An AI-based planning land site selection multi-objective optimization method and system

Abstract

The application relates to an AI-based land planning site selection multi-objective optimization method and system, which comprises the following steps: constructing a city data cube; constructing a cellular automaton model and generating a land use potential map; constructing a game decision model and generating a preliminary site selection scheme set; inputting the preliminary site selection scheme set into the cellular automaton model to generate an evolution simulation data set; constructing a multi-objective evaluation model and calculating the relative closeness of each scheme in the preliminary site selection scheme set to an ideal solution; generating a site selection decision report according to the relative closeness; and in summary, the application integrates multi-source real-time and historical data to construct a city data cube, applies a cellular automaton model to generate a dynamic land use potential map, introduces multi-agent game decision to realize multi-objective scientific trade-off, realizes automatic processing and consistency guarantee of planning rules, and provides a scientific and quantitative trade-off mechanism in a multi-objective conflict scenario, and has the effect of improving the data timeliness and dynamic response capability of site selection decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an AI-based planning land site selection multi-objective optimization method and system. BACKGROUND

[0002] With the deepening of smart city and territorial space planning, using information means to assist construction project land site selection has become a core link to improve the scientific nature and decision-making efficiency of planning. In the existing technical practice, the mainstream site selection support method mainly relies on the spatial analysis function of geographic information system, and through overlay analysis, buffer zone analysis and other operations, the preset screening conditions (such as slope, geological conditions, ecological sensitivity, traffic convenience, etc.) are applied to the target area to exclude unsuitable land for construction, and on this basis, combined with expert experience, manual selection is carried out to determine the final site selection. Some improved methods introduce a single optimization algorithm, such as genetic algorithm or particle swarm optimization, to optimize for specific targets (such as minimum cost or optimal traffic accessibility) while meeting basic constraints.

[0003] However, the above methods have the following defects in practical application: first, the processing of data sources is limited to static historical data, which cannot effectively fuse and process real-time dynamic data streams, making it difficult to capture the instantaneous changes of city operation and seriously lacking in timeliness; second, the interpretation of planning provisions highly depends on manual operation, with low automation, which not only causes the efficiency of rule generation to lag, but also causes rule conflicts due to subjective understanding differences, affecting the objective consistency of site selection results; third, in the face of multi-objective conflict scenarios such as economic benefit, comprehensive benefit and ecological benefit, there is a lack of systematic quantitative trade-off mechanism, and the multi-objective is usually simplified into a single objective by using simple weighted summation, which cannot generate a multi-dimensional balanced scheme set, limiting the decision support capability and making it difficult to meet the scientific decision-making needs of complex planning scenarios. SUMMARY

[0004] In order to solve the above defects, the present application provides an AI-based planning land site selection multi-objective optimization method and system.

[0005] The above invention purpose of the present application is achieved by the following technical scheme:

[0006] An AI-based planning land site selection multi-objective optimization method, comprising the steps of:

[0007] obtaining a multi-source real-time data set and a multi-source historical data set, and constructing a city data cube based on the multi-source real-time data set and the multi-source historical data set;

[0008] constructing and training a cellular automaton model based on the city data cube, and generating a land potential map based on the trained cellular automaton model;

[0009] According to the land use potential map, a game decision model based on a multi-agent reinforcement learning framework is constructed, project parameters are obtained and input into the game decision model, and the game decision model generates a preliminary site selection scheme set through a preset multi-level decision strategy;

[0010] The preliminary site selection scheme set is input into the cellular automaton model for multi-scenario dynamic deduction to generate an evolutionary simulation data set;

[0011] A multi-objective evaluation model is constructed based on the evolutionary simulation data set, and the relative closeness of each scheme in the preliminary site selection scheme set to the ideal solution is calculated through the multi-objective evaluation model;

[0012] According to the obtained relative closeness, the schemes in the preliminary site selection scheme set are sorted and graded to generate a site selection decision report.

[0013] In a preferred example, the application can be further configured to: the step of obtaining a multi-source real-time data set and a multi-source historical data set, and constructing a city data cube based on the multi-source real-time data set and the multi-source historical data set, includes the steps of:

[0014] Based on the multi-source real-time data set and the multi-source historical data set, an initial spatio-temporal topology graph is constructed, and the multi-order neighborhood relationship between nodes in the initial spatio-temporal topology graph is extracted;

[0015] Data integrity recognition is performed on the initial spatio-temporal topology graph, when data missing is identified, the data missing is completed and semantic consistency verification is performed through the pre-constructed generative adversarial network, and an improved spatio-temporal topology graph is generated;

[0016] Based on the improved spatio-temporal topology graph, a city data cube is constructed.

[0017] In a preferred example, the application can be further configured to: the step of constructing and training a cellular automaton model based on the city data cube, and generating a land use potential map based on the trained cellular automaton model, includes the steps of:

[0018] Based on the city data cube, a multi-scale feature set is extracted, and an enhanced feature vector of cellular state description is constructed according to the multi-scale feature set, the multi-scale feature set including microscopic attribute features, mesoscopic environmental features, and macroscopic regional features;

[0019] Based on the enhanced feature vector, the cellular state is initialized, a cellular automaton engine containing a state transition rule library is constructed, and the cellular automaton engine is trained through the city data cube to obtain a cellular automaton model;

[0020] A multi-rule-oriented scenario parameter library is constructed, and the scenario parameters in the scenario parameter library are injected into the cellular automata model as conditional inputs. Land use evolution simulation is performed through a parallel computing architecture to generate a set of land use potential distributions.

[0021] A pre-set attention-weighted fusion algorithm is used to adaptively integrate the potential distributions of multiple scenarios to generate a preliminary application map.

[0022] Calculate the confidence interval of the potential of each grid cell in the preliminary land potential map, and generate the land potential map through a pre-established multidimensional assessment system.

[0023] In a preferred embodiment, this application can be further configured as follows: the step of constructing a multi-rule-oriented scenario parameter library, injecting the scenario parameters in the scenario parameter library as conditional inputs into a cellular automata model, and performing land use evolution simulation through a parallel computing architecture to generate a set of land use potential distributions includes the following steps:

[0024] Acquire spatial planning text information, extract planning guidance keywords using natural language processing technology, and establish a planning knowledge network;

[0025] A multi-objective-oriented scenario parameter template is constructed based on a planning knowledge network, and the scenario parameter template is quantified into a scenario parameter vector;

[0026] Scenario parameter vectors are injected into cellular automata models, and land use evolution simulations are performed at multiple time scales through a parallel computing architecture.

[0027] The results of land use evolution simulations are integrated to generate a probabilistic land use potential distribution, and its uncertainty is quantified through Monte Carlo simulation.

[0028] Based on the uncertainty quantification results, a spatial significance test is conducted on the probabilistic land use potential distribution to identify potential hotspot areas and high-risk areas.

[0029] Spatial overlay analysis is performed on potential hotspot areas and high-risk areas to generate a land use potential distribution set, and each potential unit in the land use potential distribution set is labeled with recommendation priority and risk level.

[0030] In a preferred embodiment, this application can be further configured as follows: the step of constructing a game decision model based on a multi-agent reinforcement learning framework according to a land use potential map, obtaining site selection project parameters and inputting them into the game decision model, so that the game decision model generates a preliminary set of site selection schemes through a preset multi-level decision strategy, includes the following steps:

[0031] Based on land use potential maps, a game-theoretic decision-making model is constructed, which includes spatially constrained agents, economic utility agents, comprehensive utility agents, and ecological utility agents. Corresponding objective functions are designed for each agent in the game-theoretic decision-making model.

[0032] High-potential candidate areas are identified in the land use potential map, and the agents of the game decision model are driven to play multiple rounds of games in the high-potential candidate areas until they converge to Nash equilibrium, thus obtaining the initial equilibrium solution set.

[0033] Sensitivity analysis is performed on the initial equilibrium solution set, and unstable solutions that are sensitive to parameter fluctuations are removed to obtain the optimized equilibrium solution set. A preliminary location scheme set is then generated based on the optimized equilibrium solution set.

[0034] In a preferred embodiment, this application can be further configured as follows: the step of inputting the preliminary site selection scheme set into a cellular automata model for multi-scenario dynamic deduction to generate an evolutionary simulation dataset includes the following steps:

[0035] Identify the spatial parameters of each location scheme in the preliminary location scheme set, and map the identified spatial parameters to the initial state of the cellular automata model;

[0036] Configure multi-scenario inference strategy parameters for the cellular automaton model and drive the cellular automaton model to perform iterative calculations at multiple time steps;

[0037] Multidimensional time-series data at each time step during the iterative calculation process are recorded, and an evolutionary simulation dataset is constructed based on the recorded multidimensional time-series data. The multidimensional time-series data includes the transformation process of cell states, the evolution trajectory of spatial patterns, and the dynamic values ​​of key indicators.

[0038] In a preferred embodiment, this application can be further configured as follows: the step of constructing a multi-objective evaluation model based on an evolutionary simulation dataset and calculating the relative closeness between each scheme in the preliminary site selection scheme set and the ideal solution through the multi-objective evaluation model includes the following steps:

[0039] Multidimensional evaluation indicators were extracted based on evolutionary simulation datasets, and a multi-objective evaluation system including positive and negative indicators was established.

[0040] The optimal and worst values ​​of each evaluation index based on the multidimensional evaluation index are selected from the preliminary site selection scheme set to form the positive ideal solution and the negative ideal solution.

[0041] Calculate the Euclidean distance between each scheme in the preliminary site selection scheme set and the positive ideal solution and the negative ideal solution, and calculate their relative proximity.

[0042] The second objective of this invention is achieved through the following technical solution:

[0043] An AI-based multi-objective optimization system for land use planning and site selection includes:

[0044] The data acquisition module is used to acquire multi-source real-time datasets and multi-source historical datasets, and to construct a city data cube based on the multi-source real-time datasets and multi-source historical datasets.

[0045] The map generation module is used to build and train a cellular automaton model based on urban data cubes, and generate a land use potential map based on the trained cellular automaton model.

[0046] The first model building module is used to build a game decision model based on a multi-agent reinforcement learning framework according to the land use potential map, obtain the site selection project parameters and input them into the game decision model, so that the game decision model can generate a preliminary site selection scheme set through a preset multi-level decision strategy.

[0047] The deduction module is used to input the preliminary site selection scheme set into the cellular automata model for multi-scenario dynamic deduction and generate an evolutionary simulation dataset;

[0048] The second model building module is used to build a multi-objective evaluation model based on the evolutionary simulation dataset, and to calculate the relative closeness between each scheme in the preliminary site selection scheme set and the ideal solution through the multi-objective evaluation model;

[0049] The report generation module is used to sort and classify the schemes in the preliminary site selection scheme set according to the obtained relative proximity, and generate a site selection decision report.

[0050] This application also relates to a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described AI-based multi-objective optimization method for planning and land use site selection.

[0051] This application also relates to a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned AI-based multi-objective optimization method for land use planning.

[0052] In summary, this application provides an AI-based multi-objective optimization method and system for planning land use site selection. By integrating multi-source real-time and historical data to construct an urban data cube, applying a cellular automata model to generate a dynamic land use potential map, and introducing multi-agent game decision-making to achieve scientific trade-offs among multiple objectives, it effectively solves the problems of insufficient data timeliness, low rule generation efficiency, and lack of multi-objective conflict quantification mechanisms in existing technologies. It achieves automated processing and consistency assurance of planning rules and provides a scientific quantitative trade-off mechanism in multi-objective conflict scenarios, thus improving the data timeliness and dynamic response capability of site selection decisions. Attached Figure Description

[0053] Figure 1 This is a flowchart of an embodiment of an AI-based multi-objective optimization method for planning land use site selection according to this application;

[0054] Figure 2 This is a flowchart of step S10 in an embodiment of an AI-based multi-objective optimization method for planning land use selection in this application.

[0055] Figure 3 This is a flowchart of step S20 in an embodiment of an AI-based multi-objective optimization method for planning and land use selection in this application. Detailed Implementation

[0056] The following is in conjunction with the appendix Figures 1-3 This application will be described in further detail.

[0057] In one embodiment, such as Figure 1 As shown, this application discloses an AI-based multi-objective optimization method for land use planning, which specifically includes the following steps:

[0058] S10: Acquire multi-source real-time datasets and multi-source historical datasets, and construct a city data cube based on the multi-source real-time datasets and multi-source historical datasets;

[0059] In this embodiment, multi-source real-time datasets refer to dynamic data streams acquired through various sensing devices, monitoring networks, and internet applications. Multi-source real-time datasets can reflect the current or recent operational status and dynamic changes of the urban system. Furthermore, multi-source real-time datasets include traffic dynamic data, population flow and activity data, environmental monitoring data, and public facility usage data. Multi-source historical datasets refer to relatively static or periodically updated data accumulated over a period of time. Multi-source historical datasets record the history, long-term patterns, and relatively stable characteristics of urban development. Urban data cubes refer to multi-dimensional data structures used to organize and process spatiotemporal data. Unlike simple databases or datasets, they discretize urban space into regular three-dimensional grids. Each grid cell contains a series of attribute values, such as population density, land price, and noise level. These attribute values ​​change over time. Specifically, the essence of an urban data cube is a multi-dimensional array that supports efficient slicing, dicing, and aggregation operations, providing a unified and well-organized data base for subsequent spatiotemporal analysis.

[0060] S20: Construct and train a cellular automaton model based on urban data cubes, and generate a land use potential map based on the trained cellular automaton model;

[0061] In this embodiment, the cellular automaton model refers to a discrete model used to simulate the spatiotemporal evolution of complex systems. It can be implemented using a predefined set of state transition rules, such as setting fixed threshold rules based on changes in the states of neighboring units, or training the state transition function through a supervised learning algorithm. Its main purpose is to achieve the automated generation of land use potential distribution. The land use potential map is a spatial analysis map constructed by the cellular automaton model to characterize the land use suitability level of different regions. Its essence is equivalent to a heat map of land development value. Furthermore, in the land use potential map, the target area is divided into several small grid units, and each unit is assigned a value. This value represents the potential suitability or future development probability of the grid unit under a specific goal orientation (such as commercial development, ecological protection, etc.). The higher the value, the greater the development potential; the lower the value, the stronger the restriction or the lower the potential.

[0062] S30: Construct a game decision model based on a multi-agent reinforcement learning framework according to the land use potential map, obtain the site selection project parameters and input them into the game decision model, so that the game decision model can generate a preliminary site selection scheme set through a preset multi-level decision strategy;

[0063] In this embodiment, the multi-agent reinforcement learning framework refers to an AI paradigm that simulates multiple decision-making agents (intelligent agents) learning the optimal strategy through trial and error in a shared environment; the game-theoretic decision model refers to a decision system based on the multi-agent reinforcement learning framework, which can use Q-learning algorithms to optimize agent policies, such as iteratively updating the decision policy through a state-action value function, or directly optimizing policy parameters using the policy gradient method. Its main purpose is to achieve the generation of a set of solutions under multi-objective interactions. Through the game-theoretic decision model based on the multi-agent reinforcement learning framework, the multi-objective optimization problem is transformed into a multi-agent... The game theory process automatically generates a preliminary set of site selection schemes that balance multiple objectives such as economy, society, and environment by simulating the dynamic game of multiple parties' interests, rather than relying on subjective weighting. The site selection project parameters refer to the specific requirements and constraints for this particular site selection project input by the user. The preset multi-level decision strategy refers to a hierarchical processing algorithm framework preset within the game decision model, used to generate the preliminary set of site selection schemes. The preliminary set of site selection schemes refers to the intermediate result generated after the game decision model runs the multi-level decision strategy. It is a collection containing multiple optional site selection schemes, which is usually a Pareto optimal solution set.

[0064] Furthermore, site selection parameters typically include:

[0065] Target area information: Specifies the geographical range of the selected location, such as a region or the radius of radiation around a central point;

[0066] Attribute requirements information: Describes the nature and requirements of the project itself, such as: project type (e.g., school, logistics center or park), construction scale, investment budget, functional requirements (whether it needs to be near a highway), etc.

[0067] S40: Input the preliminary site selection scheme set into the cellular automata model for multi-scenario dynamic simulation and generate an evolutionary simulation dataset;

[0068] In this embodiment, multi-scenario dynamic simulation is a simulation analysis method. Specifically, each scheme in the preliminary site selection scheme set is placed in several different possible future development scenarios to simulate its long-term impact. For example, different scenarios such as "normal development," "high-speed economic growth," and "ecological protection priority" can be set, each with different parameters to test the robustness and adaptability of the scheme. As a preferred implementation method, multi-scenario dynamic simulation can be implemented through a time-step iteration method, such as sequentially executing the simulation process of different scenarios on a general-purpose computing device, or using the elastic scheduling of cloud computing resources for multiple rounds of simulation. The evolutionary simulation dataset is a detailed data record of the multi-scenario dynamic simulation process and is also the direct output of the simulation activity. Specifically, the evolutionary simulation dataset is a structured dataset containing time series data. For each scheme in each scenario, the evolutionary simulation dataset records the data changes at each step from start to finish, such as: land use type conversion, traffic flow growth, changes in environmental indicators, and infrastructure load.

[0069] S50: Construct a multi-objective evaluation model based on the evolutionary simulation dataset, and calculate the relative closeness between each scheme in the preliminary site selection scheme set and the ideal solution through the multi-objective evaluation model;

[0070] In this embodiment, the multi-objective evaluation model refers to a mathematical model used to evaluate and compare the comprehensive performance of multiple schemes on different objectives. The core task of the multi-objective evaluation model is to quantify the gap between each scheme and the ideal optimal solution. In this embodiment, the multi-objective evaluation model receives an evolutionary simulation dataset and calculates the relative closeness of each scheme to the ideal solution to achieve objective quantitative evaluation of multi-objective decision-making. The ideal solution is a theoretically perfect reference point used for measurement in multi-objective optimization, also known as the positive ideal solution. It refers to a virtual, theoretically optimal solution. For example, in a land use selection problem, evaluation indicators may include "lowest cost," "highest economic benefit," "widest infrastructure service coverage," and "minimum environmental impact." Therefore, the ideal solution is a scheme that achieves "absolutely lowest cost, absolutely highest benefit, absolutely widest coverage, and absolutely minimum impact." The relative closeness is a quantitative ranking indicator used in multi-objective decision-making, which accurately measures the degree of closeness between a scheme to be evaluated and the ideal solution. The value of the relative closeness is usually between 0 and 1; the larger the value, the better the overall performance of the scheme.

[0071] S60: Based on the obtained relative proximity, sort and classify the schemes in the preliminary site selection scheme set, and generate a site selection decision report.

[0072] In this embodiment, the site selection decision report is the final output of this solution, which is usually a structured report document delivered to the user.

[0073] Furthermore, the site selection decision report includes:

[0074] Scheme ranking list: Scheme priority is given based on relative proximity;

[0075] Detailed analysis of each solution: including its spatial location, simulation performance in multiple scenarios, advantages and disadvantages analysis, potential risk warnings, etc.;

[0076] Visual charts: such as land use potential maps, comparison charts of the impact of different plans, and trend charts of indicators;

[0077] Data source information: the basis and data sources for key conclusions.

[0078] Specifically, a multi-source real-time dataset and a multi-source historical dataset are acquired and integrated to construct a city data cube. This city data cube stores data in a multi-dimensional spatiotemporal structure, ensuring that site selection analysis can dynamically reflect the real-time status of urban operations, thereby overcoming the timeliness defects caused by traditional methods relying solely on static data. Furthermore, based on the city data cube, a cellular automata model is constructed and trained to simulate spatial evolution and generate a land use potential map. This land use potential map objectively represents the suitability of land use in each area, avoiding subjective biases from manual interpretation of planning provisions, thus improving the automation level and consistency of planning rule processing. Subsequently, based on the land use potential map, a game-theoretic decision-making model based on a multi-agent reinforcement learning framework is constructed. Multiple agents engage in multiple rounds of games within high-potential candidate areas until convergence to a Nash equilibrium, generating a preliminary set of site selection schemes. This process utilizes agents similar to those with spatial constraints. The interaction of economic utility agents, comprehensive utility agents, and ecological utility agents enables a systematic and quantitative trade-off among multiple objectives, effectively avoiding the subjective weighting of simple weighted summation. Next, the initial site selection scheme set is input into a cellular automata model for multi-scenario dynamic simulation, generating an evolutionary simulation dataset. This multi-scenario dynamic simulation records the transition process of cellular states and the evolution trajectory of spatial patterns through iterative calculations at multiple time steps, evaluating the long-term adaptability of schemes in dynamic environments. Then, based on the evolutionary simulation dataset, a multi-objective evaluation model is constructed to calculate the Euclidean distance between each scheme in the initial site selection scheme set and the positive and negative ideal solutions, and further calculates the relative proximity, providing an objective basis for scheme ranking. Finally, based on the relative proximity, the schemes in the initial site selection scheme set are ranked and graded, generating a site selection decision report, providing decision-makers with a multi-dimensional trade-off reference for optimal solutions.

[0079] For example, in practical applications, this application can be implemented as follows: The multi-source real-time dataset can specifically include real-time traffic flow data collected by the urban traffic monitoring system and real-time weather data released by the meteorological bureau; the multi-source historical dataset can specifically include land use change records and population census statistics from the past ten years; the urban data cube is constructed as a three-dimensional structure containing longitude, latitude, and time dimensions to store and associate the aforementioned data; the cellular automata model is trained using a convolutional neural network architecture to learn spatial evolution rules and generate land use potential distributions; in the game-theoretic decision-making model, each agent uses a reinforcement learning algorithm... Interactions occur within the candidate area. For example, in a new urban area site selection project, the economic utility agent uses the rate of return on investment as the objective function, while the ecological utility agent uses the green coverage rate as the objective function. Through multiple rounds of game theory, a set containing multiple candidate solutions is generated. During multi-scenario dynamic simulation, scenario parameters such as different economic development speeds or climate change scenarios are set. The cellular automata model performs iterative calculations at multiple time steps to generate evolutionary simulation data containing the trajectory of spatial pattern evolution. Based on this evolutionary simulation data, the multi-objective evaluation model extracts indicators such as traffic accessibility and environmental impact, and calculates the relative closeness of each solution to the ideal solution.

[0080] This application addresses the timeliness deficiencies caused by insufficient real-time data fusion, the inefficiencies and inconsistencies arising from the lack of automated processing of planning rules, and the problem of scheme homogenization caused by the subjectivity of weights in multi-objective decision-making, all through the aforementioned technical means. Specifically, it achieves dynamic integration of multi-source data by constructing a city data cube, avoiding the limitations of relying solely on static data; it automatically learns spatial evolution patterns through the training and application of cellular automata models, eliminating subjective biases from manual interpretation of planning provisions; it establishes a systematic multi-objective quantitative trade-off mechanism through a multi-agent interactive game decision-making model with multiple scheme sets; it evaluates the adaptability of schemes in dynamic environments through multi-scenario dynamic simulations; and finally, it provides objective ranking criteria through a multi-objective evaluation model, supporting scheme optimization in complex decision-making scenarios. Thus, the above steps ultimately generate a site selection decision report, providing scientific support for planned land use site selection.

[0081] In some of the embodiments described above in this application, steps are proposed to obtain multi-source real-time datasets and multi-source historical datasets and construct a city data cube. However, in this process, multi-source data often suffers from missing data and semantic inconsistencies, resulting in incomplete or logically conflicting spatiotemporal topology maps initially constructed. This, in turn, affects the accuracy and reliability of the city data cube and fails to effectively support the training of subsequent cellular automata models and the generation of land use potential maps.

[0082] In this regard, this application further proposes: in one embodiment, as... Figure 2 As shown, step S10 includes:

[0083] S11: Construct an initial spatiotemporal topology graph based on multi-source real-time datasets and multi-source historical datasets, and extract multi-order neighborhood relationships between nodes in the initial spatiotemporal topology graph;

[0084] In this embodiment, the initial spatiotemporal topology graph is a data structure used to represent the relationships between entities in spatiotemporal data. It abstracts spatial objects (such as sensors and geographical regions) as nodes and spatiotemporal relationships as edges. This graph structure can simultaneously capture spatial adjacency and temporal continuity, providing a foundation for analyzing complex spatiotemporal dependencies. Multi-order neighborhood relationships refer to the associations between nodes established through multi-hop connections. First-order neighborhoods are directly connected nodes, second-order neighborhoods are nodes connected through an intermediate node, and so on. Specifically, it can be extracted through breadth-first search or depth-first search algorithms. Its purpose is to capture deep data relationships to provide a refined structural basis.

[0085] S12: Perform data integrity identification on the initial spatiotemporal topology graph. When data missing is identified, use a pre-built generative adversarial network to complete the missing data and perform semantic consistency verification to generate an improved spatiotemporal topology graph.

[0086] In this embodiment, data integrity identification refers to the process of detecting missing data states, which can be achieved, for example, based on the statistical method of missing value distribution, with the aim of accurately locating incomplete regions; Generative Adversarial Network (GAN) refers to a deep learning generative model. Typically, a GAN contains two neural networks: a generator and a discriminator. The generator is used to generate completed data, and the discriminator is used to distinguish between real data and generated data. The purpose is to learn potential patterns from historical data distributions to generate statistically accurate missing values; Semantic consistency verification refers to the verification process to ensure the semantic rationality of the completed data. It can be understood as a mechanism to verify the logical compliance of the data. Specifically, it can be implemented based on a domain knowledge rule base or ontology reasoning, with the aim of ensuring that the completed data conforms to professional logic and avoids semantic conflicts; Improved spatiotemporal topology graph refers to the high-fidelity graph structure after completion and verification, with the aim of providing a complete and consistent data input foundation.

[0087] S13: Constructing a city data cube based on an improved spatiotemporal topology graph.

[0088] In this embodiment, the city data cube refers to a multidimensional data organization model that structures city data according to spatial, temporal, and thematic dimensions, and supports efficient multidimensional query and analysis operations.

[0089] Specifically, firstly, an initial spatiotemporal topology graph is constructed based on multi-source real-time datasets and multi-source historical datasets, and multi-order neighborhood relationships between nodes are extracted to provide a structured analytical basis for data integrity identification. Subsequently, data integrity identification is performed on the initial spatiotemporal topology graph. When data missing is detected, a generative adversarial network is triggered to complete the data. Simultaneously, semantic consistency verification is used to validate the logical correctness of the completed data, thereby generating an improved spatiotemporal topology graph. Finally, a city data cube is constructed based on the improved spatiotemporal topology graph, transforming complete and consistent data into a multi-dimensional cube structure. This application, through the above-described systematic integration of data completion and semantic verification mechanisms, effectively solves the problems of missing and inconsistent multi-source data, ensuring the spatiotemporal coherence and logical rigor of the city data cube, and providing a reliable data foundation for subsequent cellular automata model training.

[0090] For example, as a specific implementation method, the solution of this application is implemented as follows: In a smart land planning project, real-time satellite remote sensing data and historical land change survey data are acquired. When constructing the initial spatiotemporal topology map, the Neo4j graph database is used to store the node and edge relationships, and multi-order neighborhoods are extracted through graph traversal algorithms. When geological data missing in a certain area is identified, a pre-trained generative adversarial network model is used to generate complete data, and semantic verification is performed using an urban planning knowledge graph. Finally, the improved spatiotemporal topology map is imported into the OLAP system to construct an urban data cube to support land use potential analysis.

[0091] Through the above technical solutions, this application effectively improves the construction quality of urban data cubes, reduces model prediction bias caused by data missingness and semantic conflicts, and enhances the adaptability and decision reliability of multi-objective optimization methods for planning land use in dynamic data environments.

[0092] In some of the embodiments described above in this application, a cellular automata model is proposed to be constructed and trained based on urban data cubes to generate land use potential maps. However, in the process of implementation, there may be problems such as feature extraction being limited to a single scale dimension and the uncertainty of potential distribution not being quantified. This will result in insufficient accuracy and poor adaptability of the generated land use potential maps, making it difficult to effectively capture the complexity of urban spatial evolution and the differences in multiple scenarios, thereby affecting the scientific nature of subsequent site selection decisions.

[0093] In this regard, this application further proposes: in one embodiment, as... Figure 3 As shown, step S20 includes:

[0094] S21: Extract a multi-scale feature set based on the city data cube, and construct an enhanced feature vector describing the cell state based on the multi-scale feature set. The multi-scale feature set includes micro-attribute features, meso-environmental features, and macro-regional features.

[0095] In this embodiment, the multi-scale feature set refers to a collection of urban spatial features extracted from different spatial granularity levels. The multi-scale feature set includes micro-attribute features, meso-environmental features, and macro-regional features, aiming to comprehensively characterize the multi-level influencing factors of urban spatial evolution. The enhanced feature vector refers to a strengthened feature representation formed by integrating the multi-scale feature set through feature fusion technology. It typically employs methods such as feature weighting and dimensionality compression to provide richer and more accurate feature inputs for the cellular state. Micro-attribute features refer to plot-level attributes, such as natural attributes like elevation, slope, and geological conditions. Meso-environmental features refer to neighborhood-level environmental indicators, such as accessibility and service facility coverage. Macro-regional features refer to regional-level development indicators, such as economic density, population distribution, and industrial layout.

[0096] S22: Initialize the cell state based on the enhanced feature vector, construct a cellular automaton engine containing a state transition rule base, and train it through a city data cube to obtain a cellular automaton model;

[0097] In this embodiment, the state transition rule base refers to the set of rules that define the evolution logic of cellular states. It can be constructed based on historical land use change data mining or expert experience rules. Its purpose is to ensure that the cellular automata model conforms to the evolution law of urban land use. The rule types in the state transition rule base include: spatial rules, temporal rules, and constraint rules.

[0098] S23: Construct a multi-rule-oriented scenario parameter library, and inject the scenario parameters in the scenario parameter library as conditional inputs into the cellular automata model. Simulate land use evolution through a parallel computing architecture to generate a set of land use potential distributions.

[0099] In this embodiment, the multi-rule-oriented scenario parameter library refers to a set of parameters storing different planning assumptions. It can be constructed based on a planning knowledge graph or an expert rule base, and its purpose is to support land use evolution simulation under diverse scenarios. Among them, the parameter types of scenario parameters include development-oriented parameters (such as the weights of strategies such as ecological priority and economic priority), constraint parameters (such as development intensity, ecological protection zone and other restrictive indicators), and target weight parameters (weight configuration for multi-objective optimization). Land use evolution simulation refers to using a computational model to predict the spatial changes of regional land use types or cover types over a future period of time. The land use potential distribution set refers to the distribution set of results generated after the land use evolution simulation runs under different scenarios. Each potential distribution in the land use potential distribution set can be understood as a probability map or a weighted scoring map. The value of the grid cell represents the relative probability or suitability of development at that location. When the cellular automaton model runs to a future point in time under each scenario, it outputs the development probability or state transition probability of each cell at that time, thus forming the land use potential distribution map.

[0100] S24: Adaptively integrate the potential distribution of multiple scenarios using a pre-set attention-weighted fusion algorithm to generate a preliminary application map.

[0101] In this embodiment, the pre-set attention-weighted fusion algorithm is a multi-source information fusion method based on the attention mechanism, which can adaptively adjust the weight of the prediction results under different scenarios. It can be implemented using an attention module based on Transformer, with the aim of highlighting the contribution of key scenarios to the potential distribution. The preliminary map is an intermediate result map generated after the cellular automata model completes the simulation of land use evolution under multiple scenarios and undergoes data fusion and optimization processing.

[0102] S25: Calculate the confidence interval of the potential of each grid cell in the preliminary land potential map, and generate the land potential map through a pre-established multidimensional assessment system.

[0103] In this embodiment, confidence interval calculation refers to a statistical method for quantifying the fluctuation range of potential estimates, which can be implemented based on Bootstrap resampling technology. Its purpose is to assess the reliability of potential values. The multidimensional evaluation system refers to a framework for verifying the quality of potential maps by integrating multiple dimensions. It may include indicators such as spatial consistency, temporal stability, institutional compliance, and risk controllability. Its purpose is to ensure the decision applicability of the generated map.

[0104] Specifically, enhanced feature vectors are constructed through multi-scale feature extraction to provide a comprehensive environmental description for cell state initialization. Based on the enhanced feature vectors, a cellular automaton engine is built and trained, enabling the state transition rule base to dynamically adapt to the historical and real-time data patterns of the urban data cube. On this basis, parameters from a multi-rule-oriented scenario parameter library are injected into the model, driving a parallel computing architecture to perform land use evolution simulations at multiple time scales, generating a potential distribution set covering different planning assumptions. Subsequently, an attention-weighted fusion algorithm is used to adaptively integrate the multi-scenario distributions, dynamically adjusting the weights according to the importance of the scenarios to avoid dilution of key information. Finally, the uncertainty of the potential of each grid cell is quantified through confidence intervals, and a highly reliable land use potential map is generated by combining a multi-dimensional evaluation system. This process forms a closed loop through multi-scale feature fusion, dynamic scenario deduction, and uncertainty quantification, solving the problems of one-sided feature extraction, rigid scenario simulation, and lack of uncertainty handling.

[0105] For example, as a specific implementation method, the solution of this application is implemented as follows: The constructed urban data cube integrates satellite remote sensing imagery, IoT sensor network data, and urban planning documents; from the urban data cube, micro-scale plot slope and building height, meso-scale road density and green space coverage, and macro-scale regional population density and economic indicators are extracted to construct an enhanced feature vector; the enhanced feature vector is used to initialize the cell state, and a cellular automaton engine is trained based on historical land use change data; scenario parameters from the scenario parameter library are injected into the cellular automaton model as conditional inputs. The scenario parameter library includes scenarios such as high-density development and ecological protection priority, and the parameter vectors are quantified through expert rules; after the parameters are injected, the model performs land use evolution simulation on a parallel computing architecture to generate a probabilistic potential distribution; an attention mechanism is used to adaptively integrate the potential distributions of multiple scenarios to generate a preliminary land use map; the confidence interval of the potential of each grid cell is calculated, and a land use potential map is generated through a multi-dimensional evaluation system that includes spatial accuracy, temporal consistency, and planning compliance.

[0106] Through the above technical solutions, this application effectively improves the accuracy and adaptability of land use potential maps, avoids model distortion caused by single-scale features, overcomes the shortcomings of static simulation in responding to planning intentions, and ensures that the map reflects both potential values ​​and decision confidence, providing a comprehensive and robust data foundation for subsequent site selection decisions.

[0107] In one embodiment, step S23 includes:

[0108] S231: Obtain spatial planning text information, extract planning guidance keywords through natural language processing technology, and establish a planning knowledge network;

[0109] In this embodiment, spatial planning text information refers to a collection of effective normative documents within the national spatial planning system, encompassing multi-level and multi-type structured and unstructured text data. Spatial planning text information serves as the original information carrier and authoritative basis for this analysis process. Typically, spatial planning text information includes overall planning texts, special planning texts, control detailed plans, and regulatory documents. Natural language processing technology refers to the technical means of semantically parsing unstructured text using computers. This can be achieved using pre-trained language models or rule matching algorithms, aiming to automatically extract key elements from planning texts and reduce subjective bias caused by human interpretation. Planning guidance keywords refer to core vocabulary that reflects the strategic intent and regulatory tendencies of the plan, including strategic positioning, development system, spatial control, and implementation guarantee categories. The planning knowledge network refers to a semantic network graph structure model formed by structuring the extracted planning elements. It can be constructed as a directed graph containing nodes and edges, where nodes represent planning elements and edges represent constraints or relationships between elements. Its purpose is to transform unstructured text into structured knowledge representation.

[0110] S232: Construct multi-objective-oriented scenario parameter templates based on planning knowledge networks, and quantify the scenario parameter templates into scenario parameter vectors;

[0111] In this embodiment, the scenario parameter template refers to a standardized parameter framework that transforms planning knowledge into a model readable form. The template design elements include parameter type, value range, and association rules. The scenario parameter vector specifically refers to a numerical expression that describes the characteristics of the planning scenario. It can be represented as a multi-dimensional vector containing economic, social, and ecological dimensions. Its purpose is to transform multi-objective planning requirements into computable input parameters.

[0112] S233: Inject scenario parameter vectors into cellular automata models and simulate land use evolution across multiple time scales using a parallel computing architecture;

[0113] In this embodiment, multi-timescale land use evolution simulation refers to a predictive analysis method based on dynamic systems theory. By establishing computational models with different time dimensions, it simulates the evolution patterns and spatial expressions of land use or cover changes at different time stages, in order to reveal the process characteristics and long-term trends of land use changes.

[0114] S234: Integrate the land use evolution simulation results to generate a probabilistic land use potential distribution, and quantify its uncertainty through Monte Carlo simulation;

[0115] In this embodiment, probabilistic land use potential distribution refers to a land use suitability evaluation method based on uncertainty theory. It typically uses a probabilistic model to express the suitability of a specific plot for a certain development type, aiming to quantify the spatial distribution characteristics of different development possibilities through probability values. Monte Carlo simulation is a computational method based on probability statistics and random sampling. It solves numerical solutions to problems through a large number of repeated random experiments. Its essence is to explore the behavioral characteristics of deterministic systems through randomness. Uncertainty quantification is a technical system for measuring the credibility of analysis results. Uncertainty includes data quality uncertainty, model parameter uncertainty, and external environmental uncertainty. Through the uncertainty quantification method based on Monte Carlo simulation, more scientific and reliable technical support is provided for land use potential analysis, significantly improving the scientificity and reliability of planning decisions.

[0116] Furthermore, in this embodiment, the implementation path of Monte Carlo simulation includes:

[0117] Parameter space sampling: Establish probability distribution models for key parameters affecting land use potential (such as population growth rate, land price fluctuation coefficient, etc.);

[0118] Stochastic process simulation: generating a large number of random number combinations that conform to a specific distribution using a computer;

[0119] Repeated calculation iteration: The land potential calculation model is run each time with a different combination of random parameters;

[0120] Statistical analysis: Statistical analysis of a large number of simulation results yields the probability distribution characteristics of land use potential.

[0121] S235: Based on the uncertainty quantification results, conduct spatial significance tests on the distribution of probabilistic land use potential to identify potential hotspot areas and high-risk areas;

[0122] In this embodiment, spatial significance testing refers to identifying regions with significant spatial clustering characteristics through statistical methods. This can be achieved using global spatial autocorrelation (such as Moran's index, Gillley index, etc.), local spatial autocorrelation (such as LISA analysis, hotspot analysis, etc.), and spatial regression analysis (such as geographically weighted regression, etc.). The purpose is to objectively distinguish between real potential hotspots and randomly fluctuating regions. Potential hotspot regions refer to spatial clusters that show significant development advantages in land use potential analysis. Potential hotspot regions typically have the following characteristics: statistical significance (significantly high-value clusters identified through spatial autocorrelation analysis), development suitability (showing high scores in multiple evaluation index systems), and spatial continuity (forming spatial clusters of a certain scale, rather than isolated distributions). High-risk regions refer to spatial units with significant uncertainties and potential losses in the process of land use development. Their main risk types include: ecological risk, market risk, and institutional risk.

[0123] S236: Perform spatial overlay analysis on potential hotspot areas and high-risk areas to generate a land use potential distribution set, and label each potential unit in the land use potential distribution set with recommendation priority and risk level.

[0124] In this embodiment, spatial overlay analysis specifically refers to spatial data fusion technology in the GIS field, which can be achieved based on algebraic operations of raster data or topological overlay of vector data. Its purpose is to comprehensively evaluate the overlay effect of potential and risk. Potential unit refers to the smallest spatial analysis unit in land use potential analysis. It is the basic spatial unit that constitutes the land use potential distribution map. Each potential unit can represent a geographical area with independent development potential and has clear boundaries and location attributes in space.

[0125] Specifically, by acquiring spatial planning text information and automatically extracting planning guidance keywords using natural language processing technology, unstructured text is transformed into a structured planning knowledge network. Based on this planning knowledge network, a multi-objective-oriented scenario parameter template is constructed and quantified into an input-capable numerical vector. The quantified scenario parameter vector is then injected as conditional input into a cellular automata model, and land use evolution simulations at multiple time scales are executed synchronously through a parallel computing architecture. The simulation results are integrated to generate a probabilistic land use potential distribution, and its spatial uncertainty is quantified using Monte Carlo simulation. Based on the quantification results, a spatial significance test is performed on the probability distribution to objectively identify potential hotspots and high-risk areas with statistical significance. Finally, spatial overlay analysis is performed on the identified key areas to generate a set of land use potential distributions that include recommendation priorities and risk level labels. This technical process forms a complete closed loop from text parsing to potential assessment: the planning knowledge network provides a structured foundation for the scenario parameter template, the quantified scenario parameter vector drives the dynamic deduction of the cellular automata model, the quantification of the uncertainty of the simulation results supports the reliability of the spatial significance test, and the test results directly guide the input conditions for the spatial overlay analysis, thereby ensuring the automation and objectivity of the entire process.

[0126] For example, as a specific implementation method, the spatial planning text information can be specifically urban master plan text; natural language processing technology can use the BERT pre-trained model for keyword extraction; the planning knowledge network can be specifically constructed as a graph structure containing planning element nodes and constraint relationship edges; the scenario parameter template can be specifically quantified as a three-dimensional vector containing economic intensity, social demand, and ecological sensitivity dimensions; the cellular automata model can be specifically deployed on a GPU cluster for parallel computing; the spatial saliency test can specifically use the Z-test method to identify areas with a confidence level greater than 95%; and the spatial overlay analysis can specifically use GIS software to perform layer overlay operations on potential hotspots and high-risk areas to generate a potential distribution set containing high, medium, and low recommendation priorities and corresponding risk levels.

[0127] Through the above technical solutions, this application achieves automated parsing and structured transformation of spatial planning texts, reducing inconsistencies in rules caused by manual interpretation; it transforms multi-dimensional planning objectives into quantifiable numerical expressions, providing an objective basis for multi-objective optimization; it accelerates the dynamic extrapolation process across multiple time scales through a parallel computing architecture, effectively reflecting real-time changes in urban operations; it utilizes Monte Carlo simulation to quantify the uncertainty of simulation results, providing a risk assessment basis for decision-making; it objectively identifies potential hotspots and high-risk areas through statistical methods, ensuring the accuracy of key area identification; and finally, it generates a potential distribution set labeled with recommendation priorities and risk levels through spatial overlay analysis, supporting multi-objective trade-off decisions and significantly improving the scientific rigor and reliability of site selection schemes.

[0128] In some of the embodiments described above in this application, a game-theoretic decision-making model based on a multi-agent reinforcement learning framework is proposed to generate a preliminary set of location schemes. However, in this process, the generated initial equilibrium solution set may contain unstable solutions that are sensitive to parameter fluctuations, resulting in insufficient robustness of the location scheme in practical applications. It cannot effectively cope with the uncertainty in multi-objective decision-making, thereby affecting the scientificity and reliability of the final location decision.

[0129] In this regard, this application further proposes that, in one embodiment, step S30 includes:

[0130] S31: Construct a game decision model based on land use potential maps, including spatially constrained agents, economic utility agents, comprehensive utility agents, and ecological utility agents, and design corresponding objective functions for each agent in the game decision model;

[0131] In this embodiment, the spatial constraint agent refers to the decision-making unit representing spatial planning constraints, which can be implemented using spatial analysis rules based on geographic information systems or topological relationship reasoning models. Its purpose is to ensure that the site selection scheme conforms to spatial layout specifications. The economic utility agent refers to the decision-making unit quantifying economic benefits, which can be implemented using return on investment calculation models or cost-benefit analysis frameworks. Its purpose is to optimize the efficiency of economic resource allocation. The comprehensive utility agent refers to the decision-making unit assessing social impact, which can be implemented using population density weighted models or infrastructure service accessibility index systems. Its purpose is to improve comprehensive utility. The ecological utility agent refers to the agent that measures... The decision-making unit for ecological impact can be implemented using ecological sensitivity index assessment methods or biodiversity conservation models, with the aim of minimizing the risk of ecological damage. The game-theoretic decision-making model refers to a distributed decision-making framework based on a multi-agent system, which models the complex land use selection problem as a strategic interaction process among multiple decision-making agents. This game-theoretic decision-making model achieves the synergistic optimization of the interests of multiple parties by simulating the behavioral logic of different stakeholders in the real world. The objective function refers to the mathematical expression that defines the optimization objective of each agent, which can be implemented using linear functions, quadratic functions, or piecewise nonlinear functions. Its purpose is to transform multi-dimensional decision objectives into quantifiable optimization criteria.

[0132] S32: Identify high-potential candidate areas in the land use potential map and drive the agents of the game decision model to play multiple rounds of games in the high-potential candidate areas until they converge to Nash equilibrium, thus obtaining the initial equilibrium solution set;

[0133] In this embodiment, a high-potential candidate area refers to a region in the land use potential map whose potential value exceeds a dynamic threshold. It can be determined based on the statistical quantiles of the potential distribution or spatial clustering algorithms. The purpose is to focus on high-potential areas to improve computational efficiency. Nash equilibrium refers to a state in which no agent can improve its own utility by unilaterally changing its strategy when the game reaches equilibrium. The initial equilibrium solution set is the primary stable state reached in the process of multi-agent game. It is the set of strategy combinations formed by the players under the existing constraints.

[0134] S33: Perform sensitivity analysis on the initial equilibrium solution set, remove unstable solutions that are sensitive to parameter fluctuations, obtain the optimized equilibrium solution set, and generate a preliminary location scheme set based on the optimized equilibrium solution set.

[0135] In this embodiment, sensitivity analysis refers to assessing the impact of parameter perturbations on solution stability. This can be achieved using the finite difference method or parameter perturbation simulation techniques, with the aim of identifying vulnerable solutions sensitive to parameter fluctuations. Furthermore, key indicators for sensitivity analysis include: the first-order sensitivity index (the independent influence of a single parameter) and the total-order sensitivity index (the influence of interactions between parameters). Unstable solutions sensitive to parameter fluctuations refer to equilibrium solutions in multi-objective optimization games that exhibit high sensitivity to small changes in input parameters or environmental conditions. Although these solutions reach equilibrium under the current parameter settings, their stability is severely flawed, making it difficult to maintain robustness in practical applications. The optimal equilibrium solution set is a high-quality solution set obtained through sensitivity analysis and robustness screening. The preliminary site selection scheme set refers to the result of transforming the optimal equilibrium solution set into an operable planning scheme.

[0136] Specifically, by decomposing multi-objective decision-making into multiple interrelated optimization sub-problems, spatially constrained agents, economic utility agents, comprehensive utility agents, and ecological utility agents interact strategically within high-potential candidate areas defined by land use potential maps. Each agent continuously adjusts its decision-making strategy based on an independently designed objective function, converging to a Nash equilibrium state after multiple rounds of iterative game, forming an initial equilibrium solution set. Subsequently, by quantifying the impact of parameter fluctuations on the solution set, stable solutions that are insensitive to parameter changes are selected, effectively filtering out fragile schemes susceptible to external disturbances, thereby obtaining a robust optimized equilibrium solution set and ensuring that the generated preliminary site selection scheme set can adapt to the uncertainties in the actual decision-making environment.

[0137] For example, as a preferred embodiment, the solution of this application is implemented as follows: In a site selection project for an industrial park in a certain city, the top 25% of the potential value areas are extracted as high-potential candidate areas based on the land potential map; a game decision model is constructed, which includes a spatially constrained agent (using a spatial topology relationship rule library), an economic utility agent (using a dynamic investment return rate model), a comprehensive utility agent (using an infrastructure service coverage radius weighted algorithm), and an ecological utility agent (using an ecological sensitivity index evaluation module); each agent is driven to perform multiple rounds of strategy optimization in the candidate areas, and when the strategy change is less than a preset convergence threshold for 10 consecutive rounds, it is determined that a Nash equilibrium has been reached, generating an initial equilibrium solution set containing 15 schemes; a ±3% parameter perturbation is introduced for sensitivity analysis, and 9 unstable solutions that are sensitive to parameter fluctuations are identified and eliminated by calculating the solution set stability index, and 6 robust schemes are retained to form an optimized equilibrium solution set, finally generating a preliminary site selection scheme set.

[0138] Through the above technical solutions, this application effectively improves the stability and robustness of the site selection scheme set, enabling the generated schemes to adapt to parameter fluctuations in actual decision-making, thereby improving the scientificity and reliability of site selection decisions.

[0139] In some of the embodiments described above in this application, a preliminary set of site selection schemes is proposed to be input into a cellular automata model for multi-scenario dynamic extrapolation to generate an evolutionary simulation dataset. However, in its implementation, the extrapolation process may be oversimplified, failing to fully consider the fine configuration of multi-scenario parameters, the dynamic iteration mechanism of multiple time steps, and the systematic recording of multi-dimensional time series data. This results in the generated evolutionary simulation dataset failing to fully capture the spatiotemporal dynamic characteristics and uncertainties of urban land use evolution, thereby affecting the accuracy of subsequent assessments and the adaptability of site selection decisions.

[0140] In this regard, this application further proposes that, in one embodiment, step S40 includes:

[0141] S41: Identify the spatial parameters of each location scheme in the preliminary location scheme set, and map the identified spatial parameters to the initial state of the cellular automata model;

[0142] In this embodiment, spatial parameters refer to the spatial attribute characteristics of the site selection scheme, which can be implemented using vector polygon data or rasterized coordinate matrices in a geographic information system. The purpose is to accurately characterize the spatial range and geometric characteristics of the site selection area. Spatial parameters include geometric features, location features, and relational features. Geometric parameters include land boundary coordinates, area, perimeter, and shape index. Location parameters include center point coordinates, elevation, slope, and aspect. Relational parameters include distance to main traffic arteries, distance to ecological protection zones, and connectivity with built-up areas. Initial state mapping refers to transforming abstract spatial parameters into an initial configuration state that can be recognized by the cellular automata model. Its mapping rules include: attribute mapping (mapping land use properties to cellular state codes), intensity mapping (mapping development intensity to cellular state intensity values), and relational mapping (mapping spatial relationships to neighborhood influence weights).

[0143] S42: Configure multi-scenario inference strategy parameters for the cellular automaton model and drive the cellular automaton model to perform iterative calculations at multiple time steps;

[0144] In this embodiment, the multi-scenario simulation strategy parameters refer to the set of configuration variables for different simulation scenarios. These parameters can be implemented using predefined scenario template vectors or dynamically adjusted parameter matrices, with the aim of supporting the simulation of diverse institutional orientations and environmental conditions. The multi-time-step iterative calculation specifically refers to the phased time-series simulation process, which can be implemented using discrete step sizes with fixed time intervals or adaptive step sizes based on event triggers, with the aim of capturing short-term fluctuations and long-term trends in urban land use evolution.

[0145] S43: Record multidimensional time-series data for each time step during the iterative calculation process, and construct an evolutionary simulation dataset based on the recorded multidimensional time-series data. The multidimensional time-series data includes the transformation process of cell states, the evolution trajectory of spatial patterns, and the dynamic values ​​of key indicators.

[0146] In this embodiment, multidimensional time-series data refers to holographic data of system state recorded in a time sequence, including three dimensions: space, attributes, and relationships. Multidimensional time-series data includes the transformation process of cell states, the evolution trajectory of spatial patterns, and the dynamic values ​​of key indicators. The transformation process of cell states refers to the evolution trajectory of states at the individual level; the evolution trajectory of spatial patterns refers to the change path of the overall spatial structure of the region, which can be characterized using morphological indicator sequences or spatial autocorrelation indices, with the aim of quantifying the evolutionary laws of macro-patterns; the dynamic values ​​of key indicators refer to the time-series change data of evaluation indicators, which can be stored using indicator time series or dynamic scoring curves, with the aim of monitoring the evolution trend of core performance; the evolutionary simulation dataset refers to a structured spatiotemporal data set generated after multi-scenario dynamic extrapolation through a cellular automata model. This evolutionary simulation dataset fully records the dynamic evolution trajectory of land use selection schemes in the long-term development process.

[0147] Specifically, by identifying the spatial parameters of the preliminary site selection scheme and accurately mapping them to the initial state of the cellular automata model, the spatial configuration of the simulation starting point is ensured to be highly consistent with that of the actual site selection scheme. On this basis, by configuring multi-scenario simulation strategy parameters for the model and driving it to perform iterative calculations at multiple time steps, the dynamic process of land use evolution under different institutional orientations is continuously tracked. Finally, by recording multi-dimensional time-series data at each time step, including cellular state transitions, spatial pattern evolution, and key indicator dynamics, a complete evolution evidence chain is constructed, enabling the evolution simulation dataset to comprehensively reflect the spatiotemporal dynamic characteristics and uncertainties of urban land use evolution.

[0148] For example, as a preferred embodiment, the solution of this application is implemented as follows: In a new urban area planning project, the spatial parameters (specifically, the polygonal data of the plot boundaries) of the generated preliminary site selection scheme set are mapped to the initial state of the cellular automaton model; a multi-scenario inference strategy parameter vector containing different transportation development systems and population growth scenarios is configured; the model is driven to perform iterative calculations for ten years with quarterly time steps; during the calculation process, the cellular state transition log, spatial pattern morphology index sequence, and dynamic values ​​of key indicators (such as development intensity and ecological sensitivity) for each quarter are recorded synchronously; finally, an evolutionary simulation dataset containing complete time-series information is constructed based on these records.

[0149] Through the above technical solution, this application effectively solves the problem of oversimplification in the extrapolation process, enabling the generated evolutionary simulation dataset to fully capture the spatiotemporal dynamic characteristics and uncertainties of urban land use evolution, thereby providing a high-fidelity data foundation for subsequent multi-objective assessment and significantly improving the adaptability and scientific nature of site selection decisions.

[0150] In one embodiment, step S50 includes:

[0151] S51: Extract multidimensional evaluation indicators based on evolutionary simulation datasets and establish a multi-objective evaluation system that includes positive and negative indicators;

[0152] In this embodiment, the evolutionary simulation dataset refers to a collection of spatiotemporal sequence data generated through multi-scenario dynamic extrapolation using cellular automata, recording the long-term evolution of land use schemes under different development paths. Extracting multidimensional evaluation indicators refers to extracting core evaluation dimensions from complex simulation data using methods such as principal component analysis and factor analysis. These multidimensional evaluation indicators are a quantitative indicator system used to evaluate the comprehensive effectiveness of land use selection schemes. They construct a complete evaluation framework from multiple dimensions, including economic, social, environmental, and spatial aspects, providing a comprehensive basis for scheme comparison. Furthermore, the multidimensional evaluation indicators adopt a hierarchical structure, decomposing complex evaluation objectives into quantifiable components. Specific indicators; a multi-objective evaluation system refers to a structured framework used to systematically evaluate complex decision-making problems. By establishing a multi-level, multi-angle indicator system, it achieves a comprehensive evaluation of the plan. The multi-objective evaluation system can simultaneously consider multiple conflicting or competing objectives, providing a comprehensive and scientific basis for decision-making; positive indicators are evaluation indicators whose larger values ​​indicate better performance, also known as benefit-type indicators or high-value indicators. Positive indicators directly reflect the positive effects and benefit levels of the plan; negative indicators are evaluation indicators whose smaller values ​​indicate better performance, also known as cost-type indicators or low-value indicators. Negative indicators mainly reflect the negative impacts and cost expenditures of the plan.

[0153] S52: Select the optimal and worst values ​​of each evaluation index from the preliminary site selection scheme set based on the multi-dimensional evaluation index to form the positive ideal solution and the negative ideal solution;

[0154] In this embodiment, a positive ideal solution refers to a virtual ideal solution consisting of the optimal values ​​of all evaluation indicators in the solution set; a negative ideal solution refers to a virtual worst solution consisting of the worst values ​​of all evaluation indicators in the solution set.

[0155] S53: Calculate the Euclidean distance between each scheme in the preliminary site selection scheme set and the positive ideal solution and the negative ideal solution, and calculate their relative proximity.

[0156] In this embodiment, Euclidean distance refers to the straight-line distance between two points in space, which is the most intuitive distance measurement method. In the multidimensional evaluation of this embodiment, each solution is regarded as a point in a high-dimensional space. The similarity is quantified by calculating the straight-line distance between points. The smaller the Euclidean distance, the closer the solution is to the ideal state. Specifically, the Euclidean distance is represented as follows: each solution corresponds to a coordinate point in space, each evaluation index represents a coordinate axis, and the index value determines the coordinate value of the point on the corresponding axis. The Euclidean distance reflects the comprehensive difference between solutions. Relative proximity refers to the degree of closeness to the relative position of the positive and negative ideal solutions, rather than the absolute degree of closeness. It evaluates the quality of the solution by comparing the relative relationship between the solution and the "optimal" and "worst". Specifically, the relative proximity is defined as follows: the positive ideal solution and the negative ideal solution are regarded as two anchor points in space. The relative proximity describes the projection position of the solution on the line connecting these two anchor points. The closer to the positive ideal solution, the larger the proximity value, and the closer to the negative ideal solution, the smaller the proximity value.

[0157] Specifically, by systematically analyzing the spatiotemporal dynamic information in the evolutionary simulation dataset, multidimensional evaluation indicators that comprehensively reflect the long-term performance of the schemes are extracted, and positive and negative indicator classification systems are constructed according to benefit-type and cost-type attributes. On this basis, by identifying the extreme value distribution of each indicator in the scheme set, positive ideal solutions representing the theoretical optimal situation and negative ideal solutions representing the theoretical worst situation are constructed respectively, providing an absolute reference benchmark for scheme evaluation. Finally, by calculating the Euclidean distance between each scheme and the positive and negative ideal solutions in the high-dimensional indicator space, and quantifying the degree of closeness between the schemes and the ideal solutions based on the relative closeness model, the schemes are accurately ranked and classified.

[0158] For example, as a preferred embodiment, the specific implementation of the scheme in this application is as follows: In a new city development project, based on an evolutionary dataset containing 20 years of simulated data, indicators such as economic benefits (e.g., land appreciation rate of return, infrastructure investment efficiency), comprehensive benefits (e.g., infrastructure service coverage, job-housing balance index), and ecological benefits (e.g., carbon sink capacity change rate, biodiversity maintenance degree) are extracted; the optimal and worst values ​​of each indicator in the scheme set are determined through extreme value analysis, and positive ideal solutions (e.g., highest rate of return, maximum coverage, strongest carbon sink capacity) and negative ideal solutions (e.g., lowest rate of return, minimum coverage, weakest carbon sink capacity) are constructed; after determining the indicator weights using the entropy weight method, the weighted Euclidean distance between each scheme and the ideal solution is calculated, and finally the scheme ranking results are obtained through relative closeness calculation, where the relative closeness of the optimal scheme reaches 0.87, which is significantly better than other schemes.

[0159] Through the above technical solutions, this application establishes a complete technical chain from dynamic simulation to comprehensive evaluation. By extracting multi-dimensional indicators, constructing ideal solutions, and calculating relative proximity, it achieves full life-cycle performance evaluation of site selection schemes, providing quantitative basis for scientific decision-making. This has the effect of improving the systematicness and accuracy of scheme comparison and selection, and effectively reducing decision-making risks.

[0160] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0161] In one embodiment, an AI-based multi-objective optimization system for planned land use site selection is provided. This AI-based multi-objective optimization system for planned land use site selection corresponds one-to-one with the AI-based multi-objective optimization method for planned land use site selection described in the previous embodiment. The AI-based multi-objective optimization system for planned land use site selection includes:

[0162] The data acquisition module is used to acquire multi-source real-time datasets and multi-source historical datasets, and to construct a city data cube based on the multi-source real-time datasets and multi-source historical datasets.

[0163] The map generation module is used to build and train a cellular automaton model based on urban data cubes, and generate a land use potential map based on the trained cellular automaton model.

[0164] The first model building module is used to build a game decision model based on a multi-agent reinforcement learning framework according to the land use potential map, obtain the site selection project parameters and input them into the game decision model, so that the game decision model can generate a preliminary site selection scheme set through a preset multi-level decision strategy.

[0165] The deduction module is used to input the preliminary site selection scheme set into the cellular automata model for multi-scenario dynamic deduction and generate an evolutionary simulation dataset;

[0166] The second model building module is used to build a multi-objective evaluation model based on the evolutionary simulation dataset, and to calculate the relative closeness between each scheme in the preliminary site selection scheme set and the ideal solution through the multi-objective evaluation model;

[0167] The report generation module is used to sort and classify the schemes in the preliminary site selection scheme set according to the obtained relative proximity, and generate a site selection decision report.

[0168] Specific limitations regarding the AI-based multi-objective optimization system for land use planning and site selection can be found in the limitations of the AI-based multi-objective optimization method for land use planning and site selection described above, and will not be repeated here. Each module in the aforementioned AI-based multi-objective optimization system for land use planning and site selection can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0169] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements an AI-based multi-objective optimization method for planning and land use site selection.

[0170] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements an AI-based multi-objective optimization method for site selection in planning.

[0171] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A multi-objective optimization method for land use planning based on AI, characterized in that, Including the following steps: Acquire multi-source real-time datasets and multi-source historical datasets, and construct a city data cube based on the multi-source real-time datasets and multi-source historical datasets. The multi-source real-time datasets include traffic dynamics data, population flow and activity data, environmental monitoring data, and public facility usage data. A cellular automaton model is constructed and trained based on urban data cubes, and a land use potential map is generated based on the trained cellular automaton model. Based on the land use potential map, a game decision model based on a multi-agent reinforcement learning framework is constructed. The parameters of the site selection project are obtained and input into the game decision model, so that the game decision model generates a preliminary set of site selection schemes through a preset multi-level decision strategy. The initial site selection scheme set is input into the cellular automata model for multi-scenario dynamic simulation to generate an evolutionary simulation dataset; A multi-objective evaluation model was constructed based on the evolutionary simulation dataset, and the relative closeness of each scheme in the preliminary site selection scheme set to the ideal solution was calculated through the multi-objective evaluation model. Based on the obtained relative proximity, the preliminary site selection schemes are ranked and graded to generate a site selection decision report; The step of constructing and training a cellular automaton model based on urban data cubes, and generating a land use potential map based on the trained cellular automaton model, includes the following steps: Multi-scale feature sets are extracted based on urban data cubes, and enhanced feature vectors describing cell states are constructed based on the multi-scale feature sets. The multi-scale feature sets include micro-attribute features, meso-environmental features, and macro-regional features. Cellular states are initialized based on enhanced feature vectors, a cellular automaton engine containing a state transition rule base is constructed, and the engine is trained using a city data cube to obtain a cellular automaton model. A multi-rule-oriented scenario parameter library is constructed, and the scenario parameters in the scenario parameter library are injected into the cellular automata model as conditional inputs. Land use evolution simulation is performed through a parallel computing architecture to generate a set of land use potential distributions. A pre-set attention-weighted fusion algorithm is used to adaptively integrate the potential distributions of multiple scenarios to generate a preliminary application map. Calculate the confidence interval of the potential of each grid cell in the preliminary land use potential map, and generate the land use potential map through a pre-established multidimensional assessment system; The step of inputting the preliminary site selection scheme set into the cellular automata model for multi-scenario dynamic inference and generating an evolutionary simulation dataset includes the following steps: Identify the spatial parameters of each location scheme in the preliminary location scheme set, and map the identified spatial parameters to the initial state of the cellular automata model; Configure multi-scenario inference strategy parameters for the cellular automaton model and drive the cellular automaton model to perform iterative calculations at multiple time steps; Multidimensional time-series data at each time step during the iterative calculation process are recorded, and an evolutionary simulation dataset is constructed based on the recorded multidimensional time-series data. The multidimensional time-series data includes the transformation process of cell states, the evolution trajectory of spatial patterns, and the dynamic values ​​of key indicators.

2. The AI-based multi-objective optimization method for land use planning as described in claim 1, characterized in that: The steps of acquiring multi-source real-time datasets and multi-source historical datasets, and constructing a city data cube based on the multi-source real-time datasets and multi-source historical datasets, include the following steps: An initial spatiotemporal topology graph is constructed based on multi-source real-time datasets and multi-source historical datasets, and multi-order neighborhood relationships between nodes in the initial spatiotemporal topology graph are extracted. Data integrity is identified in the initial spatiotemporal topology graph. When data is missing, a pre-built generative adversarial network is used to fill in the missing data and perform semantic consistency verification to generate an improved spatiotemporal topology graph. Constructing a city data cube based on an improved spatiotemporal topology graph.

3. The AI-based multi-objective optimization method for land use planning as described in claim 1, characterized in that: The steps of constructing a multi-rule-oriented scenario parameter library, injecting scenario parameters from the library as conditional inputs into a cellular automata model, simulating land use evolution through a parallel computing architecture, and generating a set of land use potential distributions include the following steps: Acquire spatial planning text information, extract planning guidance keywords using natural language processing technology, and establish a planning knowledge network; A multi-objective-oriented scenario parameter template is constructed based on a planning knowledge network, and the scenario parameter template is quantified into a scenario parameter vector; Scenario parameter vectors are injected into cellular automata models, and land use evolution simulations are performed at multiple time scales through a parallel computing architecture. The results of land use evolution simulations are integrated to generate a probabilistic land use potential distribution, and its uncertainty is quantified through Monte Carlo simulation. Based on the uncertainty quantification results, a spatial significance test is conducted on the probabilistic land use potential distribution to identify potential hotspot areas and high-risk areas. Spatial overlay analysis is performed on potential hotspot areas and high-risk areas to generate a land use potential distribution set, and each potential unit in the land use potential distribution set is labeled with recommendation priority and risk level.

4. The AI-based multi-objective optimization method for land use planning and site selection according to claim 1, characterized in that: The steps of constructing a game-theoretic decision-making model based on a multi-agent reinforcement learning framework according to the land use potential map, obtaining site selection project parameters and inputting them into the game-theoretic decision-making model, and enabling the game-theoretic decision-making model to generate a preliminary set of site selection schemes through a preset multi-level decision-making strategy, include the following steps: Based on land use potential maps, a game-theoretic decision-making model is constructed, which includes spatially constrained agents, economic utility agents, comprehensive utility agents, and ecological utility agents. Corresponding objective functions are designed for each agent in the game-theoretic decision-making model. High-potential candidate areas are identified in the land use potential map, and the agents of the game decision model are driven to play multiple rounds of games in the high-potential candidate areas until they converge to Nash equilibrium, thus obtaining the initial equilibrium solution set. Sensitivity analysis is performed on the initial equilibrium solution set, and unstable solutions that are sensitive to parameter fluctuations are removed to obtain the optimized equilibrium solution set. A preliminary location scheme set is then generated based on the optimized equilibrium solution set.

5. The AI-based multi-objective optimization method for land use planning as described in claim 1, characterized in that: The steps of constructing a multi-objective evaluation model based on an evolutionary simulation dataset and calculating the relative closeness of each scheme in the preliminary site selection scheme set to the ideal solution using the multi-objective evaluation model include the following steps: Multidimensional evaluation indicators were extracted based on evolutionary simulation datasets, and a multi-objective evaluation system including positive and negative indicators was established. The optimal and worst values ​​of each evaluation index based on the multidimensional evaluation index are selected from the preliminary site selection scheme set to form the positive ideal solution and the negative ideal solution. Calculate the Euclidean distance between each scheme in the preliminary site selection scheme set and the positive ideal solution and the negative ideal solution, and calculate their relative proximity.

6. An AI-based multi-objective optimization system for planning land use site selection, used to execute the steps of the AI-based multi-objective optimization method for planning land use site selection as described in any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire multi-source real-time datasets and multi-source historical datasets, and to construct a city data cube based on the multi-source real-time datasets and multi-source historical datasets. The map generation module is used to build and train a cellular automaton model based on urban data cubes, and generate a land use potential map based on the trained cellular automaton model. The first model building module is used to build a game decision model based on a multi-agent reinforcement learning framework according to the land use potential map, obtain the site selection project parameters and input them into the game decision model, so that the game decision model can generate a preliminary site selection scheme set through a preset multi-level decision strategy. The deduction module is used to input the preliminary site selection scheme set into the cellular automata model for multi-scenario dynamic deduction and generate an evolutionary simulation dataset; The second model building module is used to build a multi-objective evaluation model based on the evolutionary simulation dataset, and to calculate the relative closeness between each scheme in the preliminary site selection scheme set and the ideal solution through the multi-objective evaluation model; The report generation module is used to sort and classify the schemes in the preliminary site selection scheme set according to the obtained relative proximity, and generate a site selection decision report.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the AI-based multi-objective optimization method for land use planning as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the AI-based multi-objective optimization method for land use planning as described in any one of claims 1-5.

Images