CN116523415A - Urban extension simulation method and system based on urban extension deep learning CA model - Google Patents
Urban extension simulation method and system based on urban extension deep learning CA model Download PDFInfo
- Publication number
- CN116523415A CN116523415A CN202310221025.3A CN202310221025A CN116523415A CN 116523415 A CN116523415 A CN 116523415A CN 202310221025 A CN202310221025 A CN 202310221025A CN 116523415 A CN116523415 A CN 116523415A
- Authority
- CN
- China
- Prior art keywords
- city
- matrix
- data
- space
- land
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004088 simulation Methods 0.000 title claims abstract description 57
- 238000013135 deep learning Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 29
- 239000011159 matrix material Substances 0.000 claims abstract description 79
- 238000013145 classification model Methods 0.000 claims abstract description 32
- 230000000694 effects Effects 0.000 claims abstract description 21
- 238000011161 development Methods 0.000 claims abstract description 9
- 238000005070 sampling Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 14
- 238000010586 diagram Methods 0.000 claims description 10
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000018109 developmental process Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000005065 mining Methods 0.000 claims description 7
- 230000001413 cellular effect Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 5
- 230000011712 cell development Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- 230000009028 cell transition Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 230000007704 transition Effects 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 abstract description 11
- 210000004027 cell Anatomy 0.000 description 42
- 238000000605 extraction Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013079 data visualisation Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000005251 gamma ray Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 230000005919 time-dependent effect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
- Y02A30/60—Planning or developing urban green infrastructure
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a city extension simulation method and a system based on a city extension deep learning CA model, wherein the city extension simulation method and the system acquire multi-period land utilization grid data and reclassify the multi-period land utilization grid data; superposing the data of the last two periods after classification to obtain the increasing range of the town land, and rasterizing the space variable influencing the expansion of the town land, randomly sampling in the increasing range and the non-increasing range according to a certain picture size and a certain proportion to obtain sample data and a corresponding label thereof; establishing a land utilization matrix, wherein matrix elements correspond to cells, inputting multi-stage driving data in a certain space range corresponding to each cell into a classification model, obtaining conditional probability of classifying each cell into urban land under the influence of multi-stage historical space variables, taking the conditional probability as CA model cell conversion probability, and constructing an urban expansion CA model by combining neighborhood constraint on the basis. The CA model constructed by the method considers the spatial multi-scale neighborhood effect and the time dependence, and is helpful for more accurately simulating urban development.
Description
Technical Field
The invention belongs to the technical field of geographic simulation, relates to a city expansion simulation method and system, and in particular relates to a city expansion simulation method and system for utilizing deep learning to mine a cellular automaton of a city expansion space multi-scale neighborhood effect and time dependence.
Background
Cellular automata (cellular automata, CA) is a "bottom-up" model that simulates spatially discrete, time discrete complexity phenomena by local operations, and can be integrated with other models to model and predict urban land utilization evolution due to its open architecture. CA has been widely used to model the spatio-temporal evolution of complex nonlinear systems, well suited for the simulation and prediction of complex geographic processes. The conversion rule is a core component of the CA model, and students adopt various methods to mine the conversion rule of the CA model, so as to improve the simulation precision of the CA model.
However, the following disadvantages exist in the research of the conversion rule:
(1) The model is constructed with insufficient expression of the characteristics of the multi-scale area of the neighborhood. The change of the state of the cells of the urban land is not only influenced by the information of the single cells of the urban land, but also the space variable information of the adjacent cells influences the change of the state of the cells on different scales through different characteristic effects, and the research is less and complete to express the influence of the multi-scale space effect of the expansion of the urban land;
(2) The conversion rules are constructed without taking into account the time-dependent effects of the long time series. At present, scholars simulate urban land expansion based on a Markov process, consider that land utilization evolution is only influenced by the last time land utilization and driving factors, and consider insufficient time dependence under long time sequence.
Therefore, a new method is needed to solve the two problems, and a more accurate and reliable simulation method is provided for city expansion.
Disclosure of Invention
In order to solve the technical problems, the invention provides a city CA model which considers space multiscale neighborhood effect and time dependence, namely a city extended Deep learning CA model (3 DCNN-ConvLSTM-CA, simply called Deep-CA), wherein the neighborhood effect under different space scales is mined along the time dimension by combining 3DCNN of common convolution and cavity convolution, the common convolution layer is used for extracting the close space neighborhood effect of a modeling cell, and the cavity convolution can obtain a larger neighborhood range in a mode of setting intervals, so that the running efficiency of the model is ensured, and the aggregation of multiscale neighborhood characteristics is realized. And filtering and fusing time and space information on the spatial feature map obtained by excavation through ConvLSTM to obtain space-time dependency information influencing future urban land utilization change, and obtaining a conversion rule of urban expansion through a fully-connected network. The model enables the mining and extraction of CA conversion rules to be more perfect and reliable, and provides a more accurate and reliable simulation method for city expansion.
The technical scheme adopted by the method is as follows: a city extension simulation method based on a city extension deep learning CA model comprises a deep learning module and a CA module; the deep learning module is trained to obtain a city growth classification model, the city growth classification model extracts space neighborhood effect and time dependency information of city expansion, a city development probability layer is output, and the CA module is used for final simulation of city expansion;
the method comprises the following steps:
step 1: land utilization space raster data and space variable driving data preprocessing and sample acquisition;
step 2: converting the multi-period land utilization space raster data and the space variable driving data raster data in the step 1 into matrixes respectively, wherein matrix elements correspond to cells, and matrix spaces correspond to cell spaces;
step 3: calculating the conditional probability of each matrix element attribute classification in the land utilization state matrix P as the urban land by using the city growth classification model, and obtaining a probability map;
the city growth classification model is formed by sequentially connecting 3DCNN, convLSTM with a full-connection layer; the 3DCNN is used for data reduction and mining of a space neighborhood effect and consists of a convolution layer and a pooling layer; the ConvLSTM realizes the transmission of time information under the condition of ensuring that the space information is not lost, and consists of ConvLSTM and a standardization layer; the full-connection layer is used for outputting space-time information to obtain a conditional probability layer, takes ConvLSTM output data as input data, and outputs the data through a plurality of full-connection linear layers after being unfolded;
step 4: and the probability layer is input to a CA module, and a final simulation result diagram is obtained after the set iteration termination condition is reached.
The system of the invention adopts the technical proposal that: an urban extension simulation system based on an urban extension deep learning CA model, wherein the urban extension deep learning CA model consists of a deep learning module and a CA module; the deep learning module is trained to obtain a city growth classification model, the city growth classification model extracts space neighborhood effect and time dependency information of city expansion, a city development probability layer is output, and the CA module is used for final simulation of city expansion;
the system comprises the following modules:
the module 1 is used for preprocessing land utilization space raster data and space variable driving data and acquiring samples;
the module 2 is used for converting the multi-period land utilization space raster data and the space variable driving data raster data in the module 1 into matrixes respectively, wherein matrix elements correspond to cells, and matrix spaces correspond to cell spaces;
the module 3 is used for calculating the conditional probability of each matrix element attribute classification in the land utilization state matrix P as the urban land by using the urban growth classification model, and obtaining a probability map;
the city growth classification model is formed by sequentially connecting 3DCNN, convLSTM with a full-connection layer; the 3DCNN is used for data reduction and mining of a space neighborhood effect and consists of a convolution layer and a pooling layer; the ConvLSTM realizes the transmission of time information under the condition of ensuring that the space information is not lost, and consists of ConvLSTM and a standardization layer; the full-connection layer is used for outputting space-time information to obtain a conditional probability layer, takes ConvLSTM output data as input data, and outputs the data through a plurality of full-connection linear layers after being unfolded;
and the module 4 is used for inputting the probability layer into the CA module, and obtaining a final simulation result diagram after the set iteration termination condition is reached.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) The sample data of the invention is space data with a certain size, rather than single unit cell, and the 3DCNN performs feature extraction on space information along the time dimension by combining common convolution and cavity convolution, thereby being very suitable for mining the urban expansion space features
(2) The invention utilizes ConvLSTM to carry out stricter space-time feature extraction on the space feature map obtained in the step (1) to realize assimilation output of time and space information, and obtains the conditional probability of classifying each cell into a land cell for towns through a full-connection layer, thereby obtaining the conversion rule taking the space multiscale neighborhood effect and long-time sequence time dependency feature into consideration.
(3) The method constructs an urban extended deep learning CA model: after the cell classification conditional probability is standardized, the cell classification conditional probability is used as the conversion probability of cells in the CA model, and a neighborhood constraint is added to be applied to town expansion simulation, so that the urban expansion can be accurately and objectively simulated.
Accessory description
FIG. 1 is a schematic diagram of a study area location according to an embodiment of the present invention, wherein (a) is Beijing city location and (b) is Beijing city 1995-2015 city expansion every 10 years;
FIG. 2 is a diagram of a city extension deep learning CA model structure in accordance with an embodiment of the present invention;
fig. 3 is a schematic diagram of an extended range of town land in 2015 versus 2010 according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the result of processing spatial variables affecting town land expansion according to an embodiment of the present invention;
FIG. 5 is a training curve of the city growth classification model according to an embodiment of the present invention;
fig. 6 is a schematic diagram of simulation results of an extended deep learning CA model of an embodiment of the present invention, in which: (1) For practical city expansion, (2) - (5) are simulation results obtained using Deep-CA, maximum entropy-CA, ANN-CA, logic-CA, respectively.
Detailed Description
In order to facilitate the understanding and practice of the invention, those of ordinary skill in the art will now make further details with reference to the drawings and examples, it being understood that the examples described herein are for the purpose of illustration and explanation only and are not intended to limit the invention thereto.
Referring to fig. 1, in this embodiment, city expansion simulation is performed using beijing as an experimental area. The embodiment adopts an urban extension simulation method based on an urban extension deep learning CA model, and the urban extension deep learning CA model (3 DCNN-ConvLSTM-CA) of the embodiment consists of a deep learning module and a CA module; training by the deep learning module to obtain a city growth classification model, extracting space neighborhood effect and time dependency information of city expansion by the city growth classification model, outputting a city development probability layer, and finally simulating the city expansion by the CA module;
please refer to fig. 2, the city extension deep learning CA model of the present embodiment includes a city growth classification model (formed by sequentially connecting 3DCNN, convLSTM and a full connection layer) and a CA module; the 3DCNN is used for data reduction and mining of space neighborhood effects and consists of a convolution layer and a pooling layer (more than 2 are generally set); under the condition of ensuring that the space information is not lost, the ConvLSTM realizes the transmission of time information, and consists of ConvLSTM and a standardized layer; the full-connection layer is used for outputting space-time information to obtain a conditional probability layer, takes ConvLSTM output data as input data, and outputs the data through a plurality of full-connection linear layers after being unfolded;
the ConvLSTM is connected to 3DCNN blocks, the 3DCNN performs strict spatial feature extraction on different convolution scales along the time dimension, the 3DCNN performs multi-step convolution, standardization and maximum pooling operation on an input sample image, and the extracted spatial multi-scale neighborhood feature map is conveyed to the ConvLSTM; after ConvLSTM receives the feature map, more strict space-time feature extraction is performed; finally, the final characteristic map is input into 512 neuron full-connection layers after being flattened, a final probability value is output through a plurality of full-connection layers, a probability map is obtained, and a dropout layer with the probability of 0.5 is used between the layers so as to avoid overfitting; and the probability map is input to a CA module to obtain a final simulation result map.
The method of the embodiment specifically comprises the following steps:
step 1: land utilization space raster data and space variable driving data preprocessing and sample acquisition;
the specific implementation comprises the following substeps:
step 1.1: reclassifying the land utilization space raster data (more than or equal to 4) of a plurality of periods, including town land, expandable land and water area;
step 1.2: overlapping and calculating the grid data of the land utilization space in the last two periods to obtain the expansion range of the town land in the last period relative to the last period; the results are shown in FIG. 3;
step 1.3: rasterizing the space variable driving data influencing the expansion of the town land, and combining the space variable driving data of each period into multi-band raster data, wherein each band corresponds to one space variable driving data;
if the space variable driving data affecting the expansion of the town land is a space entity, such as a road, a water area, a POI (point of interest) and the like, performing Euclidean distance processing on the space variable driving data, and keeping the grid size, the coordinate projection and the position consistent with the land utilization data; if the space variable driving data affecting the expansion of the town land is a non-space entity element, such as GDP space distribution, population density and the like, interpolation processing is carried out on the space variable driving data, and the grid size, coordinate projection and position are kept consistent with land utilization data; the final drive data visualization is shown in fig. 4;
step 1.4: and randomly sampling according to a certain picture size and a certain proportion in a growing range and a non-growing range to obtain space variable driving data sample data (space driving variable) and corresponding labels thereof, wherein 0 represents growth and 1 represents non-growth.
Step 2: converting the multi-period land utilization space raster data and the space variable driving data raster data in the step 1 into matrixes respectively, wherein matrix elements correspond to cells, and matrix spaces correspond to cell spaces;
converting the classified land utilization space raster data into a land utilization state matrix S, wherein three values exist in the matrix, the value of the land for town is 1, the value of the expandable land is 0, and the value of the water area is 2; converting the space variable driving data rasterized data into space variable influence matrixes A, B, C and …, wherein each matrix corresponds to one period of space driving data;
since the grid size position is kept consistent in step 1, the space coordinates (y) of the matrix elements are determined in the matrix space, so that the state of the land utilization state matrix P at the position can be determined: s is S x,y =0 or 1 or 2, and the influence values of the multiple spatial variable influence matrices a, B, C … at and around (x, y).
Step 3: calculating the conditional probability of each matrix element attribute classification in the land utilization state matrix P as the urban land by using the city growth classification model, and obtaining a probability map;
the specific implementation method comprises the following steps:
step 3.1: inputting the space variable driving data of a plurality of periods into a city growth classification model, calculating to obtain the conditional probability of classifying each cell into a town land, and outputting a conditional probability matrix
Step 3.2: the conditional probability matrixNormalization processing is carried out to obtain a matrix P D Cell transition probability matrix as CA module:
where i is the number of rows of the matrix, j is the number of columns of the matrix, W (i,j) A transition probability for a cell of matrix coordinates (i, j);
step 3.3: calculating a neighborhood constraint matrix P according to the rule of the CA module N Taking into account water constraints P C Performing matrix operation to obtain a cell development probability matrix P of the CA module;
the example neighborhood takes the form of a 7×7 extended molar neighborhood, and in one 7×7 neighborhood, the greater the number of town cells, the greater the probability that the center cell will develop into a town cell, described in mathematical language as:
wherein P is N Representing the influence of a neighborhood configuration for constraining city morphology and receiving surrounding cellular influences, wherein n represents the neighborhood size, d represents the distance between the domain unit and the central unit, d = 1,2,3, …, (n+1)/2; CS (d) represents a neighborhood cell state function when the distance is d, if the cell state is city, returning a value of 1, otherwise, returning 0; obtaining a neighborhood constraint matrix P through matrix calculation N ；
Where i is the number of rows of the matrix, j is the number of columns of the matrix, N (i,j) A neighborhood constraint influence value obtained by calculation for matrix elements with matrix coordinates of (i, j);
then the cell development probability matrix of CA module p=p D *P N *P C ；
Wherein P is C As a constraint factor, when a cell is an urban cell or a water cell, CON () returns 0, and others return 1; i is the row number of the matrix, j is the column number of the matrix, P (i,j) The probability is developed for cells with matrix coordinates (i, j).
Step 4: and inputting the probability layer into the CA module, and obtaining a final simulation result diagram after the set iteration termination condition is reached.
The specific implementation method comprises the following steps:
step 4.1: calculating the total number M (M= 493585 in this embodiment) of the end of the stage relative to the initial town cell increment, simulating a time span T (T=5 in this embodiment, and 10 iterations in half a year in combination with the actual situation in this embodiment), and increasing the number of town cells per iteration to beTaking the constraint condition as a constraint condition of each iteration ending of the CA module, recalculating the neighborhood constraint after each iteration ending, and simulating the rest of the last generation;
step 4.2: and sequentially screening the maximum value in the development probability matrix P, setting matrix element attributes corresponding to the matrix coordinate positions as town land in a simulation result matrix, stopping iteration until iteration conditions are met, namely, the number of converted town land cells reaches the number of newly increased town land cells in a simulation stage, and outputting an image.
The city growth classification model adopted in the embodiment is a city growth classification model trained by the deep learning module; during training, firstly, a training sample data set is built, the training sample data set is input into a city extension deep learning CA model (3 DCNN-ConvLSTM-full connection) for classification model training, so that a city extension classification model is obtained, 80% of the sample set is used for training, and the rest 20% of the verification set is used for adjusting model super-parameters and checking stop conditions;
in this embodiment, the study dimensions include space, time and the number of driving factors, so that the required data type is 4D tensor, since the deep learning model is trained in batches, a batch dimension is added, the final model input data is 5D tensor of "batch size time feature height width", the shape is (b, t, c, h, w), and it can be understood that b sample data is t-period driving data, c driving factors are present in each period, each driving factor is a picture of h x w, and b sample data corresponds to b labels. The sample data set is [ (B, t, c, h, w), L ] and B is the total number of samples, L corresponds to the number of labels, and obviously B=L.
The sample set is input into a city growth classification model for training, and the optimal parameters are obtained by adjusting the parameters, setting the stopping conditions and observing the training curve (figure 5), and the city growth classification model required by the invention is obtained by inputting the sample set into the model.
The city growth classification model and CA module combination of the present embodiment may perform city expansion simulation.
Table 1 dataset
Land utilization data and driving factor data of 1995-2015 of Beijing are used for verifying the scientificity and applicability of the CA model, and data of 1995-2010 are used for model training to simulate the urban range of 2015. Experimental data and descriptions are presented in table 1. And (5) carrying out rasterization pretreatment on the data based on the ArcGIS 10.7 platform. For example, at 30m resolution, each layer contains 6110×5492 cells. In order to acquire training data required by experiments, space superposition analysis, space interpolation analysis and Euclidean distance calculation are carried out on historical data by using an ArcGIS tool, urban land change and space variable data are extracted, data standardization processing is carried out, then the data are converted into random sampling points in the range of urban land in the 2005-2015 time period, final space training samples and space influence variable constraint conditions are obtained by checking and eliminating error points, and finally the maximum entropy classification model is adopted to calculate land utilization development probability P.
The construction and operation of the city extension deep learning CA model are realized in a Python 3.7 environment, pytorch is used as a back end, and Python libraries such as Numpy, visdom, gdal and the like are also used. The computer is configured as Intel i7-10 generation, 16G memory, 4G GTX 1650Ti display card, and the model training and running adopts GPU parallel acceleration. In order to represent the advantage of the city extended deep learning CA model in improving the model precision, the model is compared with a logic CA model, a maximum entropy CA and an ANN-CA.
At present, a common town expansion simulation result precision test mode is available: the whole Accuracy (OA), kappa coefficient and Figure of Merit (FoM) are all in the range of 0 to 1, and the larger the value, the better the simulation result. Wherein Kappa coefficient is a common method for measuring the quantity consistency between land utilization simulation and observation; foM focuses on measuring the consistency of variation between simulation and observation. The calculation formula is as follows:
OA＝righ/
wherein, high is the number of cells simulating correct; n is the total number of all cells; a, a 0 And a 1 The number of urban cells and non-urban cells in the observation result are respectively; b 0 And b 1 The number of urban and non-urban cells in the simulation result, respectively. Wherein a is the number of cells observed as varying but simulated as remaining unchanged; b is the number of cells that are city-extended in both observation and simulation; c is the number of cells observed as city expansion but simulated as other land class conversion, and in the embodiment, only the conversion from non-city to city land is considered, and then C is 0; d is a model of non-cityThe number of cells intended for city expansion.
Kappa coefficients fall between 0 and 1, and can be divided into five groups to represent different levels of consistency. As shown in table 2.
TABLE 2Kappa coefficient level Classification
Table 3 2015 Beijing city extension model simulation accuracy evaluation
Table 3 shows that compared with other models, the city extension deep learning CA model obviously improves the simulation precision, the FoM reaches 0.304, and the simulation precision is improved by about 4 percent, so that compared with the models such as logistic regression CA and the like, the city extension deep learning CA model is improved, and the simulation result is more accurate.
In this example 3, the simulation result of the DCNN-ConvLSTM-CA model and the local comparison with logistic regression-CA, maximum entropy-CA and ANN-CA are shown in FIG. 6; the 3DCNN-ConvLSTM-CA is shown to have finer control on urban growth, the edge of a simulation result is closer to reality, and the influence of a spatial multiscale neighborhood effect and time dependence on urban growth morphology is proved. In particular, for the filled city extension (fig. 6 (a)), the 3DCNN-ConvLSTM-CA and the maximum entropy-CA have better effects, the logic-CA and the ANN-CA are still edge extensions, and the filling is not thorough enough; it can be seen from fig. 6 (b) and 6 (c) that 3DCNN-ConvLSTM-CA is better for urban edge constraint, and no blind urban edge expansion occurs.
In addition, a set of landscape indices was used to evaluate the landscape pattern similarity between actual and simulated urban land, the number of patches (Number of Patches, NP), average euclidean nearest neighbor (Euclidean Nearest Neighbor Index, enn_mn), average Shape Index (shape_mn), and maximum patch occupancy landscape area ratio (Largest Patch Index, LPI) were selected from the density and difference, the proximity Index, the Shape Index, and the area Index, respectively, for the number of patches, enn_mn for the patch distribution, shape_mn for the patch Shape complexity, LPI for the dominant patch dominance, and all landscape indices were calculated by FRAGSTATS 4.2. Similarity calculation formula:
wherein I is si And I oi The value of the ith landscape index, delta I, for simulating the urban land and the actual urban land respectively i Is the normalized difference of the ith pair of simulated and actual landscape indices, the original units of LPI are already percentages, therefore ΔI of LPI i Is the absolute value of the difference; gamma ray I Is the landscape pattern similarity between the actual and simulated urban lands, and n is the number of landscape indexes.
Table 4 View of simulation results for different models of Beijing city in 2015
Table 4 shows the results of the calculation of the landscape similarity for the four models. Compared with ANN-CA, maximum entropy-CA and logic-CA, the city extended Deep learning CA model has the highest landscape similarity with the actual city land (0.871), which shows that Deep-CA has better control over city morphology.
The analysis verifies a CA model which is constructed by the invention and reasonably considers the spatial multi-scale neighborhood effect and the time dependence of urban expansion, and effectively improves the cellular automaton algorithm of the model simulation precision. 3DCNN and ConvLSTM were first used for city extension simulation, which set aside the Markov assumption premise of city extension simulation, i.e., only the start and end points of the study period were focused, and the time-dependent features of long-time-series data were considered into city extension simulation. The 3DCNN performs strict space neighborhood feature extraction on different convolution scales along the time dimension, the ConvLSTM performs stricter space-time feature extraction on the 3DCNN extracted space feature map, and assimilates and outputs time and space information, so that urban land utilization change is simulated more accurately.
It should be understood that the foregoing description of the preferred embodiments is not intended to limit the scope of the invention, but rather to limit the scope of the claims, and that those skilled in the art can make substitutions or modifications without departing from the scope of the invention as set forth in the appended claims.
Claims (7)
1. A city extension simulation method based on a city extension deep learning CA model is characterized in that; the city extension deep learning CA model consists of a deep learning module and a CA module; the deep learning module is trained to obtain a city growth classification model, the city growth classification model extracts space neighborhood effect and time dependency information of city expansion, a city development probability layer is output, and the CA module is used for final simulation of city expansion;
the method comprises the following steps:
step 1: land utilization space raster data and space variable driving data preprocessing and sample acquisition;
step 2: converting the multi-period land utilization space raster data and the space variable driving data raster data in the step 1 into matrixes respectively, wherein matrix elements correspond to cells, and matrix spaces correspond to cell spaces;
step 3: calculating the conditional probability of each matrix element attribute classification in the land utilization state matrix P as the urban land by using the city growth classification model, and obtaining a probability map;
the city growth classification model is formed by sequentially connecting 3DCNN, convLSTM with a full-connection layer; the 3DCNN is used for data reduction and mining of a space neighborhood effect and consists of a convolution layer and a pooling layer; the ConvLSTM realizes the transmission of time information under the condition of ensuring that the space information is not lost, and consists of ConvLSTM and a standardization layer; the full-connection layer is used for outputting space-time information to obtain a conditional probability layer, takes ConvLSTM output data as input data, and outputs the data through a plurality of full-connection linear layers after being unfolded;
step 4: and the probability layer is input to a CA module, and a final simulation result diagram is obtained after the set iteration termination condition is reached.
2. The city extension simulation method based on city extension deep learning CA model of claim 1, wherein the specific implementation of step 1 comprises the following sub-steps:
step 1.1: reclassifying the land utilization space raster data of a plurality of periods, including town land, expandable land and water areas;
step 1.2: overlapping and calculating the grid data of the land utilization space in the last two periods to obtain the expansion range of the town land in the last period relative to the last period;
step 1.3: rasterizing the space variable driving data influencing the expansion of the town land, and combining the space variable driving data of each period into multi-band raster data, wherein each band corresponds to one space variable driving data;
if the space variable driving data affecting the expansion of the town land is a space entity, performing Euclidean distance processing on the space variable driving data, and keeping the grid size, coordinate projection and position consistent with the land utilization data; if the space variable driving data affecting the expansion of the town land is a non-space entity element, carrying out interpolation processing on the space variable driving data, and keeping the grid size, coordinate projection and position consistent with land utilization data;
step 1.4: and randomly sampling in a certain picture size and a certain proportion in a growing range and a non-growing range to obtain space variable drive data sample data and corresponding labels thereof, wherein 0 represents growth and 1 represents non-growth.
3. The city extension simulation method based on the city extension deep learning CA model according to claim 1, wherein: in the step 2, the classified land utilization space raster data is converted into a land utilization state matrix S, three values exist in the matrix, the value of the land for towns is 1, the value of the expandable land is 0, and the value of the water area is 2; the space variable driving data rasterized data is converted into space variable influence matrices A, B, C, … and N, and each matrix corresponds to one period of space driving data.
4. The city extension simulation method based on the city extension deep learning CA model according to claim 1, wherein: in the step 3, the city extension deep learning CA model is a trained city extension deep learning CA model; the data used in training is the 5D tensor of "batch size time feature height width", noted (b, t, c, h, w); representing t periods of driving data of b sample data, wherein each period has c driving factors, each driving factor is a picture of h x w, and each b sample data corresponds to b labels; the sample data set is [ (B, t, c, h, w), L ] and B is the total number of samples, L corresponds to the number of labels, and B=L.
5. The city extension simulation method based on the city extension deep learning CA model of claim 1, wherein the specific implementation of step 3 comprises the following steps:
step 3.1: inputting the space variable driving data of a plurality of periods into a city growth classification model, calculating to obtain the conditional probability of classifying each cell into a town land, and outputting a conditional probability matrix
Step 3.2: the conditional probability matrixNormalization processing is carried out to obtain a matrix P D Cell transition probability matrix as CA module:
where i is the number of rows of the matrix, j is the number of columns of the matrix, W (i,j) A transition probability for a cell of matrix coordinates (i, j);
step 3.3: calculating a neighborhood constraint matrix P according to the rule of the CA module N Taking into account water constraints P C Performing matrix operation to obtain a cell development probability matrix P of the CA module;
wherein P is N Representing the influence of a neighborhood configuration for constraining city morphology and receiving surrounding cellular influences, wherein n represents the neighborhood size, d represents the distance between the domain unit and the central unit, d = 1,2,3, …, (n+1)/2; CS (d) represents a neighborhood cell state function when the distance is d, if the cell state is city, returning a value of 1, otherwise, returning 0; obtaining a neighborhood constraint matrix P through matrix calculation N ；
Where i is the number of rows of the matrix, j is the number of columns of the matrix, N (i,j) A neighborhood constraint influence value obtained by calculation for matrix elements with matrix coordinates of (i, j);
then the cell development probability matrix of CA module p=p D *P N *P C ；
Wherein P is C As a constraint factor, when a cell is an urban cell or a water cell, CON () returns 0, and others return 1; i is the row number of the matrix, j is the column number of the matrix, P (i,j) Cell development for matrix coordinates (i, j)Probability.
6. The city extension simulation method based on the city extension deep learning CA model according to any one of claims 1-5, wherein the specific implementation of step 4 comprises the following steps:
step 4.1: calculating the total number M of the increment of the urban land cells at the end of the stage relative to the initial stage, simulating a time span T, and increasing the number of the urban land cells for each iteration to beTaking the constraint condition as a constraint condition of each iteration ending of the CA module, recalculating the neighborhood constraint after each iteration ending, and simulating the rest of the last generation;
step 4.2: and sequentially screening the maximum value in the development probability matrix P, setting matrix element attributes corresponding to the matrix coordinate positions as town land in a simulation result matrix, stopping iteration until iteration conditions are met, namely, the number of converted town land cells reaches the number of newly increased town land cells in a simulation stage, and outputting an image.
7. The city extension simulation system based on the city extension deep learning CA model is characterized in that the city extension deep learning CA model consists of a deep learning module and a CA module; the deep learning module is trained to obtain a city growth classification model, the city growth classification model extracts space neighborhood effect and time dependency information of city expansion, a city development probability layer is output, and the CA module is used for final simulation of city expansion;
the system comprises the following modules:
the module 1 is used for preprocessing land utilization space raster data and space variable driving data and acquiring samples;
the module 2 is used for converting the multi-period land utilization space raster data and the space variable driving data raster data in the module 1 into matrixes respectively, wherein matrix elements correspond to cells, and matrix spaces correspond to cell spaces;
the module 3 is used for calculating the conditional probability of each matrix element attribute classification in the land utilization state matrix P as the urban land by using the urban growth classification model, and obtaining a probability map;
the city growth classification model is formed by sequentially connecting 3DCNN, convLSTM with a full-connection layer; the 3DCNN is used for data reduction and mining of a space neighborhood effect and consists of a convolution layer and a pooling layer; the ConvLSTM realizes the transmission of time information under the condition of ensuring that the space information is not lost, and consists of ConvLSTM and a standardization layer; the full-connection layer is used for outputting space-time information to obtain a conditional probability layer, takes ConvLSTM output data as input data, and outputs the data through a plurality of full-connection linear layers after being unfolded;
and the module 4 is used for inputting the probability layer into the CA module, and obtaining a final simulation result diagram after the set iteration termination condition is reached.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310221025.3A CN116523415A (en) | 2023-03-08 | 2023-03-08 | Urban extension simulation method and system based on urban extension deep learning CA model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310221025.3A CN116523415A (en) | 2023-03-08 | 2023-03-08 | Urban extension simulation method and system based on urban extension deep learning CA model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116523415A true CN116523415A (en) | 2023-08-01 |
Family
ID=87405376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310221025.3A Pending CN116523415A (en) | 2023-03-08 | 2023-03-08 | Urban extension simulation method and system based on urban extension deep learning CA model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116523415A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117251892A (en) * | 2023-08-29 | 2023-12-19 | 北京建筑大学 | Urban space growth simulation system for coupling water resource environment bearing |
-
2023
- 2023-03-08 CN CN202310221025.3A patent/CN116523415A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117251892A (en) * | 2023-08-29 | 2023-12-19 | 北京建筑大学 | Urban space growth simulation system for coupling water resource environment bearing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112861722B (en) | Remote sensing land utilization semantic segmentation method based on semi-supervised depth map convolution | |
| Roodposhti et al. | Towards automatic calibration of neighbourhood influence in cellular automata land-use models | |
| CN110991497B (en) | BSVC (binary sequence video coding) -method-based urban land utilization change simulation cellular automaton method | |
| CN113807210A (en) | Remote sensing image semantic segmentation method based on pyramid segmentation attention module | |
| CN108376183B (en) | City CA model construction method based on maximum entropy principle | |
| CN110909924A (en) | City expansion multi-scenario simulation cellular automata method based on cross entropy optimizer | |
| CN106228185A (en) | A kind of general image classifying and identifying system based on neutral net and method | |
| CN104732240A (en) | Hyperspectral image waveband selecting method applying neural network to carry out sensitivity analysis | |
| CN107590515A (en) | The hyperspectral image classification method of self-encoding encoder based on entropy rate super-pixel segmentation | |
| CN114821217B (en) | Image recognition method and device based on quantum classical hybrid neural network | |
| CN110245683B (en) | Residual error relation network construction method for less-sample target identification and application | |
| CN114662414B (en) | Oil reservoir production prediction method based on graph wavelet neural network model | |
| CN110349170B (en) | Full-connection CRF cascade FCN and K mean brain tumor segmentation algorithm | |
| CN112560966B (en) | Polarized SAR image classification method, medium and equipment based on scattering map convolution network | |
| CN102222313A (en) | Urban evolution simulation structure cell model processing method based on kernel principal component analysis (KPCA) | |
| CN112036249B (en) | Method, system, medium and terminal for end-to-end pedestrian detection and attribute identification | |
| CN113297174B (en) | Land utilization change simulation method based on deep learning | |
| CN113222316A (en) | Change scene simulation method based on FLUS model and biodiversity model | |
| CN111639587A (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN116523415A (en) | Urban extension simulation method and system based on urban extension deep learning CA model | |
| CN114283285A (en) | Cross consistency self-training remote sensing image semantic segmentation network training method and device | |
| CN114821340A (en) | Land utilization classification method and system | |
| CN114863348A (en) | Video target segmentation method based on self-supervision | |
| CN116524369A (en) | Remote sensing image segmentation model construction method and device and remote sensing image interpretation method | |
| CN113642764B (en) | Village and town aggregation space evolution simulation prediction method and computer equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |