Method and device for determining land planning scheme, and nonvolatile storage medium

Abstract

The invention discloses a land planning scheme determination method and device and a nonvolatile storage medium. The method comprises the steps that related data of target geographic grids of a to-be-planned area are acquired from a database, and each target geographic grid represents a sub-area contained in the to-be-planned area; the related data of the multiple target geographic grids are processed and analyzed through a multi-target optimization model, a prediction planning scheme of the to-be-planned area is obtained, the prediction planning scheme comprises the land type of each sub-area, and in the training process, each conversion scheme corresponding to each land type is scored according to a multi-target reward function. The technical problem that a land utilization planning method adopted in the related technology cannot output a planning scheme meeting various preset constraint conditions at the same time is solved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method and apparatus for determining land planning schemes, and a non-volatile storage medium. Background Technology

[0002] In related technologies, land use planning often relies on rule-driven models or professional manual formulation of plans. However, the above-mentioned methods for implementing land use planning have the problem of being unable to balance multiple objectives, and the output planning scheme cannot simultaneously meet multiple pre-set constraints.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method and apparatus for determining land planning schemes, as well as a non-volatile storage medium, to at least solve the technical problem that land use planning methods used in related technologies cannot output planning schemes that simultaneously meet multiple pre-set constraints.

[0005] According to one aspect of the embodiments of this application, a method for determining a land planning scheme is provided, comprising: obtaining relevant data of target geographic grids of a region to be planned from a database, wherein each target geographic grid represents a sub-region contained in the region to be planned, and the relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region, the attribute information including: ecological information and land application information; processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the region to be planned, wherein the predicted planning scheme includes: the land type of each sub-region, the multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules, the preset conversion rules are used to indicate multiple conversion schemes supported by each land type, the conversion schemes are used to indicate the conversion of the land type to another land type, and during the training process, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0006] Optionally, the target optimization model is iteratively trained using the following method: At each iteration, the loss function of the target optimization model is updated. If the loss function converges, the iteration stops; if the loss function fails to converge, the update continues. Updating the loss function includes: for each known planning area, the policy network in the target optimization model determines the possible planning schemes corresponding to the known planning area based on the land types included in the known planning area and preset conversion rules. The possible planning schemes record: multiple geographic grids obtained by dividing the known planning area, the planned land type of each geographic grid, and the planned land type is one of several other land types that the land type supports conversion from. One approach involves the value network in the objective optimization model evaluating the available planning schemes for a known planning area from multiple dimensions based on a multi-objective reward function. These multi-dimensional scores include: an ecological reward score indicating the ecological value of the known planning area; a spatial compactness score indicating the land use density of the known planning area; an economic score indicating the planning cost of the known planning area; and a difference score indicating the difference between land types before and after applying the available planning schemes. The model parameters and loss function of the multi-objective optimization model are updated based on these multi-dimensional scores. The model parameters include: the weights of the value network, the weights of the policy network, and the weights of each type of objective reward function.

[0007] Optionally, the ecological incentive score is determined by the following method: All geographic grids constituting the known planning area are classified into Category I and Category II geographic grids based on the planned land type corresponding to each geographic grid. Category I geographic grids are those whose corresponding planned land type belongs to an ecological type, and Category II geographic grids are those whose corresponding planned land type does not belong to an ecological type. The total supply capacity of multiple Category I geographic grids and the total demand capacity of multiple Category II geographic grids are determined. The ecological incentive score of the known planning area is determined based on the total supply capacity, total demand capacity, preset spatial attenuation coefficient, distance between every two Category I and Category II geographic grids, relevant data of Category I and Category II geographic grids, and relevant data of Category II geographic grids. The relevant data of Category I geographic grids includes: the vegetation index and area of ​​each Category I geographic grid; the relevant data of Category II geographic grids includes: the population density and area of ​​each Category II geographic grid.

[0008] Optionally, the spatial compactness score is determined by the following method: classifying geographic grids with planned land type as urban construction land into Category III geographic grids, geographic grids with planned land type as cultivated land into Category IV geographic grids, and geographic grids with planned land type as ecological land into Category V geographic grids; for Category III geographic grids, determining the first spatial compactness score of the first region composed of all Category III geographic grids based on the concentrated distribution of multiple Category III geographic grids and the urban construction land compatibility coefficient; for Category IV geographic grids, determining the second spatial compactness score of the second region composed of all Category IV geographic grids based on the concentrated distribution of multiple Category IV geographic grids and the cultivated land compatibility coefficient; for Category V geographic grids, determining the third spatial compactness score of the third region composed of all Category V geographic grids based on the concentrated distribution of multiple Category V geographic grids and the continuity of multiple Category V geographic grids; and determining the spatial compactness score of the known planned area based on the first, second, and third spatial compactness scores.

[0009] Optionally, a first spatial compactness score for a first region composed of all third-type geographic grids is determined based on the concentrated distribution of multiple third-type geographic grids and the urban construction land compatibility coefficient. This includes: determining a first spatial compactness bonus for the first region based on the planned land type corresponding to each third-type geographic grid and the planned land type corresponding to each adjacent geographic grid of the third-type geographic grid; determining a first compatibility bonus for the first region based on multiple urban construction land compatibility coefficients corresponding to multiple first-type geographic grid groups, wherein each first-type geographic grid group contains two adjacent third-type geographic grids; determining a first spatial compactness score based on the first spatial compactness bonus and the first compatibility bonus; and a second spatial compactness score for a second region composed of all fourth-type geographic grids is determined based on the concentrated distribution of multiple fourth-type geographic grids and the cultivated land compatibility coefficient. This includes: determining a first spatial compactness score for the second region based on the planned land type corresponding to each fourth-type geographic grid and the planned land type corresponding to each adjacent geographic grid of the fourth-type geographic grid. The second spatial compactness reward is determined based on the farmland compatibility coefficients of multiple second geographic grid groups, where each second geographic grid group contains two adjacent fourth-type geographic grids. The second spatial compactness score is determined based on the second spatial compactness reward and the second compatibility reward. The third spatial compactness score of the third region, composed of all fifth-type geographic grids, is determined based on the concentrated distribution and continuity of multiple fifth-type geographic grids, including: determining the third spatial compactness reward of the third region based on the planned land type corresponding to each fifth-type geographic grid and the planned land type corresponding to each adjacent fifth-type geographic grid; grouping multiple adjacent fifth-type geographic grids into a set, and determining the ecological stability reward of the third region based on the number of sets, the number of fifth-type geographic grids in each set, and a preset ecological reward coefficient; and determining the third spatial compactness score of the third region based on the third spatial compactness reward and the ecological stability reward.

[0010] Optionally, the economic score is determined by the following method: the economic score is determined based on the distance from multiple geographic grids to the central geographic grid, the preset payment amount, and the traffic cost sensitivity coefficient corresponding to each geographic grid, wherein the central geographic grid is the grid representing the center of the known planning area.

[0011] Optionally, the difference score is determined by the following method: classifying geographic grids whose original land type is the same as the planned land type into the sixth category of geographic grids, and classifying geographic grids whose original land type and the planned land type belong to the same land application classification into the seventh category of geographic grids. Here, the original land type is the land type before the application of the optional planning scheme in the known planning area, and each land application classification contains multiple land types; determining the first number of sixth category geographic grids and the second number of seventh category geographic grids among all geographic grids constituting the known planning area; and determining the difference score based on the first preset difference reward and the first number corresponding to the sixth category geographic grids, and the second preset difference reward and the second number corresponding to the seventh category geographic grids.

[0012] Optionally, the model parameters of the multi-objective optimization model are updated based on the multi-dimensional scores, including: when the initial reward value is 0, the comprehensive score is determined based on the multi-dimensional scores and the multiple weight coefficients corresponding to the multi-dimensional scores, wherein the comprehensive score is used to guide the updating of model parameters, and the initial reward value is determined based on the multi-dimensional scores calculated for the first time; when the initial reward value is not 0, the comprehensive score after normalization using the initial reward value is determined as the comprehensive score.

[0013] Optionally, before processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model, the method includes: for non-numerical type first-class data, converting the first-class data into corresponding codes according to preset coding rules; for numerical type second-class data, normalizing the second-class data to obtain normalization results; generating spatial feature vectors for each target geographic grid based on the relative position information of the target geographic grids, and generating state vectors for the target geographic grids based on the spatial feature vectors, codes, and normalization results, wherein the relative position information of the target geographic grids is used to indicate the positional relationship between the target geographic grids and other target geographic grids contained in the area to be planned.

[0014] Optionally, after obtaining the predicted planning scheme, the method further includes: converting the predicted planning scheme into a visual diagram, wherein different land types are displayed in different colors in the visual diagram.

[0015] According to another aspect of the embodiments of this application, a land planning scheme determination apparatus is also provided, comprising: an acquisition module, configured to acquire relevant data of target geographic grids of a region to be planned from a database, wherein each target geographic grid represents a sub-region contained in the region to be planned, and the relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region, the attribute information including: ecological information and land application information; and a planning module, configured to process and analyze the relevant data of multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the region to be planned, wherein the predicted planning scheme includes: the land type of each sub-region, the multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules, the preset conversion rules are used to indicate multiple conversion schemes supported by each land type, and the conversion schemes are used to indicate the conversion of the land type to another land type, and during the training process, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0016] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, which stores a computer program, wherein the above-described method for determining a land planning scheme is executed by running the computer program on the device where the non-volatile storage medium is located.

[0017] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to execute the above-described method for determining a land planning scheme through the computer program.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, including computer instructions, which, when executed by a processor, implement the steps of the above-described method for determining a land planning scheme.

[0019] In this embodiment, relevant data of the target geographic grid of the area to be planned are obtained from a database. Each target geographic grid represents a sub-region within the area to be planned. The relevant data of each target geographic grid includes: geographic location information and attribute information of the sub-region. The attribute information includes: ecological information and land use information. A multi-objective optimization model is used to process and analyze the relevant data of multiple target geographic grids to obtain a predictive planning scheme for the area to be planned. Automatic planning of the area to be planned is achieved by using a multi-objective optimization model trained based on a multi-objective reward function to process the geographic grid data of the area to be planned. The predictive planning scheme output by the multi-objective optimization model includes: the soil of each sub-region... For land types, the multi-objective optimization model is trained based on a multi-objective reward function, known planning areas with multiple planning schemes, and preset conversion rules. The preset conversion rules indicate the multiple conversion schemes supported by each land type, and the conversion schemes indicate how to convert a land type to another land type. During training, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function, so that the output predicted planning scheme can simultaneously meet the constraints of the multi-objective reward function. This achieves the technical effect of outputting a land planning scheme that simultaneously meets multiple constraints, thereby solving the technical problem that the land use planning methods used in related technologies cannot output planning schemes that simultaneously meet multiple preset constraints. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 This is a hardware structure block diagram of a computer terminal for implementing a method for determining land planning schemes, according to an embodiment of this application.

[0022] Figure 2 This is a flowchart illustrating the steps of a method for determining a land planning scheme according to an embodiment of this application;

[0023] Figure 3 This is a land type conversion table according to an embodiment of this application;

[0024] Figure 4 This is a conversion condition limitation table according to an embodiment of this application;

[0025] Figure 5 It is a correlation table of related data of a geographic grid according to an embodiment of this application;

[0026] Figure 6This is a visual diagram of a predictive planning scheme for a planned area A according to an embodiment of this application;

[0027] Figure 7 This is a structural diagram of a land planning scheme determination device according to an embodiment of this application;

[0028] Figure 8 This is a flowchart of a land planning scheme determination device generating a land planning scheme according to an embodiment of this application. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] To better understand the embodiments of this application, the technical terms involved in the embodiments of this application are explained below:

[0032] Spatial heterogeneity: the phenomenon that environmental characteristics and physical properties differ at different locations in a geographic space.

[0033] A geographic grid is a plot of land in geographic space. It is a series of grid units with specific attribute information after the geographic space has been discretized. These grid units usually constitute the spatial representation of the corresponding geographic region.

[0034] In related technologies, land use planning is conducted through rule-driven models or manual methods. However, these two methods struggle to handle geographic grid data in areas with spatial heterogeneity, and it is difficult to ensure that multiple constraints (i.e., multiple objectives) are simultaneously met during land use planning. Therefore, there is a problem of not being able to output land planning schemes that simultaneously satisfy multiple constraints, and a difficulty in collaboratively optimizing for multiple constraints. To address this issue, this application provides relevant solutions, which are detailed below.

[0035] According to an embodiment of this application, a method embodiment for determining a land planning scheme is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0036] The methods and embodiments provided in this application can be executed on mobile terminals, computer terminals, or similar computing devices. Figure 1 A hardware block diagram of a computer terminal for implementing a method for determining land planning schemes is shown. Figure 1 As shown, the computer terminal 10 may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0037] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as a form of processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0038] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the land planning scheme determination method in this embodiment of the application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned land planning scheme determination method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0039] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0040] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0041] This application provides a method for determining a land planning scheme that can operate under the above-described operating environment. Figure 2 This is a flowchart illustrating the steps of a method for determining a land planning scheme according to an embodiment of this application, as shown below. Figure 2 As shown, the method includes the following steps:

[0042] Step S202: Obtain relevant data of the target geographic grid of the area to be planned from the database. Each target geographic grid represents a sub-region contained in the area to be planned. The relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region. The attribute information includes: ecological information and land application information.

[0043] The method provided in this application applies reinforcement learning to land planning. By setting various constraints and complex objectives in planning scenarios, it flexibly trains various model parameters of a multi-objective optimization model. Therefore, the trained multi-objective optimization model has multi-objective trade-off capabilities, self-learning capabilities, and spatial collaborative optimization capabilities. Using the trained multi-objective optimization model, automatic planning of the area to be planned can be achieved. In step S202, relevant data of the geographic grid (i.e., target geographic grid) of the area to be planned is obtained from the database. Each target geographic grid is a grid unit in the area to be planned. An area to be planned is divided into multiple grid units, and each grid unit is a sub-region (i.e., a sub-region) contained within the area to be planned. For example, when using a 1 km × 1 km grid unit to divide the area to be planned, the area of ​​each target geographic grid is 1 square kilometer. The target geographic grid data obtained in step S202 is information related to land planning and utilization. In this embodiment, the information related to land planning and utilization is classified into: plot location information and plot attribute information. Therefore, the target geographic grid data includes: the geographic location information of the sub-region (i.e., sub-region) represented by each target geographic grid, including: coordinates composed of longitude and latitude, boundary information of the sub-region, and attribute information of the sub-region (i.e., sub-region) represented by each target geographic grid. Each attribute information is used to describe the ecological information, land use information, and human... Information on various activities, including ecological information such as the National Distance Index (NDVI), and land use information such as land price (e.g., land price per square kilometer), road density (e.g., road density per square kilometer), population density (e.g., population density per square kilometer), and land use type. Land type describes the use of land in its corresponding area. In this embodiment, land use types include: urban construction land (land used for constructing residential, commercial, industrial, public management and service facilities, etc.), arable land (e.g., land used for growing crops), grassland, forest land, water area, and unused land.The relevant data for the aforementioned target geographic grid are typically recorded in publicly available databases. For example, land application information can be obtained by querying the Multi-Period Land Use Remote Sensing Monitoring Dataset (CNLUCC) and the Land Use Type Dataset (EULUC); population density can be obtained by querying the Global Population Distribution Dataset (LandScan Global); land price can be obtained by querying the land price monitoring website; vegetation index (NDVI) can be obtained by querying the vegetation index dataset; and road density can be obtained by querying the data stored in the road data dataset (Open Street Map). Therefore, the database used to provide the relevant data for the target geographic grid in step S202 contains at least all of the aforementioned datasets. Alternatively, the device / apparatus performing the method provided in this application embodiment can obtain the data by calling the browser through an interface connected to the browser. Or, the relevant data for the target geographic grid in step S202 can also be manually obtained from publicly available channels and then manually entered into the interactive interface of the device / apparatus performing the method provided in this application embodiment.

[0044] Step S204: A multi-objective optimization model is used to process and analyze the relevant data of multiple target geographic grids to obtain a predicted planning scheme for the area to be planned. The predicted planning scheme includes the land type of each sub-region. The multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules. The preset conversion rules are used to indicate multiple conversion schemes supported by each land type. The conversion schemes are used to indicate the conversion of the land type to another land type. During the training process, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0045] After obtaining the geographic data of the area to be planned (i.e., the relevant data of the target geographic grid) in step S202, in step S204, a multi-objective optimization model trained using reinforcement learning is used to process the geographic data of the area to be planned, resulting in a predicted planning scheme output by the multi-objective optimization model. The predicted planning scheme indicates how to apply the area to be planned, and it records the land type (i.e., land use type) of each target geographic grid (i.e., each sub-region) of the area to be planned. In this embodiment, during land planning, different sub-regions (i.e., sub-regions) within an area to be planned can be planned as different land types. The multi-objective optimization model used in step S204 is trained based on a multi-objective reward function, preset conversion rules, and a planning area (i.e., a known planning area) where multiple planning schemes are consistent. During training, the multi-objective optimization model learns how to make land type conversion decisions while satisfying multiple planning objectives. Therefore, the trained multi-objective optimization model can output a land planning scheme (i.e., a predicted planning scheme) for the area to be planned that satisfies the multi-objective trade-offs. During the training of the multi-objective optimization model, a conversion scheme is executed for each land type in the known planning area according to preset conversion rules. One conversion scheme indicates the conversion of this land type to another. After the conversion is completed, the conversion scheme is scored using a multi-objective reward function. The multi-objective reward function is a numerical signal form that multiple planning objectives (i.e., multiple constraints) can be understood by the multi-objective optimization model. The multi-objective reward function is generated by multiple objective reward functions, each of which is a function of the reward mechanism of a planning objective. The multi-objective reward function integrates all planning objectives through mathematical operations (such as weighted summation) to form a single reward signal, which guides the decision-making process of the multi-objective optimization model.

[0046] In step S204, the multi-objective optimization model can be loaded into memory. For example, the raw data of the multi-objective optimization model can be loaded from non-volatile memory into volatile memory so that the processor can run the multi-objective optimization model. The raw data of the multi-objective optimization model refers to unprocessed data, which typically includes the parameters and structural data of the multi-objective optimization model. The structural data can be the computational relationships based on the parameters, such as the forward propagation computational relationships between intermediate layers and between neurons. Specifically, the structural data can include the structure-related code of the multi-objective optimization model, such as code used to perform related calculations between intermediate layers and between neurons.

[0047] In one implementation, a region can be partitioned in memory for loading the multi-objective optimization model, which may include a structure data storage area and a parameter storage area. The structure data storage area stores structure-related code, and the parameters referenced by it can be accessed via pointers pointing to the addresses of specific parameters in the parameter storage area. During the training of the multi-objective optimization model, frequent parameter updates may be required; in this case, updating the parameter values ​​in the parameter storage area is sufficient.

[0048] Figure 3 It is a land type conversion table. Figure 4 This is a conversion condition constraint table. The preset conversion rules in this embodiment include: Figure 3 The conversion conditions defined in the land type conversion table shown are as follows: Figure 4 The transformation conditions defined in the transformation condition constraint table shown are as follows: Figure 3 As shown, the preset conversion rules define multiple other land types that each land type supports conversion to, such as... Figure 3 As shown, in the solutions provided in this application embodiment, areas with land type arable land can be converted to other land types including: urban construction land, rural settlements, and grassland; areas with land type forest land can be converted to other land types including: arable land and urban construction land; areas with land type grassland can be converted to other land types including: urban construction land and arable land; areas with land type unused land can be converted to other land types including: grassland, forest land, and urban construction land; areas with land type residential can be converted to other land types including: commercial, public management, and other urban construction; areas with land type commercial can be converted to other land types including: residential, public management, and other urban construction; areas with land type industrial can be converted to other land types including: commercial, park, and other urban construction; areas with land type park can be converted to other land types including: forest land, grassland, and public management; areas with land type public management can be converted to other land types including: commercial, park, and other urban construction; areas with land type other urban construction land can be converted to other land types including: arable land, grassland, commercial, residential, and public management. Figure 4 The preset conversion rules also define the ecological and transportation conditions that land type conversion should follow, such as... Figure 4As shown, other land types corresponding to this land type include parks. To convert a land type to a park, an NDVI ≥ 0.5 is required. Other land types corresponding to this land type include forest land. To convert a land type to forest land, an NDVI ≥ 0.8 is required. Other land types corresponding to this land type include arable land. To convert a land type to arable land, an NDVI ≥ 0.6 is required. Other land types corresponding to this land type include grassland. To convert a land type to grassland, an NDVI ≥ 0.7 is required. Other land types corresponding to this land type include residential / commercial / industrial land. To convert a land type to residential / commercial / industrial land, an NDVI ≤ 0.7 is required. Other land types corresponding to this land type include commercial land. To convert a land type to commercial land, a road network density of 6 is required. Other land types corresponding to this land type include industrial land. To convert a land type to industrial land, a road network density of 3 is required. Other land types corresponding to this land type include residential land. To convert a land type to residential land, a road network density of 4 is required. Other land types corresponding to this land type include public management land. To convert a land type to public management land, a road network density of 5 is required. Other land types corresponding to this land type include parks. To convert a land type to parks, a road network density of 3 is required.

[0049] According to some optional embodiments of this application, the target optimization model is iteratively trained using the following method: During each iteration, the loss function of the target optimization model is updated; if the loss function converges, the iteration stops; if the loss function does not converge, the loss function continues to be updated. Updating the loss function includes: for each known planning area, the policy network in the target optimization model determines the possible planning schemes corresponding to the known planning area based on the land types contained in the known planning area and preset conversion rules. The possible planning schemes record: multiple geographic grids obtained by dividing the known planning area, the planned land type of each geographic grid, and the planned land type being one of several other land types that support conversion. One of the land types; the value network in the objective optimization model scores the possible planning schemes corresponding to the known planning area from multiple dimensions according to the multi-objective reward function. The multi-dimensional scores include: ecological reward score indicating the ecological value of the known planning area, spatial compactness score indicating the land use density of the known planning area, economic score indicating the planning cost of the known planning area, and difference score indicating the difference between the land types before and after applying the possible planning schemes. The model parameters and loss function of the multi-objective optimization model are updated according to the multi-dimensional scores. The model parameters include: the weights of the value network, the weights of the policy network, and the weights of each type of objective reward function.

[0050] In this embodiment, a multi-objective optimization model is trained using reinforcement learning. The iteration is stopped by determining whether the loss function converges. Specifically, iteration stops when the loss function converges, and updates the loss function and continues iterating when it fails to converge. In this embodiment, the loss function of the multi-objective optimization model is composed of the loss function of the value network and the loss function of the policy network. The policy network is a neural network in the multi-objective optimization model used to predict the target land use type (i.e., the selectable planning scheme) to be selected in the current training state (the training state constructed from the known land types of the planning area). The value network is a neural network in the multi-objective optimization model used to calculate the scores of the selectable planning schemes. The value network uses the given multi-objective neural network to perform multi-dimensional scoring on the selectable planning schemes output by the policy network. During training, the loss function is updated and iteration continues when it fails to converge. Each iteration proceeds as follows: the current training state is constructed using relevant data from the geographic grid of the known planning area. This relevant data includes land type, NDVI, population density, land price, road density, etc. The relevant data of the geographic grid of the known planning area are used as input data for the policy network. Based on the analysis of this input data and under the constraints of preset transformation rules, the policy network generates an optional planning scheme. This optional planning scheme records the planned land type for each geographic grid of the known planning area. As described above, the preset transformation rules are used to constrain the policy network to follow... Figure 3The land type conversion table shown generates optional planning schemes. For example, under the constraints of preset conversion rules, arable land can be converted into urban construction land but cannot be directly converted into water area. Next, the value network will score the optional planning schemes output by the above-mentioned policy network based on the multi-objective reward function. As mentioned in the above embodiment, the multi-objective reward function refers to the function generated by multiple objective reward functions, and each objective reward function corresponds to a planning objective. Therefore, when using the multi-objective reward function to score the optional planning schemes, it actually scores the optional planning schemes from multiple scoring dimensions corresponding to multiple objective reward functions. In this embodiment, one objective reward function corresponds to one scoring dimension. The scoring dimensions include: economic dimension, ecological dimension, spatial compactness dimension, and initial difference dimension. Among them, the score of the ecological dimension (i.e., the ecological reward score) is determined based on the ecological value of the known planning area after implementing the optional planning scheme. The score of the economic dimension (i.e., the economic score) is determined based on the planning cost generated by implementing the optional planning scheme in the known planning area. The score of the spatial compactness dimension (i.e., the spatial compactness score) is determined based on the land use density of the known planning area after implementing the optional planning scheme. The score of the initial difference dimension (i.e., the difference score) is determined based on the degree of difference in land type of each sub-region in the known planning area before and after implementing the optional planning scheme. After obtaining the multi-dimensional scores, the scores are weighted and summed to obtain the total reward value (i.e., the comprehensive score). Next, the Proximal Policy Optimization (PPO) algorithm is used to calculate the loss functions of the policy network and the value network. The total reward value and advantage estimation are used to update the parameters of the whole model. In the process of updating the model parameters, the weights of the value network and the policy network are updated through the backpropagation algorithm, and the weights of the reward functions of each type of objective are dynamically adjusted. The weights of the objective reward functions are used to distinguish the importance of different planning objectives under different environments.

[0051] The training process of the aforementioned multi-objective optimization model can be summarized into four main components: regional state representation, policy network structure, value assessment mechanism, and iterative optimization process. Regional state representation discretizes the known planning area into multiple geographic grids in a two-dimensional grid format. The relevant data for each geographic grid includes land use type, population density, NDVI, and land price. In each decision step, the relevant data of all geographic grids in the entire known planning area are vectorized. Specifically, vectorization is achieved by encoding the relevant data of the geographic grids. The encoding methods include: 1) Land use encoding: converting the land use type (i.e., land type) of each grid into a one-hot encoded vector and embedding it using a fully connected layer; 2) Grid location embedding: constructing a location embedding layer using the grid ID (gridid) of the geographic grid to learn location-related policy preferences. Policy network structure: The policy network consists of two independent sub-modules: a land use type decision head and a land use location decision head. The land use type decision head predicts other land types to be selected in the current state, while the land use location decision head selects the optimal allocation location from all available locations. Each decision distribution is processed by a normalized exponential function (softmax) and then sampled to generate alternative planning schemes. During training, the joint distribution of the two strategies is used to calculate the objective function of the PPO. Value evaluation mechanism: The value network calculates a score for each alternative planning scheme based on a multi-objective reward function, guiding policy updates. The value network structure includes a state embedding layer and a regression head output layer composed of a multilayer perceptron (MLP). In the learning and optimization process of the multi-objective optimization model, a multi-environment parallel sampling + iterative optimization strategy is adopted. The core process includes: 1) State sampling and policy execution: a) Sampling land use types and target grids from the policy network; b) Interacting with the environment to update the grid state and calculating the corresponding composite rewards (economic / ecological / spatial compactness / differentiation). 2) Advantage estimation and policy optimization: a) Estimating the advantage function using generalized advantage estimation (GAE); b) Jointly optimizing the policy network using a truncated PPO policy loss function and an entropy regularization term; c) Introducing a pruning term for the quality assessment value (Critic value) of alternative planning schemes to prevent overfitting. 3) Dynamic scheduling mechanism: a) Introducing a gradient norm monitoring mechanism and an adaptive learning rate adjuster; b) Simultaneously supporting an early stopping strategy to accelerate convergence. Furthermore, the multi-objective optimization model provided in this application supports dynamic combinations of multiple sub-objective rewards (economic, ecological, compact, and differentiated). Therefore, during the training phase, the importance of each sub-objective can be adjusted through weighting coefficients, and each type of reward is automatically normalized during training to ensure fairness in comparisons between different dimensions.

[0052] Optionally, the ecological incentive score is determined by the following method: All geographic grids constituting the known planning area are classified into Category I and Category II geographic grids based on the planned land type corresponding to each geographic grid. Category I geographic grids are those whose corresponding planned land type belongs to an ecological type, and Category II geographic grids are those whose corresponding planned land type does not belong to an ecological type. The total supply capacity of multiple Category I geographic grids and the total demand capacity of multiple Category II geographic grids are determined. The ecological incentive score of the known planning area is determined based on the total supply capacity, total demand capacity, preset spatial attenuation coefficient, distance between every two Category I and Category II geographic grids, relevant data of Category I and Category II geographic grids, and relevant data of Category II geographic grids. The relevant data of Category I geographic grids includes: the vegetation index and area of ​​each Category I geographic grid; the relevant data of Category II geographic grids includes: the population density and area of ​​each Category II geographic grid.

[0053] In this embodiment, an ecological reward function is used to determine the ecological reward score (i.e., the ecological reward value). The core calculation principle of the ecological reward function is based on the ecosystem service value stream model. By calculating the supply capacity of ecological grids and the demand capacity of non-ecological grids, and combining the distance decay effect between them, it measures the spatial flow and value of ecosystem services. Specifically, the ecological reward score is the product of the supply value of ecological grids and the demand value of non-ecological grids, multiplied by the distance decay factor, and finally summed over all combinations of ecological and non-ecological grids. The ecological reward score is calculated using the following formula: Among them, R eco This is an ecological reward score, where i represents the first type of geographic grid, j represents the second type of geographic grid, S is the set of geographic grids of ecological type (i.e., the first type of geographic grid), D is the set of geographic grids of non-ecological type (i.e., the second type of geographic grid), and μ s It is the total supply capacity, μ d It is the total demand capacity, α is the preset space attenuation coefficient, A i Let A represent the area of ​​grid i. j d represents the area of ​​grid j. ij Represents the distance between grids i and j, POP j This refers to the population density of grid j. In this embodiment, whether a geographic grid is an ecological or non-ecological type depends on whether the planned land type corresponding to the grid belongs to ecological land. In this embodiment, land types such as grassland and water bodies are classified as ecological land, while other land types are classified as non-ecological land. The population density (POP) used in the spatial compactness scoring mentioned above... jNDVI (Vegetation Index) i It can be retrieved from the database in step S202. Figure 5 It is a relational table of related data in a geographic grid. In a database, it can be used as follows: Figure 5 The table shown records relevant data for each geographic grid, such as land type, population density, and vegetation index, in a single row. This ensures that data representing different meanings within the same row are all associated with a single geographic grid. Therefore, all relevant data for a specific geographic grid, such as population density (POP), can be retrieved from the database. j NDVI (Vegetation Index) i Land prices, etc.

[0054] According to some alternative embodiments of this application, the spatial compactness score is determined by the following method: classifying geographic grids with planned land type of urban construction land into third-class geographic grids, classifying geographic grids with planned land type of cultivated land into fourth-class geographic grids, and classifying geographic grids with planned land type of ecological land into fifth-class geographic grids; for third-class geographic grids, determining a first spatial compactness score for a first region composed of all third-class geographic grids based on the concentrated distribution of multiple third-class geographic grids and the urban construction land compatibility coefficient; for fourth-class geographic grids, determining a second spatial compactness score for a second region composed of all fourth-class geographic grids based on the concentrated distribution of multiple fourth-class geographic grids and the cultivated land compatibility coefficient; for fifth-class geographic grids, determining a third spatial compactness score for a third region composed of all fifth-class geographic grids based on the concentrated distribution of multiple fifth-class geographic grids and the continuity of multiple fifth-class geographic grids; and determining the spatial compactness score of a known planned area based on the first, second, and third spatial compactness scores.

[0055] The spatial compactness reward aims to encourage the concentrated layout of various land uses in land planning and avoid excessive land dispersion. Its core calculation principle is to evaluate the compactness of urban land use at multiple levels and accumulate the reward values ​​at each level to obtain the final spatial compactness reward. In this embodiment, the spatial compactness reward value (i.e., the spatial compactness score) is divided into four levels: compactness within a major category, internal rules of urban construction land, internal rules of cultivated land, and ecological patch continuity reward. Therefore, for a geographic grid whose planned land type is urban construction land (i.e., the third type of geographic grid), the spatial compactness score (i.e., the first spatial compactness score) of the area (i.e., the first area) composed of multiple third type geographic grids is jointly determined by the compactness within a major category and the internal rules of urban construction land. In this embodiment, the internal rules of urban construction land are predefined, including a predefined urban construction land compatibility coefficient. The urban construction land compatibility coefficient reflects the synergistic effect and compatibility of different land types contained in urban construction land; for example, it can reflect the compatibility between commercial land and residential land. For geographic grids with cultivated land as the planned land type (i.e., the fourth type of geographic grid), the spatial compactness score (i.e., the second spatial compactness score) of the region (i.e., the second region) composed of multiple fourth type geographic grids is determined jointly from two levels: compactness within the major category and internal rules within cultivated land. In this embodiment, the internal rules within cultivated land are predefined, including a predefined cultivated land compatibility coefficient. The cultivated land compatibility coefficient reflects the compatibility between different types of cultivated land, for example, it can reflect the mutual influence between paddy fields and dry land. For geographic grids with ecological land as the planned land type (i.e., the fifth type of geographic grid), the spatial compactness score (i.e., the third region) of the region composed of multiple fifth type geographic grids is determined jointly from two levels: compactness within the major category and ecological patch continuity reward. In this embodiment, the ecological patch continuity reward is predefined. The continuity (i.e., the concentration distribution) between ecological land grids is detected using a disjoint-set data structure algorithm. Based on the concentration distribution of the fifth type geographic grids, the spatial compactness score (i.e., the third region) of the region composed of multiple fifth type geographic grids is determined. As can be seen from the above, in the solution provided in this application embodiment, when determining the compactness score of a known planning area, it is necessary to simultaneously consider the spatial compactness score of urban construction land (i.e., the first spatial compactness score), the spatial compactness score of cultivated land (i.e., the second spatial compactness score), and the spatial compactness score of ecological land (i.e., the third spatial compactness score) in the known planning area after implementing the optional planning scheme.

[0056] According to some optional embodiments of this application, a first spatial compactness score of a first region composed of all third-class geographic grids is determined based on the concentrated distribution of multiple third-class geographic grids and the urban construction land compatibility coefficient, including: determining a first spatial compactness reward of the first region based on the planned land type corresponding to the third-class geographic grid and the planned land type corresponding to the adjacent geographic grid of each third-class geographic grid; determining a first compatibility reward of the first region based on multiple urban construction land compatibility coefficients corresponding to multiple first geographic grid groups, wherein each first geographic grid group contains two adjacent third-class geographic grids; determining a first spatial compactness score based on the first spatial compactness reward and the first compatibility reward; and determining a second spatial compactness score of a second region composed of all fourth-class geographic grids based on the concentrated distribution of multiple fourth-class geographic grids and the cultivated land compatibility coefficient, including: determining a first spatial compactness score based on the planned land type corresponding to the fourth-class geographic grid and the planned land type corresponding to the adjacent geographic grid of each fourth-class geographic grid. The second spatial compactness reward for the second region; the second compatibility reward for the second region is determined based on the farmland compatibility coefficients corresponding to multiple second geographic grid groups, wherein each second geographic grid group contains two adjacent fourth-type geographic grids; the second spatial compactness score is determined based on the second spatial compactness reward and the second compatibility reward; the third spatial compactness score for the third region, composed of all fifth-type geographic grids, is determined based on the concentrated distribution and continuity of multiple fifth-type geographic grids, including: determining the third spatial compactness reward for the third region based on the planned land type corresponding to the fifth-type geographic grid and the planned land type corresponding to the adjacent geographic grids of each fifth-type geographic grid; classifying multiple adjacent fifth-type geographic grids into a set, and determining the ecological stability reward for the third region based on the number of sets, the number of fifth-type geographic grids contained in each set, and the preset ecological reward coefficient; the third spatial compactness score for the third region is determined based on the third spatial compactness reward and the ecological stability reward.

[0057] The "category" (or "cat") in the compactness reward within a category mentioned in the previous embodiment refers to three major categories: urban construction land, arable land, and ecological land. The concentrated spatial distribution of land within the same major category helps improve infrastructure utilization efficiency and reduce transportation costs. For each geographic grid... Define its adjacent mesh set as If a certain adjacent grid The type belongs to the same major category as i (such as urban construction, arable land, or ecology), denoted as: The formula for calculating the major categories of tightness rewards (first-space tightness rewards, second-space tightness rewards, and third-space tightness rewards) is as follows: Among them, R categoryβ represents the compactness reward for the first space, the second space, and the third space, which is determined based on the specific calculation scenario (i.e., the planned land type corresponding to geographic grids i and j). β is a pre-set basic reward coefficient (e.g., it can be set to 0.5). Its function is to eliminate duplicate counts.

[0058] The urban construction land compatibility bonus (i.e., the first compatibility bonus) can be calculated using a formula. The calculation yields, where R urban This represents the first compatibility bonus, and β is a pre-set base bonus coefficient (e.g., it can be set to 0.5). This indicates that the planned land type for geographic grid i is urban construction land. This indicates that geographic grid j belongs to the adjacent grid of geographic grid i. Furthermore, the planned land type is urban construction land. It is the compatibility coefficient of urban construction land. It can be found in the predefined urban construction land compatibility matrix. Each element in the urban construction land compatibility matrix is...

[0059] The farmland internal compatibility bonus (i.e., the second compatibility bonus) can be calculated using a formula. The calculation yields, where R farm Let β represent the second compatibility reward, be the pre-set base reward coefficient (e.g., 0.5), and let i∈F represent the planned land type of geographic grid i as arable land (F). Similar to urban construction land, there are also compatibility relationships between different types of arable land (such as paddy fields and dry land). Therefore, when calculating the compatibility reward within arable land, the arable land compatibility coefficient also needs to be used. In the above formula, the coefficient is... Indicates the compatibility coefficient of arable land. It can be found in a predefined farmland compatibility matrix, where each element is...

[0060] Since the integrity and stability of an ecosystem depend on the continuity of ecological patches, in this embodiment, when determining the spatial compactness score (i.e., the third spatial compactness score) of ecological land, in addition to considering the category compactness reward (i.e., the third spatial compactness reward) of ecological land, it is also necessary to consider the ecological patch continuity reward (i.e., the ecological stability reward). Before determining the ecological patch continuity reward (i.e., the ecological stability reward), a disjoint-set data structure algorithm is used to detect whether there are continuous patches (i.e., adjacent fifth-category geographic grids) in the areas of ecological land type (i.e., the third area). All fifth-category geographic grids belonging to the same ecological type and spatially adjacent are merged into the same patch, with each patch corresponding to a set, resulting in multiple sets. Each set It is a continuous ecological patch, the size of which is Representative set The number of geographic grids included; further, based on the multiple sets obtained above, the ecological patch continuity reward (i.e., ecological stability reward) is determined, specifically according to the formula. Determine the ecological patch continuity reward (i.e., ecological stability reward), where R eco cluster K represents the ecological stability reward, K represents the number of sets, and γ is the preset ecological patch unit reward coefficient (i.e., the preset ecological reward coefficient). In this embodiment, the value of γ can be set to 0.3.

[0061] Optionally, the economic score is determined by the following method: the economic score is determined based on the distance from multiple geographic grids to the central geographic grid, the preset payment amount, and the traffic cost sensitivity coefficient corresponding to each geographic grid, wherein the central geographic grid is the grid representing the center of the known planning area.

[0062] In the bidding model, the transportation cost and land rent at a distance t from the city center (i.e., the central geographic grid) are k(t) and p(t), respectively, q is the land use quantity at t, p is the commodity price (a constant), and z is the commodity consumption. Therefore, for a land bidder with income y, the budget constraint equation is: y = k(t) + p(t)·q + p·z. Transforming this equation, we can obtain: Given the varying sensitivities of different land use types to the location of the city center, the calculation of transportation costs must incorporate location sensitivity factors. The transportation cost k(t) = θ·t, where θ is the location sensitivity (i.e., the transportation cost sensitivity coefficient). The land price payment Y can then be determined using the formula: Y = yp·z. In this embodiment, Y is the land price that the land bidder is willing to pay (i.e., the preset payment amount). In this embodiment, θ and Y are hyperparameters of the model, which can be initialized manually and adjusted during model training. Furthermore, in the method provided in this application embodiment, when introducing a reinforcement learning model for land use planning, we adopted a simplified but effective strategy: focusing the calculation of land rent on the unit area, that is, treating q as 1 (the area of ​​each grid cell is considered 1). Therefore, according to the formula... The land rent amount p per unit area (area of ​​each grid cell) can be calculated without needing q. (t) Therefore, in this embodiment, when the value of q is set to 1, the reward function for land value (i.e., the formula for calculating the economic score) is defined as: P = Y - βd i Where P represents the economic score, d i The distance from geographic grid i to the city center (i.e., the central geographic grid) represents the land value reward function (i.e., the economic score). The optimal land use distribution (i.e., the land type that can be paid for the highest land price Y at the same distance from the city center) is when all planned land types in the known planning area are the land types that can be paid for the highest land price Y at the same distance from the city center (e.g., forest land has the highest payable land price, so the economic score is highest when the available planning schemes in the known planning area indicate that all land types in the known planning area are converted to forest land). This is a land use distribution that conforms to the theory of competitive renting.

[0063] According to an optional embodiment of this application, the difference score is determined by the following method: classifying geographic grids whose original land type is the same as the planned land type into a sixth type of geographic grid, and classifying geographic grids whose original land type and planned land type belong to the same land application category into a seventh type of geographic grid, wherein the original land type is the land type before the application of the optional planning scheme in the known planning area, and each land application category contains multiple land types; determining a first number of sixth type geographic grids and a second number of seventh type geographic grids among all geographic grids constituting the known planning area; and determining the difference score based on the first preset difference reward and the first number corresponding to the sixth type geographic grid, and the second preset difference reward and the second number corresponding to the seventh type geographic grid.

[0064] The initial difference reward (i.e., difference score) aims to measure the degree of difference between the land type layout of a known area after implementing alternative planning schemes and the initial environment (i.e., the land type layout of a known area before implementing alternative planning schemes), and is a numerical value determined based on the degree of difference. Its core idea is to encourage land use planning to develop in a direction that aligns with or is reasonably consistent with the initial environment. In this embodiment, for a given geographic grid, a higher reward is given when its corresponding planned land type is consistent with the initial type (i.e., the original land type); if they belong to the same broad category (i.e., land use classification) but have different specific types, a relatively lower reward is given, thereby guiding land use planning to follow the layout characteristics of the initial environment to a certain extent. Specifically, in this embodiment, for geographic grids whose corresponding planned land type is exactly the same as the initial type (i.e., the original land type) (i.e., the sixth type of geographic grid), the corresponding preset difference reward (i.e., the first preset difference reward) is set to 1, so that the original land type and the planned land type of the geographic grid are completely consistent, and 1 point is obtained; for geographic grids whose corresponding planned land type is not the same as the initial type (i.e., the original land type) but belong to the same land application classification (i.e., the seventh type of geographic grid), the corresponding preset difference reward is set to 0.5, so that the original land type and the planned land type of the geographic grid are not completely consistent but belong to the same major category, and 0.5 points are obtained. In this embodiment, the land application classification includes three categories: urban construction land, agricultural land, and ecological land. Among them, the urban construction land land application classification includes the following land types: residential land, commercial land, industrial land, park, and public management land; agricultural land includes the following land types: cultivated land and forest land; and ecological land includes the following land types: grassland and water area. In this embodiment, the formula for determining the difference score is: Among them, R diff Representing the difference score, i∈allocated represents any geographic grid i that has been allocated a planned land type. The original land type representing geographic grid i, The planned land type represents geographic grid i. As the initial type, C(·) represents the major category to which the original land type / planned land type of geographic grid i belongs (i.e., land use classification, such as urban land, agricultural land, or ecological land).

[0065] According to some optional embodiments of this application, updating the model parameters of a multi-objective optimization model based on multi-dimensional scores includes: when the initial reward value is 0, determining a comprehensive score based on the multi-dimensional scores and multiple weight coefficients corresponding to the multi-dimensional scores, wherein the comprehensive score is used to guide the updating of model parameters, and the initial reward value is determined based on the first calculated multi-dimensional scores; when the initial reward value is not 0, determining the comprehensive score after normalization using the initial reward value as the comprehensive score.

[0066] In this embodiment, when updating the model parameters of the multi-objective optimization model based on multi-dimensional scores, a comprehensive score is first determined based on the multi-dimensional scores, and then the model parameters are updated under the guidance of the comprehensive score. For example, the model parameters are updated with the goal of maximizing the comprehensive score. Determining the comprehensive score includes two schemes: if the multi-dimensional scores are calculated first at the start of model training, and if all multi-dimensional scores are 0, then the original reward value is directly returned when calculating the comprehensive score subsequently. The original reward value is a comprehensive score obtained by weighted summation of the multi-dimensional scores calculated each time and the multiple weight coefficients corresponding to the multi-dimensional scores. The multi-dimensional scores include economic scores, ecological scores, spatial compactness scores, and difference scores. Each dimension corresponds to a weight coefficient, which reflects the importance of the planning objective corresponding to that dimension in the overall evaluation. If the initial reward value is not zero, meaning that there are non-zero scores when calculating the multi-dimensional scores (i.e., scores across multiple dimensions) for the first time, the comprehensive score is normalized by dividing the currently calculated comprehensive score by the first calculated comprehensive score (i.e., the initial reward value). The normalized multi-dimensional scores are then multiplied by their respective reward coefficients (i.e., weight coefficients) and summed to obtain the comprehensive score. During the model parameter update process, the generalized advantage estimation (GAE) strategy is used to process the obtained comprehensive score, yielding the advantage function for each planning objective's corresponding planning land type. Then, a truncated PPO strategy loss function and an entropy regularization term are used to jointly optimize the strategy network, adjusting the model parameters with the goal of maximizing the comprehensive score. The generalized advantage estimation (GAE) strategy, PPO strategy loss function, and entropy regularization term in this embodiment are the same as those used in related technologies for model training, as can be found by consulting relevant technologies.

[0067] According to some optional embodiments of this application, before processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model, the method includes: for non-numerical type first-class data, converting the first-class data into corresponding codes according to preset coding rules; for numerical type second-class data, normalizing the second-class data to obtain normalization results; generating spatial feature vectors for each target geographic grid based on the relative position information of the target geographic grids, and generating state vectors for the target geographic grids based on the spatial feature vectors, codes, and normalization results, wherein the relative position information of the target geographic grids is used to indicate the positional relationship between the target geographic grids and other target geographic grids contained in the area to be planned.

[0068] In this embodiment, before using a multi-objective optimization model for urban land use planning, the non-numerical data (i.e., the first type of data) and numerical data (i.e., the second type of data) in the geographic grid are preprocessed to process the relevant data of the target geographic grid into a form that the model can understand. Specifically, in this embodiment, the relevant data of the target geographic grid is converted into a state vector representing the current land use status of the planned area through encoding, and the state vector of the planned area is processed by a multi-objective optimization model. In generating state vectors, for non-numerical data (i.e., the first type of data), such as land use type, the first type of data is converted into numerical encoding according to preset encoding rules (such as the encoding rules defined in one-hot encoding). For numerical data (i.e., the second type of data), such as population density, land price, NDVI, etc., the second type of data is normalized to map the values ​​to the range of 0 to 1, reducing the impact of data magnitude differences on model learning. The result obtained after normalization of each second type of data (i.e., the normalization result) is the encoding form of the second type of data. Next, spatial feature vectors are constructed using the relative position information of the target geographic grids (such as the relative positions of neighboring target geographic grids (such as adjacency in the north, south, east, and west directions) and the distance between each target geographic grid and the central grid). A location embedding layer is constructed using the number of each target geographic grid. Through the location embedding layer, the grid number is mapped to a set of location features to capture the spatial relationship between grids. Ultimately, the state vector of each generated geographic grid contains the following information: relevant data in the geographic grid (represented in the form of encoded or normalized results), spatial feature vector, and location features.

[0069] Optionally, after obtaining the predicted planning scheme, the method further includes: converting the predicted planning scheme into a visual diagram, wherein different land types are displayed in different colors in the visual diagram.

[0070] Figure 6This is a visual representation of the predicted planning scheme for the planned area A. In this embodiment, the relevant data for the planned area A is recorded in the form of Figure A. The map of the planned area A is processed into a two-dimensional grid with sides of 1 km, dividing the planned area A into multiple geographic grids. The original land type of the planned area A is determined by intersecting the two-dimensional gridded map of the planned area A with images from the Multi-Period Land Use Remote Sensing Monitoring Dataset (CNLUCC) and the Land Use Type Dataset (EULUC). Figure 6 As shown in Figure A, the original land use types of the area to be planned, A, are as follows: residential land (marked yellow 11), commercial land (marked red 12), industrial land (marked brown 13), parkland (marked green 14), public and administrative service land (marked purple 15), urban and other construction land (marked gray 16), paddy fields (marked orange 21, a type of arable land), dry land (marked light orange 22, a type of arable land), woodland (marked dark green 31), grassland (marked slightly dark 32), water area (marked blue 33), unused land (marked gray 34), and rural settlements (marked light gray). After processing the geographic grid data of the area to be planned, A, using a multi-objective optimization model, a predicted planning scheme for the area to be planned, A, was obtained. The visualization image corresponding to the predicted planning scheme for the area to be planned, A, is shown below. Figure 6 Figure B in the image, as shown Figure 6 As shown, in the predicted planning scheme for area A, the planned land types in area A are consistent with the original land types in terms of total number, but the distribution is different. In this embodiment, the prediction scheme can be converted into a visual image using Geographic Information System (GIS) software. In GIS software, the predicted planning scheme is converted into a visual image, and each geographic grid will be displayed in a different color according to its predicted land use type. In the process of transforming the planning scheme of the area to be planned, A, from Figure A to Figure B using a multi-objective optimization model, the multi-objective optimization model aims to maximize the multi-objective function while considering multiple planning objectives corresponding to various objective reward functions. In the multi-dimensional score of the predicted planning scheme shown in Figure B, the ecological score is 1, the spatial compactness score is 1, the economic score is 1.074, the difference score is 1.024, and the comprehensive score is 1.075812. Compared with the multi-dimensional score of the original planning scheme shown in Figure A (ecological score 1, spatial compactness score 1, economic score 1, difference score 1, and comprehensive score 1), the predicted planning scheme shown in Figure B improves the economic value of the predicted planning scheme and the rationality of its spatial layout compared with the original planning scheme while maintaining ecological balance and spatial compactness, thus achieving the technical effect of multi-objective trade-off.

[0071] Through the above steps, a multi-objective optimization model with multi-objective trade-offs and multi-spatial collaborative optimization capabilities can be provided. The planning scheme output by the multi-objective optimization model for the area to be planned reduces the time for generating land planning schemes and improves the efficiency of land planning.

[0072] Figure 7 This is a structural diagram of a land planning scheme determination device provided according to an embodiment of this application, such as... Figure 7 As shown, the land planning scheme determination device includes: an acquisition module 70, used to acquire relevant data of target geographic grids of the area to be planned from a database, wherein each target geographic grid represents a sub-region contained in the area to be planned, and the relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region, the attribute information including: ecological information and land application information; and a planning module 72, used to process and analyze the relevant data of multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the area to be planned, wherein the predicted planning scheme includes: the land type of each sub-region, the multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules, the preset conversion rules are used to indicate multiple conversion schemes supported by each land type, and the conversion schemes are used to indicate the conversion of the land type to another land type. During the training process, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0073] Figure 8 This is a flowchart of the land planning scheme determination device generating land planning schemes, such as... Figure 8As shown, when generating a land planning scheme for a planned area, the land planning scheme determination device initializes the planning environment based on the relevant data of the target geographic grid of the planned area. This data is obtained by the acquisition module 70 from a database. The target geographic grid data includes the geographic location information of the region (sub-region) represented by each target geographic grid and the attribute information of the sub-region, including ecological information and land application information. Next, the land planning scheme determination device executes the model loading step. After loading the multi-objective optimization model, the planning module 72 calls the multi-objective optimization model to perform iterative planning for the planned area. Each iteration of iterative planning outputs a conversion scheme for a land type. After each iteration, the multi-objective optimization model uses a multi-objective reward function to score the output conversion scheme from multiple dimensions. The completion of the planning is determined based on the multi-dimensional scoring results. In this embodiment, planning is considered complete when the multi-dimensional scores reach their respective maximum values ​​while maintaining balance. At this point, the land planning scheme determination device saves the predicted planning scheme output from the last iteration and converts it into a visual image, ending the planning process.

[0074] It should be noted that, Figure 7 Preferred embodiments of the shown examples can be found in [reference needed]. Figure 2 The relevant descriptions of the embodiments shown will not be repeated here.

[0075] This application embodiment also provides a non-volatile storage medium storing a computer program, wherein the above method for determining the land planning scheme is executed by running the computer program on the device where the non-volatile storage medium is located.

[0076] The aforementioned non-volatile storage medium is used to store programs that perform the following functions: Retrieving relevant data of target geographic grids for the area to be planned from a database, wherein each target geographic grid represents a sub-region within the area to be planned, and the relevant data of each target geographic grid includes: geographic location information and attribute information of the sub-region, including: ecological information and land application information; Processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the area to be planned, wherein the predicted planning scheme includes: the land type of each sub-region; The multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules; The preset conversion rules indicate multiple conversion schemes supported by each land type; The conversion schemes indicate the conversion of a land type to another land type; During training, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0077] This application also provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor is configured to execute the above method for determining the land planning scheme through the computer program.

[0078] The processor in the aforementioned electronic device is used to run a program that performs the following functions: Retrieving relevant data of target geographic grids for the area to be planned from a database, wherein each target geographic grid represents a sub-region within the area to be planned, and the relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region, including: ecological information and land application information; Processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the area to be planned, wherein the predicted planning scheme includes: the land type of each sub-region; the multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules; the preset conversion rules indicate multiple conversion schemes supported by each land type; the conversion schemes indicate the conversion of a land type to another land type; during training, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

[0079] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the above-described method for determining land planning schemes.

[0080] It should be noted that the modules in the above-mentioned land planning scheme determination device can be program modules (for example, a set of program instructions to implement a certain function) or hardware modules. For the latter, it can be manifested in the following forms, but is not limited to them: each of the above modules is manifested as a processor, or the functions of each of the above modules are implemented by a processor.

[0081] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0082] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0087] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining a land planning scheme, characterized in that, include: Retrieve relevant data of target geographic grids for the area to be planned from the database. Each target geographic grid represents a sub-region contained in the area to be planned. The relevant data of each target geographic grid includes: geographic location information of the sub-region and attribute information of the sub-region. The attribute information includes: ecological information and land application information. A multi-objective optimization model is used to process and analyze relevant data from multiple target geographic grids to obtain a predicted planning scheme for the area to be planned. The predicted planning scheme includes the land type of each sub-region. The multi-objective optimization model is trained based on a multi-objective reward function, known planning areas of multiple planning schemes, and preset conversion rules. The preset conversion rules are used to indicate multiple conversion schemes supported by each land type. The conversion schemes are used to indicate the conversion of the land type to another land type. During training, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

2. The method according to claim 1, characterized in that, The target optimization model is iteratively trained using the following method: In each iteration, the loss function of the target optimization model is updated. If the loss function converges, the iteration stops; if the loss function fails to converge, the update of the loss function continues. Updating the loss function includes: For each known planning area, the strategy network in the target optimization model determines the possible planning scheme corresponding to the known planning area based on the land types contained in the known planning area and the preset conversion rules. The possible planning scheme includes: multiple geographic grids obtained by dividing the known planning area, and the planned land type of each geographic grid. The planned land type is one of a variety of other land types that the land type supports conversion to. The value network in the target optimization model performs multi-dimensional scoring on the alternative planning schemes corresponding to the known planning area according to the multi-objective reward function. The multi-dimensional scoring includes: an ecological reward score indicating the ecological value of the known planning area, a spatial compactness score indicating the land use density of the known planning area, an economic score indicating the planning cost of the known planning area, and a difference score indicating the difference between the land types before and after applying the alternative planning scheme in the known planning area. The model parameters of the multi-objective optimization model and the loss function are updated based on the multi-dimensional score, wherein the model parameters include: the weights of the value network, the weights of the policy network, and the weights of the objective reward function for each class.

3. The method according to claim 2, characterized in that, The ecological reward score is determined using the following method: Based on the planned land type corresponding to each geographic grid, all geographic grids constituting the known planning area are classified into a first type of geographic grid and a second type of geographic grid. The first type of geographic grid is the geographic grid whose corresponding planned land type belongs to an ecological type, and the second type of geographic grid is the geographic grid whose corresponding planned land type does not belong to an ecological type. Determine the total supply capacity of multiple first-type geographic grids and the total demand capacity of multiple second-type geographic grids; The ecological incentive score of the known planning area is determined based on the total supply capacity, the total demand capacity, the preset spatial attenuation coefficient, the distance between every two first-type geographic grids and second-type geographic grids, the relevant data of the first-type geographic grids, and the relevant data of the second-type geographic grids. The relevant data of the first-type geographic grids includes: the vegetation index of each first-type geographic grid and the area of ​​each first-type geographic grid. The relevant data of the second-type geographic grids includes: the population density of each second-type geographic grid and the area of ​​each second-type geographic grid.

4. The method according to claim 2, characterized in that, The space compactness score was determined by the following method: The geographic grids for the planned land type of urban construction land are classified as the third type of geographic grid, the geographic grids for the planned land type of cultivated land are classified as the fourth type of geographic grid, and the geographic grids for the planned land type of ecological land are classified as the fifth type of geographic grid. For the third type of geographic grid, the first spatial compactness score of the first region composed of all the third type of geographic grids is determined based on the concentrated distribution of multiple third type of geographic grids and the urban construction land compatibility coefficient; For the fourth type of geographic grid, the second spatial compactness score of the second region composed of all the fourth type of geographic grids is determined based on the concentrated distribution of multiple fourth type of geographic grids and the farmland compatibility coefficient; For the fifth type of geographic grid, the third spatial compactness score of the third region composed of all the fifth type of geographic grids is determined based on the concentrated distribution of the multiple fifth type of geographic grids and the continuity of the multiple fifth type of geographic grids. The spatial compactness score of the known planning area is determined based on the first spatial compactness score, the second spatial compactness score, and the third spatial compactness score.

5. The method according to claim 4, characterized in that, The first spatial compactness score of a first region composed of all the third-type geographic grids is determined based on the concentrated distribution of multiple third-type geographic grids and the urban construction land compatibility coefficient. This includes: determining a first spatial compactness reward for the first region based on the planned land type corresponding to each third-type geographic grid and the planned land type corresponding to each adjacent geographic grid of the third-type geographic grid; determining a first compatibility reward for the first region based on multiple urban construction land compatibility coefficients corresponding to multiple first geographic grid groups, wherein each first geographic grid group contains two adjacent third-type geographic grids; and determining the first spatial compactness score based on the first spatial compactness reward and the first compatibility reward. The second spatial compactness score of the second region, composed of all the fourth-type geographic grids, is determined based on the concentrated distribution of multiple fourth-type geographic grids and the farmland compatibility coefficient. This includes: determining the second spatial compactness reward of the second region based on the planned land type corresponding to each fourth-type geographic grid and the planned land type corresponding to each adjacent geographic grid of the fourth-type geographic grid; determining the second compatibility reward of the second region based on the farmland compatibility coefficient corresponding to multiple second geographic grid groups, wherein each second geographic grid group contains two adjacent fourth-type geographic grids; and determining the second spatial compactness score based on the second spatial compactness reward and the second compatibility reward. The third spatial compactness score of the third region, composed of all the fifth-type geographic grids, is determined based on the concentrated distribution and continuity of the multiple fifth-type geographic grids. This includes: determining the third spatial compactness reward of the third region based on the planned land type corresponding to each fifth-type geographic grid and the planned land type corresponding to each adjacent geographic grid of the fifth-type geographic grid; grouping multiple adjacent fifth-type geographic grids into a set, and determining the ecological stability reward of the third region based on the number of sets, the number of fifth-type geographic grids contained in each set, and a preset ecological reward coefficient; and determining the third spatial compactness score of the third region based on the third spatial compactness reward and the ecological stability reward.

6. The method according to claim 2, characterized in that, The economic efficiency score was determined using the following method: The economic score is determined based on the distance from the multiple geographic grids to the central geographic grid, the preset payment amount, and the traffic cost sensitivity coefficient corresponding to each geographic grid, wherein the central geographic grid is the grid representing the center of the known planning area.

7. The method according to claim 2, characterized in that, The difference score was determined by the following method: Geographic grids whose original land type is the same as the planned land type are classified as sixth type of geographic grids, and geographic grids whose original land type and the planned land type belong to the same land application classification are classified as seventh type of geographic grids. The original land type is the land type of the known planning area before the application of the optional planning scheme. Each land application classification contains multiple land types. Determine the first number of the sixth type of geographic grids and the second number of the seventh type of geographic grids among all the geographic grids that make up the known planning area; The difference score is determined based on the first preset difference reward corresponding to the sixth type of geographic grid, the first quantity, the second preset difference reward corresponding to the seventh type of geographic grid, and the second quantity.

8. The method according to claim 2, characterized in that, The model parameters of the multi-objective optimization model are updated based on the multi-dimensional scores, including: With an initial reward value of 0, a comprehensive score will be determined based on the multi-dimensional score and the multiple weight coefficients corresponding to the multi-dimensional score. The comprehensive score is used to guide the updating of the model parameters, and the initial reward value is determined based on the multi-dimensional score calculated for the first time. If the initial reward value is not 0, the comprehensive score after normalization using the initial reward value will be determined as the comprehensive score.

9. The method according to claim 1, characterized in that, Before processing and analyzing the relevant data of multiple target geographic grids using a multi-objective optimization model, the method includes: For the first type of data that is not numerical, the first type of data is converted into the corresponding code according to the preset encoding rules; For the second type of numerical data, normalization is performed on the second type of data to obtain the normalized result; A spatial feature vector for each target geographic grid is generated based on the relative position information of the target geographic grid, and a state vector for the target geographic grid is generated based on the spatial feature vector, the encoding, and the normalization result. The relative position information of the target geographic grid is used to indicate the positional relationship between the target geographic grid and other target geographic grids contained in the area to be planned.

10. The method according to claim 1, characterized in that, After obtaining the predicted planning scheme, the method further includes: converting the predicted planning scheme into a visual diagram, wherein different land types are displayed in different colors in the visual diagram.

11. A device for determining a land planning scheme, characterized in that, include: The acquisition module is used to acquire relevant data of the target geographic grid of the area to be planned from the database. Each target geographic grid represents a sub-region contained in the area to be planned. The relevant data of each target geographic grid includes: the geographic location information of the sub-region and the attribute information of the sub-region. The attribute information includes: ecological information and land application information. The planning module is used to process and analyze relevant data from multiple target geographic grids using a multi-objective optimization model to obtain a predicted planning scheme for the area to be planned. The predicted planning scheme includes the land type of each sub-region. The multi-objective optimization model is trained based on a multi-objective reward function, known planning areas for multiple planning schemes, and preset conversion rules. The preset conversion rules indicate multiple conversion schemes supported by each land type, and the conversion schemes indicate the conversion of the land type to another land type. During training, each conversion scheme corresponding to each land type is scored according to the multi-objective reward function.

12. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores a computer program, wherein the device containing the non-volatile storage medium executes the method for determining the land planning scheme according to any one of claims 1 to 10 by running the computer program.

13. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method for determining the land planning scheme according to any one of claims 1 to 10 through the computer program.

14. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method for determining the land planning scheme as described in any one of claims 1 to 10.

Images