that: Includes the following steps: S1. Collect data and build a dataset; S2, Data Preprocessing; S3. Construct the state transition matrix; S4. Establish a grey system model; S5. Combine Markov model to simulate and predict the changing trends of land use types; S6. Using the ANN module of deep learning, train and evaluate the suitability probability of land use types, and use the CA module to simulate the relationship between various land use types. S7. Based on land use transfer and change and the driving factors of land use change, construct a land use change prediction model; S8. Use the trained model to predict future land use changes under multiple scenarios; Step S4, in detail Includes the following steps, S401. Input data and construct a 7×5 matrix X, with units of km. 2 Its expression is, （3）； 7 represents 7 types of land, 5 represents 5 years, and 5 years constitute 1 year. (i) (t) represents the area of ​​the i-th type of land in year t, where i=1,…,7 correspond to paddy fields, dry land, forest land, grassland, water area, construction land, and unused land, respectively. S402. Data standardization to eliminate dimensional differences. （4）； z (i) (t) represents the standardized area with a mean of 1; S403, Accumulate and generate the sequence. （5）； Among them, z (1,i) (t k ) represents the land of type i up to the year. The cumulative standardized area; S404. Establish a grey differential equation to predict the area of ​​one type of land use, with the remaining six types as related variables; （6）； Where a is the development gray number, reflecting its own decay rate; b i Z represents the association weight between the i-th type of land and changes in paddy fields. (1,1) z represents the standardized area of ​​the first land use type. (1,i) The area of ​​the i-th type of land after standardization; S405. Use the least squares method to estimate the model parameters, parameter vector. , （7）； （8）； S406. Set up a sliding window. The initial window contains data from 1995-2000-2005, indicating that the land use status in 2005 is predicted using real data from 1995 and 2000. Slide to the next window: data from 2000-2005-2010, to predict the land use status in 2010, and so on. The time response function is obtained by accumulating the predicted values ​​of the generated series. , in Let be the standardized area of ​​the i-th type of land at time k. Step S5 specifically includes the following steps: S501. Area residual calculation and parameter update: Fit GM(1,N) to the data within the window, predict the next time point, and obtain the area residual. （9）； e (i) (t k+1 ) represents the i-th type of land in t k+1 Annual area residual; S502, Parameter Update （10）； （11）； Where η is the learning rate, controlling the update magnitude; e (1) For the predicted area residual of the model for data points of land use type 1, sign(e (1) The direction of the area residual determines the increase or decrease of the parameter, e. (1) >0, sign(e (1) )=+1;e (1) =0, sign(e (1) )=0, e (1) <0, sign(e (1) )=-1; a new Here, 'a' represents the updated gray number of development, and 'b' represents the gray number of development before the update. i,new Let b be the updated association weight for the i-th land use type. i The association weight of the i-th land use type before the update; S503. Based on autoregressive correction of area residuals, the historical area residuals of each land type are fitted. （12）； These are the autoregressive coefficients, estimated using the least squares method; This represents the mean of the area residuals; For the next time step (t) of the i-th type of land in the time series k+1 Predicted values ​​of area residuals; S504. Based on the current state and the transition probability matrix, correct the predicted values ​​of the original sequence. Its formula is, （13）； S505. Set the transfer probability between different land types according to different development scenarios, and then calculate the number of pixels for future land use by combining the transfer probability matrix and the corrected prediction value.

2. The method for predicting land use change in mountainous cities based on grey model and deep learning as described in claim 1, characterized in that: In step S1, the collected data includes land use imagery data of the prediction area and data on four aspects of land use change drivers: natural environment, socio-economic factors, accessibility, and policy restrictions.

3. The method for predicting land use change in mountainous cities based on grey model and deep learning as described in claim 1, characterized in that: Step S2, specifically Includes the following steps, S201. Determine the classification standards for land use types based on the attributes of the data source, land classification standards, the actual situation of the prediction area, and the requirements of the prediction objectives; S202. Unify the data coordinate system of the raster data and transform it into the same projected coordinate system. Trim the data range from overlapping with the study area so that the number of raster cells in all raster data is equal. S203. Determine the resolution of the raster data based on the size of the prediction area, the model operation time, and the calculation accuracy.

4. The method for predicting land use change in mountainous cities based on grey model and deep learning as described in claim 3, characterized in that: Step S3, specifically Includes the following steps, S301. Using the Markov model, calculate the transition probability matrix between different land use types based on historical data. The expression is as follows: （1）； （2）； Where, p ij (m) represents the probability of transitioning from land class i to land class j after m time units, p ij p is the probability of changing from type i to type j from the initial to the final stage, 0 ≤ p ij ≤1, S ij This represents the area that changes from type i to type j during the study period. Let represent the total area of ​​land use type i, and n be the number of land use data categories.

5. The method for predicting land use change in mountainous cities based on grey model and deep learning as described in claim 4, characterized in that: Step S6 specifically involves: S601. Select the driving factors affecting land use change, normalize the driving factor data using fuzzy membership, and obtain the suitability probability map of land use in the base year in the ANN module of the FLUS model. Its expression is: （14）； （15）； Where i represents land use type; s represents hidden layer; r represents raster; and t represents time. For suitability probability; w j,i As weight, ent s (r,t) represents the feedback received by the grid r of the s-th hidden layer at training time t; S602. Using the land use data for the forecast year and the constructed suitability probability atlas for the base year, the spatial pattern of land use for the forecast year is predicted in the CA module of the FLUS model. The expression is as follows: (16); Among them, TP t r,k Let r be the total probability of converting to land use type r in the t-th iteration. This represents the inertia coefficient of land use type k at time t. Indicates the cost of space type conversion. denoted by the number of grid cells generated by land use type i after the iteration, N is the Moore neighborhood in CA, and w is the variable neighborhood weight for each land use type; S603. Select the initial year of land use, the land use change transition matrix, the land use transition suitability image set, and the prediction period to conduct land use change simulation. Compare the simulation results with actual land use data, and use the Kappa coefficient to evaluate the accuracy of the simulation results. The formula for calculating the Kappa coefficient is: （17）； （18）； （19）； Where n is the total number of rasters; u1 is the number of rasters that are consistent with the simulation; U is the number of land types; and P0 represents the proportion of rasters that are consistent with the simulation. S604. Perform error analysis on the simulation results, adjust the land use transfer rules according to the analysis results, re-run the model simulation and evaluate the simulation results until simulation results that meet the accuracy requirements are obtained. S605. Analyze and predict the results, including changes in the quantity and spatial shifts of various land use types.

Images