A block green space structure intelligent optimization method and system

Abstract

This invention discloses an intelligent optimization method and system for the structure of green spaces in urban blocks, relating to the fields of urban planning, landscape architecture, and artificial intelligence. The method includes: acquiring a planning area; delineating units based on the planning area to obtain multiple block units; acquiring relevant indicator data for each block unit; inputting the relevant indicator data for each block unit into a pre-established regression model; and outputting an optimized regression model. The method further includes: calculating the importance of variables in the optimized regression model; performing weight correction and coupling degree calculation based on variable importance to obtain corrected weights and block unit coupling degree; acquiring images of vacant land; performing raster recognition on the vacant land images and overlaying them to obtain an expanded vacant land area; predicting the vacancy probability of the land use raster based on the optimized regression model to obtain the vacancy probability; generating an initial database of urban blocks based on the vacancy probability; generating an initial population from the initial database using a genetic algorithm; and determining the optimized green space structure of the urban blocks based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm. This method can provide diverse options for urban green space planning.

Description

Technical Field

[0001] This invention relates to the fields of urban planning, landscape architecture, and artificial intelligence, specifically to a method and system for intelligent optimization of the structure of green spaces in urban blocks. Background Technology

[0002] In recent years, urban shrinkage has spread across China and is expected to become the norm for urban development in the future. This process has led to an increase in vacant urban land, triggering multiple problems related to the ecological environment and socio-economic aspects. Green spaces have proven to be an effective means of addressing the problems caused by vacant urban land, playing a multifaceted role in improving ecosystem services, alleviating social problems, and driving economic development.

[0003] Currently, existing technologies have explored intelligent planning methods for green spaces or other similar planning elements (such as urban open spaces). However, these methods lack both the constraints on the preconditions for green space planning and the assessment of the coupling degree between the generated scheme and the built environment, thus making it difficult to meet the planning requirements of limited spatial resources under existing planning. Furthermore, the use of random methods for green space layout in intelligent planning results in high computational costs and slow processing speeds. Summary of the Invention

[0004] To address the shortcomings mentioned in the background art, the present invention aims to provide a method and system for intelligent optimization of the structure of green spaces in urban blocks.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a method for intelligent optimization of the structure of green spaces in urban blocks, the method comprising the following steps:

[0006] Obtain the planning area, delineate units based on the planning area to obtain multiple street blocks, obtain relevant indicator data for each street block, input the relevant indicator data for each street block into a pre-established regression model, and output the optimized regression model.

[0007] The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm.

[0008] The importance of variables is calculated for the optimized regression model. Based on the importance of variables, the weights are corrected and the coupling degree is calculated to obtain the corrected weights and the coupling degree of the block unit. Images of vacant land are obtained. After raster recognition of the vacant land images, the vacant land expansion area is obtained by overlaying. Based on the optimized regression model, the vacancy probability of land use raster is predicted to obtain the vacant land probability.

[0009] An initial database of neighborhoods is generated based on the probability of vacant land use. An initial population is generated from the initial database of neighborhoods based on a genetic algorithm. Finally, the optimization result of the green space structure of the neighborhoods is determined by the fitness of the best individual in each generation of the initial population based on a reinforcement learning algorithm.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of obtaining relevant indicator data for each block unit:

[0011] Calculate the corresponding target performance indicators and potential influencing factor indicators for each block unit in each period. Subtract the previous year's indicator from the current year's indicator to obtain the target performance change value {ΔP} and the potential influencing factor indicator change value {ΔE}. For each block unit i, {ΔP, ΔE} are included. Perform correlation analysis on all potential influencing factor indicator change values, eliminate influencing factors with multicollinearity, and summarize the remaining indicators of all block units into a database I{ΔP, ΔE}.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the formula for the random forest regression algorithm is as follows:

[0013]

[0014] Where E = {E1, E2, ... E} n Let} be a feature vector containing the change values ​​of n potential influencing indicators, M be the number of decision trees, and T be the number of trees. m (X) is the prediction value of the m-th decision tree for the input feature vector X. Each decision tree T m (X) processes the input features according to its own tree structure and splitting rules, and gives the final predicted value;

[0015] The evaluation metrics for the regression model include: mean squared error (MSE), mean absolute error (RMSE), and R-squared. 2 The formula is as follows:

[0016]

[0017] In the formula, y i For the actual values ​​of the test set, These are values ​​calculated based on the random forest regression model. is the average value of the test set, and n is the number of samples in the test set.

[0018] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: calculating the variable importance (VIM) of the optimized regression model. e The calculation process:

[0019]

[0020] In the formula, M represents the number of decision trees in the output regression model M, and T... m It is the set of all nodes in the m-th decision tree, ΔMSE t,e It is the amount of impurity reduction caused by feature e at node t;

[0021] The process of adjusting weights and calculating coupling degree based on variable importance:

[0022] For each target performance model, variables with an independent variable importance (VIM) greater than a preset threshold are selected. The VIM importance is used as the initial weight W, and the initial weight W is then adjusted. ′ As a weight in the coupling model, for each block unit i, the coupling degree D for target J is... ij The model formula is the sum of the products of variables greater than the threshold and their adjusted weights;

[0023]

[0024] Among them, D ij E represents the coupling degree of block unit i in target j. im W′ represents the value of variable m (whose importance VIM is greater than a threshold) in block unit i. m The corrected weights;

[0025] The method for adjusting the weights is as follows:

[0026]

[0027] Among them, W m To determine the importance of independent variables (VIM) in the target performance model e Importance score of variable m exceeding the threshold (VIM) e , where n is the total number of variables m.

[0028] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the optimized regression model RF expression is as follows:

[0029]

[0030] Where X = {X1, X2, ... X} n Let} be a feature vector containing n driving factors, M be the number of final decision trees, and T be the number of decision trees. m The depth of (X) and the number of node split samples are determined based on the training results.

[0031] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the vacant land being completely vacant land within an urban area without development purpose, as well as land abandoned, vacant, or underutilized from construction purposes, the vacant land categories including bare land, natural grassland, natural shrubland, inefficient construction land, areas under construction, and natural wetlands.

[0032] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the generation process of the initial street database is as follows:

[0033] Construct an M×M matrix D0 corresponding to the street grid division, where each element V ij Let represent the grid cell in the i-th row and j-th column of the block, where 1≤i≤M, 1≤j≤M;

[0034]

[0035] Each element V ij All are represented in the form of attribute vectors. For an element V containing n attributes... ij ={V ij1 V ij2 …V ijn};

[0036] V ij1 It is the current land use type, V ij1 ∈{0,1,2,3…,9}, representing bare land, natural grassland, artificial grassland, natural shrubland, woodland, residential land, commercial and office land, industrial land, hard plaza and water body respectively;

[0037] V ij2 It represents the probability of vacant land, with a value range of 0 ≤ V. ij2 ≤1.

[0038] V ij3 This indicates whether green space needs to be added; 0 means no addition, and 1 means addition.

[0039] V ij4 This represents the type of green space addition, with 0 indicating natural green space and 1 indicating artificial green space.

[0040] V ij5 This refers to land conversion costs, including land acquisition costs C1 and land consolidation costs C2, based on the current land use type V. ij1 Based on the project's geographical location, each land use type corresponds to a land conversion cost. The land acquisition cost C1 for each type is C11, C12...C110, and the land consolidation cost C2 is C21, C22...C210.

[0041] Vij6 It is the cost of green space construction, according to V ij4 The types of green space additions are determined, and the construction costs of natural green spaces and artificial green spaces are C31 and C32, respectively;

[0042] V ij7 This is the total cost of green space construction, and its calculation formula is:

[0043]

[0044] V ij8 This is the potential value for green space development, calculated using the following formula:

[0045] V ij8 =V ij2 *(1-V ij5 ′)

[0046] Among them, V ij5 ′ is the standardized V after the maximum and minimum values. ij5 The value;

[0047] Determine the demand level of each planning objective for different land use types, and calculate the weight of each planning objective based on the grid proportion of land use types within the block unit:

[0048] Define the demand level for each land use type l for each planning objective m. in, <1, for each land use type, the sum of the demand degrees of all planning objectives is 1, that is Calculate the area proportion P of each land use type in the block unit. area_ln For the entire block unit, the weight W of the planning objective m m For all land use types L n The proportion of land occupied by a unit multiplied by its corresponding demand level The sum is calculated using the following formula:

[0049]

[0050] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of generating an initial population based on the initial database of the block using a genetic algorithm, and determining the optimization result of the green space structure of the block based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm, as follows:

[0051] Initialization of the green space supplementation population in the neighborhood: Determine the number N of individuals to be generated in the initial population. Based on the principle of genetic algorithm, generate an initial population containing N schemes on the basis of the initial database of the neighborhood. The principle is as follows: Determine the green space supplementation probability of each grid according to the green space construction potential value. Use random numbers to compare with the supplementation probability to determine whether to supplement the green space. Among them, the grids confirmed to be supplemented are randomly assigned the type of green space supplementation. Based on the results, update the green space construction cost and the total green space construction cost.

[0052] Green space supplementary cost accounting: Calculate the sum of the total cost of all grid construction for each individual in the initial population, exclude individuals with costs exceeding the budget, re-initialize the population to generate a new scheme, until the number of all individuals meeting the budget cost reaches N;

[0053] Fitness assessment of generated schemes: The coupling degree of each individual's sub-goals is calculated using the green space coupling degree calculation formula. The coupling degree of each sub-goal is multiplied by the corresponding planning goal weight to obtain the fitness of each individual.

[0054] Reward function calculation: Based on the principle of reinforcement learning, the reward function is calculated to guide the direction of change of the genetic algorithm. The calculation method is to compare the fitness of the new generation of individuals with the fitness of the best individual in the previous generation. If the fitness of each individual in the current generation is higher than the highest fitness among all individuals in the previous generation, a positive reward is given; otherwise, a negative reward is given. For the first operation, the fitness is compared with the fitness of the original state.

[0055] The selection, crossover, and mutation scheme iteration is as follows: Following a tournament selection method, random selection is performed in the initial population, choosing the individual with the highest candidate probability from each group as the next generation, until N individuals are selected. The candidate probability is directly proportional to fitness and the reward function. For the selected N individuals, 2 / N parent pairs are randomly generated by pairwise selection. The crossover probability is adjusted based on the sum of the rewards for each pair and compared with a random number to determine whether to perform a crossover operation. If a crossover operation is performed, the attribute fragments of the two parent individuals are randomly swapped to generate two offspring schemes, replacing the parent schemes. The crossover probability is negatively correlated with the sum of the reward functions of the parent individuals. The offspring individuals updated after the crossover operation are mutated. The mutation probability is adjusted based on the individual's reward value and compared with a random value to determine whether to change the attribute: whether to supplement the green space and the type of green space supplementation, and then mutate the attribute to generate a new scheme.

[0056] Supplementation cost calculation: Calculate the supplementation cost of the N offspring individuals after mutation, remove schemes that exceed the budget. If the number of schemes is less than N, return to the previous step and repeat the selection, crossover, and mutation until N individual schemes are generated. At this point, the N generated schemes are the individual population results of this round of iteration.

[0057] Optimization of the green space structure scheme in the neighborhood: Calculate the fitness F of the best individual in each generation of the population. If the change in fitness of the best individual in multiple consecutive generations is less than a pre-set threshold, the fitness is considered to have converged. Output the scheme of the last round, which is the optimal scheme.

[0058] Secondly, in order to achieve the above objectives, this invention discloses an intelligent optimization system for the structure of green spaces in urban blocks, comprising:

[0059] The model optimization module is used to obtain the planning area, delineate the units based on the planning area to obtain multiple street blocks, obtain the relevant indicator data of each street block, input the relevant indicator data of each street block into the pre-established regression model, and output the optimized regression model.

[0060] The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm.

[0061] The probability prediction module is used to calculate the importance of variables in the optimized regression model, perform weight correction and coupling degree calculation based on the importance of variables, and obtain the corrected weights and the coupling degree of the block unit; acquire vacant land images, perform raster recognition on the vacant land images and overlay them to obtain the vacant land expansion area, and perform land use raster vacancy probability prediction based on the optimized regression model to obtain the vacant land probability.

[0062] The structure optimization module is used to generate an initial database of blocks based on the probability of vacant land, generate an initial population from the initial database of blocks based on a genetic algorithm, and determine the structural optimization result of the green space of the blocks based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm.

[0063] In another aspect of the present invention, in order to achieve the above-mentioned objective, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor, and when the processor loads and executes the computer program, it employs a street green space structure intelligent optimization method as described above.

[0064] The beneficial effects of this invention are:

[0065] This invention can effectively overcome the shortcomings of strong subjectivity and randomness in the current construction of urban green spaces. By constructing a street block database, it enables the formulation of intelligent green space supplementation schemes in highly restricted urban environments. Based on the current land use types in the city, it determines the planning goals and needs of urban green spaces according to local conditions and generates optimized structures based on these needs. Furthermore, based on reinforcement learning and genetic algorithms, it integrates the above key steps to generate a series of optimized schemes in multiple iterations, providing diversified choices for urban green space planning. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0068] Figure 2 This is a flowchart illustrating the technical process of intelligently generating a scheme for optimizing the structure of green spaces in urban blocks based on reinforcement learning and genetic algorithms, as described in this invention.

[0069] Figure 3 This is a schematic diagram of the system structure of the present invention;

[0070] Figure 4 This is a diagram illustrating the calculation process for the probability of street vacancy in an embodiment of the present invention.

[0071] Figure 5 This is a partial initial individual graph generated based on a genetic algorithm in an embodiment of the present invention;

[0072] Figure 6 This is the final preferred individual map obtained from fitness evaluation in an embodiment of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Example 1:

[0075] like Figure 1 As shown, a method for intelligent optimization of the structure of green spaces in urban blocks includes the following steps:

[0076] Obtain the planning area, delineate units based on the planning area to obtain multiple street blocks, obtain relevant indicator data for each street block, input the relevant indicator data for each street block into a pre-established regression model, and output the optimized regression model.

[0077] The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm.

[0078] Specifically, in this embodiment, after determining the planning area, the study area is divided into square street-scale planning units, hereinafter referred to as street units, based on the actual situation and planning needs of the planning area. In this example, Hengyang City, Hunan Province, is selected as the planning area. Based on the general road spacing and street size in the planning area, the planning area is divided into 1km*1km street units.

[0079] Based on the planning guidance, select the target performance P{P1,P2,...,Pm}, and list the potential influencing factors E{E1,E2,En,...,En} based on the relationship between green space structure and built environment.

[0080] In this example, based on the coordination requirements for production, living and ecological spaces in the national land spatial planning, P1 living performance, P2 production performance and P3 ecological performance are proposed.

[0081] Furthermore, based on relevant literature, neighborhood activity, average housing price, and neighborhood green space are used to reflect P1 living performance, P2 production performance, and P3 ecological performance, respectively.

[0082] Furthermore, for neighborhood activity, the calculation method is the average of the daily cumulative population within a week in the study unit. The higher the value, the higher the living performance. For average house price, the calculation method is the average house price of all neighborhoods in the study unit. The higher the value, the higher the productivity performance. For three-dimensional green volume, the calculation method is the sum of the NDVI values ​​within the study unit. The higher the value, the higher the productivity performance.

[0083] Furthermore, based on previous research, the relationship between green space structure and the built environment includes three categories: scale, form, and spatial layout. A series of representative indicators were selected as potential influencing factors. In this example, for scale, per capita green space area (E1), per capita natural green space area (E2), and per capita artificial green space area (E3) were selected; for form, artificial green space synergistic entropy (E4) and natural green space synergistic entropy (E5) were selected; for spatial layout, green space 5-minute service coverage rate (E6), average distance to the nearest green space (E7), average distance to the nearest artificial green space (E8), and average distance to the nearest natural green space (E9) were selected.

[0084] The specific calculation method is shown in Table 1 below:

[0085] Table 1. Potential Influencing Factors and Corresponding Indicators and Calculation Methods

[0086]

[0087]

[0088] Based on the target performance and potential influencing factors, data on the current year and the previous year of the regional planning were collected. For each block unit in each period, the corresponding target performance indicators and potential influencing factor indicators were calculated. The current year's indicator was subtracted from the previous year's indicator to obtain the target performance change value {ΔP} and the potential influencing factor indicator change value {ΔE}. Each block unit i contains {ΔP, ΔE}. Correlation analysis was performed on all potential influencing factor indicator changes to eliminate influencing factors with multicollinearity. The remaining indicators of all block units were then aggregated into a database I{ΔP, ΔE}.

[0089] In this example, considering data availability and urban changes, 2024 and 2019 were selected as target years, and data for these two years were collected. Neighborhood activity data came from real-time population data from Baidu Heatmap, average house prices came from housing price data platforms such as Lianjia and Anjuke, neighborhood green space data came from NDVI data, green space distribution data and building distribution data came from remote sensing imagery data, and the number of people in each neighborhood unit came from census data.

[0090] Using the change in target performance (ΔP) as the dependent variable and the change in potential influencing factor indicators (ΔE) as the independent variable, a regression model M is constructed for each target performance. The regression model with the best explanatory power is output based on the evaluation indicators. The importance of each independent variable is calculated.

[0091] In this example, a regression model is constructed using the random forest algorithm. In the random forest regression model, the database I is randomly divided into a training set and a test set in an 8:2 ratio. The training set is used as the training set, and the test set is used to evaluate the training results.

[0092] The importance of independent variables can be extracted using both principal component regression and least squares regression. The random forest model is chosen because it has the following advantages:

[0093] 1. Stability: Random forests integrate multiple decision trees, which has a good resistance to overfitting. Least squares and principal component regression are prone to overfitting when the amount of data is small or there are too many features, which leads to a decrease in the model's generalization ability on the test set.

[0094] 2. Scalability: Principal component regression and least squares regression have high computational complexity and long training time when dealing with large-scale data, and the implementation of parallel computing is relatively complicated. Their efficiency and scalability are not as good as the random forest regression algorithm.

[0095] 3. Adapting to high-dimensional data: Random forests use random feature selection during the construction of decision trees to automatically screen out important features and reduce the problem of collinearity between features; Principal component regression requires principal component analysis to reduce dimensionality before processing data, which may result in the loss of original data; Least squares regression is unstable in high-dimensional data and in the presence of multicollinearity.

[0096] During training, the training set is used for training, and the performance of the random forest regression model is adjusted by adjusting the number of decision trees (n_estimators), the maximum depth of decision trees (max_depth), the minimum number of samples for node splits (min_samples_split), the minimum number of samples for leaf nodes (min_samples_leaf), and the number of features for node splits (max_features).

[0097] The basic formula for the random forest regression algorithm is:

[0098]

[0099] Where E = {E1, E2, ... E} n Let} be a feature vector containing the change values ​​of n potential influencing indicators, M be the number of decision trees, and T be the number of trees. m (X) is the prediction value of the m-th decision tree for the input feature vector X. Each decision tree T... m (X) processes the input features according to its own tree structure and splitting rules, and gives the final predicted value.

[0100] In the actual training process, the RandomForestRegressor function of Python's scikitlearn library is directly called for calculation.

[0101] The test set was used for evaluation, and the explanatory power of the constructed regression model was assessed using metrics such as mean squared error (MSE), mean absolute error (RMSE), and R-squared (R2).

[0102] The formulas for calculating the mean squared error (MSE), mean absolute error (RMSE), and R-squared (R²) are as follows:

[0103]

[0104] In the formula, y i For the actual values ​​of the test set, These are values ​​calculated based on the random forest regression model. is the average value of the test set, and n is the number of samples in the test set.

[0105] In this example, it is assumed that the regression model has better explanatory power when R-squared is greater than 0.7 and the MSE and RMSE indices are relatively small, and the regression model M is output in this case.

[0106] The importance of variables is calculated for the optimized regression model. Based on the importance of variables, the weights are corrected and the coupling degree is calculated to obtain the corrected weights and the coupling degree of the block unit. Images of vacant land are obtained. After raster recognition of the vacant land images, the vacant land expansion area is obtained by overlaying. Based on the optimized regression model, the vacancy probability of land use raster is predicted to obtain the vacant land probability.

[0107] Based on the output regression model M, the importance of independent variables is calculated using variable importance measures (VIM).

[0108] For each independent variable e, the Variable Importance Score (VIMe) is calculated as follows:

[0109]

[0110] In the formula, M represents the number of decision trees in the output regression model M, and T... m It is the set of all nodes in the m-th decision tree, ΔMSE t,e It is the amount of impurity reduction caused by feature e at node t.

[0111] In this example, using Python tools, the forest.feature_importances function in the sklearn package can be called to directly calculate the importance of the independent variables.

[0112] For each target performance model, variables with an independent variable importance (VIM) greater than a certain threshold are selected. These VIM values ​​are used as initial weights (W), which are then adjusted. The adjusted weights W are then used as the final weights. ′ As the weight of the coupling degree model. For each block unit i, the coupling degree D for target J. ij The model formula is the sum of the products of variables greater than the threshold and their adjusted weights.

[0113]

[0114] Among them, D ij E represents the coupling degree of block unit i in target j. im W represents the value of variable m in block unit i for which the variable importance (VIM) is greater than a certain threshold.′ m The corrected weights.

[0115] The method for adjusting the weights is as follows:

[0116]

[0117] Among them, W m To determine the importance of independent variables (VIM) in the target performance model e Importance score (VIMe) of variable m that is greater than a certain threshold, where n is the total number of variables m.

[0118] In this example, the VIM threshold is set to 10%.

[0119] Based on the definition of vacant urban land and the actual conditions of the research site, the vacant land categories were determined, and the land use of each grid in the current year of the planning status of the research area and the previous year was identified.

[0120] In this example, based on the definition of vacant land—that is, completely vacant land within an urban area without any development purpose, as well as land abandoned, vacant, or underutilized from construction use—and considering the actual conditions of the study site, the categories of vacant land include: bare land, natural grassland, natural shrubland, and inefficient construction land. In other study areas, vacant land may also include areas under construction, natural wetlands, etc.

[0121] The grid size for land use identification, vacancy probability prediction, land acquisition cost assessment, and land acquisition potential assessment should be determined based on research needs. A grid size with a side length of 10m-20m is recommended. In this example, a side length of 20m is used as the grid size; however, the appropriate size should be selected based on the specific requirements of different research sites.

[0122] Furthermore, differentiated identification methods are adopted for different types of vacant land characteristics. For bare land, natural grassland, and natural shrubland, machine learning algorithms are used to interpret remote sensing imagery to obtain urban land use distribution maps, and then the types of bare land, natural grassland, and natural shrubland are extracted. For inefficient construction land, remote sensing imagery is first interpreted to obtain the construction land distribution area, and then inefficient construction land areas are extracted using inefficient land use assessment indicators. After completing land use identification, the land use type for each raster is obtained using mode resampling.

[0123] In this example, a Level 17 Google Maps remote sensing image with a spatial resolution of 2.15m was selected. Other high-resolution data such as Esri World Image, Tianditu, Gaofen series, GeoEye, Quickbird, and Ikonos can also be used. ArcGIS Pro image classification tools were used, with 50 samples per type, evenly distributed within the target area. Support Vector Machine (SVM) was chosen as the classifier. Random Forest, Maximum Likelihood, or other machine learning or deep learning algorithms with image recognition and classification capabilities, such as VGG-16 or U-net, can also be used. The accuracy evaluation tool was used to assess the recognition results. In this example, the output model achieved an accuracy of over 80% in land use type identification, recognizing 10 types: bare land, natural grassland, artificial grassland, natural shrubland, woodland, residential land, commercial / office land, industrial land, hard plazas, and water bodies. The land use data was resampled to raster size using the mode resampling method. The distribution areas of bare land, natural grassland, and natural shrubland types were extracted.

[0124] Extract construction land lattice data for residential, commercial / office, and industrial land. For each type of land (residential, commercial / office, and industrial), use land inefficiency assessment indicators to evaluate its inefficiency. Land with inefficiency below a certain threshold is selected and included in the vacant land category. The inefficiency calculation method is as follows: set n evaluation indicators (X1, X2, ... X...). n For this type of land use grid, the inefficiency of the grid is the grid's evaluation index value (x1, x2, ... x). n The equal-weighted superposition of ) is expressed as follows:

[0125]

[0126] In the formula, C represents the inefficiency of the construction land type grid, and n represents the number of evaluation indicators;

[0127] In this example, the evaluation indicators for residential land include road intersection density, service facility density, and nighttime light index. Road intersection data is from the OSM road network, with intersections extracted using ArcGIS. Service facility density is from Amap (Gaode Maps), selecting POIs categorized as "catering services, shopping services, living services, sports and leisure services, and healthcare services." The nighttime light index is from NPP / VIIRS data. The evaluation indicators for commercial land include popularity index and commercial facility density. Popularity index is from Baidu heatmap, and commercial facility density data is from Amap. The evaluation indicators for industrial land include building density and GDP. Building density is from remote sensing satellite imagery, and GDP data is from the Resource and Environmental Science Data Platform. After completing the inefficiency assessment for each land use type, the inefficiency was normalized, and the units ranking in the bottom 10% for each type of inefficiency evaluation were selected as inefficient construction land.

[0128] The locations of the two vacant land parcels were superimposed to obtain the areas of vacant land expansion. A regression algorithm was used to construct the relationship between driving factors and vacant land expansion. For grid i under study, its driving factors constitute an n-dimensional vector. All driving factors were normalized. The study grids for expanding vacant land were assigned a value of 1, while the other study grids were assigned a value of 0. A regression relationship was constructed between an n-dimensional vector and the expansion of vacant land.

[0129] In this example, the "Erase" tool in ArcGIS is used to output the locations of vacant land expansions. The normalized driving factors are then converted into raster layers.

[0130] Furthermore, in this example, referring to relevant research on the evolution mechanism of construction land change in shrinking cities, the following factors were selected as driving factors: slope (x1), water sensitivity (x2), distance from the city center (x3), commercial facility density (x4), transportation facility density (x5), housing price (x6), population density (x7), and hard surface ratio (x8).

[0131] The regression algorithm used can be random forest regression, logistic regression, or artificial neural network. This yields m factors {x1, x2, ..., x} that have a significant impact on the expansion of vacant land. m} and its corresponding weights {W x1 W x2, …W xm}

[0132] In this example, considering the advantage of the random forest regression algorithm in accurately analyzing complex nonlinear relationships, the random forest regression algorithm is selected.

[0133] Besides the random forest regression algorithm, the Markov transformation matrix algorithm or artificial neural network algorithm can also be selected. However, the artificial neural network algorithm has a long training time and is prone to overfitting, making it unsuitable for conditions with limited data and difficult to explain the influence of different factors on the vacancy probability. The Markov transformation matrix algorithm predicts the conversion situation in the next period based on the conversion probability of different land use types to vacant land in the previous period. This prediction method ignores the influence of other factors besides land use factors on land vacancy. Therefore, in this example, the random forest regression algorithm is used to predict the future land vacancy status.

[0134] The regression model RF is as follows:

[0135]

[0136] Where X = {X1, X2, ... X} n Let} be a feature vector containing n driving factors, M be the number of final decision trees, and T be the number of decision trees. m The depth of (X), the number of node split samples, etc., are determined based on the training results.

[0137] An initial database of neighborhoods is generated based on the probability of vacant land use. An initial population is generated from the initial database of neighborhoods based on a genetic algorithm. Finally, the optimization result of the green space structure of the neighborhoods is determined by the fitness of the best individual in each generation of the initial population based on a reinforcement learning algorithm.

[0138] Specifically, the process of generating the initial database for the neighborhood is as follows:

[0139] First, construct an M×M matrix D0 corresponding to the street grid division, where each element V ij (1≤i≤M, 1≤j≤M) represents the grid cell in the i-th row and j-th column of the block.

[0140]

[0141] Each element V ij All are represented in the form of attribute vectors. For an element V containing n attributes... ij ={V ij1 V ij2 …V ijn}

[0142] In this example, each element contains 8 attributes. In other cases, attributes can be added as needed (such as terrain, slope, etc., to reflect the basic land use conditions), but at least the following attributes must be included, and the specific values ​​within each attribute can be adjusted as needed.

[0143] Specifically, V ij1 This refers to the current land use type, with data sourced from land use data interpreted from S31, V.ij1 ∈{0,1,2,3…,9} represents 10 land use types: bare land, natural grassland, artificial grassland, natural shrubland, woodland, residential land, commercial and office land, industrial land, hard plaza, and water body. In other application cases, the types should be adjusted according to the actual situation.

[0144] V ij2 This is the vacancy probability, data source: S33 land vacancy probability, with a value range of 0 ≤ V. ij2 ≤1.

[0145] V ij3 This indicates whether green space needs to be added; 0 means no addition and 1 means addition.

[0146] V ij4 This represents the type of green space addition, with 0 representing natural green space and 1 representing artificial green space.

[0147] V ij5 This refers to land conversion costs, which include land acquisition costs (C1) and land consolidation costs (C2). Based on the current land use type V... ij1 Based on the specific conditions of the project area, each land use type corresponds to a land conversion cost. Specifically, the land acquisition costs (C1) for 10 types of land, including bare land, natural grassland, artificial grassland, natural shrubland, woodland, residential land, commercial and office land, industrial land, hard plaza, and water body, are C11, C12...C110, respectively, and the land consolidation costs (C2) are C21, C22,...C210, respectively.

[0148] V ij6 It is the cost of green space construction, according to V ij4 The types of green space additions are determined, and the construction costs for natural green spaces and artificial green spaces are C31 and C32, respectively.

[0149] V ij7 This is the total cost of green space construction, and its calculation formula is:

[0150]

[0151] V ij8 This is the potential value for green space development, and its calculation formula is:

[0152] V ij8 =V ij2 *(1-V ij5 ′).

[0153] Among them, V ij5 ′ is the standardized V after the maximum and minimum values. ij5 The value of .

[0154] Figure 4This is a diagram illustrating the process of calculating the vacancy probability of a street block.

[0155] Determine the demand level of each planning objective for different land use types, and calculate the weight of each planning objective based on the grid proportion of land use types within the block unit.

[0156] First, define the demand level for each land use type l for each planning objective m. For example, for land use type l1, the demand for planning objective m1 is expressed as: For each land use type, the sum of the demand degrees of all its planning objectives is 1, that is... =1. Next, calculate the area proportion P of each land use type within the block unit. area_ln Therefore, for the entire block unit, the weight of the planning objective m is equal to the weight of all land use types L. n The proportion of land occupied by a unit multiplied by its corresponding demand level The sum of . Its calculation formula is:

[0157]

[0158] Specifically, the following embodiments further illustrate the solution of the present invention: The process of generating an initial population based on the initial database of the block using a genetic algorithm, and determining the optimization result of the green space structure of the block based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm, is as follows:

[0159] Based on the principles of genetic algorithms, an initial population (scheme set) D1 is generated from the initial neighborhood database (D0). The basic principle is based on the green space development potential value (V) of each element. ij8 Determine whether to supplement green space (i.e., determine V) ij3 Then determine the form of green space addition (i.e., V) ij4 Then update other corresponding attribute values. The specific steps are as follows:

[0160] Determine the number N of individuals (scheme) to generate the initial population. Each individual is an M×M matrix identical to the initial block database D0, where each element is a vector containing 8 attributes. In this example, the number of individuals (scheme) to generate the initial population is set to 30. Other numbers can be chosen in other schemes, but are typically set between 20 and 50.

[0161] Figure 5 These are some of the initial individuals (schemes) generated based on a genetic algorithm.

[0162] Based on the green space development potential value (V) of each element ij8 ), determine its green space replenishment probability (P) ijThis will determine whether to supplement green space (V) ij3 For element V ij The probability of it supplementing green space (P) ij The formula for calculating ) is:

[0163]

[0164] Where S represents the sum of the green space construction potential values ​​of all elements of the current individual, and its calculation formula is:

[0165]

[0166] Use a random number generator to generate a random number r between 0 and 1, if r ≤ P ij Then for element V ij Its attribute V ij3 Setting it to 1 indicates that green space will be added; otherwise, the current value (0) will be retained.

[0167] When green space is added (i.e., V) ij3 =1), use a random number generator to randomly generate a value of 0 or 1 and assign it to V. ij4 When no green space is added (i.e., V) ij3 =0)V ij4 It is 0.

[0168] According to the type of green space construction (V) ij4 Cost of updating green space construction (V) ij6 ), and according to Formula Update V ij7 .

[0169] Green space supplementary cost accounting

[0170] Calculate the value of each individual (scheme) D in the initial population (set of schemes) D1. 1_n Calculate its total optimization cost C D1n The calculation method is the total green space construction cost V of all elements in the current individual. ij7 The sum of the terms is calculated using the following formula:

[0171]

[0172] Compare the supplementary total cost with the budgeted cost C budge If C D1n ≤C budge If the generated individual (solution) is successful, then retain it; otherwise, exclude it and generate a new individual (solution). Repeat the above steps until the number of generated individuals reaches N.

[0173] Generate Scheme Fitness Evaluation

[0174] The coupling degree of each individual (scheme) is calculated using the formula for measuring the coupling degree of green space structure. The overall coupling degree is then calculated based on the planning baseline weights of the planning objectives, and the overall coupling degree is used as the fitness F of that individual (scheme). The calculation formula is as follows:

[0175]

[0176] Among them, D j The coupling degree of target j is calculated using the structural coupling degree measurement model, W. j The planning weights are the sub-objective j;

[0177] Reward function calculation

[0178] A reward function is calculated to guide the direction of the genetic algorithm's changes. The reward value is calculated based on the changes in fitness between generations. For example, if the fitness of each individual in the current generation is higher than the highest fitness among all individuals in the previous generation, a positive reward is given; if the fitness decreases, a negative reward is given. The reward value is calculated as follows. For the first operation, a comparison is made with the original state.

[0179] R = α × (F current -F previosu_best )

[0180] Where α is a positive zero adjustment coefficient constant, F current For the fitness of the current generation of individuals, F previosu_best F represents the highest fitness among all individuals in the previous generation, and is calculated during the first reward function calculation. previosu_best The fitness of the initial state of the block.

[0181] Iterative solutions based on selection, crossover, and mutation

[0182] Random selection is performed using a "tournament selection" method. Each time, K individuals (K < N, typically 3-5) are randomly selected from the population to form a tournament group. The individual with the highest candidate probability is selected for the next generation of the population, and this process continues until N individuals are selected. The candidate probability P of individual k is... k The calculation formula is as follows:

[0183]

[0184] In a tournament group of size K, F k The fitness of individual k is calculated as R. k Let be the reward value for individual k, derived from the reward function. β is an adjustment constant used to control the influence of the reward coefficient.

[0185] After selecting a population of N individuals, randomly pairwise select parent individuals to form N / 2 parent pairs. The sum of the rewards R for each pair of individuals (the two individuals participating in the crossover) is then used to determine the next pair. sum To adjust the crossover probability P C The calculation formula is as follows:

[0186]

[0187] Among them, R sum The sum of the rewards for the two individuals participating in the crossover. It is the sum of the highest reward values ​​among all individual pairs, and δ is a constant adjustment coefficient.

[0188] Generate a random number r, if r ≤ P C Perform crossover operations on individuals; otherwise, simply retain the individual.

[0189] The crossover operation involves randomly selecting n distinct crossover points, dividing the individual into n+1 parts, with each new offspring consisting of its parent. For example, offspring C might consist of the first segment of parent A, the second segment of parent B, the third segment of parent A, and the fourth segment of parent B; conversely, offspring D might consist of the first segment of parent B, the second segment of parent A, the third segment of parent B, and the fourth segment of parent A. This operation is performed on N / 2 groups of parent individuals, with each group generating two offspring, resulting in a total of N offspring.

[0190] For the N offspring individuals that have undergone crossover, a mutation operation is performed. An initial mutation probability P is set. m0 And adjust the mutation probability P based on the individual's reward value. m0 The adjustment method is as follows:

[0191]

[0192] Among them, P m0 R is the initial mutation probability. j It is the reward value for individual j. It represents the highest reward value among offspring individuals. If an individual's reward value is low, its mutation probability is increased; if an individual's reward value is high, its mutation probability is decreased.

[0193] For each element of each individual, generate a random number r1 between 0 and 1. If r1 ≤ P m0 ′, then V ij3 Perform a 0 / 1 transition. Then generate a random number r2 between 0 and 1. If r2 ≤ P m0 ′and V ij3 =1, then V ij4 Perform a 0 / 1 conversion. Recalculate V based on this new value. ij7.

[0194] Based on the calculated cost of supplementing the mutated individuals, schemes exceeding the budget are removed. If the number of schemes is less than N, individuals are selected, crossovered, and mutated again until N individual schemes are generated. At this point, the N generated schemes represent the individual population results for this round of iteration.

[0195] Preferred Scheme for Green Space Structure in Urban Blocks

[0196] Calculate the fitness F of the best individual in each generation of the population. If the change in fitness of the best individual is less than a pre-set threshold over several consecutive generations (e.g., 20 generations), then the fitness is considered to have converged. Output the solution for the last round, which is the optimal solution.

[0197] Figure 6 The final preferred individual (scheme) obtained based on fitness assessment.

[0198] Example 2: Second aspect, such as Figure 3 As shown, in order to achieve the above objectives, this invention discloses an intelligent optimization system for the structure of urban green spaces, comprising:

[0199] Model optimization module 11 is used to obtain the planning area, delineate units based on the planning area to obtain multiple street blocks, obtain relevant indicator data for each street block, input the relevant indicator data for each street block into the pre-established regression model, and output the optimized regression model.

[0200] The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm.

[0201] The probability prediction module 12 is used to calculate the importance of variables in the optimized regression model, perform weight correction and coupling degree calculation based on the importance of variables, and obtain the corrected weight and the coupling degree of the block unit; acquire vacant land images, perform raster recognition on the vacant land images and overlay them to obtain the vacant land expansion area, and perform land use raster vacancy probability prediction based on the optimized regression model to obtain the vacant land probability.

[0202] The structure optimization module 13 is used to generate an initial database of blocks based on the probability of vacant land, generate an initial population from the initial database of blocks based on a genetic algorithm, and determine the structural optimization result of the green space of the blocks based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm.

[0203] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0204] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0205] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0206] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A method for intelligent optimization of the structure of green spaces in urban blocks, characterized in that, The method includes the following steps: Obtain the planning area, delineate units based on the planning area to obtain multiple street blocks, obtain relevant indicator data for each street block, input the relevant indicator data for each street block into a pre-established regression model, and output the optimized regression model. The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm. The importance of variables is calculated for the optimized regression model. Based on the importance of variables, the weights are corrected and the coupling degree is calculated to obtain the corrected weights and the coupling degree of the block unit. Images of vacant land are obtained. After raster recognition of the vacant land images, the vacant land expansion area is obtained by overlaying. Based on the optimized regression model, the vacancy probability of land use raster is predicted to obtain the vacant land probability. An initial database of neighborhoods is generated based on the probability of vacant land use. An initial population is generated from the initial database of neighborhoods based on a genetic algorithm. Finally, the optimization result of the green space structure of the neighborhoods is determined by the fitness of the best individual in each generation of the initial population based on a reinforcement learning algorithm.

2. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The process of obtaining relevant indicator data for each street block unit: Calculate the corresponding target performance indicators and potential influencing factor indicators for each block unit in each period. Subtract the previous year's indicator from the current year's indicator to obtain the target performance change value {ΔP} and the potential influencing factor indicator change value {ΔE}. For each block unit i, {ΔP, ΔE} are included. Perform correlation analysis on all potential influencing factor indicator change values, eliminate influencing factors with multicollinearity, and summarize the remaining indicators of all block units into a database I{ΔP, ΔE}.

3. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The formula for the random forest regression algorithm is as follows: in, Let M be a feature vector containing the change values ​​of n potential influencing indicators, and M be the number of decision trees. It is the prediction value of the m-th decision tree for the input feature vector X. Each decision tree... The system processes the input features based on its own tree structure and splitting rules, and then gives the final predicted value. The evaluation metrics for the regression model include: mean squared error (MSE), mean absolute error (RMSE), and R-squared. 2 The formula is as follows: In the formula, For the actual values ​​of the test set, These are values ​​calculated based on the random forest regression model. is the average value of the test set, and n is the number of samples in the test set.

4. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The importance of variables is calculated for the optimized regression model. The calculation process: In the formula, M represents the number of decision trees in the output regression model M. It is the set of all nodes in the m-th decision tree. It is the amount of impurity reduction caused by feature e at node t; The process of adjusting weights and calculating coupling degree based on variable importance: For each target performance model, select the variable importance (VIM) of variable m. m Variables exceeding a preset threshold are assigned a variable importance (VIM) based on variable m. m As the initial weight W, and then correcting the initial weight W to obtain the corrected weight. As a weight in the coupling model, for each block unit i, the coupling degree with respect to target j is... The model formula is the sum of the products of variables greater than the threshold and their adjusted weights; in, Let be the coupling degree of block unit i in target j. The importance of variable m (VIM) m The value of variable m in block unit i that is greater than the threshold. The corrected weights; The method for adjusting the weights is as follows: ； in, VIM for variable importance of variable m in the target performance model m Importance score for variable m that exceeds the threshold, where n is the total number of variables m.

5. The method for intelligent optimization of the structure of green space in a street block according to claim 4, characterized in that, The optimized regression model The expression is as follows: in, Given a feature vector containing n driving factors, and M being the final number of decision trees, the decision tree... The depth and the number of node split samples are determined based on the training results.

6. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The vacant land refers to land within the city that is completely vacant and has no development purpose, as well as land that has been abandoned, is vacant, or is underutilized from its construction purpose. The categories of vacant land include bare land, natural grassland, natural shrubland, inefficient construction land, areas under construction, and natural wetlands.

7. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The process of generating the initial database for the neighborhood is as follows: Constructing a grid system corresponding to the street blocks matrix Each element Let represent the grid cell in the i-th row and j-th column of the block, where ; Each element All are represented in the form of attribute vectors. For an element containing n attributes... ; It is the current land use type. These represent bare land, natural grassland, artificial grassland, natural shrubland, woodland, residential land, commercial and office land, industrial land, hard plazas, and water bodies, respectively. It represents the probability of vacant land use, with a value range of [value missing]. ; This indicates whether green space needs to be added; 0 means no addition, and 1 means addition. This represents the type of green space addition, with 0 indicating natural green space and 1 indicating artificial green space. This refers to land conversion costs, including land acquisition costs (C1) and land consolidation costs (C2), based on the current land use type. Based on the project's geographical location, each land use type corresponds to a land conversion cost. The land acquisition cost C1 for each type is C11, C12...C110, and the land consolidation cost C2 is C21, C22...C210. It is the cost of green space construction, according to The types of green space additions are determined, and the construction costs of natural green spaces and artificial green spaces are C31 and C32, respectively; This is the total cost of green space construction, and its calculation formula is: This is the potential value for green space development, calculated using the following formula: = (1- ) in, It is the standardized maximum and minimum values. The value; Determine the demand level of each planning objective for different land use types, and calculate the weight of each planning objective based on the grid proportion of land use types within the block unit: Define the demand level for each land use type l for each planning objective m. ,in, For each land use type, the sum of the demand for all planning objectives is 1, that is... Calculate the area ratio of each land use type in the block unit. For the entire block unit, the weight of the planning objective m For all land use types The proportion of land occupied by a unit multiplied by its corresponding demand level The sum is calculated using the following formula: 。 8. The method for intelligent optimization of the structure of green space in a street block according to claim 1, characterized in that, The process of generating an initial population from the initial database of the neighborhood based on a genetic algorithm, and determining the optimized green space structure of the neighborhood based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm, is as follows: Initialization of the green space supplementation population in the neighborhood: Determine the number N of individuals to be generated in the initial population. Based on the principle of genetic algorithm, generate an initial population containing N schemes on the basis of the initial database of the neighborhood. The principle is as follows: Determine the green space supplementation probability of each grid according to the green space construction potential value. Use random numbers to compare with the supplementation probability to determine whether to supplement the green space. Among them, the grids confirmed to be supplemented are randomly assigned the type of green space supplementation. Based on the results, update the green space construction cost and the total green space construction cost. Green space supplementary cost accounting: Calculate the sum of the total cost of all grid construction for each individual in the initial population, exclude individuals with costs exceeding the budget, re-initialize the population to generate a new scheme, until the number of all individuals meeting the budget cost reaches N; Fitness assessment of generated schemes: The coupling degree of each individual's sub-goals is calculated using the green space coupling degree calculation formula. The coupling degree of each sub-goal is multiplied by the corresponding planning goal weight to obtain the fitness of each individual. Reward function calculation: Based on the principle of reinforcement learning, the reward function is calculated to guide the direction of change of the genetic algorithm. The calculation method is to compare the fitness of the new generation of individuals with the fitness of the best individual in the previous generation. If the fitness of each individual in the current generation is higher than the highest fitness among all individuals in the previous generation, a positive reward is given; otherwise, a negative reward is given. For the first operation, the fitness is compared with the fitness of the original state. The selection, crossover, and mutation scheme iteration is as follows: Following a tournament selection method, random selection is performed in the initial population, choosing the individual with the highest candidate probability from each group as the next generation, until N individuals are selected. The candidate probability is directly proportional to fitness and the reward function. For the selected N individuals, 2 / N parent pairs are randomly generated by pairwise selection. The crossover probability is adjusted based on the sum of the rewards for each pair and compared with a random number to determine whether to perform a crossover operation. If a crossover operation is performed, the attribute fragments of the two parent individuals are randomly swapped to generate two offspring schemes, replacing the parent schemes. The crossover probability is negatively correlated with the sum of the reward functions of the parent individuals. The offspring individuals updated after the crossover operation are mutated. The mutation probability is adjusted based on the individual's reward value and compared with a random value to determine whether to change the attribute: whether to supplement the green space and the type of green space supplementation, and then mutate the attribute to generate a new scheme. Supplementation cost calculation: Calculate the supplementation cost of the N offspring individuals after mutation, remove schemes that exceed the budget. If the number of schemes is less than N, return to the previous step and repeat the selection, crossover, and mutation until N individual schemes are generated. At this point, the N generated schemes are the individual population results of this round of iteration. Optimization of the green space structure scheme in the neighborhood: Calculate the fitness F of the best individual in each generation of the population. If the change in fitness of the best individual in multiple consecutive generations is less than a pre-set threshold, the fitness is considered to have converged. Output the scheme of the last round, which is the optimal scheme.

9. A smart optimization system for the structure of green spaces in urban blocks, characterized in that, include: The model optimization module is used to obtain the planning area, delineate the units based on the planning area to obtain multiple street blocks, obtain the relevant indicator data of each street block, input the relevant indicator data of each street block into the pre-established regression model, and output the optimized regression model. The relevant indicator data for each block unit includes the corresponding target performance indicators and potential influencing factor indicators for each block unit, and the regression model is constructed based on the random forest regression algorithm. The probability prediction module is used to calculate the importance of variables in the optimized regression model, perform weight correction and coupling degree calculation based on the importance of variables, and obtain the corrected weights and the coupling degree of the block unit; acquire vacant land images, perform raster recognition on the vacant land images and overlay them to obtain the vacant land expansion area, and perform land use raster vacancy probability prediction based on the optimized regression model to obtain the vacant land probability. The structure optimization module is used to generate an initial database of blocks based on the probability of vacant land, generate an initial population from the initial database of blocks based on a genetic algorithm, and determine the structural optimization result of the green space of the blocks based on the fitness of the best individual in each generation of the initial population using a reinforcement learning algorithm.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on a processor. When the processor loads and executes the computer program, it implements a method for intelligent optimization of the structure of green space in a neighborhood as described in any one of claims 1 to 8.

Images