A multi-scale urban garden carbon sink improvement and ecological synergistic regulation system

Abstract

The present application relates to a multi-scale urban garden carbon sink improvement and ecological collaborative regulation system, specifically relates to the field of urban gardens, by constructing a computable and evolvable knowledge graph, deeply integrating multi-source data and field rules, and dynamically driving spatial optimization decisions, the core innovation lies in upgrading static planning to a "prediction-simulation-learning" closed loop process: the optimization algorithm uses the knowledge graph for efficient optimization, producing a multi-objective balanced scheme; the digital twin performs high-fidelity, uncertain long-term dynamic deduction and risk assessment on the scheme; the simulation results are fed back to calibrate the knowledge rules through meta-reinforcement learning, and successful patterns are extracted from the knowledge base, this process realizes the collaborative self-evolution of planning knowledge, optimization model and simulation environment, thereby continuously improving the accuracy, robustness and foresight of the decision, and ultimately achieving multi-scale collaborative optimization of carbon sink, ecology, social services and economic cost.

Description

Technical Field

[0001] This invention relates to the field of urban landscaping, and more specifically, to a multi-scale urban landscaping carbon sequestration enhancement and ecological synergistic regulation system. Background Technology

[0002] In the current green and low-carbon-oriented urban renewal process, integrating green spaces to synergistically enhance carbon sequestration capacity, ecological service functions, and social welfare has become a core issue. Especially in the renovation projects of old urban areas, it is often necessary to use limited plots of land to build multi-functional community parks. Their planning and construction face multiple and interconnected goals: not only do they need to quickly form a considerable carbon sink to offset the carbon emissions generated by regional development and meet the carbon assessment requirements of higher authorities, but they also need to meet the urgent needs of residents in high-density residential areas for recreational spaces, undertake the ecological function of building nodes of local biodiversity corridors, and achieve low-cost sustainable maintenance throughout the entire life cycle under external constraints such as annual fiscal budgets and fluctuations in the seedling market. These multiple goals work together on the same limited land, funds, seedlings, and water resources, making it a challenge involving complex environmental, social, and economic trade-offs to accurately allocate various resource units to different functional areas in the initial planning stage and subsequent management.

[0003] However, existing landscape greening management and planning technologies are insufficient to effectively support the complex decision-making processes involving multi-objective collaborative optimization. Mainstream landscape information management systems primarily focus on asset ledgers, daily inspections, and transactional process management. Even when integrated with geographic information systems, they mainly achieve spatial information visualization and simple queries, lacking embedded intelligent decision support modules. Although academia has developed relatively independent carbon sequestration models or ecosystem service assessment tools, these models are often disconnected from actual resource planning, allocation, and scheduling processes. The fundamental technical problem lies in the fact that existing systems cannot quantitatively integrate and solve multi-objective optimization problems for heterogeneous indicators such as carbon sequestration increment, biodiversity index, per capita recreational area, and life-cycle cost within a unified framework, thus failing to provide support for "in special circumstances." Specific decisions such as "what plant combination to configure in a given space and how much maintenance resources to allocate to achieve the best overall benefits" require data- and model-based scheme simulation and comparison. At the same time, the cross-domain data required to support the decision-making, such as high-precision carbon sink prediction data, community population dynamic data, and historical data on seedling costs and adaptability, are scattered across different management departments, resulting in data barriers and format differences. This makes it difficult to integrate in real time and drive dynamic analysis. As a result, there are gaps in the process chain from planning and design to maintenance. The initial planning scheme is often determined based on experience and cannot be dynamically adjusted and resources reallocated after implementation based on real-time information such as plant growth monitoring and public feedback. Ultimately, this may lead to a series of risks such as failure to achieve carbon sink targets, low public satisfaction, insufficient ecological benefits, or excessive long-term maintenance costs. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system. It solves the problems mentioned in the background art through a decision factor library construction module, an optimization scheme generation module, a scheme dynamic deduction module, and a co-evolution module.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: specifically, it includes: a decision factor library construction module, an optimization scheme generation module, a scheme dynamic deduction module, and a co-evolution module connected in sequence, wherein;

[0006] Decision factor library construction module: In response to planning task triggers, it integrates multi-source heterogeneous data, including soil, meteorology, remote sensing, community population, historical cost data and real-time IoT monitoring data, to construct a knowledge graph with site conditions, plant species, ecosystem services and resource consumption as entities. It also dynamically expresses the quantitative relationships between entities through computable association rules, forming and outputting a structured decision factor library knowledge package containing static attributes and dynamic association logic.

[0007] The optimization scheme generation module dynamically assembles multi-objective optimization functions by calling the association rules in the decision factor library knowledge package, discretizes the planning space into grid cells and defines plant configuration and maintenance level as decision variables, and uses the improved NSGA-III algorithm based on knowledge graph pre-screening solution to solve the function with carbon sink, ecological diversity, social services and economic cost as objective vectors, and outputs the Pareto optimal scheme set and its corresponding resource allocation blueprint and multi-dimensional benefit estimate;

[0008] Dynamic simulation module: For each Pareto optimal solution in the Pareto optimal solution set, an independent digital twin is established. The digital twin is a hybrid simulation system that integrates plant growth model, carbon cycle model and human flow simulation model. Under the condition of accessing real-time IoT monitoring data and introducing random disturbance factors to simulate uncertain events, the hybrid simulation system is run to simulate the dynamic changes and risk status of carbon sink accumulation, ecological indicators, human flow distribution and maintenance costs of each Pareto optimal solution in the future preset period, and generates dynamic simulation results of the solution including the simulation results of each solution.

[0009] Co-evolution module: The dynamic deduction results of each Pareto optimal solution are compared with the corresponding multidimensional benefit predictions in the optimization solution generation module to generate a deviation matrix. Based on the deviation matrix, the confidence and weight of the corresponding association rules in the decision factor knowledge package are updated through the meta-reinforcement learning framework. The features of the solutions with excellent performance in the dynamic deduction results are abstracted into new association rules and fed back to the knowledge graph. This realizes the closed-loop co-evolution of the decision factor knowledge package, the multi-objective optimization function and the improved NSGA-III algorithm in the optimization solution generation module, and the digital twin and hybrid simulation system in the dynamic deduction module.

[0010] In a preferred embodiment, the specific operation of integrating multi-source heterogeneous data in the decision factor library construction module is as follows:

[0011] In response to the planning task, the system acquires and locks the geographical boundaries of the target planning area; collects and integrates the following categories of data: soil pH, organic matter content, and bulk density data; historical precipitation, temperature series, sunshine hours, and real-time precipitation and temperature monitoring data; remote sensing data on vegetation index, surface temperature, and land use classification; community population density distribution, age structure, and mobile signaling heat map data; cost data on seedling purchase price, transportation costs, and historical maintenance records; and real-time monitoring data from IoT sensors, including real-time precipitation and temperature data from meteorological sensors and real-time soil volumetric moisture content, leaf surface humidity, and stem flow rate data from soil and plant sensors; and performs spatiotemporal alignment and normalization on all collected multi-source heterogeneous data to form a standardized dataset.

[0012] In a preferred embodiment, the specific operation of constructing entities based on site conditions, plant species, ecosystem services, and resource consumption, and dynamically expressing the quantitative relationships between entities through computable association rules, is as follows:

[0013] Based on the standardized dataset, site condition entities, plant species entities, ecosystem service entities, and resource consumption entities are defined in the knowledge graph. Each site condition entity is assigned an attribute vector containing soil pH, organic matter content, bulk density, precipitation, temperature, sunshine duration, vegetation index, surface temperature, soil volumetric water content, leaf surface humidity, and trunk flow rate of the corresponding spatial grid. Each plant species entity is assigned an attribute vector containing the photosynthetic rate, transpiration coefficient, and shade tolerance of the corresponding species.

[0014] This process establishes applicable, provisional, and consumption relationships between entities, transforming domain knowledge and data patterns into computable production rules. A computable list of association rules is then created, containing all rules. Each computable production rule is an independent data structure comprising a condition part, an output part, and a confidence part. The condition part is the logic for determining the supply of a specific ecosystem service or the demand for a specific resource consumption. This quantitative estimate, called the ecosystem service potential estimate or resource consumption demand estimate, is associated with the specific algorithm used to calculate it. The confidence part records the rule's information. The confidence and weight parameters; the calculation process for the estimated value of ecosystem service potential or resource consumption demand is as follows: First, determine the degree of membership of temperature, precipitation, and soil volumetric water content to the optimal range of plant photosynthesis to obtain the membership degree of temperature, precipitation, and soil moisture. Then, determine a conversion factor based on the physiological characteristics of plants, which is determined by the specific leaf area and biomass allocation coefficient. Finally, multiply the product of the membership degree of temperature, precipitation, and soil moisture by the photosynthetic rate of the plant species and the conversion factor, and then multiply by the time period to complete the calculation. Based on this calculation process, a set of computable association rules containing multiple production rules is generated.

[0015] In a preferred embodiment, the specific process of forming a decision factor library containing static attributes and dynamic correlation logic is as follows:

[0016] The defined site condition entities, plant species entities, ecosystem service entities, and resource consumption entities, the attribute vectors assigned to these entities, the applicable, providing, and consuming relationships established between entities, and the generated set of computable association rules are all encapsulated into a structured decision factor library knowledge package.

[0017] This decision factor knowledge package provides an application programming interface (API). The API supports two types of calls: the first call is to query and return the corresponding estimates of various ecosystem service potentials and resource consumption demands, as well as the attribute vectors of the associated site condition entities and plant species entities, based on the input identification information of the site condition entities and plant species entities; the second call is to return a list of computable association rules. Each rule in this list exists in its data structure form, including condition judgment logic, the specific algorithm used to calculate the ecosystem service potential estimate or resource consumption demand estimate, confidence parameters, and weight parameters.

[0018] In a preferred embodiment, the specific process of dynamically assembling a multi-objective optimization function by calling association rules from the decision factor knowledge package in the optimization scheme generation module is as follows:

[0019] The system invokes the application programming interface (API) provided by the decision factor knowledge package. Through its first invocation, it obtains estimates of various ecosystem service potentials and resource consumption demands encapsulated within the knowledge package. Through its second invocation, it obtains a complete list of computable association rules. The planning space is discretized into grid cells of equal area to form a set of grid cells. For each grid cell in this set, its decision variable is defined as a triple containing plant species ID, planting density, and maintenance intensity level. Based on the decision variable and the obtained estimates of ecosystem service potentials and resource consumption demands, four objective functions are dynamically constructed: a carbon sink objective function, an ecodiversity objective function, a social service objective function, and an economic cost objective function.

[0020] The calculation process of the carbon sink objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output carbon sink and whose conditions are partially applicable to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a rule selection result set. Then, calculate the product of the carbon sink potential estimate of each rule in this rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this rule selection result set to obtain a weighted sum of carbon sink contribution for the grid cell.

[0021] Next, the area of ​​this grid cell, the planting density in its decision variables, and the calculated carbon sink contribution are multiplied together to obtain the contribution of this grid cell to the total carbon sink target.

[0022] Finally, the contributions of all grid cells in the grid cell set are summed, and the result is the carbon sink objective function value.

[0023] The calculation process of the ecological diversity objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output ecological diversity-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming an ecological diversity rule selection result set; then, calculate the product of the ecological diversity potential estimate corresponding to each rule in this ecological diversity rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the weighted sum of the ecological diversity contribution of the grid cell; the ecological diversity potential estimate is calculated from the corresponding association rule in the decision factor library; next, multiply the area of ​​this grid cell, the planting density in its decision variable, and the calculated weighted sum of ecological diversity contribution to obtain the contribution of the grid cell to the total ecological diversity objective; finally, sum the contributions of all grid cells, and the result is the ecological diversity objective function value;

[0024] The calculation process of the social service objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output social service-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a social service rule selection result set; then, calculate the product of the social service potential estimate corresponding to each rule in this social service rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the weighted sum of the social service contribution of the grid cell; the social service potential estimate is calculated from the corresponding association rule in the decision factor library; next, multiply the area of ​​the grid cell, the planting density in its decision variable, and the calculated weighted sum of social service contributions, and then multiply by a service demand weight coefficient determined based on community population distribution data to obtain the contribution of the grid cell to the overall social service objective; finally, sum the contributions of all grid cells, and the result is the value of the social service objective function;

[0025] The calculation process of the economic cost objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output resource consumption demand and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a resource consumption rule selection result set; then, calculate the product of the estimated resource consumption demand corresponding to each rule in this resource consumption rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the comprehensive resource consumption weighted demand of the grid cell; next, multiply the area of ​​the grid cell, the planting density and maintenance intensity level conversion coefficient in its decision variables, and the calculated comprehensive resource consumption weighted demand to obtain the resource consumption cost of the grid cell; finally, sum the resource consumption costs of all grid cells and combine them with the initial cost of seedling procurement and transportation, and the result is the value of the economic cost objective function;

[0026] The dynamically constructed carbon sink objective function, ecodiversity objective function, social service objective function, and economic cost objective function together constitute a multi-objective optimization function.

[0027] In a preferred embodiment, the specific process of solving the function with carbon sinks, biodiversity, social services, and economic costs as objective vectors using the improved NSGA-III algorithm based on knowledge graph pre-screening is as follows:

[0028] From the list of computable association rules, mandatory constraints on plant site tolerance are extracted. These mandatory constraints include the requirement that the soil pH and salinity for the plant species be within their tolerance range, and these constraints are used as hard constraints for the optimization problem. During each generation of population evolution in the improved NSGA-III algorithm, a knowledge graph pre-screening operation is performed on the newly generated candidate solutions.

[0029] For candidate solutions pre-screened through the knowledge graph, the four objective function values ​​of carbon sink, biodiversity, social services, and economic cost are calculated. The non-dominated sorting, reference point association, and niche preservation operations of the NSGA-III algorithm are then executed to guide population evolution. After the algorithm terminates, a set of Pareto optimal solutions is output. Each Pareto optimal solution in this set contains two parts: one part is the plant species number, planting density, and maintenance intensity level determined for each grid cell, which is called the resource allocation blueprint of the solution; the other part is the corresponding objective function values ​​of carbon sink, biodiversity, social services, and economic cost, which together are called the multidimensional benefit estimate of the solution.

[0030] In a preferred embodiment, the specific process of establishing an independent digital twin for each Pareto optimal solution in the Pareto optimal solution set in the dynamic derivation module is as follows:

[0031] For each Pareto optimal solution in the Pareto optimal solution set output by the optimization solution generation module, extract the resource allocation blueprint corresponding to that Pareto optimal solution.

[0032] Based on this resource allocation blueprint, a separate digital twin is instantiated for the Pareto optimal solution; this digital twin contains a hybrid simulation system consisting of three coupled models: a plant growth model, a carbon cycle model, and a pedestrian flow simulation model;

[0033] The initial state of the plant growth model is set according to the plant species number and planting density of each grid cell in the resource allocation blueprint. The plant growth model is a hybrid model that combines mechanism and data-driven approaches. Its mechanism part is based on the theory of photosynthesis, respiration and biomass allocation. Its data-driven part obtains the photosynthetic rate, transpiration coefficient and physiological and ecological parameters of the configured plant species, as well as the association rules that affect plant growth, by calling the application programming interface of the decision factor library knowledge package.

[0034] The initial parameters of the carbon cycle model are derived from the attribute vectors of the site condition entities in the corresponding grid in the decision factor base knowledge package. The carbon cycle model is connected to the output of the plant growth model to receive biomass change data generated by plant growth simulation.

[0035] The pedestrian flow simulation model constructs the basic environment based on the plant space configuration described in the resource allocation blueprint and combined with geographic information system data.

[0036] In a preferred embodiment, the specific process of generating the dynamic simulation results of the scheme, which include all simulation results, is as follows:

[0037] During the simulation process, real-time monitoring data from IoT sensors is continuously connected to each digital twin, and the real-time monitoring data is input into each model of the hybrid simulation system as an environmental driving variable in real time.

[0038] Simultaneously, a random disturbance factor is systematically introduced into the simulation timeline. This random disturbance factor is a random event sequence generator that simulates uncertain events such as extreme drought, floods, and pest outbreaks. It generates event sequences according to a preset spatiotemporal distribution pattern, introducing a disturbance impact calculation process into the plant growth model: First, the theoretical total primary productivity is calculated based on the model state; then, an environmental stress function value determined by real-time temperature, precipitation, and soil volumetric moisture content is calculated; next, for each random disturbance event triggered at each moment on the simulation timeline, the current intensity value of the event is obtained and multiplied by the vulnerability coefficient of the plant species to such events, obtained from the decision factor knowledge package by querying the attributes or association rules of the configured plant species, to obtain the proportion of potential productivity loss caused by the event at the current moment.

[0039] The proportion of potential productivity loss caused by all triggered random disturbance events is accumulated and applied to the theoretical total primary productivity together with the environmental stress function value, thereby calculating the net primary productivity after the disturbance.

[0040] The hybrid simulation system couples the plant growth model, carbon cycle model, and pedestrian flow simulation model in a time-step manner within a preset future time period. This process is called hybrid simulation operation. It dynamically simulates the changes in carbon sink accumulation, ecological indicators, social service effects represented by pedestrian flow distribution, and cumulative maintenance costs over time for each scheme.

[0041] Generating dynamic simulation results of the scheme, which include the results of each simulation, is accomplished through the following three sub-steps:

[0042] Sub-step E1, Data Acquisition and Aggregation: During the hybrid simulation, the following four types of time series data are continuously collected and recorded: the first type is the time series data of carbon sink accumulation; the second type is the time series data of one or more ecological indicators characterizing the ecological status; the third type is the time series data of cumulative maintenance costs; the fourth type is the real-time temperature and precipitation time series data that serve as the simulation driving source; at the same time, the distribution data of people flow expressed in the form of spatial heat maps for evaluating the effectiveness of social services are continuously collected and recorded, as well as information on all random disturbance events triggered during the simulation, forming a disturbance event log;

[0043] Sub-step E2, Risk Quantification: For the carbon sink objective function value, biodiversity objective function value, social service objective function value, and economic cost objective function value in the multidimensional benefit estimate corresponding to the current scheme, output from the optimization scheme generation module, the following quantification operation is performed: Through multiple independent simulation runs with introduced random disturbance factors, the probability distribution of the carbon sink accumulation, ecological indicators, social service quantitative indicators calculated from population distribution data, and cumulative maintenance costs corresponding to the above objective function values ​​collected in sub-step E1 at the end of the preset simulation period is calculated; based on this probability distribution, the risk probability value that the value at the end of the simulation period is lower than the preset percentage of the corresponding multidimensional benefit estimate is calculated.

[0044] Sub-step E3, Result Encapsulation: For each Pareto optimal solution, the four types of time series data, pedestrian flow distribution data, and disturbance event logs collected and recorded in sub-step E1, together with the various risk probability values ​​calculated in sub-step E2, are encapsulated into a structured dynamic simulation result of the solution.

[0045] In a preferred embodiment, the specific process of generating a deviation matrix by comparing the dynamic derivation results of each Pareto optimal solution with its corresponding multidimensional benefit prediction in the optimization solution generation module in the co-evolution module is as follows:

[0046] For each Pareto optimal solution output by the dynamic simulation module, the following are extracted: the cumulative carbon sink amount statistically obtained at the end of the preset simulation period; the final values ​​of the time series data of the second type of ecological indicators recorded in sub-step E1; the quantitative indicators of social services calculated from the population distribution data; and the actual results of the cumulative maintenance costs. Simultaneously, from the multi-dimensional benefit forecast value corresponding to the solution output by the optimization solution generation module, the carbon sink objective function value, the biodiversity objective function value, the social service objective function value, and the economic cost objective function value are extracted as corresponding expected values. The relative deviation between the actual and expected values ​​for each benefit dimension—carbon sink, ecology, social services, and economic costs—is calculated. For carbon sink, ecology, and social service indicators, the relative deviation is the actual value minus the expected value, then divided by the expected value. For economic costs, the relative deviation is the expected value minus the actual value, then divided by the expected value.

[0047] Subsequently, a deviation tracing operation is performed: for each Pareto optimal solution, the relative deviation calculated for each benefit dimension is determined based on the calculation process of the carbon sink objective function, ecodiversity objective function, social service objective function, and economic cost objective function in the optimization solution generation module. The contribution of each association rule in the decision factor base knowledge package that contributes to the expected value of this benefit dimension is then determined. Next, this relative deviation value is multiplied by the ratio of the contribution of this association rule to the expected value of this benefit dimension, yielding a deviation signal generated by this association rule in this benefit dimension due to the solution. This process is repeated for all Pareto optimal solutions and all their benefit dimensions. For each association rule, all deviation signals obtained across all benefit dimensions of all solutions are accumulated, thereby calculating a comprehensive deviation of this association rule in each benefit dimension.

[0048] All association rules and their combined deviations across various benefit dimensions are organized into a two-dimensional table, forming a deviation matrix.

[0049] In a preferred embodiment, the specific process of updating the confidence and weight of the corresponding association rules in the decision factor base knowledge package based on the bias matrix through a meta-reinforcement learning framework, and abstracting the features of schemes with excellent performance in the dynamic deduction results into new association rules to feed back to the knowledge graph includes:

[0050] First, based on the deviation matrix, a loss function for each association rule is constructed using the comprehensive deviation of each rule across the dimensions of carbon sink, biodiversity, social services, and economic costs. The value of the loss function is the sum of the products of the preset preference weights for each dimension and the squares of the corresponding comprehensive deviations.

[0051] Secondly, a meta-policy network is introduced. This network takes the feature vector of the historical performance data of each association rule as input and dynamically outputs an adaptive learning rate for updating the parameters of the rule. The adaptive learning rate includes a first learning rate for updating the confidence parameter and a second learning rate for updating the weight parameter.

[0052] Then, a parameter update operation is performed, which includes the following steps: calculating the partial derivative of the loss function with respect to the confidence parameter of the association rule, as the update gradient of the confidence parameter; calculating the partial derivative of the loss function with respect to the weight parameter of the association rule, as the update gradient of the weight parameter; subtracting the product of the first learning rate and the update gradient of the confidence parameter from the current confidence parameter of the association rule to obtain an intermediate value of the confidence parameter; subtracting the product of the second learning rate and the update gradient of the weight parameter from the current weight parameter of the association rule to obtain an intermediate value of the weight parameter; and restricting the intermediate values ​​of the confidence parameter and the intermediate values ​​of the weight parameter to a valid value range of zero to one, thereby obtaining the updated confidence parameter and weight parameter.

[0053] Simultaneously, the feature abstraction and rule feedback operation for high-performing solutions is performed, which includes four sub-steps executed sequentially:

[0054] Sub-step F1, Selection of excellent solutions: Based on the risk probability values ​​recorded in the dynamic simulation results of the solutions and the actual results obtained at the end of the preset simulation period, set the achievement rate thresholds for carbon sink, biodiversity, and social service indicators and the control rate thresholds for economic cost indicators, and select all Pareto optimal solutions that meet the threshold conditions to form a set of excellent solutions.

[0055] Sub-step F2, Feature Pattern Mining: Data mining is performed on the resource allocation blueprints of all schemes in the selected set of high-performing schemes to identify the plant species combinations that appear repeatedly, the range of soil and climate conditions corresponding to the combination, and the stable ecosystem service potential, social service effects and resource consumption demand characteristics of the combination in the simulation.

[0056] Sub-step F3, New Rule Generation and Formalization: The stable patterns identified in sub-step F2 are formalized into new computable association rules. The condition part of the rule is defined based on the identified range of soil and climate conditions and plant species combinations. The output part includes the estimated values ​​of the ecosystem service potential, social service potential, and resource consumption demand for the combination. The new rule is assigned an initial confidence parameter and an initial weight parameter.

[0057] Sub-step F4, Knowledge Base Feedback: Add all the new computable association rules generated in sub-step F3 to the computable association rule list of the decision factor base knowledge package, thereby completing the expansion of the knowledge graph.

[0058] The beneficial effects of this invention are as follows: by constructing a computable and evolvable knowledge graph, multi-source data and domain rules are deeply integrated to dynamically drive spatial optimization decisions. Its core innovation lies in upgrading static planning into a closed-loop process of "prediction-simulation-learning": the optimization algorithm uses the knowledge graph to efficiently find the best solution and generate a multi-objective equilibrium solution; the digital twin performs high-fidelity, uncertain long-term dynamic extrapolation and risk assessment of the solution; the simulation results are used to calibrate the knowledge rules through meta-reinforcement learning feedback, and successful patterns are extracted to feed back into the knowledge base. This process realizes the collaborative self-evolution of planning knowledge, optimization models and simulation environment, thereby continuously improving the accuracy, robustness and foresight of decisions, and ultimately achieving multi-scale synergistic optimization of carbon sinks, ecology, social services and economic costs. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method of the present invention;

[0060] Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0063] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application. Example 1

[0064] This embodiment provides, for example Figure 1-2 The system shown is a multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system, which specifically includes: a decision factor library construction module, an optimization scheme generation module, a scheme dynamic deduction module, and a synergistic evolution module connected in sequence.

[0065] Decision factor library construction module: In response to planning task triggers, it integrates multi-source heterogeneous data, including soil, meteorology, remote sensing, community population, historical cost data and real-time IoT monitoring data, to construct a knowledge graph with site conditions, plant species, ecosystem services and resource consumption as entities. It also dynamically expresses the quantitative relationships between entities through computable association rules, forming and outputting a structured decision factor library knowledge package containing static attributes and dynamic association logic.

[0066] The optimization scheme generation module dynamically assembles multi-objective optimization functions by calling the association rules in the decision factor library knowledge package, discretizes the planning space into grid cells and defines plant configuration and maintenance level as decision variables, and uses the improved NSGA-III algorithm based on knowledge graph pre-screening solution to solve the function with carbon sink, ecological diversity, social services and economic cost as objective vectors, and outputs the Pareto optimal scheme set and its corresponding resource allocation blueprint and multi-dimensional benefit estimate;

[0067] Dynamic simulation module: For each Pareto optimal solution in the Pareto optimal solution set, an independent digital twin is established. The digital twin is a hybrid simulation system that integrates plant growth model, carbon cycle model and human flow simulation model. Under the condition of accessing real-time IoT monitoring data and introducing random disturbance factors to simulate uncertain events, the hybrid simulation system is run to simulate the dynamic changes and risk status of carbon sink accumulation, ecological indicators, human flow distribution and maintenance costs of each Pareto optimal solution in the future preset period, and generates dynamic simulation results of the solution including the simulation results of each solution.

[0068] The co-evolution module compares the dynamic derivation results of each Pareto optimal solution with the corresponding multidimensional benefit predictions in the optimization solution generation module to generate a deviation matrix. Based on the deviation matrix, the confidence and weight of the corresponding association rules in the decision factor knowledge package are updated through a meta-reinforcement learning framework. The features of the solutions that perform well in the dynamic derivation results are abstracted into new association rules and fed back to the knowledge graph. This realizes the closed-loop co-evolution of the decision factor knowledge package, the multi-objective optimization function and the improved NSGA-III algorithm in the optimization solution generation module, and the digital twin and hybrid simulation system in the dynamic derivation module.

[0069] In this embodiment, it is necessary to specifically explain the operation of integrating multi-source heterogeneous data in the decision factor library construction module as follows:

[0070] In response to planning task triggers, the system acquires and locks the geographical boundaries of the target planning area; collects and integrates the following categories of data: soil pH, organic matter content, and bulk density data; historical precipitation, temperature series, sunshine hours, and real-time precipitation and temperature monitoring data; remote sensing data on vegetation index, surface temperature, and land use classification; community population density distribution, age structure, and mobile signaling heat map data; cost data on seedling purchase price, transportation costs, and historical maintenance records; and real-time monitoring data from IoT sensors, including real-time precipitation and temperature data from meteorological sensors and real-time soil volumetric moisture content, leaf surface humidity, and stem flow rate data from soil and plant sensors; and performs spatiotemporal alignment on all collected multi-source heterogeneous data. With normalization processing, a standardized dataset with unified spatiotemporal benchmarks and standardized units of measurement is formed; spatiotemporal alignment specifically refers to uniformly registering data from different sources with different spatiotemporal resolutions to a spatiotemporal grid framework with the geographical boundary of the target planning area as the range, a preset grid size (e.g., 10m×10m) as the basic spatial unit, and a preset time interval (e.g., 1 hour) as the basic time unit; normalization processing uses the maximum-minimum value normalization or Z-score standardization method to map the numerical range of each data sequence to the [0,1] interval or transform it into a standard normal distribution with a mean of 0 and a standard deviation of 1, so as to eliminate the influence of dimensions. For example, for soil pH, its effective value range can be set to pH 4.0 to 9.0, and linear mapping can be performed within this range;

[0071] The specific operations for constructing entities based on site conditions, plant species, ecosystem services, and resource consumption, and dynamically expressing the quantitative relationships between entities through computable association rules, are as follows:

[0072] Based on the standardized dataset, site condition entities, plant species entities, ecosystem service entities, and resource consumption entities are defined in the knowledge graph. Each site condition entity is assigned an attribute vector containing soil pH, organic matter content, bulk density, precipitation, temperature, sunshine duration, vegetation index, surface temperature, soil volumetric water content, leaf surface humidity, and trunk flow rate of the corresponding spatial grid. Each plant species entity is assigned an attribute vector containing the photosynthetic rate, transpiration coefficient, and shade tolerance of the corresponding species.

[0073] This approach establishes "applicable," "providing," and "consuming" relationships between entities, transforming domain knowledge and data patterns into computable production rules. This results in a computable list of association rules, each a separate data structure comprising a condition part, an output part, and a confidence part. The condition part is the logic for judging based on attribute matching between the site condition entity and the plant species entity. This condition judgment logic is specified as a series of threshold comparisons. For example, the conditions for judging an "applicable" relationship include: the soil pH value in the site condition entity's attribute vector is within the suitable pH range threshold defined in the plant species entity's attribute vector, and the annual average temperature value is within its suitable temperature range threshold. The output part is a quantitative estimate of a specific ecosystem service supply or resource consumption demand, referred to as the ecosystem service potential estimate or resource consumption demand estimate, and is associated with the specific algorithm used to calculate this value. The confidence part records the confidence level of the rule. The calculation process for the estimated value of ecosystem service potential or resource consumption demand is as follows: First, determine the degree of membership of temperature, precipitation, and soil volumetric water content to the optimal range of plant photosynthesis to obtain the membership degree of temperature, precipitation, and soil moisture. The "degree of membership" is calculated using a fuzzy membership function. For example, for temperature, a trapezoidal or triangular membership function can be used. The optimal temperature range is set as [T_opt_low, T_opt_high]. When the temperature is below T_min or above T_max, the membership degree is 0, and it linearly changes to 1 within this range. Then, a conversion factor determined by specific leaf area and biomass allocation coefficient is determined according to the physiological characteristics of the plant. Finally, the product of the membership degree of temperature, precipitation, and soil moisture is multiplied by the photosynthetic rate of the plant species and the conversion factor, and then multiplied by the time period to complete the calculation. Based on this calculation process, a set of computable association rules containing multiple production rules is generated.

[0074] The specific process for forming a decision factor library that includes static attributes and dynamic correlation logic is as follows:

[0075] The defined site condition entities, plant species entities, ecosystem service entities, and resource consumption entities, the attribute vectors assigned to these entities, the applicable, providing, and consuming relationships established between entities, and the generated set of computable association rules are all encapsulated into a structured decision factor library knowledge package.

[0076] This decision factor knowledge package provides an application programming interface (API). The API supports two types of calls: the first call queries and returns the corresponding estimates of ecosystem service potential and resource consumption demand, along with the attribute vectors of the associated site condition entities and plant species entities, based on the input identifiers of the site condition entities and plant species entities. The identifiers can be unique ID numbers or name strings for the entities. The second call returns a list of computable association rules. Each rule in this list exists in its data structure form, including conditional judgment logic, the specific algorithm used to calculate the ecosystem service potential or resource consumption demand estimates, confidence parameters, and weight parameters. The API can be implemented using RESTful API or gRPC interfaces, and query requests and response data are encapsulated in JSON or Protocol Buffers format.

[0077] The encapsulated decision factor knowledge package outputs the ecosystem service potential estimate, resource consumption demand estimate, and entity attribute vector obtained through the first call operation, as well as the computable association rule list obtained through the second call operation. These are then called by the optimization scheme generation module, the scheme dynamic deduction module, and the co-evolution module for subsequent calculations. Specifically, the optimization scheme generation module uses the ecosystem service potential estimate and resource consumption demand estimate obtained through the first call operation to construct the objective function and constraints, and calls the second call operation to obtain the computable association rule list. The scheme dynamic deduction module uses the entity attribute vector obtained through the first call operation and the association rules obtained through the second call operation as initial parameters and local rules for the simulation model. The co-evolution module calls the second... The system retrieves a list of computable association rules and updates the confidence and weight parameters of the rules in the list, which are stored in the decision factor knowledge package. The update operation is implemented using a meta-reinforcement learning framework. For example, the deviation between the predicted value of an association rule and the simulation result in a single planning-simulation cycle is used as a loss signal, and gradient descent is employed to update the weight parameters of the associated rules. The confidence parameters of the associated rules can be Bayesian-updated based on their historical accuracy verified in multiple cycles. The effective range for the weight parameters of each association rule is set to [0.0, 1.0], and the effective range for the confidence parameters is [0.5, 1.0]. When the confidence parameter of an association rule falls below a preset failure threshold (e.g., 0.6), the system issues a prompt that the rule needs to be reviewed.

[0078] In this embodiment, it is necessary to specifically explain the process by which the multi-objective optimization function is dynamically assembled by calling the association rules in the decision factor knowledge package in the optimization scheme generation module:

[0079] The system invokes the application programming interface (API) provided by the decision factor knowledge package. Through its first invocation, it obtains estimates of various ecosystem service potentials and resource consumption demands encapsulated within the knowledge package. Through its second invocation, it obtains a complete list of computable association rules. The planning space is discretized into grid cells of equal area to form a set of grid cells. For each grid cell in this set, its decision variable is defined as a triple containing plant species ID, planting density, and maintenance intensity level. Based on the decision variable and the obtained estimates of ecosystem service potentials and resource consumption demands, four objective functions are dynamically constructed: a carbon sink objective function, an ecodiversity objective function, a social service objective function, and an economic cost objective function.

[0080] The calculation process of the carbon sink objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output carbon sink and whose conditions are partially applicable to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a rule selection result set. Then, calculate the product of the carbon sink potential estimate of each rule in this rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this rule selection result set to obtain a weighted sum of carbon sink contribution for the grid cell.

[0081] Next, the area of ​​this grid cell, the planting density in its decision variables, and the calculated carbon sink contribution are multiplied together to obtain the contribution of this grid cell to the total carbon sink target.

[0082] Finally, the contributions of all grid cells in the grid cell set are summed, and the result is the carbon sink objective function value.

[0083] The calculation process of the ecological diversity objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output ecological diversity-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable for that grid cell, forming an ecological diversity rule selection result set; then, calculate the product of the ecological diversity potential estimate corresponding to each rule in this ecological diversity rule selection result set with the rule's confidence parameter and weight parameter, and sum the product results of all rules in this result set to obtain the ecological diversity of that grid cell. The ecological diversity potential is estimated by calculating the weighted sum of ecological diversity contributions. Ecological diversity-related indicators may include, but are not limited to, habitat suitability index and plant species diversity index, whose corresponding ecological diversity potential estimates are calculated by the corresponding association rules in the decision factor library. Next, the area of ​​this grid cell, the planting density in its decision variables, and the calculated weighted sum of ecological diversity contributions are multiplied to obtain the contribution of this grid cell to the overall ecological diversity objective. Finally, the contributions of all grid cells are summed, and the result is the ecological diversity objective function value.

[0084] The calculation process of the social service objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output social service-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a social service rule selection result set; then, calculate the product of the social service potential estimate corresponding to each rule in this social service rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the weighted sum of the social service contribution of the grid cell; the social service potential estimate is calculated from the corresponding association rules in the decision factor library; social service-related indicators... The objectives may include recreational suitability, aesthetic value of the landscape, and shade and cooling effect. The service demand weighting coefficient, determined based on community population distribution data, can be obtained by normalizing the population density within a preset radius (e.g., 500 meters) around the grid unit. The higher the population density, the larger the coefficient, indicating a stronger demand for the social service function of that grid. Next, the area of ​​this grid unit, the planting density in its decision variables, and the calculated social service contribution are multiplied together, and then multiplied by a service demand weighting coefficient determined based on community population distribution data to obtain the contribution of this grid unit to the overall social service objective. Finally, the contributions of all grid units are summed, and the result is the social service objective function value.

[0085] The calculation process of the economic cost objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output resource consumption demand and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a resource consumption rule selection result set; then, calculate the product of the estimated resource consumption demand corresponding to each rule in this resource consumption rule selection result set, the confidence parameter and the weight parameter of the rule, and sum the product results of all rules in this result set to obtain the comprehensive resource consumption weighted demand of the grid cell; the resource consumption demand includes irrigation water demand, fertilizer demand, and estimated labor maintenance hours, etc., and the maintenance intensity... The coefficient for the degree level conversion maps the maintenance intensity level (e.g., low, medium, high) to a preset numerical multiplier (e.g., corresponding to 1.0, 1.5, 2.0 respectively) to adjust the resource consumption cost. Next, the area of ​​this grid cell, the planting density in its decision variables, the coefficient for the maintenance intensity level conversion, and the calculated comprehensive weighted demand for resource consumption are multiplied to obtain the resource consumption cost of this grid cell. Finally, the resource consumption costs of all grid cells are summed and combined with the initial cost of seedling procurement and transportation. The result is the economic cost objective function value. The initial cost is calculated based on the plant species number and planting density in the decision variables by querying the historical cost data encapsulated in the decision factor knowledge package.

[0086] The dynamically constructed carbon sink objective function, ecodiversity objective function, social service objective function, and economic cost objective function together constitute a multi-objective optimization function;

[0087] The specific process of solving a function with carbon sinks, biodiversity, social services, and economic costs as objective vectors using the improved NSGA-III algorithm based on knowledge graph pre-screening is as follows:

[0088] From the list of computable association rules, mandatory constraints regarding plant site tolerance are extracted. These mandatory constraints include ensuring that the soil pH and salinity required by the plant species are within their tolerance range, and these are used as hard constraints in the optimization problem. During each generation of population evolution in the improved NSGA-III algorithm, a knowledge graph pre-screening operation is performed on newly generated candidate solutions. This knowledge graph pre-screening operation includes two verification steps:

[0089] Step Q1: Verify whether the plant species number defined on each grid cell in the candidate solution satisfies the mandatory constraint conditions corresponding to the site conditions of that grid extracted from the computable association rules;

[0090] Step Q2: Based on the network of supply and consumption relationships between entities in the decision factor base knowledge package, quickly estimate the overall profile of the candidate solution in terms of ecological benefits and resource consumption, and compare it with the Pareto optimal solution set currently maintained by the algorithm. The quick estimation is achieved by directly obtaining the typical ecological service potential and resource consumption demand empirical values ​​of the main plant configuration of the candidate solution from the decision factor base. Specifically, if all index values ​​in the estimated ecological benefit profile of the candidate solution are lower than 90% of the corresponding index value of a certain solution in the current Pareto optimal solution set, and all index values ​​in its estimated resource consumption profile are higher than 110% of the corresponding index value of that solution, then it is determined to be "significantly low in ecological benefits or significantly high in resource consumption", and candidate solutions that are significantly low in ecological benefits or significantly high in resource consumption are eliminated early.

[0091] For candidate solutions pre-screened through the knowledge graph, the four objective function values ​​of carbon sink, biodiversity, social services, and economic cost are calculated. The non-dominated sorting, reference point association, and niche preservation operations of the NSGA-III algorithm are then executed to guide population evolution. After the algorithm terminates, a set of Pareto optimal solutions is output. The algorithm terminates when a preset maximum number of generations (e.g., 200 generations) is reached, or when the set of Pareto optimal solutions remains unchanged for multiple consecutive generations (e.g., 20 generations). Each Pareto optimal solution in this set contains two parts: one part is the plant species number, planting density, and maintenance intensity level determined for each grid cell, referred to as the resource allocation blueprint of the solution; the other part is the corresponding objective function values ​​for carbon sink, biodiversity, social services, and economic costs, collectively referred to as the multidimensional benefit estimate of the solution. The multidimensional benefit estimate is used as a baseline in the dynamic simulation module for comparison with the simulation results.

[0092] In this embodiment, it is necessary to specifically explain the process of establishing an independent digital twin for each Pareto optimal solution in the Pareto optimal solution set in the dynamic deduction module:

[0093] For each Pareto optimal solution in the Pareto optimal solution set output by the optimization solution generation module, the resource allocation blueprint corresponding to the Pareto optimal solution is extracted. The resource allocation blueprint defines three decision variables for each spatial grid cell in the Pareto optimal solution: the plant species number configured in the grid, the planting density of the plant species, and the preset maintenance intensity level for the grid.

[0094] Based on this resource allocation blueprint, a separate digital twin is instantiated for the Pareto optimal solution; this digital twin contains a hybrid simulation system consisting of three coupled models: a plant growth model, a carbon cycle model, and a pedestrian flow simulation model;

[0095] The initial state of the plant growth model is set according to the plant species number and planting density of each grid cell in the resource allocation blueprint. The plant growth model is a hybrid model combining mechanism and data-driven approaches. Its mechanism part is based on the theories of photosynthesis, respiration, and biomass allocation. For example, the Farquhar photosynthesis biochemical model or its improved model can be used as the core mechanism framework. Its data-driven part obtains the photosynthetic rate, transpiration coefficient, and other physiological and ecological parameters of the configured plant species, as well as the association rules affecting plant growth, such as the water stress response function, by calling the application programming interface of the decision factor library knowledge package. The obtained parameters and rules are used as the internal parameters and local logic of the model. The data-driven part is specifically implemented through a lightweight machine learning correction module (such as a gradient boosting tree or a shallow neural network). This module receives the initial output of the mechanism model and real-time environmental data, and fine-tunes the final growth prediction.

[0096] The initial parameters of the carbon cycle model are derived from the attribute vectors of the site condition entities in the corresponding grid in the decision factor base knowledge package. The carbon cycle model is connected to the output of the plant growth model to receive biomass change data generated by plant growth simulation.

[0097] The pedestrian flow simulation model constructs a basic environment based on the plant spatial configuration described in the resource allocation blueprint and combined with geographic information system data. The behavior rules of its agents are affected by the landscape features and microclimate information generated by the dynamic simulation of the plant growth model. The behavior rules of the agents can be based on the discrete choice model, and its utility function includes factors such as spatial attractiveness, environmental comfort (such as temperature and humidity) and distance cost determined by the plant configuration.

[0098] The specific process for generating dynamic simulation results of the scheme that include all simulation results is as follows:

[0099] During the simulation process, real-time monitoring data from IoT sensors is continuously connected to each digital twin, including real-time precipitation and temperature data provided by meteorological sensors, and real-time soil volumetric moisture content, leaf surface humidity, and trunk stem flow rate data provided by soil and plant sensors. The real-time monitoring data is then input into each model of the hybrid simulation system as environmental driving variables.

[0100] Simultaneously, a random perturbation factor is systematically introduced into the simulation timeline. This random perturbation factor is a random event sequence generator that simulates uncertain events such as extreme drought, flooding, and pest outbreaks. It generates event sequences according to a preset spatiotemporal distribution pattern. For example, for pest outbreaks, this preset spatiotemporal distribution pattern can be described by a composite random process that follows a specific kernel function diffusion in space and a Poisson process arrival in time. To quantify the impact of these random perturbation events on plant growth, a perturbation impact calculation process is introduced into the plant growth model: First, the theoretical total primary productivity is calculated based on the model state; then, a perturbation impact calculation is performed based on real-time temperature and precipitation. The environmental stress function value is jointly determined by soil volumetric moisture content; the environmental stress function can be constructed as the product of various factor stress sub-functions, where the temperature stress sub-function is described by a symmetrical or asymmetrical curve outside the optimal temperature range, and the water stress sub-function is described by an S-shaped curve based on soil moisture content; then, for each random disturbance event triggered at each moment on the simulation time axis, the current intensity value of the event is obtained and multiplied by the vulnerability coefficient of the plant species to such events obtained from the decision factor knowledge package by querying the attributes or association rules of the configured plant species, to obtain the proportion of potential productivity loss caused by the event at the current moment;

[0101] The proportion of potential productivity loss caused by all triggered random disturbance events is accumulated and applied to the theoretical total primary productivity together with the environmental stress function value, thereby calculating the net primary productivity after the disturbance.

[0102] The hybrid simulation system couples plant growth models, carbon cycle models, and pedestrian flow simulation models in a time-step manner within a preset future time period. This process is called hybrid simulation operation. It dynamically simulates the changes in carbon sink accumulation, ecological indicators, social service effects represented by pedestrian flow distribution, and cumulative maintenance costs over time for each scheme. The preset future time period can be set according to planning needs, such as 10 years or 20 years, and the simulation time step can be 1 day or 1 week.

[0103] Generating dynamic simulation results of the scheme, which include the results of each simulation, is accomplished through the following three sub-steps:

[0104] Sub-step E1, Data Acquisition and Aggregation: During the hybrid simulation, the following four types of time series data are continuously collected and recorded: The first type is the time series data of carbon sink accumulation; the second type is the time series data of one or more ecological indicators characterizing the ecological status; ecological indicators may include, but are not limited to, the Simpson diversity index, the Shannon-Wiener index, etc.; the third type is the time series data of cumulative maintenance costs; the fourth type is the real-time temperature and precipitation time series data used as the simulation driving source; at the same time, the distribution data of people flow expressed in the form of spatial heat maps for evaluating the effectiveness of social services are continuously collected and recorded, as well as information on all random disturbance events triggered during the simulation, forming a disturbance event log;

[0105] Sub-step E2, Risk Status Quantification: For the carbon sink objective function value, biodiversity objective function value, social service objective function value, and economic cost objective function value in the multidimensional benefit estimate corresponding to the current scheme, output from the optimization scheme generation module, the following quantification operation is performed: Through multiple independent simulation runs with introduced random disturbance factors, the number of independent simulation runs can be set according to the accuracy requirements, for example, 1000 times. The probability distribution of the carbon sink accumulation, ecological indicators, social service quantitative indicators calculated from the population distribution data, and cumulative maintenance costs corresponding to the above objective function values ​​collected in sub-step E1 at the end of the preset simulation period is calculated. Based on this probability distribution, the risk probability value of the value at the end of the simulation period being lower than the preset percentage of the corresponding multidimensional benefit estimate is calculated. The preset percentage can be set according to the risk tolerance, for example, 90%, that is, if the carbon sink accumulation at the end of the simulation period is lower than 90% of its initial estimate, it is considered that there is a risk of insufficient carbon sink.

[0106] Sub-step E3, Result Encapsulation: For each Pareto optimal solution, the four types of time series data, pedestrian flow distribution data, and disturbance event logs collected and recorded in sub-step E1, together with the various risk probability values ​​calculated in sub-step E2, are encapsulated into a structured dynamic deduction result of the solution. The encapsulation can adopt a standardized data format, such as JSON or XML, to facilitate the parsing and processing of subsequent modules.

[0107] In this embodiment, it is specifically necessary to explain the process in the co-evolution module of comparing the dynamic derivation results of each Pareto optimal solution with their corresponding multidimensional benefit predictions in the optimization solution generation module to generate a deviation matrix as follows:

[0108] For each Pareto optimal solution output by the dynamic simulation module, the following are extracted: the cumulative carbon sink amount statistically obtained at the end of the preset simulation period; the final values ​​of the time series data of the second type of ecological indicators recorded in sub-step E1; the quantitative indicators of social services calculated from the population distribution data; and the actual results of the cumulative maintenance costs. Simultaneously, from the multi-dimensional benefit forecast value corresponding to the solution output by the optimization solution generation module, the carbon sink objective function value, the biodiversity objective function value, the social service objective function value, and the economic cost objective function value are extracted as corresponding expected values. The relative deviation between the actual and expected values ​​for each benefit dimension—carbon sink, ecology, social services, and economic costs—is calculated. For carbon sink, ecology, and social service indicators, the relative deviation is the actual value minus the expected value, then divided by the expected value. For economic costs, the relative deviation is the expected value minus the actual value, then divided by the expected value.

[0109] Subsequently, a deviation tracing operation is performed: for each Pareto optimal solution, the relative deviation calculated for each benefit dimension is analyzed. Based on the calculation process of the carbon sink objective function, biodiversity objective function, social service objective function, and economic cost objective function in the optimization solution generation module, the contribution of each association rule in the decision factor base knowledge package that contributes to the expected value of this benefit dimension of the solution is determined. The contribution is specifically determined by tracing back the weight ratio of the association rule in the calculation of the corresponding objective function value of the solution. For example, for the carbon sink objective function, the contribution of a rule to a certain grid cell is its carbon sink potential estimate. The product of the evaluation value, confidence parameter, and weighting parameter represents the contribution of this rule to the overall carbon sink objective function value of the scheme, i.e., the proportion of the sum of its contributions across all grid cells to the total carbon sink objective function value of the scheme. Then, this relative deviation value is multiplied by the ratio of the contribution of the association rule to the expected value of this benefit dimension to obtain a deviation signal of the association rule on this benefit dimension due to the scheme. By traversing all Pareto optimal schemes and all their benefit dimensions, for each association rule, all deviation signals obtained by it on all benefit dimensions of all schemes are accumulated to calculate a comprehensive deviation of the association rule on each benefit dimension.

[0110] All association rules and their comprehensive deviations across each benefit dimension are organized into a two-dimensional table, forming a deviation matrix. The columns of the deviation matrix represent each association rule, and the columns represent each benefit dimension (carbon sink, biodiversity, social services, and economic costs). The cells contain the corresponding comprehensive deviations. This matrix serves as the input data for the meta-reinforcement learning framework.

[0111] The specific process of updating the confidence and weight of corresponding association rules in the decision factor base knowledge package based on the bias matrix using a meta-reinforcement learning framework, and abstracting the features of schemes with excellent performance in the dynamic deduction results into new association rules to feed back to the knowledge graph includes:

[0112] First, based on the deviation matrix, a loss function for each association rule is constructed using the comprehensive deviation of each rule across the dimensions of carbon sink, biodiversity, social services, and economic costs. The value of the loss function is the sum of the products of the preset preference weights for each dimension and the squares of the corresponding comprehensive deviations. The preset preference weights can be used to reflect the importance of different benefit dimensions in system optimization. For example, in regions with high carbon sink assessment pressure, the carbon sink dimension can be given a higher weight.

[0113] Secondly, a meta-policy network is introduced. This network takes the historical performance data feature vector of each association rule as input and dynamically outputs an adaptive learning rate for updating the parameters of that rule. The historical performance data feature vector is formed by summarizing the comprehensive deviation records of the association rule in previous planning-simulation cycles before the current update operation and calculating its mean, variance, and trend statistics. The adaptive learning rate includes a first learning rate for updating the confidence parameter and a second learning rate for updating the weight parameter. The meta-policy network can be a multilayer perceptron model, whose parameters are obtained by training on a large number of simulated rule update tasks. The goal is to make the overall loss of the association rule base decrease faster and more stably after applying its output learning rate.

[0114] Then, a parameter update operation is performed, which includes the following steps: calculating the partial derivative of the loss function with respect to the confidence parameter of the association rule, as the update gradient of the confidence parameter; calculating the partial derivative of the loss function with respect to the weight parameter of the association rule, as the update gradient of the weight parameter; subtracting the product of the first learning rate and the update gradient of the confidence parameter from the current confidence parameter of the association rule to obtain an intermediate value of the confidence parameter; subtracting the product of the second learning rate and the update gradient of the weight parameter from the current weight parameter of the association rule to obtain an intermediate value of the weight parameter; and restricting the intermediate values ​​of the confidence parameter and the intermediate values ​​of the weight parameter to a valid value range of zero to one, thereby obtaining the updated confidence parameter and weight parameter.

[0115] Simultaneously, the feature abstraction and rule feedback operation for high-performing solutions is performed, which includes four sub-steps executed sequentially:

[0116] Sub-step F1: Selection of Excellent Solutions: Based on the risk probability values ​​recorded in the dynamic simulation results and the actual results obtained at the end of the preset simulation period, set achievement rate thresholds for carbon sink, biodiversity, and social service indicators, as well as a control rate threshold for economic cost indicators. Select all Pareto optimal solutions that meet the threshold conditions to form a set of high-performing solutions. The thresholds can be set according to management requirements. For example, the carbon sink achievement rate threshold, biodiversity achievement rate threshold, and social service achievement rate threshold can all be set to 90%, and the economic cost control rate threshold can be set to 85%, meaning that the actual cost does not exceed 115% of the estimated cost.

[0117] Sub-step F2, Feature Pattern Mining: Data mining is performed on the resource allocation blueprints of all schemes in the selected set of high-performing schemes to identify the plant species combinations that appear repeatedly, the range of soil and climate conditions corresponding to the combination, and the stable ecosystem service potential, social service effects and resource consumption demand characteristics of the combination in the simulation. Data mining can be performed using frequent pattern mining algorithms (such as Apriori or FP-Growth).

[0118] Sub-step F3, New Rule Generation and Formalization: The stable patterns identified in sub-step F2 are formalized into new computable association rules. The condition part of the rule is defined based on the identified soil and climate condition ranges and plant species combinations. The output part includes the estimated values ​​of the ecosystem service potential, social service potential, and resource consumption demand for the combination. Initial confidence and weight parameters are assigned to the new rule. The output values, such as the ecosystem service potential estimate, can be obtained by statistically analyzing the mean and variance of the simulation results when the pattern appears in the set of superior solutions. The initial confidence and weight parameters can be set according to the frequency and support of the pattern in the superior solutions. For example, the higher the frequency of occurrence, the higher the initial confidence can be set, such as 0.7; the initial weight can be set to the default value, such as 0.5.

[0119] Sub-step F4, Knowledge Base Feedback: Add all the new computable association rules generated in sub-step F3 to the computable association rule list of the decision factor base knowledge package, thereby completing the expansion of the knowledge graph.

[0120] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0121] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0122] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0123] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0124] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0125] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0126] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system, characterized in that, Specifically, it includes: The decision factor library construction module, optimization scheme generation module, scheme dynamic deduction module, and co-evolution module are connected sequentially, among which; Decision factor library construction module: In response to planning task triggers, it integrates multi-source heterogeneous data, including soil, meteorology, remote sensing, community population, historical cost data and real-time IoT monitoring data, to construct a knowledge graph with site conditions, plant species, ecosystem services and resource consumption as entities. It also dynamically expresses the quantitative relationships between entities through computable association rules, forming and outputting a structured decision factor library knowledge package containing static attributes and dynamic association logic. The optimization scheme generation module dynamically assembles multi-objective optimization functions by calling the association rules in the decision factor library knowledge package, discretizes the planning space into grid cells and defines plant configuration and maintenance level as decision variables, and uses the improved NSGA-III algorithm based on knowledge graph pre-screening solution to solve the function with carbon sink, ecological diversity, social services and economic cost as objective vectors, and outputs the Pareto optimal scheme set and its corresponding resource allocation blueprint and multi-dimensional benefit estimate; Dynamic simulation module: For each Pareto optimal solution in the Pareto optimal solution set, an independent digital twin is established. The digital twin is a hybrid simulation system that integrates plant growth model, carbon cycle model and human flow simulation model. Under the condition of accessing real-time IoT monitoring data and introducing random disturbance factors to simulate uncertain events, the hybrid simulation system is run to simulate the dynamic changes and risk status of carbon sink accumulation, ecological indicators, human flow distribution and maintenance costs of each Pareto optimal solution in the future preset period, and generates dynamic simulation results of the solution including the simulation results of each solution. Generating dynamic simulation results of the scheme, which include the results of each simulation, is accomplished through the following three sub-steps: Sub-step E1, Data Acquisition and Aggregation: During the hybrid simulation, the following four types of time series data are continuously collected and recorded: the first type is the time series data of carbon sink accumulation; the second type is the time series data of one or more ecological indicators characterizing the ecological status; the third type is the time series data of cumulative maintenance costs; the fourth type is the real-time temperature and precipitation time series data that serve as the simulation driving source; at the same time, the distribution data of people flow expressed in the form of spatial heat maps for evaluating the effectiveness of social services are continuously collected and recorded, as well as information on all random disturbance events triggered during the simulation, forming a disturbance event log; Sub-step E2, Risk Quantification: For the carbon sink objective function value, biodiversity objective function value, social service objective function value, and economic cost objective function value in the multidimensional benefit estimate corresponding to the current scheme, output from the optimization scheme generation module, the following quantification operation is performed: Through multiple independent simulation runs with introduced random disturbance factors, the probability distribution of the carbon sink accumulation, ecological indicators, social service quantitative indicators calculated from population distribution data, and cumulative maintenance costs corresponding to the above objective function values ​​collected in sub-step E1 at the end of the preset simulation period is calculated; based on this probability distribution, the risk probability value of the value at the end of the preset simulation period being lower than the preset percentage of the corresponding multidimensional benefit estimate is calculated. Sub-step E3, Result Encapsulation: For each Pareto optimal solution, the four types of time series data, pedestrian flow distribution data, and disturbance event logs collected and recorded in sub-step E1, together with the various risk probability values ​​calculated in sub-step E2, are encapsulated into a structured dynamic deduction result of the solution. Co-evolution module: The dynamic deduction results of each Pareto optimal solution are compared with the corresponding multidimensional benefit predictions in the optimization solution generation module to generate a deviation matrix. Based on the deviation matrix, the confidence and weight of the corresponding association rules in the decision factor knowledge package are updated through the meta-reinforcement learning framework. The features of the solutions with excellent performance in the dynamic deduction results are abstracted into new association rules and fed back to the knowledge graph. This realizes the closed-loop co-evolution of the decision factor knowledge package, the multi-objective optimization function and the improved NSGA-III algorithm in the optimization solution generation module, and the digital twin and hybrid simulation system in the dynamic deduction module. The specific process of generating a deviation matrix by comparing the dynamic simulation results of each Pareto optimal solution with its corresponding multidimensional benefit prediction in the optimization solution generation module is as follows: For each Pareto optimal solution output by the dynamic simulation module, the following are extracted: the cumulative carbon sink amount statistically obtained at the end of the preset simulation period; the final values ​​of the time series data of the second type of ecological indicators recorded in sub-step E1; the quantitative indicators of social services calculated from the population distribution data; and the actual results of the cumulative maintenance costs. Simultaneously, from the multi-dimensional benefit forecast value corresponding to the solution output by the optimization solution generation module, the carbon sink objective function value, the biodiversity objective function value, the social service objective function value, and the economic cost objective function value are extracted as corresponding expected values. The relative deviation between the actual and expected values ​​for each benefit dimension—carbon sink, ecology, social services, and economic costs—is calculated. For carbon sink, ecology, and social service indicators, the relative deviation is the actual value minus the expected value, then divided by the expected value. For economic costs, the relative deviation is the expected value minus the actual value, then divided by the expected value. Subsequently, a deviation tracing operation is performed: for each Pareto optimal solution, the relative deviation calculated for each benefit dimension is determined based on the calculation process of the carbon sink objective function, ecodiversity objective function, social service objective function, and economic cost objective function in the optimization solution generation module. The contribution of each association rule in the decision factor base knowledge package that contributes to the expected value of this benefit dimension is then determined. Next, this relative deviation value is multiplied by the ratio of the contribution of this association rule to the expected value of this benefit dimension, yielding a deviation signal generated by this association rule in this benefit dimension due to the solution. This process is repeated for all Pareto optimal solutions and all their benefit dimensions. For each association rule, all deviation signals obtained across all benefit dimensions of all solutions are accumulated, thereby calculating a comprehensive deviation of this association rule in each benefit dimension. All association rules and their combined deviations across various benefit dimensions are organized into a two-dimensional table, forming a deviation matrix. The specific process of updating the confidence and weight of corresponding association rules in the decision factor base knowledge package based on the bias matrix using a meta-reinforcement learning framework, and abstracting the features of schemes with excellent performance in the dynamic deduction results into new association rules to feed back to the knowledge graph includes: First, based on the deviation matrix, a loss function for each association rule is constructed using the comprehensive deviation of each rule across the dimensions of carbon sink, biodiversity, social services, and economic costs. The value of the loss function is the sum of the products of the preset preference weights for each dimension and the squares of the corresponding comprehensive deviations. Secondly, a meta-policy network is introduced. This network takes the feature vector of the historical performance data of each association rule as input and dynamically outputs an adaptive learning rate for updating the parameters of the rule. The adaptive learning rate includes a first learning rate for updating the confidence parameter and a second learning rate for updating the weight parameter. Then, a parameter update operation is performed, which includes the following steps: calculating the partial derivative of the loss function with respect to the confidence parameter of the association rule, as the update gradient of the confidence parameter; calculating the partial derivative of the loss function with respect to the weight parameter of the association rule, as the update gradient of the weight parameter; subtracting the product of the first learning rate and the update gradient of the confidence parameter from the current confidence parameter of the association rule to obtain an intermediate value of the confidence parameter; subtracting the product of the second learning rate and the update gradient of the weight parameter from the current weight parameter of the association rule to obtain an intermediate value of the weight parameter; and restricting the intermediate values ​​of the confidence parameter and the intermediate values ​​of the weight parameter to a valid value range of zero to one, thereby obtaining the updated confidence parameter and weight parameter. Simultaneously, the feature abstraction and rule feedback operation for high-performing solutions is performed, which includes four sub-steps executed sequentially: Sub-step F1, Selection of excellent solutions: Based on the risk probability values ​​recorded in the dynamic simulation results of the solutions and the actual results obtained at the end of the preset simulation period, set the achievement rate thresholds for carbon sink, biodiversity, and social service indicators and the control rate thresholds for economic cost indicators, and select all Pareto optimal solutions that meet the threshold conditions to form a set of excellent solutions. Sub-step F2, Feature Pattern Mining: Data mining is performed on the resource allocation blueprints of all schemes in the selected set of high-performing schemes to identify the plant species combinations that appear repeatedly, the range of soil and climate conditions corresponding to the combination, and the stable ecosystem service potential, social service effects and resource consumption demand characteristics of the combination in the simulation. Sub-step F3, New Rule Generation and Formalization: The stable patterns identified in sub-step F2 are formalized into new computable association rules. The condition part of the rule is defined based on the identified range of soil and climate conditions and plant species combinations. The output part includes the estimated values ​​of the ecosystem service potential, social service potential, and resource consumption demand for the combination. The new rule is assigned an initial confidence parameter and an initial weight parameter. Sub-step F4, Knowledge Base Feedback: Add all the new computable association rules generated in sub-step F3 to the computable association rule list of the decision factor base knowledge package, thereby completing the expansion of the knowledge graph.

2. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 1, characterized in that: The specific operation of integrating multi-source heterogeneous data in the decision factor library construction module is as follows: In response to planning task triggers, the system acquires and locks the geographical boundaries of the target planning area; it collects and integrates the following categories of data: soil pH, organic matter content, and bulk density data; historical precipitation, temperature series, sunshine hours, and real-time precipitation and temperature monitoring data; remote sensing data on vegetation index, surface temperature, and land use classification; community population density distribution, age structure, and mobile signaling heat map data; cost data on seedling purchase price, transportation costs, and historical maintenance records; and real-time monitoring data from IoT sensors, including real-time precipitation and temperature data from meteorological sensors and real-time soil volumetric moisture content, leaf surface humidity, and trunk flow rate data from soil and plant sensors. All collected multi-source heterogeneous data were spatiotemporally aligned and normalized to form a standardized dataset.

3. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 2, characterized in that: The specific operations for constructing entities based on site conditions, plant species, ecosystem services, and resource consumption, and dynamically expressing the quantitative relationships between entities through computable association rules, are as follows: Based on the standardized dataset, site condition entities, plant species entities, ecosystem service entities, and resource consumption entities are defined in the knowledge graph. Each site condition entity is assigned an attribute vector containing soil pH, organic matter content, bulk density, precipitation, temperature, sunshine duration, vegetation index, surface temperature, soil volumetric water content, leaf surface humidity, and trunk flow rate of the corresponding spatial grid. Each plant species entity is assigned an attribute vector containing the photosynthetic rate, transpiration coefficient, and shade tolerance of the corresponding species. This process establishes applicable, provisional, and consumption relationships between entities, transforming domain knowledge and data patterns into computable production rules. A computable list of association rules is then created, containing all rules. Each computable production rule is an independent data structure comprising a condition part, an output part, and a confidence part. The condition part is the logic for determining the supply of a specific ecosystem service or the demand for a specific resource consumption. This quantitative estimate, called the ecosystem service potential estimate or resource consumption demand estimate, is associated with the specific algorithm used to calculate it. The confidence part records the rule's information. The confidence and weight parameters; the calculation process for the estimated value of ecosystem service potential or resource consumption demand is as follows: First, determine the degree of membership of temperature, precipitation, and soil volumetric water content to the optimal range of plant photosynthesis to obtain the membership degree of temperature, precipitation, and soil moisture. Then, determine a conversion factor based on the physiological characteristics of plants, which is determined by the specific leaf area and biomass allocation coefficient. Finally, multiply the product of the membership degree of temperature, precipitation, and soil moisture by the photosynthetic rate of the plant species and the conversion factor, and then multiply by the time period to complete the calculation. Based on this calculation process, a set of computable association rules containing multiple production rules is generated.

4. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 3, characterized in that: The specific process for forming a decision factor library that includes static attributes and dynamic correlation logic is as follows: The defined site condition entities, plant species entities, ecosystem service entities, and resource consumption entities, the attribute vectors assigned to these entities, the applicable, providing, and consuming relationships established between entities, and the generated set of computable association rules are all encapsulated into a structured decision factor library knowledge package. This decision factor knowledge package provides an application programming interface; The application programming interface supports two types of calls: The first call is to query and return the corresponding estimates of various ecosystem service potentials and resource consumption demands, as well as the attribute vectors of the associated site condition entities and plant species entities, based on the input identification information of the site condition entities and plant species entities; The second call is to return a list of computable association rules. Each rule in this list exists in the form of its data structure, which includes condition judgment logic, the specific algorithm used to calculate the estimates of ecosystem service potentials or resource consumption demands, confidence parameters, and weight parameters.

5. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 4, characterized in that: In the optimization scheme generation module, the specific process of dynamically assembling a multi-objective optimization function by calling the association rules in the decision factor knowledge package is as follows: The application programming interface provided by the decision factor library knowledge package is invoked. Through its first invocation operation, the estimated values ​​of various ecosystem service potentials and resource consumption demands encapsulated by the decision factor library knowledge package are obtained, and through its second invocation operation, a complete list of computable association rules is obtained. The planning space is discretized into grid cells of equal area to form a set of grid cells. For each grid cell in this set, its decision variable is defined as a triple containing plant species number, planting density, and maintenance intensity level. Based on decision variables, the obtained estimates of ecosystem service potential and resource consumption demand, four objective functions are dynamically constructed: carbon sink objective function, ecosystem diversity objective function, social service objective function and economic cost objective function. The calculation process of the carbon sink objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output carbon sink and whose conditions are partially applicable to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a rule selection result set. Then, calculate the product of the carbon sink potential estimate of each rule in this rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this rule selection result set to obtain a weighted sum of carbon sink contribution for the grid cell. Next, the area of ​​this grid cell, the planting density in its decision variables, and the calculated carbon sink contribution are multiplied together to obtain the contribution of this grid cell to the total carbon sink target. Finally, the contributions of all grid cells in the grid cell set are summed, and the result is the carbon sink objective function value. The calculation process of the ecological diversity objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output ecological diversity-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming an ecological diversity rule selection result set; then, calculate the product of the ecological diversity potential estimate corresponding to each rule in this ecological diversity rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the weighted sum of the ecological diversity contribution of the grid cell; the ecological diversity potential estimate is calculated from the corresponding association rule in the decision factor library; next, multiply the area of ​​this grid cell, the planting density in its decision variable, and the calculated weighted sum of ecological diversity contribution to obtain the contribution of the grid cell to the total ecological diversity objective; finally, sum the contributions of all grid cells, and the result is the ecological diversity objective function value; The calculation process of the social service objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output social service-related indicators and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a social service rule selection result set; then, calculate the product of the social service potential estimate corresponding to each rule in this social service rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the weighted sum of the social service contribution of the grid cell; the social service potential estimate is calculated from the corresponding association rule in the decision factor library; next, multiply the area of ​​the grid cell, the planting density in its decision variable, and the calculated weighted sum of social service contributions, and then multiply by a service demand weight coefficient determined based on community population distribution data to obtain the contribution of the grid cell to the overall social service objective; finally, sum the contributions of all grid cells, and the result is the value of the social service objective function; The calculation process of the economic cost objective function is as follows: For each grid cell in the grid cell set, firstly, from the list of computable association rules, select all rules that output resource consumption demand and whose conditions partially apply to the site conditions and the selected plant species number of the decision variable of the grid cell, forming a resource consumption rule selection result set; then, calculate the product of the estimated resource consumption demand corresponding to each rule in this resource consumption rule selection result set with the confidence parameter and weight parameter of the rule, and sum the product results of all rules in this result set to obtain the comprehensive resource consumption weighted demand of the grid cell; next, multiply the area of ​​the grid cell, the planting density and maintenance intensity level conversion coefficient in its decision variables, and the calculated comprehensive resource consumption weighted demand to obtain the resource consumption cost of the grid cell; finally, sum the resource consumption costs of all grid cells and combine them with the initial cost of seedling procurement and transportation, and the result is the value of the economic cost objective function; The dynamically constructed carbon sink objective function, ecodiversity objective function, social service objective function, and economic cost objective function together constitute a multi-objective optimization function.

6. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 5, characterized in that: The specific process of solving the function with carbon sink, biodiversity, social services, and economic costs as the objective vector using the improved NSGA-III algorithm based on knowledge graph pre-screening is as follows: From the list of computable association rules, mandatory constraints on plant site tolerance are extracted. These mandatory constraints include the requirement that the soil pH and salinity for the plant species be within their tolerance range, and these constraints are used as hard constraints for the optimization problem. During each generation of population evolution in the improved NSGA-III algorithm, a knowledge graph pre-screening operation is performed on the newly generated candidate solutions. For candidate solutions pre-screened through the knowledge graph, the four objective function values ​​of carbon sink, biodiversity, social services, and economic cost are calculated. The non-dominated sorting, reference point association, and niche preservation operations of the NSGA-III algorithm are then executed to guide population evolution. After the algorithm terminates, a set of Pareto optimal solutions is output. Each Pareto optimal solution in this set contains two parts: one part is the plant species number, planting density, and maintenance intensity level determined for each grid cell, which is called the resource allocation blueprint of the solution; the other part is the corresponding objective function values ​​of carbon sink, biodiversity, social services, and economic cost, which together are called the multidimensional benefit estimate of the solution.

7. A multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 6, characterized in that: In the dynamic derivation module of the proposed scheme, the specific process of establishing an independent digital twin for each Pareto optimal solution in the Pareto optimal solution set is as follows: For each Pareto optimal solution in the Pareto optimal solution set output by the optimization solution generation module, extract the resource allocation blueprint corresponding to that Pareto optimal solution. Based on this resource allocation blueprint, a separate digital twin is instantiated for the Pareto optimal solution; this digital twin contains a hybrid simulation system consisting of three coupled models: a plant growth model, a carbon cycle model, and a pedestrian flow simulation model; The initial state of the plant growth model is set according to the plant species number and planting density of each grid cell in the resource allocation blueprint. The plant growth model is a hybrid model that combines mechanism and data-driven approaches. Its mechanism part is based on the theory of photosynthesis, respiration and biomass allocation. Its data-driven part obtains the photosynthetic rate, transpiration coefficient and physiological and ecological parameters of the configured plant species, as well as the association rules that affect plant growth, by calling the application programming interface of the decision factor library knowledge package. The initial parameters of the carbon cycle model are derived from the attribute vectors of the site condition entities in the corresponding grid in the decision factor base knowledge package. The carbon cycle model is connected to the output of the plant growth model to receive biomass change data generated by plant growth simulation. The pedestrian flow simulation model constructs the basic environment based on the plant space configuration described in the resource allocation blueprint and combined with geographic information system data.

8. The multi-scale urban garden carbon sequestration enhancement and ecological synergistic regulation system according to claim 7, characterized in that: The specific process for generating dynamic simulation results of the scheme that include all simulation results is as follows: During the simulation process, real-time monitoring data from IoT sensors is continuously connected to each digital twin, and the real-time monitoring data is input into each model of the hybrid simulation system as an environmental driving variable in real time. Simultaneously, a random disturbance factor is systematically introduced into the simulation timeline. This random disturbance factor is a random event sequence generator that simulates uncertain events such as extreme drought, floods, and pest outbreaks. It generates event sequences according to a preset spatiotemporal distribution pattern, introducing a disturbance impact calculation process into the plant growth model: First, the theoretical total primary productivity is calculated based on the model state; then, an environmental stress function value determined by real-time temperature, precipitation, and soil volumetric moisture content is calculated; next, for each random disturbance event triggered at each moment on the simulation timeline, the current intensity value of the event is obtained and multiplied by the vulnerability coefficient of the plant species to such events, obtained from the decision factor knowledge package by querying the attributes or association rules of the configured plant species, to obtain the proportion of potential productivity loss caused by the event at the current moment. The proportion of potential productivity loss caused by all triggered random disturbance events is accumulated and applied to the theoretical total primary productivity together with the environmental stress function value, thereby calculating the net primary productivity after the disturbance. The hybrid simulation system couples the plant growth model, carbon cycle model, and pedestrian flow simulation model in a time-step manner within a preset future time period. This process is called hybrid simulation operation. It dynamically simulates the changes in carbon sink accumulation, ecological indicators, social service effects represented by pedestrian flow distribution, and cumulative maintenance costs over time for each scheme.

Images