Multi-agent urban renewal land management and control multi-objective optimization method

Abstract

The application discloses a multi-agent urban renewal land management and control multi-objective optimization method, which comprises the following steps: collecting multi-source spatio-temporal data and planning management and control requirements of a target urban renewal area, and constructing a multi-agent system and a multi-objective land management and control scheme population; each agent evaluates a scheme individual in parallel, generates a multi-agent evaluation state diagram and a differentiated modification operation pool; based on the differentiated modification operation pool, a multi-round structured interaction and spatial perception closed-loop review are carried out, a consensus modification operation set or a marked irreconcilable plot is obtained; the consensus modification operation set is converted into a structured editing operation, a directional variation and compliance correction are carried out on the scheme individual, a candidate child scheme set is generated and iterative calculation is carried out, and a Pareto optimal solution set is obtained. The application can reduce spatial semantic pseudo-consistency, and improve the optimization efficiency and stability of the land management and control scheme.

Description

Technical Field

[0001] This invention belongs to the field of geographic information and urban renewal computing, and more specifically, relates to a multi-agent multi-objective optimization method for urban renewal land use management. Background Technology

[0002] Urban renewal is a crucial area in my country's urban development strategy, shifting from incremental expansion to improving the quality of existing urban resources. Urban renewal area planning typically requires coordinating multiple objectives simultaneously, including floor area ratio control, land use adjustment, public service facility allocation, and spatial quality optimization. Therefore, methods such as multi-agent systems and multi-objective evolutionary computation are increasingly being introduced to simulate the interests of different stakeholders and assist in generating and optimizing land use control schemes.

[0003] However, existing methods lack close linkage between multi-agent evaluation, negotiation, and scheme optimization. Problems identified by different agents are difficult to directly translate into differentiated spatial modification operations, and negotiation results are also difficult to further drive the directed evolution of individual schemes. Furthermore, for local conflict areas that cannot be resolved after multiple rounds of negotiation, there is a lack of cross-generational constraint mechanisms, making them prone to repeated adjustments in subsequent iterations, thus affecting optimization efficiency and scheme stability. In addition, existing multi-agent negotiation methods typically do not feed back the accessibility effects caused by spatial topology changes to the negotiation process in real time, lacking corresponding spatial perception closed-loop verification mechanisms. In urban renewal area land use management, in response to the community user agent's request for additional green space, the manager agent may need to adjust land boundaries, road network connections, or facility site selection. These spatial adjustments may cause changes in road topology, causing the newly added green space to disrupt key pedestrian paths relied upon by residents for daily travel, creating a paradoxical situation where compensatory green space increases but pedestrian network continuity is compromised. In this case, although the agents reach a consensus on the negotiation results, this consensus does not guarantee the effectiveness of the adjusted spatial relationships, forming a "pseudo-consistency" at the spatial semantic level. The pseudo-consistent solutions generated may form pseudo-optimal solutions and be mixed into the Pareto front after entering the multi-objective optimization process. This may affect the selection of real non-dominated solutions and the distribution of the front, reduce the operability and credibility of the final recommended solution, and may lead to problems such as increased walking detour distances for residents and degradation of accessibility of public service facilities.

[0004] The existing invention patent with publication number CN121094346A proposes a method and system for assessing the accessibility of public facilities based on geographic multi-agent systems. This method acquires user distribution data and public facility data within the study area, constructs a multi-agent system including user agents and facility agents, establishes a set of neighboring interactive objects for user agents, and constructs a dynamic utility function based on evolutionary game theory. It iteratively updates the probability distribution of user agents choosing different facility agents by simulating their exploration and learning behaviors. After system convergence, it calculates the service demand and supply-demand ratio of each facility agent and combines spatial decay relationships to calculate the public facility accessibility evaluation value. However, this existing technology mainly focuses on the dynamic assessment of public facility accessibility. Its multi-agent system primarily simulates the selection and supply-demand relationship between users and facilities, without addressing the differentiated evaluation and modification operations of multiple agents for the same land use control scheme. It also fails to address the issue of continuous feedback and cross-generational optimization of multi-agent negotiation results, gene-edited directed mutations, and multi-generational convergence judgments. Therefore, it struggles to solve the problems of continuous feedback and cross-generational optimization of multi-agent conflicts in urban renewal land use control. Summary of the Invention

[0005] To overcome the problems of insufficient continuous feedback of multi-agent conflicts and lack of cross-generational constraint mechanisms in the process of urban renewal land use management in the prior art, this invention provides a multi-agent multi-objective optimization method for urban renewal land use management.

[0006] The primary objective of this invention is to solve the aforementioned technical problems. The technical solution of this invention is as follows: The first aspect of this invention provides a multi-agent urban renewal land use management multi-objective optimization method, comprising the following steps: S1: Collect multi-source spatiotemporal data and planning control requirements of the target urban renewal area, and construct a multi-agent system including manager intelligent agents, development and operation intelligent agents, community user intelligent agents, and a multi-objective land use control scheme population; S2: Each agent performs parallel evaluation of each scheme in the multi-objective land use control scheme population to obtain the multi-agent evaluation state diagram corresponding to each scheme; S3: Based on the multi-agent evaluation state diagram, each agent generates a first type of structural modification instruction set, a second type of optimization proposal set, and a third type of compensation requirement set. The first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation requirement set are combined to obtain a problem-oriented differentiated modification operation pool. S4: For each individual scheme in the scheme population, initiate multi-round structured interaction between multiple agents based on a preset negotiation protocol based on the differentiated modification operation pool, and perform spatial perception closed-loop verification during the multi-round structured interaction. When a valid consensus on spatial semantic consistency is reached, a negotiated and agreed modification operation set is obtained. When a valid consensus on spatial semantic consistency is not reached, mark the irreconcilable plots and attach a temporary development intensity freeze mark to the corresponding individual scheme. S5: Transform the consensus-based modification operation set into structured editing operations that can be directly applied to the individual gene coding of the scheme through preset mapping rules; S6: Using structured editing operations combined with a differentiated mutation strategy based on land parcel type, targeted mutation is performed on individual schemes that have reached a valid consensus on spatial semantic consistency, and compliance correction is performed on the new individual schemes obtained from the targeted mutations to obtain new individual schemes after compliance correction. S7: Merge the new scheme individuals with the scheme individuals with the additional temporary development intensity freeze mark to obtain a candidate offspring scheme set. Iteratively calculate the multi-objective land use control scheme population through the candidate offspring scheme set to obtain the set of all Pareto optimal solutions of the multi-objective land use control scheme population.

[0007] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention employs a managerial intelligent agent, a development and operation intelligent agent, and a community user intelligent agent to conduct differentiated evaluations of the same multi-objective land use management scheme. Based on the evaluation results, differentiated modification operations are generated, enabling the multi-agent evaluation results to be directly used for subsequent scheme adjustments. A consensus modification operation set is generated through multi-agent negotiation and spatial perception closed-loop verification. This consensus modification operation set is then used to drive gene editing-oriented mutations, achieving effective linkage between multi-agent evaluation, negotiation, and scheme optimization. Simultaneously, a temporary development intensity freeze mark is applied to irreconcilable land parcels and passed to subsequent iterations. Combined with external archive updates and multi-generation convergence judgments, repeated adjustments to irreconcilable land parcels are avoided, achieving continuous feedback of multi-agent conflict information and cross-generational optimization, thereby improving the optimization efficiency and stability of urban renewal land use management schemes. Attached Figure Description

[0008] The present invention is provided with the following figures and is described below: Figure 1 This is a flowchart of a multi-agent urban renewal land use management multi-objective optimization method provided in an embodiment of the present invention; Figure 2 This is a flowchart of the multi-agent evaluation and differential modification operation pool generation process provided in this embodiment of the invention; Figure 3 Detailed flowchart of multi-agent negotiation and spatial perception closed-loop verification provided in embodiments of the present invention; Figure 4 A flowchart of candidate offspring generation and compliance correction provided for embodiments of the present invention; Figure 5 The flowchart for external file update and convergence judgment provided in the embodiments of the present invention. Detailed Implementation

[0009] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0010] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0011] Example 1: This invention provides a multi-agent, multi-objective optimization method for urban renewal land use management, such as... Figure 1 The diagram shows a multi-agent urban renewal land use management multi-objective optimization method. The specific steps are as follows: S1: Collect multi-source spatiotemporal data and planning control requirements of the target urban renewal area, construct a multi-agent system including manager intelligent agents, development and operation intelligent agents, community user intelligent agents, and a population of multi-objective land use control schemes, initialize external archives independent of the populations, and use the external archives to store global non-dominated land use control scheme individuals obtained during the iteration process.

[0012] S2: Each agent performs parallel evaluations of each scheme in the multi-objective land use control scheme population to obtain the multi-agent evaluation state diagram corresponding to each scheme.

[0013] S3: Based on the multi-agent evaluation state diagram, each agent generates a first type of structural modification instruction set, a second type of optimization proposal set, and a third type of compensation requirement set. The first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation requirement set are combined to obtain a problem-oriented differentiated modification operation pool.

[0014] S4: For each individual scheme in the scheme population, initiate multi-round structured interaction between multiple agents based on a preset negotiation protocol based on the differentiated modification operation pool, and perform spatial perception closed-loop verification during the multi-round structured interaction process. When a valid consensus on spatial semantic consistency is reached, a negotiated and agreed modification operation set is obtained. When a valid consensus on spatial semantic consistency is not reached, mark the irreconcilable plots and attach a temporary development intensity freeze mark to the corresponding individual scheme.

[0015] S5: Transform the consensus-based modification operation set into structured editing operations that can be directly applied to the individual gene coding of the scheme through preset mapping rules.

[0016] S6: Using structured editing operations combined with a differentiated mutation strategy based on land parcel type, targeted mutations are performed on individual schemes that have reached a valid consensus on spatial semantic consistency. The new individual schemes obtained from the targeted mutations are then corrected for compliance, resulting in new individual schemes with improved compliance.

[0017] S7: Merge the new scheme individuals with the scheme individuals with the additional temporary development intensity freeze mark to obtain a candidate offspring scheme set. Iteratively calculate the multi-objective land use control scheme population through the candidate offspring scheme set to obtain the set of all Pareto optimal solutions of the multi-objective land use control scheme population.

[0018] This invention employs a managerial intelligent agent, a development and operation intelligent agent, and a community user intelligent agent to conduct differentiated evaluations of the same multi-objective land use management scheme. Based on the evaluation results, differentiated modification operations are generated, enabling the multi-agent evaluation results to be directly used for subsequent scheme adjustments. A consensus modification operation set is generated through multi-agent negotiation and spatial perception closed-loop verification. This consensus modification operation set is then used to drive gene editing-oriented mutations, achieving effective linkage between multi-agent evaluation, negotiation, and scheme optimization. Simultaneously, a temporary development intensity freeze mark is applied to irreconcilable land parcels and passed to subsequent iterations. Combined with external archive updates and multi-generation convergence judgments, repeated adjustments to irreconcilable land parcels are avoided, achieving continuous feedback of multi-agent conflict information and cross-generational optimization, thereby improving the optimization efficiency and stability of urban renewal land use management schemes.

[0019] Example 2: This embodiment further supplements the explanation of the multi-agent urban renewal land use management multi-objective optimization method proposed in Embodiment 1.

[0020] Step S2, which involves using agents to perform parallel evaluations of individual schemes within a multi-objective land use management scheme population, includes the following steps: S2.1: Using the local utility functions of the manager agent, the development and operation agent, and the community user agent, we evaluate each individual scheme in the current multi-objective land use control scheme population in parallel, and identify the problem plots in each scheme individual that have not met the expectations of the corresponding agent according to the satisfaction threshold of each agent, so as to obtain the set of problem plots.

[0021] More specifically, by utilizing community-based intelligent agents to calculate the walkability of individual schemes within the current multi-objective land use control scheme population, problem plots are identified. A negative exponential decay function is employed, representing the shortest path distance from the geometric center of the plot along the existing road network to the nearest public service facility. The process includes the following steps: A weighted directed graph was constructed by extracting the centerlines of individual road areas from multi-source spatiotemporal data for each scheme. ,in This is a set of nodes representing road intersections and site entrances / exits. For road segment collection, The segment weight is defined as the product of segment length and pedestrian obstruction coefficient. The pedestrian obstruction coefficient is taken as 0.8 to 1.5 based on factors such as sidewalk width, the completeness of pedestrian crossing facilities, and the presence or absence of traffic lights. Each plot of land Walkable accessibility index The expression is as follows:

[0022] in, The types of public service facilities that the community-based smart agents focus on include community green spaces, community health service stations, community cultural activity centers, and everyday commercial outlets. For facility type The weighting coefficients can be determined using the Delphi method or a resident questionnaire survey. This is the distance attenuation factor, which is set to 0.15 in this embodiment. For the plot of land To the latest type The shortest path distance to the facility is determined by Dijkstra's algorithm in the graph. Calculated in the middle; The value range is (0,1], and the larger the value, the better the accessibility. This is based on the walkability index of plot i. and per capita open space area The weighted sum is used to calculate the local utility value of plot i. The local utility value Land parcels that fall below the preset minimum satisfaction threshold are identified as problem parcels.

[0023] S2.2: Each agent uses a differentiating label to label the set of problem plots, obtaining a differentiating label matrix corresponding to each individual solution. The rows of the differentiating label matrix correspond to the plot numbers in the solutions, and the columns correspond to the manager agent, development and operation agent, and community user agent, respectively. Matrix element a... ij For a binary value, a ij =1 indicates that plot i did not reach the satisfaction threshold under the evaluation of agent j, a ij =0 indicates that the satisfaction threshold has been reached; the differential labeling matrix is ​​associated with the scheme spatial graph according to the plot number, so as to display the location of the problem plot and the agent that marks the plot in the multi-agent evaluation state diagram.

[0024] S2.3: Overlay the differential label matrix onto the spatial graph of the corresponding scheme individual to obtain the multi-agent evaluation state graph for each scheme.

[0025] like Figure 2 As shown, step S3 utilizes each agent to generate a differentiated modification operation pool based on the current population evaluation results, including the following steps: By utilizing the managerial intelligent agent based on the multi-agent evaluation state diagram and the current development intensity gradient of the area where the problem plot is located, a first type of structural modification instruction set is generated, which includes instructions for cross-plot transfer of floor area ratio, instructions for fine-tuning of the boundary of public facility land, and instructions for adjusting the scale of supporting facilities. The development and operation agent generates a second set of optimization proposals, including instructions for changing land use and instructions for adjusting the upper limit of development intensity, based on the expected internal rate of return and the capital recovery period of the problem plots. Based on the analysis results of the walkability and spatial syntactic axis diagram of the problem plots, the community uses an intelligent agent to generate a third set of compensation needs, which includes the preliminary spatial location of the proposed green space and the existing pedestrian path segments that must be protected. The first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation requirement set are combined to obtain a problem-oriented differentiated modification operation pool.

[0026] More specifically, by utilizing the managerial agent based on the multi-agent evaluation state diagram and combining it with the current development intensity gradient of the area where the problem plot is located, a first-class set of structural modification instructions is generated, including the following steps: The manager agent extracts the set of marked problem plots from the multi-agent evaluation state graph. ; Set of problem plots Current development intensity distribution data for individual projects under the current scheme Perform spatial overlay analysis to calculate the plot ratio of the problem plot i. Compared with the average target value of the area deviation The expression is as follows:

[0027] right For each plot of land, the transfer amount is calculated using the excess capacity that can be reduced from plot i, and the receiving plot is determined based on the maximum development intensity gradient of adjacent high-value areas, generating a plot ratio transfer instruction across plots. Set of problem plots The service radius coverage data of public facilities is buffered and overlaid to identify public facilities affected by insufficient coverage. The boundary of the facility land is fine-tuned according to the connectable direction of adjacent unused land, and a fine-tuning instruction for the boundary of public facility land is generated. The allocation scale is recalculated based on the population gap served by the facilities, resulting in an instruction to adjust the allocation scale; The first type of structural modification instruction set consists of instructions for transferring plot ratio across plots, instructions for fine-tuning the boundaries of public facility land, and instructions for adjusting the scale of supporting facilities.

[0028] More specifically, the development and operation agent generates a second set of optimization proposals based on the expected internal rate of return and payback period of the problem plots, including the following steps: The development and operation agent extracts the marked problem plots, their current land use, and the upper limit of development intensity from the multi-agent evaluation state map; For each problem plot Calculate the expected internal rate of return for conversion to candidate high-value types such as commercial or residential. and the return period of funds Filtering those that meet the criteria of expected internal rate of return being greater than or equal to a preset expected internal rate of return threshold and a return period of funds less than or equal to a preset maximum return period of funds. and The land use change option generates a second set of optimization proposals that includes the target land use and the recommended upper limit of development intensity.

[0029] More specifically, by utilizing community-based intelligent agents to analyze the walkability and spatial syntax axis diagrams of problem sites, a third type of compensation demand set is generated. This is achieved through a spatial syntax-assisted supplementary site selection method, including the following steps: The community uses agents to calculate the local integration degree of the spatial syntactic axis graph of the current scheme for individual areas. and global integration and the difference between the two Areas with high topological separation and poor walkable accessibility were identified as needing improvement. Larger than the total number of individual plots in the plan (Third-digit individuals); In spatial syntax, the integration index is calculated based on the adjacency matrix of the spatial syntax axis graph. For a given set of axes, the integration index is calculated based on the adjacency matrix of the spatial syntax axis graph. A single-axis model, global integration As axis global average depth The standardized reciprocal is expressed as follows:

[0030] in, The average depth is expressed as follows:

[0031] in, As axis To the axis The shortest topological distance, The standardized parameters for the diamond-shaped diagram are used to eliminate the influence of system size; Local integration Limit the average depth to the axis topological radius Within this embodiment The value is 3, meaning only the axis is considered. The remaining axes in the three-step reachable subsystem, local integration degree The expression is as follows:

[0032] For each area to be improved, unused and inefficiently used land parcels within and around it are extracted as candidate locations for supplementary green spaces. Constrained K-means clustering is then used to categorize the candidate locations according to... The values ​​are grouped, and the geometric center of each group is taken as the preliminary spatial location for the proposed supplementary green space; The community uses the smart agent to extract path segments whose usage frequency exceeds the preset minimum threshold based on the shortest path from each plot to the nearest public transportation station and public service facility in the current scheme. The path segments located inside the area to be improved or connecting the area to be improved with the core public service facilities are marked as existing pedestrian path segments that must be protected. The preliminary spatial location of the proposed additional green space and the existing pedestrian path segments that must be protected are used to form a third type of compensation demand set, which is then output to a multi-agent shared data space.

[0033] like Figure 3 As shown, step S4, which involves initiating multi-round structured interactions among multiple agents based on a preset negotiation protocol using a differentiated modification operation pool, and performing spatial awareness closed-loop verification during the multi-round structured interactions, includes the following steps: S3.1: Initialize the negotiation round counter to It initiates multi-round structured interactions between the managerial intelligent agent, the development and operation intelligent agent, and the community user intelligent agent based on a preset negotiation protocol, with a preset maximum number of negotiation rounds. .

[0034] S3.2: Utilizing the differentiated modification operation pool, the manager agent integrates the suggestions from the first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation demand set to supplement the preliminary spatial location of the green space and generate a comprehensive conflict resolution draft.

[0035] S3.3: Using a pre-built walkability accompanying agent model, a spatial perturbation simulation of the comprehensive conflict resolution draft is performed to obtain the predicted walkability change thermodynamic values ​​for each relevant plot.

[0036] More specifically, the walkability-associated surrogate model employs a graph attention network regression model (GAT). This model is used to perform spatial perturbation simulations of the comprehensive conflict resolution draft, obtaining predicted walkability change heatmap values ​​for each relevant plot. The process includes the following steps: Training samples are generated using historical iterative data and pre-executed Monte Carlo spatial perturbations. Each training sample includes a land use control scheme before the spatial perturbation, a land use control scheme after the spatial perturbation, and the corresponding precise walkability matrices before and after the spatial perturbation. The matrix elements of the precise walking reachability matrix This represents the walking time from plot i to facility type k; Based on the changes in candidate facility locations before and after spatial disturbances in each training sample, as well as the addition, deletion, or geometric adjustment of roads, the corresponding facility site selection coordinate offset vector is calculated. and local road segment weight change vector ; Calculate the thermodynamic values ​​of the changes in precise walkable accessibility for each plot of land based on the precise walkable accessibility matrix before and after the spatial disturbance. ; The land use category code, standardized development intensity, geometric center coordinates of each plot, and road connection topology between adjacent plots in the land use control plan before spatial disturbance are constructed as node feature vectors. Using facility site selection coordinate offset vector Update the node feature vector The node coordinate features corresponding to the affected facilities are used to change the local road segment weight vector. Update the edge weight attributes corresponding to the road connection topology between adjacent plots to obtain the graph attention network input for the corresponding training samples; The graph attention network is input into the first layer of the graph attention network regression model, and the attention coefficients of neighboring plots j are calculated through a multi-head attention mechanism. The expression is as follows:

[0037] in, This represents the activation function of the linear unit with leakage correction. This represents a trainable attention weight vector. , They respectively represent the use of shared linear transformation matrices For node feature vectors and The node feature representation obtained by performing a linear transformation. This represents a vector concatenation operation; The attention coefficients are Softmax normalized, and the normalized attention coefficients are then used to weight and aggregate the features of neighboring nodes to obtain the updated node embedding feature vector. The expression is as follows:

[0038] in, For the number of attention heads, For the plot of land The set of neighboring nodes, and The first The weight matrix and normalized attention coefficients of each attention head. For ELU activation function, The number of layers in the attention network propagation; The nodes, after propagation through a predetermined number of layers, are embedded into feature vectors and input into a linear output layer to obtain the predicted walkable accessibility heatmap values ​​for each relevant plot. According to the predicted changes in walkable accessibility thermodynamic values Thermodynamic values ​​of precise walkability accessibility The mean squared error loss function L is calculated as follows:

[0039] Where N represents the number of land parcel training samples participating in the loss function calculation, and i represents the land parcel training sample number participating in the loss function calculation; Backpropagation optimization is performed based on the loss function L to obtain the trained walking accessibility companion agent model; Based on the changes in the candidate locations of facilities and the addition, deletion, or geometrical adjustment of roads caused by the comprehensive conflict resolution draft relative to the individual corresponding schemes, calculate the facility site selection coordinate offset vector. and local road segment weight change vector ; Construct node feature vectors based on the corresponding schemes, and utilize facility site selection coordinate offset vectors. Update the node coordinate features corresponding to the affected facilities, and use the local road segment weight change vector. Update the edge weight attributes of the corresponding road connections to obtain the graph attention network input corresponding to the comprehensive conflict resolution draft; input the graph attention network input corresponding to the comprehensive conflict resolution draft into the trained walking accessibility adjoint agent model to obtain the predicted walking accessibility change heat value corresponding to the comprehensive conflict resolution draft.

[0040] In the process of multi-round structured negotiation between the manager agent, the development and operation agent, and the community user agent, this invention introduces a spatial perturbation pre-simulation mechanism based on the accompanying agent model. This mechanism feeds back the accessibility effect of spatial topology changes to the negotiation process in real time, solving the problem of existing technologies outputting "pseudo-consistent" solutions due to a lack of spatial awareness.

[0041] S3.4: Input the predicted heat values ​​of changes in walkable accessibility and the existing walkable route segments that must be protected from the third type of compensation demand set into the community user agent, and use the community user agent to calculate the walkable accessibility degradation rate based on the walkable accessibility of each plot before and after the adjustment. The expression is as follows:

[0042] in, This represents the predicted walkability index of plot i after adjustments based on the comprehensive conflict resolution draft for the corresponding scheme. This represents the walkability index of plot i before the individual adjustment of the corresponding scheme.

[0043] If the rate of degradation of walkability ,in, To preset the tolerance threshold, in the embodiment Set the value to 0.05 and add plot i to the set of plots with excessive accessibility degradation. .

[0044] S3.5: Traverse the set of existing walking path segments that must be protected. Each path segment Calculate path segments Connectivity indicators in the road network after adjustments to the comprehensive conflict resolution draft. , ,like Then Add to the set of critical walking path segments that must maintain connectivity The set of accessibility degradation exceeding the standard and the set of key walking path segments, namely and The union of these sets is determined as a hard constraint on spatial reachability.

[0045] S3.6: The development and operation agent calculates the expected net profit change rate of the land parcels affected by the modification operations before and after the implementation of the comprehensive conflict resolution draft. When the absolute value of the expected net profit change rate exceeds a preset profit impact tolerance threshold, an objection response is generated specifying the range for adjusting the development intensity or allocation ratio. In this embodiment, when the development and operation agent calculates the profit impact value of the draft, it uses the specific land parcels affected by the modification operations as an example. As the accounting object, the land parcels before and after the implementation of the draft were included. The expected rate of change in net profit is defined as the profit impact value. The expression is as follows:

[0046] in, This represents the expected net profit of the land parcel throughout the entire project lifecycle under the current plan. The revised draft stipulates the expected net profit for this land parcel. The formula for calculating the expected net profit is shown below:

[0047] in, The expected total revenue for the site is estimated by multiplying the saleable or leasable building area by the revenue per unit area. This represents the expected total cost, including land costs, construction costs, and additional costs incurred due to ancillary facility requirements. When the preset tolerance threshold is exceeded, the development and operation agent outputs an objection response, accompanied by... The numerical adjustment range obtained from the reverse mapping suggests that the development intensity or the proportion of supporting facilities should be adjusted accordingly to bring the impact on profits back to the tolerance range.

[0048] S3.7: If the current negotiation round is less than the preset maximum negotiation round, and there are hard constraints on spatial accessibility or objections, combine the accessibility degradation exceeding the standard, the key pedestrian route segments that must be protected, and the facilities requiring adjustment in the objections to obtain the affected element set. Extract the current land use boundaries of all plots and facilities in this set, generate the minimum bounding rectangle of the whole set, and extend the minimum bounding rectangle outward by a preset buffer distance. In the embodiments, take Meters, thus obtaining a bounded search domain; S3.8: Using the manager agent combined with the particle swarm optimization algorithm with hard constraints, perform bounded space replanning in the bounded search domain to obtain an updated comprehensive conflict resolution draft, and increase the counter value of the negotiation round. Then return the updated comprehensive conflict resolution draft to S3.3 to continue the spatial perturbation pre-play. S3.9: If the current negotiation round is less than the preset maximum negotiation round, and the spatial accessibility hard constraint and the opposition response are both empty, the community uses an intelligent agent to perform a final accessibility pre-run check on the current comprehensive conflict resolution draft. After confirming that the predicted walkability degradation of each plot has not exceeded the preset tolerance threshold, it is determined that the corresponding scheme individuals have reached a valid consensus on spatial semantic consistency, and the negotiated consensus modification operation set is output; otherwise, the relevant plot groups of the scheme individuals are marked as current irreconcilable plots, and a temporary development intensity freeze mark is applied to the scheme individuals.

[0049] In the negotiation participation logic of the intelligent agent in the community, this invention adds a mechanism for calculating walking accessibility based on the accessibility pre-simulation results and generating hard constraints on spatial accessibility. Before determining the consensus, a final accessibility verification is enforced. Any adjustment scheme that violates the condition is excluded, thereby achieving a dual guarantee of "consensus" and "valid spatial accessibility".

[0050] More specifically, the bounded space replanning is performed within a bounded search domain using a managerial agent combined with a particle swarm optimization algorithm with hard constraints, including the following steps: Using the centroid coordinates, land boundary control point coordinates, and pedestrian access node coordinates of the affected plots and adjacent facilities within the bounded search domain as decision variables, a particle swarm is constructed. The expression for the position vector X of each particle is shown below:

[0051] In the formula, and Here, x and y are the x and y coordinates of the mth spatial control point, respectively, where m is the number of spatial control points. The velocity and position of each particle are updated using inertia weights, individual historical best positions, and global best positions. The spatial reachability hard constraint violation is then calculated based on the spatial adjustment scheme corresponding to each particle after the update. , legally binding constraints violated And the number of violations of connectivity constraints of existing pedestrian path segments that must be protected. ,include: (1) Based on the hard constraint of spatial accessibility output by the community using the intelligent agent, the set of land parcels with excessive accessibility degradation is denoted as D. For any land parcel i in the set of land parcels with excessive accessibility degradation D, the spatial adjustment scheme determined by the corresponding particle is input into the walking accessibility accompanying agent model to obtain the predicted walking accessibility change thermodynamic value of land parcel i. Based on the pedestrian accessibility index of plot i before adjustment and predictive thermodynamic values ​​of changes in walkable accessibility Calculate the predicted walkability index of plot i under the spatial adjustment scheme. The expression is as follows:

[0052] Will The lower limit of the preset tolerance threshold of plot i Compare and calculate the spatial accessibility hard constraint violation of plot i. The spatial accessibility hard constraint violation amounts of each plot in the set D of plots with excessive accessibility degradation are summed to obtain the spatial accessibility hard constraint violation amount. ; (2) Based on the legally binding constraints corresponding to the upper limit of plot ratio and the lower limit of green space ratio, the plot ratio of each plot shall be determined. With floor area ratio limit Compare and determine the green space ratio of each plot. With the lower limit of green space ratio Compare and calculate the floor area ratio violation amount for plot i respectively. and green space ratio violations The sum of the violations of plot ratio and green space ratio for each plot is used to obtain the amount of violation of legally binding constraints. ; (3) Based on the set of existing pedestrian route segments that must be protected Verify the connectivity of each existing pedestrian path segment that must be protected in the road network under the corresponding spatial adjustment scheme of the particle, and adjust the path segments. The connectivity state in the adjusted road network is represented by a binary indicator. , where if path segment Maintain connectivity. If path segment Not connected Based on the binary indicator quantities corresponding to each path segment Calculate the adjusted path connectivity index The expression is as follows:

[0053] The corresponding path segments The set of existing pedestrian path segments that must be protected is identified as a violation of connectivity constraints. The connectivity constraint violations of each path segment are summed to obtain the connectivity violation amount of the current pedestrian path segments that must be protected. ; Based on the violation of spatial accessibility hard constraints , legally binding constraints violated And the number of violations of connectivity constraints of existing pedestrian path segments that must be protected. The total penalty term is calculated as follows:

[0054] in, , , The first positive penalty coefficient, the second positive penalty coefficient, and the third positive penalty coefficient are preset to ensure that particles that violate hard constraints are penalized proportionally to the amount of violation. The weighted sum is calculated by summing the absolute values ​​of the accessibility degradation degree corresponding to the community's use of intelligent agents and the expected net profit change rate corresponding to the development and operation of intelligent agents, and then the weighted sum is multiplied by the total penalty term. The fitness values ​​of each particle are then summed to obtain their fitness values. The individual historical best position and global best position are updated based on the fitness value of each particle, and the particle velocity and position updates and fitness value calculations are repeated until the preset particle swarm iteration termination condition is met. The optimal particle is determined from the particles that meet the spatial accessibility hard constraint, legal rigid constraint and the connectivity constraint of the existing pedestrian path segment that must be protected. The centroid coordinates, land boundary control point coordinates and pedestrian connecting passage node coordinates of the facilities after the bounded space replanning are determined based on the spatial control point coordinates corresponding to the optimal particle, forming an updated comprehensive conflict resolution draft.

[0055] This invention embeds a bounded space replanning step into the conflict resolution process of the manager's intelligent agent. It uses hard constraints such as spatial accessibility, legally mandated rigid planning constraints, and maintaining pedestrian path connectivity as inviolable hard constraints. It employs a particle swarm optimization method with hard constraints deployed within a limited space to jointly fine-tune facility site selection, land boundaries, and pedestrian access routes, thereby eliminating the contradictory phenomenon of "compensatory green space cutting off pedestrian paths" from the root.

[0056] like Figure 4 As shown, step S5 involves consolidating the consensus-based modification operation set into gene editing operations that can be directly applied to the gene coding of individual participants according to a preset mapping rule. Consolidation refers to transforming the dispersed, semantic modification instructions within the consensus-based modification operation set into structured editing operations that can be directly applied to the gene coding of individual participants through a preset mapping rule. The gene editing operations include: (1) Map the target plot number and new land use category code of the land use nature change instruction in the consensus modification operation to the directional assignment operation of the corresponding plot land use nature location; (2) Map the number of the land parcel being transferred, the number of the land parcel being received, and the amount of the transfer in the plot ratio transfer instruction to the reduction operation of the continuous development intensity variable of the land parcel being transferred and the increment operation of the continuous development intensity variable of the land parcel being received, respectively. The reduction of the land parcel being transferred is the amount of the transfer, and the increment of the land parcel being the product of the amount of the transfer and the preset plot ratio transfer efficiency coefficient. (3) Map the fine-tuning instructions for the boundaries of public facilities land and the instructions for adjusting the scale of supporting facilities to the displacement operation of the geometric control point coordinates of the affected plots and the assignment operation of the supporting scale parameters, respectively. Map the instructions for adding green space and pedestrian connecting passages to the assignment operation of the land use nature category code of the corresponding plots and the writing operation of the connecting passage node coordinates to the spatial topology coding layer. This study utilizes gene editing combined with a land parcel-type-based differential mutation strategy to perform targeted mutations on individuals representing schemes that achieve effective consensus on spatial semantic consistency. Specifically, based on the priority weights of the manager agents corresponding to candidate land use categories, a weighted roulette wheel is used to determine the target land use category code from the set of candidate land use categories for the corresponding land parcel. This target land use category code is then assigned to the corresponding land use locus. The priority weights of the manager agents are... The expression is as follows:

[0057] in, Indicating land price potential index, , These represent the first weighting coefficient and the second weighting coefficient, respectively (in this embodiment, they can be taken as...). , ), The difference between the maximum achievable floor area ratio (FAR) and the current FAR for a candidate land use category is expressed as:

[0058] in, This represents the maximum floor area ratio corresponding to candidate land use category j. This indicates the current average plot ratio of the corresponding land parcel; Determine the standard deviation of the adaptive decay perturbation based on the current iteration number. Wherein, the standard deviation of the disturbance corresponding to generation t. The expression is as follows:

[0059] Where t represents the current iteration number, This represents the standard deviation of the initial disturbance. The attenuation coefficient represents the standard deviation of the disturbance. Its value is determined by empirical parameter setting or cross-validation based on the expected convergence speed and search stability. The preferred value range is 0.01 to 0.1. From a mean of zero and having the aforementioned perturbation standard deviation The perturbation values ​​are extracted from the Cauchy distribution and superimposed onto the continuous development intensity variable of the corresponding plot to obtain a new scheme individual with directional variation. The compliance correction for individual new schemes includes: sequentially performing the upper limit projection of plot ratio, the lower limit projection of green space ratio, and the bottom line projection of public facility land area, and using the minimum Euclidean distance mapping to the boundary of the feasible region to map variables that exceed the legal rigid constraints to the corresponding feasible region; The newly revised schemes that have completed compliance corrections are merged with the schemes marked with temporary development intensity freeze flags to obtain a set of candidate sub-schemes. The set of candidate sub-schemes is then verified for compliance correction. Schemes that exceed the maximum floor area ratio, the minimum green space ratio, or the minimum area of ​​public facilities land are forcibly mapped to the feasible domain to obtain a set of compliant candidate sub-schemes.

[0060] like Figure 5 The diagram shows the flowchart for external file update and convergence judgment. Step S7, which iteratively calculates the multi-objective land use control scheme population using the candidate offspring scheme set, includes the following steps: S7.1: Initialize an external archive independent of the multi-objective land use control scheme population, the external archive being used to store non-dominated scheme individuals obtained during the iteration process.

[0061] S7.2: Merge the candidate offspring scheme set with the current multi-objective land use control scheme population to obtain an expanded population; S7.3: Perform non-dominated ranking on the individuals of the schemes in the extended population, and perform elite selection based on the non-dominated ranking results to obtain the next generation of multi-objective land use control scheme population.

[0062] S7.4: Merge the non-dominant scheme individuals in the expanded population with the scheme individuals in the external archive, and use the Pareto dominance relation to remove the dominated scheme individuals to obtain the updated external archive.

[0063] S7.5: Calculate the hypervolume index of the Pareto front using the updated external archive, and calculate the multigeneration moving average rate of change based on the hypervolume index over several generations.

[0064] S7.6: Convergence is determined based on the rate of change of the multi-generation moving average and the current iteration number. If the preset convergence condition (in this embodiment, a continuous preset number of iterations) is not met, the convergence is considered closed. The algebra is lower than the preset rate of change threshold, preset algebra Select 3 to 5, preset the rate of change threshold. The value is 10 -5 ~10 -4 If the current iteration number has not reached the preset maximum iteration number, then the next generation of multi-objective land use control scheme population obtained in S7.3 is used as the current multi-objective land use control scheme population and the process returns to step S2 to continue iterating. The temporary development intensity freeze flag is used as a constraint for the manager agent to generate the first type of structural modification instruction set, and the generation of adjustment instructions for irreconcilable land parcels in the scheme is prohibited. If the preset convergence condition is met or the current iteration number reaches the preset maximum iteration number, then the iteration is terminated, and the set of all Pareto optimal solutions of the multi-objective land use control scheme population is obtained according to the updated external archive.

[0065] The process of performing a non-dominated ranking of individuals in the expanded population and then selecting elites based on the non-dominated ranking results includes the following steps: The candidate offspring scheme set is merged with the current multi-objective land use control scheme population to obtain an extended population. Reference points are generated uniformly on the unit hyperplane of the M-dimensional objective space according to the preset division of each objective dimension. The generation of reference points adopts the systematic method of Das and Dennis. Based on the reference points, a fast non-dominated sort is performed on the extended population to obtain multiple non-dominated layers. For any scheme individual in the extended population, if there is no other scheme individual that is not inferior to the scheme individual in all optimization objectives and is superior to the scheme individual in at least one optimization objective, then the scheme individual is determined as a non-dominated scheme individual. For each non-dominated layer, the distance between adjacent individuals of each scheme is calculated in each target dimension. The distance between adjacent individuals is normalized based on the maximum and minimum values ​​of the current non-dominated layer in each target dimension. The normalized results corresponding to each target dimension are then aggregated to obtain the adaptive crowding distance of each scheme individual. The expression is as follows:

[0066] Where M is the objective dimension and k is the scheme number. For individual solutions In each objective The distance between adjacent individuals on the surface and The current non-dominated layer is in the th... Maximum and minimum values ​​on the target The value is 10 -6 The division constant against zero; The scheme individuals in the extended population are sorted in order of non-dominated hierarchy from low to high and adaptive congestion distance within the same non-dominated hierarchy from large to small. The scheme individuals to be selected are compared with the scheme individuals in the external archive in a joint Pareto dominance comparison. Scheme individuals that are Pareto dominated by any scheme individual in the external archive are removed. Elite individuals are selected from the remaining scheme individuals based on non-dominated hierarchy and adaptive congestion distance to form the next generation of multi-objective land use management scheme population. The non-dominated scheme individuals in the expanded population are merged with the scheme individuals in the external archive, and the dominated scheme individuals are removed using the Pareto dominance relation to obtain the updated external archive. The updated external archive stores the global non-dominated land use control scheme set obtained up to the current iteration. Scheme individuals carrying temporary development intensity freeze markers participate in the sorting and selection of the current iteration according to the non-dominated level and adaptive congestion distance. Temporary development intensity freeze markers do not participate in the sorting calculation of the current iteration. When a scheme individual carrying a temporary development intensity freeze marker enters the external archive, if the scheme individual is dominated by a new non-dominated scheme individual, it is replaced; otherwise, the scheme individual and its temporary development intensity freeze marker are retained.

[0067] Step S7.5, which involves calculating the multi-generation moving average rate of change of the hypervolume index using the updated external archive, includes the following steps: Based on the updated external archives, the hypervolume index HV(g) of the Pareto front in the g-th generation external archives is determined. This is achieved by multiplying the worst values ​​of each generation's population across all target dimensions by 1.1 to obtain the hypervolume reference point vector. The hypervolume reference point vector, along with the Lebesgue measure of the dominance region defined by the scheme individuals in the target space within the external archives, is then used to determine the hypervolume index HV(g). Based on the hypervolume index HV(g) and the preset moving window width... Calculate the moving average of the hypervolume index for generation g. The expression is as follows:

[0068] in, This determines the width of the movable window, with a value ranging from 5 to 15. The first The moving average of the hypervolume index of the generation With the The moving average of the hypervolume index of the generation Comparison, calculation of the multi-generation moving average rate of change of the hypervolume index The expression is as follows: .

[0069] This invention deeply integrates multi-agent game equilibrium search with multi-objective evolutionary population optimization, forming a complete iterative decision chain from differentiated evaluation, negotiation correction, directional mutation to reference point non-dominated ranking and hypervolume convergence criterion, which significantly improves the scientificity and operability of urban renewal area land use control schemes.

[0070] Example 3: This implementation method focuses on an old urban renewal area with a total area of ​​2.1 square kilometers in the core area of ​​a city. The area includes thirty-five plots of land numbered 1 to 35. The current land use is mixed with residential, inefficient industrial and scattered commercial use, with insufficient coverage of public service facilities and deterioration of the pedestrian environment.

[0071] 1. Data Sources and Preprocessing: Multi-source spatiotemporal data are derived from the urban land space basic information platform, real estate registration database, mobile phone signaling population heat map data, road network vector data, and public service facility POI data. Mobile phone signaling data is used to identify the distribution of residential and employed populations and pedestrian traffic flow on road segments; POI data is classified and deduplicated to obtain the locations of four types of public service facilities; in the road network vector data, the pedestrian obstruction coefficient is determined comprehensively based on factors such as sidewalk width (1.5 for <1.5m, 1.2 for 1.5-3m, and 1.0 for >3m), the presence or absence of pedestrian overpasses or underpasses (reducing the coefficient by 0.1 if present), and traffic light waiting time at intersections (increasing by 0.05 every 30 seconds).

[0072] 2. Agent goal setting: The decision-making objectives of the management intelligence agent are to achieve a balanced overall plot ratio in the area, maximize the coverage of public facilities service radius, and increase the annual tax revenue growth rate; the development and operation intelligence agent is driven by a 15-year project internal rate of return (IRR) ≥ 12% and an 8-year payback period; the community usage intelligence agent focuses on the walkability and per capita open space area (minimum 2 square meters / person) of four types of facilities: the nearest community green space, community health service station, community cultural activity center, and daily commercial outlets. Based on a survey of 200 residents in the area, the values ​​were determined using the Analytic Hierarchy Process (AHP) to be 0.4, 0.25, 0.15, and 0.20, respectively.

[0073] S1. Multi-agent parallel evaluation and problem parcel identification In the current 10th iteration, the scheme population size is 100. For each individual scheme, a road-weighted directed graph constructed from spatiotemporal data is used. Based on this, walkability is calculated. (Figure) There are a total of 412 road intersections and plot entrances / exits, comprising 688 road segments, with road segment weights. The pedestrian obstruction coefficient is calculated by multiplying the geometric length of the road segment by a factor ranging from 0.8 to 1.5, depending on the width of the sidewalk and the completeness of the crossing facilities. The community uses intelligent agents to define the set of public service facility types. Including the above four categories, weight The distance attenuation factors are 0.4, 0.25, 0.15, and 0.2, respectively. Example calculations are performed for plots numbered 5, 12, and 23: the shortest path distance from plot 5 to the nearest community health service station is calculated. meters, to the nearest community green space meters, to the nearest cultural activity center Rice, to the nearest daily commercial outlet If the distance is meters, the accessibility index of the land parcel can be calculated according to the following public notice:

[0074] The per capita open space area of ​​plot 5 under this plan Square meters per person. The local utility value of the community's intelligent agent is taken as... ,in for Values ​​normalized to [0,1]; the preset minimum satisfaction threshold is 0.55, and the utility value of plot 5 is calculated to be 0.7. 0.59 + 0.3 0.42 = 0.413 + 0.126 = 0.539 < 0.55, and is therefore marked as a problem plot. Simultaneously, the management agent uses the variance of the plot ratio deviating from the area's average plot ratio and the coverage rate of public facilities within a five-minute radius as utility functions, while the development and operation agent uses an expected internal rate of return (IRR) of less than 12% for each plot as the labeling threshold. All plots are evaluated in parallel, generating a differentiated labeling matrix. Each scheme in the population receives three columns of label vectors, forming a multi-agent evaluation state diagram.

[0075] S2. Generation of Differentiated Modification Operation Pool The management agent, considering the current development intensity gradient of the problem areas, generates a first-type structural modification instruction set: for plots 8 and 19 in the problem plot cluster, it proposes transferring the plot ratios from 2.2 and 1.8 to the high-value plots 28 and 31 proposed by the development and operation agent, with transfer amounts of 0.6 and 0.4 respectively; for public welfare facility land, the instruction slightly adjusts the location of the community health service station 12 meters northeast from the southwest corner of the current plot 14 to avoid underground pipelines, and adds 500 square meters of green space to the original site. The development and operation agent selects plots 28 and 31, generating a second-type optimization proposal set to change their land use from residential to commercial, and increase the upper limit of the development intensity plot ratio from 3.0 to 4.2, and calculates that the payback period can be shortened from 9 years to 6.5 years. The community uses intelligent execution to generate a third-type compensation demand set: calculating the local integration degree of the spatial syntax axis diagram of the current scheme. and global integration For the area where plots 5, 6, and 7 are located, The values ​​are 0.73, 0.68, and 0.75 respectively, exceeding the total value for the entire plot under this scheme. The ternary mean of the area was 0.55, identifying it as an area requiring improvement. Three unused land parcels (codes U01 and U02) and one inefficient industrial land parcel (code 10) within the area were extracted as candidate locations for supplementary green space, with coordinates (4832.5, 2910.7), (4890.2, 2928.3), (4756.8, 2873.1), and (4945.1, 2960.4), respectively. Constrained K-means clustering was applied, setting the number of clusters c=2 and imposing a minimum distance constraint of 30 meters between candidate sites, resulting in two sets of cluster center coordinates: the first suggested location for supplementary green space (4855.3, 2915.9), and the second suggested location for supplementary green space (4940.8, 2955.2). Simultaneously, the shortest paths from the four candidate sites and surrounding residents in the current plan to the nearest public transportation stops and public service facilities are extracted. The expected daily pedestrian usage for each route is calculated, and routes L12 (connecting the entrance / exit of Plot 5 to the bus stop), L15 (connecting candidate green space U01 to the community health service station), and L23 (crossing Plot 6 along the existing internal community road), with usage exceeding 1200 person-times / day, are marked as existing pedestrian routes that must be protected. The final output of the third type of compensation demand set includes: two suggested locations for additional green space, three pedestrian routes that must be protected, and a statement prohibiting the reduction of existing facility land area.

[0076] S3. Multi-round negotiation and spatial perception closed-loop verification Initial negotiation round , The manager's agent generates a draft, which specifically includes: shifting the floor area ratio by 0.6 from plot 8 to plot 28 and by 0.4 from plot 19 to plot 31; shifting the coordinates of the community health service station from (5012.3, 2941.1) to (5024.3, 2935.7); adopting the proposed locations for additional green space as the cluster centers mentioned above, covering 0.8 hectares and 0.6 hectares respectively; and adding a new pedestrian connecting path from the south side of plot 28 to the bus stop. This draft is then input into the pedestrian accessibility accompanying agent model for spatial perturbation simulation.

[0077] The accompanying proxy model is a regression model based on a graph attention network (GAT). It was previously trained using 3000 historical individual scenarios and their precise reachability matrices, employing a mean squared error loss function during training. The validation set had an average relative error of 3.2%. Specifically, the training set was divided into 80% and the validation set into 20%, resulting in 2400 training samples and 600 validation samples. The model consists of two graph attention network layers, each with four multi-head attention heads. The total training epochs are 120, using the Adam optimizer with an initial learning rate of 0.001 and a learning rate decay strategy of 0.5 every 20 epochs. The mean squared error loss function was used throughout the backpropagation parameter optimization. The model input features include: the uniquely hot coded land use category, standardized development intensity value, centroid coordinates of each plot, and road connection weights between adjacent plots. Additionally, as a spatial editing operation, a facility location coordinate offset vector is also input. In this draft And the vector of road segment weight changes due to road fine-tuning The change shortened the original L12 road segment length to 85% and reduced the obstruction coefficient by 0.2 due to the addition of a connecting passage. The model outputs predicted pedestrian accessibility change heatmap values ​​within 49 milliseconds: the predicted pedestrian accessibility degradation rate of plot 6 reached 9.7%, exceeding the preset tolerance threshold of 5%; the degradation rate of plot 7 was 3.1%, which did not exceed the threshold; and the predicted accessibility of plot 28 improved by 2.3%.

[0078] After receiving the predicted heat values, the community's agent performed a walkability calculation and found that plot 6 had exceeded the degradation threshold. Furthermore, the degradation was caused by the second suggested location of the compensatory green space (4940.8, 2955.2) precisely cutting through the original pedestrian path segment L23, which required protection, thus interrupting the L23 path. Therefore, the community's agent outputs the following hard constraint for spatial accessibility: accessibility degradation. The set of substandard land parcels {Parcel 6} must maintain connectivity. The key walking path segment set {L23}. The development and operation agent simultaneously calculated the 15-year IRRs of plots 28 and 31 under the draft to be 14.3% and 13.8% respectively, which are still higher than the threshold. However, due to the increased cost of public welfare construction requirements after the plot ratio transfer, the net profit impact of plot 31 is -6.2%, which exceeds the tolerance threshold of 5%. Therefore, an objection response with attached numerical adjustment range is issued, requiring that the upper limit of development intensity of plot 31 be increased by 0.3 to offset the cost.

[0079] At this point, t=0<5, indicating a hard constraint and an opposing response. t is incremented to 1, and a bounded spatial replanning is performed. The PSO algorithm is used, with a particle swarm size of 30 and a maximum iteration of 80 generations. A new solution is found under the constraints: the green space location is slightly adjusted 21 meters northeast, a 6-meter-wide pedestrian underpass node is added to L23, and the maximum floor area ratio of plot 31 is increased from 4.2 to 4.38. After another preview of the new draft, the degradation rate of plot 6 decreases to 4.1%, and the net profit impact of plot 31 decreases to -4.6%, both meeting the tolerance threshold. The final accessibility preview check passes, achieving a valid consensus on spatial semantic consistency. If a solution fails to reach a consensus after 5 rounds of negotiation, it is marked as an irreconcilable zone and frozen with the flag "FZ_10_001".

[0080] S4. Gene Editing and Compliance Correction For individuals who reach a consensus on a plan, the consensus-based modification operations are streamlined into a gene-editing operator. Taking plot 28 as an example, the land use category code uses a weighted roulette wheel: its candidate categories include commercial, mixed commercial / residential, and residential. Based on the priority weight of the administrator, the three categories are calculated. They are respectively , , After normalization, the probabilities are 0.6 for commercial, 0.4 for mixed commercial and residential, and 0 for residential. The sampled result is commercial. The continuous development intensity variable is based on the current value of 3.0 with Cauchy perturbation. The standard deviation of the 10th generation is 0.125. A perturbation value of +0.37 is extracted, and after projection and constraint within the interval [1.0, 4.2], a value of 3.37 is obtained. Similar operations are performed on other plots to generate new individuals. For individuals in irreconcilable zones, their original scheme is retained, and a freeze mark is added.

[0081] After merging, a total of 120 candidate offspring are obtained. The compliance correction operator is invoked to perform minimum Euclidean distance mapping projection on individuals that exceed the upper limit of plot ratio, the lower limit of green space ratio, or the bottom line of public welfare land area, so as to pull all individuals back into the feasible region.

[0082] S5. Population Merging, Non-Dominant Ranking, and File Updates The 120 candidate offspring are merged with the current population of 100 individuals to form a total of 220 individuals. The objective dimension M=3, representing the economic objectives. Annual tax revenue of the area (in ten thousand yuan, maximized), social goals : Five-minute living circle coverage (maximum), environmental goals The reciprocal (maximization) of per capita carbon emission equivalent. Reference points are generated on the unit hyperplane using the Das and Dennis methods, taking the degree of each dimension. , co-generated Reference points. After non-dominated sorting, the adaptive crowding distance for each individual in each non-dominated layer is calculated. Two individuals from the first non-dominated layer are selected. The maximum and minimum values ​​at the current layer in the dimension are 45230 and 38760, respectively. The distance between adjacent individuals in this dimension is Then its contribution to the crowding in this dimension is After aggregation, 100 elites are selected based on sorting hierarchy and crowding to form a new generation of parent population. When the external archive is updated, the non-dominated solutions of the new generation are merged with the original archive, and dominated individuals are removed, maintaining the archive size at approximately 80-95 solutions.

[0083] S6. Convergence Judgment and Output In the 10th generation, the over-volume index of external archives was calculated. Reference point Let's set the maximum value of all observations across the three target dimensions as multiplied by 1.1, i.e. Use window width Calculate the moving average and its rate of change Because the previous 9th generation had a change rate of The 10th generation is lower than the first time However, it is not yet continuous. By the 13th generation, the change rates for the four consecutive generations were as follows: , , and , satisfying continuous The iteration terminates when the value falls below a threshold. The output external archive contains 87 non-dominated options in the Pareto front, which decision-makers can choose from based on their preferences.

[0084] To verify the beneficial effects of the technical solution of this invention, a comparative experiment was designed based on the urban renewal area data described in the aforementioned specific embodiments. The experimental group adopted the full-process method of this invention, including multi-agent negotiation optimization with a built-in spatial perception closed-loop verification mechanism; the control group adopted the conventional multi-agent negotiation optimization method without this closed-loop verification, that is, omitting the accompanying agent model pre-playing, walking accessibility calculation, bounded space replanning and final accessibility verification in step S3, and directly using the initial conflict resolution draft of the manager agent as the consensus result to enter the subsequent evolutionary operation, with the remaining steps and parameter settings consistent with the experimental group.

[0085] The experimental area comprises 35 plots of land, with the optimization objective being economic goals. (Maximize annual tax revenue in the region), social goals (Maximize the coverage of the five-minute living circle), environmental goals (Maximizing the reciprocal of per capita carbon emission equivalent), population size 100, maximum number of iterations 50, each group runs independently 30 times. The graph attention network accompanying agent model is pre-trained with 3000 historical solutions and their exact reachability matrices, with an average relative error of 3.2% on the validation set. The experimental hardware environment is an Intel Xeon Gold 6248R processor and 128GB of memory. The performance evaluation metrics are as follows: metric 1 is the Pareto front hypervolume of the final external file, with the reference point being the worst value of each objective dimension multiplied by 1.1; metric 2 is the convergence generation, with the moving average rate of change of the hypervolume being less than 10 for four consecutive generations. -4 The criteria for determining convergence are as follows: Indicator 3 is the proportion of pseudo-consistent schemes in the output Pareto scheme set. A pseudo-consistent scheme is defined as a scheme in which, after being substituted into the high-precision road network model, it is found that the connectivity of at least one pedestrian path segment that must be protected is cut off due to the improper location of compensatory green space, and the pedestrian accessibility degradation of the relevant plots exceeds the preset 5% tolerance threshold; Indicator 4 is the average satisfaction of the community using the intelligent agent with the final scheme set, which is the weighted average of the accessibility satisfaction rate and the open space satisfaction rate of each plot.

[0086] The experimental results are shown in Table 1 below: Table 1

[0087] As shown in the table, the final Pareto front hypervolume index of the method of this invention is improved by about 2.2% compared with the control group, indicating that the obtained solution set has improved in terms of convergence and distribution. The average convergence generation is reduced by 4.3 generations, indicating that the spatially aware closed-loop verification mechanism can identify and eliminate invalid adjustments at the spatial semantic level in a timely manner during the negotiation process, reduce the repeated generation of invalid schemes, and thus accelerate the population's approach to the true Pareto front. Regarding the key indicator of the proportion of pseudo-consistent schemes, the method of this invention did not detect any pseudo-consistent schemes in the Pareto scheme set obtained from 30 independent runs, while the proportion of pseudo-consistent schemes in the control group reached 12.3%. Although such schemes achieve formal consistency at the multi-agent negotiation level, they may lead to a contradictory situation at the actual spatial level, such as the increase of compensatory green space but the severing of key walking paths, resulting in substantial degradation of community accessibility, thereby reducing the feasibility of the scheme. Furthermore, the method of this invention increases the average satisfaction of community agents by approximately 10.8 percentage points, indicating that it can further translate the micro-level living environment demands raised by community agents into actual improvements in the physical attributes of solutions such as land use layout and pedestrian connectivity, rather than merely remaining at a formal consensus.

[0088] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A multi-agent, multi-objective optimization method for urban renewal land use management, characterized in that, Includes the following steps: S1: Collect multi-source spatiotemporal data and planning control requirements of the target urban renewal area, and construct a multi-agent system including manager intelligent agents, development and operation intelligent agents, community user intelligent agents, and a multi-objective land use control scheme population; S2: Each agent performs parallel evaluation of each scheme in the multi-objective land use control scheme population to obtain the multi-agent evaluation state diagram corresponding to each scheme; S3: Based on the multi-agent evaluation state diagram, each agent generates a first type of structural modification instruction set, a second type of optimization proposal set, and a third type of compensation requirement set. The first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation requirement set are combined to obtain a problem-oriented differentiated modification operation pool. S4: For each individual scheme in the scheme population, initiate multi-round structured interaction between multiple agents based on a preset negotiation protocol based on the differentiated modification operation pool, and perform spatial perception closed-loop verification during the multi-round structured interaction. When a valid consensus on spatial semantic consistency is reached, a negotiated and agreed modification operation set is obtained. When a valid consensus on spatial semantic consistency is not reached, mark the irreconcilable plots and attach a temporary development intensity freeze mark to the corresponding individual scheme. S5: Transform the consensus-based modification operation set into structured editing operations that can be directly applied to the individual gene coding of the scheme through preset mapping rules; S6: Using structured editing operations combined with a differentiated mutation strategy based on land parcel type, targeted mutation is performed on individual schemes that have reached a valid consensus on spatial semantic consistency, and compliance correction is performed on the new individual schemes obtained from the targeted mutations to obtain new individual schemes after compliance correction. S7: Merge the new scheme individuals with the scheme individuals with the additional temporary development intensity freeze mark to obtain a candidate offspring scheme set. Iteratively calculate the multi-objective land use control scheme population through the candidate offspring scheme set to obtain the set of all Pareto optimal solutions of the multi-objective land use control scheme population.

2. The multi-agent urban renewal land use management multi-objective optimization method according to claim 1, characterized in that, The process involves using various agents to perform parallel evaluations of individual schemes within a population of multi-objective land use management schemes, including the following steps: S2.1: Use managerial agents, development and operation agents, and community user agents to evaluate each individual scheme in the current multi-objective land use control scheme population in parallel, and identify problem plots in each scheme individual that do not meet the expectations of the corresponding agent according to the satisfaction threshold of each agent, and obtain a set of problem plots; S2.2: Each agent uses a differentiating label to label the set of problem plots, obtaining a differentiating label matrix corresponding to each individual solution. The rows of the differentiating label matrix correspond to the plot numbers in the solutions, and the columns correspond to the manager agent, development and operation agent, and community user agent, respectively. Matrix element a... ij For a binary value, a ij =1 indicates that plot i did not reach the satisfaction threshold under the evaluation of agent j, a ij =0 indicates that the satisfaction threshold has been reached; S2.3: Overlay the differential label matrix onto the spatial graph of the corresponding scheme individual to obtain the multi-agent evaluation state graph for each scheme.

3. The multi-agent urban renewal land use management multi-objective optimization method according to claim 2, characterized in that, The evaluation of individual schemes within the current multi-objective land use management scheme population using community-based intelligent agents includes the following steps: A weighted directed graph was constructed by extracting the centerlines of individual road areas from multi-source spatiotemporal data for each scheme. ,in This is a set of nodes representing road intersections and site entrances / exits. For road segment collection, The segment weight is defined as the product of segment length and pedestrian obstruction coefficient; Each plot of land Walkability index The expression is as follows: in, This refers to the set of public service facility types that community-using agents focus on. For facility type The weighting coefficients, For distance attenuation factor, For the plot of land To the latest type The shortest path distance to the facility; Based on the walkability index of plot i and per capita open space area The weighted sum is used to calculate the local utility value of plot i. Local utility value Land parcels that fall below the preset minimum satisfaction threshold are identified as problem parcels.

4. The multi-agent urban renewal land use management multi-objective optimization method according to claim 2, characterized in that, The process involves generating a differentiated modification operation pool based on the current population evaluation results for each agent, including the following steps: By utilizing the managerial intelligent agent based on the multi-agent evaluation state diagram and the current development intensity gradient of the area where the problem plot is located, a first type of structural modification instruction set is generated, which includes instructions for cross-plot transfer of floor area ratio, instructions for fine-tuning of the boundary of public facility land, and instructions for adjusting the scale of supporting facilities. The development and operation agent generates a second set of optimization proposals, including instructions for changing land use and instructions for adjusting the upper limit of development intensity, based on the expected internal rate of return and the capital recovery period of the problem plots. Based on the analysis results of the walkability and spatial syntactic axis diagram of the problem plots, the community uses an intelligent agent to generate a third set of compensation needs, which includes the preliminary spatial location of the proposed green space and the existing pedestrian path segments that must be protected. The first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation requirement set are combined to obtain a differentiated modification operation pool; The process involves using a managerial agent to generate a first-class set of structural modification instructions based on a multi-agent evaluation state diagram and the current development intensity gradient of the area where the problem plot is located. This includes the following steps: The manager agent extracts the set of marked problem plots from the multi-agent evaluation state graph. ; Set of problem plots Current development intensity distribution data for individual projects under the current scheme Perform spatial overlay analysis to calculate the plot ratio of the problem plot i. Compared with the average target value of the area deviation The expression is as follows: right For each plot of land, the transfer amount is calculated using the excess capacity that can be reduced from plot i, and the receiving plot is determined based on the maximum development intensity gradient of adjacent high-value areas, generating a plot ratio transfer instruction across plots. Set up problem plots The service radius coverage data of public facilities is buffered and overlaid to identify public facilities affected by insufficient coverage. The boundary of the facility land is fine-tuned according to the connectable direction of adjacent unused land, and a fine-tuning instruction for the boundary of public facility land is generated. The allocation scale is recalculated based on the population gap served by the facilities, resulting in an instruction to adjust the allocation scale; The first type of structural modification instruction set is composed of instructions for transferring plot ratio across plots, instructions for fine-tuning the boundaries of public facility land, and instructions for adjusting the scale of supporting facilities. The development and operation agent generates a second set of optimization proposals based on the expected internal rate of return and payback period of the problem plots, including the following steps: The development and operation agent extracts the marked problem plots, their current land use, and the upper limit of development intensity from the multi-agent evaluation state map; For each problem plot Calculate the expected internal rate of return for conversion to candidate high-value types such as commercial or residential. and the return period of funds Filter land use change options that meet the requirements of expected internal rate of return being greater than or equal to the preset expected internal rate of return threshold and capital recovery period being less than or equal to the preset maximum capital recovery period, and generate a second set of optimized proposals that includes the target land use and the upper limit of the suggested development intensity. Based on the walkability and spatial syntactic axis diagram analysis results of the problem plots, a third set of compensation needs is generated using a community-based intelligent agent, including the following steps: The community uses agents to calculate the local integration degree of the spatial syntactic axis graph of the current scheme for individual areas. and global integration and the difference between the two Areas with high topological separation and poor walkable accessibility were identified as areas requiring improvement. For including A single-axis model, global integration As axis global average depth The standardized reciprocal is expressed as follows: in, The average depth is expressed as follows: in, As axis To the axis The shortest topological distance, Standardized parameters for the diamond pattern; Local integration Limit the average depth to the axis topological radius Within, local integration The expression is as follows: For each area to be improved, unused and inefficiently used land parcels within and around it are extracted as candidate locations for supplementary green spaces. A constrained clustering algorithm is then used to categorize these candidate locations according to... The values ​​are grouped, and the geometric center of each group is taken as the preliminary spatial location for the proposed supplementary green space; The community uses the smart agent to extract path segments whose usage frequency exceeds the preset minimum threshold based on the shortest path from each plot to the nearest public transportation station and public service facility in the current scheme. The path segments located inside the area to be improved or connecting the area to be improved with the core public service facilities are marked as existing pedestrian path segments that must be protected. The preliminary spatial location of the proposed additional green space and the existing pedestrian paths that must be protected are used to form the third set of compensation needs.

5. The multi-agent urban renewal land use management multi-objective optimization method according to claim 1, characterized in that, Based on the differentiated modification operation pool, multi-round structured interaction between multiple agents is initiated based on a preset negotiation protocol, and spatial perception closed-loop verification is performed during the multi-round structured interaction process, including the following steps: S3.1: Initialize the negotiation round counter to 0 and preset the maximum negotiation round. ; S3.2: Utilizing the differentiated modification operation pool, the manager's intelligent agent integrates the suggestions from the first type of structural modification instruction set, the second type of optimization proposal set, and the third type of compensation demand set to supplement the preliminary spatial location of the green space and generate a comprehensive conflict resolution draft. S3.3: Using a pre-built walkability accompanying agent model, a spatial perturbation simulation of the comprehensive conflict resolution draft is performed to obtain the predicted walkability change thermodynamic values ​​for each relevant plot; S3.4: Input the predicted heat values ​​of changes in walkable accessibility and the existing walkable route segments that must be protected from the third type of compensation demand set into the community user agent, and use the community user agent to calculate the walkable accessibility degradation rate based on the walkable accessibility of each plot before and after the adjustment. The expression is as follows: in, This represents the predicted walkability index of plot i after adjustments based on the comprehensive conflict resolution draft for the corresponding scheme. This represents the walkability index of plot i before the individual adjustment of the corresponding scheme; If the rate of degradation of walkability ,in, To set a preset tolerance threshold, add plot i to the set of plots with excessive accessibility degradation; S3.5: Traverse the set of existing walking path segments that must be protected. Each path segment Calculate path segments Connectivity indicators in the road network after adjustments to the comprehensive conflict resolution draft. , ,like Then The set of critical walking path segments that must maintain connectivity is included, and the set of plots with excessive accessibility degradation and the set of critical walking path segments are identified as hard constraints on spatial accessibility. S3.6: Utilize the development and operation intelligent agent to calculate the expected net profit change rate of the land parcels affected by the modification operation before and after the implementation of the comprehensive conflict resolution draft, and generate an objection response when the absolute value of the expected net profit change rate exceeds the preset profit impact tolerance threshold; S3.7: If the current negotiation round is less than the preset maximum negotiation round, and there are hard constraints on spatial accessibility or objections, combine the accessibility degradation exceeding the standard, the key pedestrian route segments, and the locations of facilities requiring adjustment in the objections to obtain the affected element set. Determine the overall minimum outer rectangle based on the current land use boundaries of each plot and facility in the affected element set, and extend the overall minimum outer rectangle outward by a preset buffer distance. Thus, a bounded search domain is obtained; S3.8: Using the manager agent combined with the particle swarm optimization algorithm with hard constraints, perform bounded space replanning in the bounded search domain to obtain an updated comprehensive conflict resolution draft, and increase the counter value of the negotiation round. Then return the updated comprehensive conflict resolution draft to S3.3 to continue the spatial perturbation pre-play. S3.9: If the current negotiation round is less than the preset maximum negotiation round, and the spatial accessibility hard constraint and the opposition response are both empty, the community uses an intelligent agent to perform a final accessibility pre-run check on the current comprehensive conflict resolution draft. After confirming that the predicted walkability degradation of each plot has not exceeded the preset tolerance threshold, it is determined that the corresponding scheme individuals have reached a valid consensus on spatial semantic consistency, and the negotiated consensus modification operation set is output; otherwise, the relevant plot groups of the scheme individuals are marked as current irreconcilable plots, and a temporary development intensity freeze mark is applied to the scheme individuals.

6. The multi-agent urban renewal land use management multi-objective optimization method according to claim 5, characterized in that, The walkability-associated surrogate model employs a graph attention network regression model. It is used to perform spatial perturbation simulations of the comprehensive conflict resolution draft, obtaining predicted walkability change heatmap values ​​for each relevant plot. The process includes the following steps: Training samples are generated using historical iterative data and pre-executed Monte Carlo spatial perturbations. Each training sample includes a land use control scheme before the spatial perturbation, a land use control scheme after the spatial perturbation, and the corresponding precise walkability matrices before and after the spatial perturbation. The matrix elements of the precise walking reachability matrix This represents the walking time from plot i to facility type k; Based on the changes in candidate facility locations before and after spatial disturbances in each training sample, as well as the addition, deletion, or geometric adjustment of roads, the corresponding facility site selection coordinate offset vector is calculated. and local road segment weight change vector ; Calculate the thermodynamic values ​​of the changes in precise walkable accessibility for each plot of land based on the precise walkable accessibility matrix before and after the spatial disturbance. ; The land use category code, standardized development intensity, geometric center coordinates of each plot, and road connection topology between adjacent plots in the land use control plan before spatial disturbance are constructed as node feature vectors. Using facility site selection coordinate offset vector Update the node feature vector The node coordinate features corresponding to the affected facilities are used to change the local road segment weight vector. Update the edge weight attributes corresponding to the road connection topology between adjacent plots to obtain the graph attention network input; The graph attention network is input into the first layer of the graph attention network regression model, and the attention coefficients of neighboring plots j are calculated through a multi-head attention mechanism. The expression is as follows: in, This represents the activation function of the linear unit with leakage correction. This represents a trainable attention weight vector. , They respectively represent the use of shared linear transformation matrices For node feature vectors and The node feature representation obtained by performing a linear transformation. This represents a vector concatenation operation; The attention coefficients are Softmax normalized, and the normalized attention coefficients are then used to weight and aggregate the features of neighboring nodes to obtain the updated node embedding feature vector. The expression is as follows: in, For the number of attention heads, For the plot of land The set of neighboring nodes, and The first The weight matrix and normalized attention coefficients of each attention head. For ELU activation function, The number of layers in the attention network propagation; The nodes, after propagation through a predetermined number of layers, are embedded into feature vectors and input into a linear output layer to obtain the predicted walkable accessibility heatmap values ​​for each relevant plot. According to the predicted changes in walkable accessibility thermodynamic values Thermodynamic values ​​of precise walkability accessibility The mean squared error loss function L is calculated as follows: Where N represents the number of land parcel training samples participating in the loss function calculation, and i represents the land parcel training sample number participating in the loss function calculation; Backpropagation optimization is performed based on the loss function L to obtain the trained walking accessibility companion agent model; Based on the changes in the candidate locations of facilities and the addition, deletion, or geometrical adjustment of roads caused by the comprehensive conflict resolution draft relative to the individual corresponding schemes, calculate the facility site selection coordinate offset vector. and local road segment weight change vector ; Construct node feature vectors based on the corresponding schemes, and utilize facility site selection coordinate offset vectors. Update the node coordinate features corresponding to the affected facilities, and use the local road segment weight change vector. Update the edge weight attributes of the corresponding road connections to obtain the graph attention network input corresponding to the comprehensive conflict resolution draft; input the graph attention network input into the trained walking accessibility adjoint agent model to obtain the predicted walking accessibility change heat value corresponding to the comprehensive conflict resolution draft.

7. The multi-agent urban renewal land use management multi-objective optimization method according to claim 5, characterized in that, Bounded space replanning is performed within a bounded search domain using a managerial intelligent agent combined with a particle swarm optimization algorithm with hard constraints, including the following steps: Particle swarm optimization is constructed using the centroid coordinates, land boundary control point coordinates, and pedestrian access node coordinates of the affected plots and adjacent facilities within the bounded search domain as decision variables. The velocity and position of each particle are updated using inertia weights, individual historical best positions, and global best positions. The spatial reachability hard constraint violation is then calculated based on the spatial adjustment scheme corresponding to each particle after the update. , legally binding constraints violated And the number of violations of connectivity constraints of existing pedestrian path segments that must be protected. ,include: (1) Based on the hard constraint of spatial accessibility, for any plot i in the set of plots with excessive accessibility degradation, the spatial adjustment scheme determined by the corresponding particle is input into the walkable accessibility surrogate model to obtain the predicted walkable accessibility change thermodynamic value of plot i. Based on the pedestrian accessibility index of plot i before adjustment and predictive thermodynamic values ​​of changes in walkable accessibility Calculate the predicted walkability index of plot i under the spatial adjustment scheme. The expression is as follows: Will The lower limit of the preset tolerance threshold of plot i Compare and calculate the spatial accessibility hard constraint violation of plot i. The spatial accessibility hard constraint violation amounts of each plot in the set D of plots with excessive accessibility degradation are summed to obtain the spatial accessibility hard constraint violation amount. ; (2) Based on the legally binding constraints corresponding to the upper limit of plot ratio and the lower limit of green space ratio, the plot ratio of each plot shall be determined. With floor area ratio limit Compare and determine the green space ratio of each plot. With the lower limit of green space ratio Compare and calculate the floor area ratio violation amount for plot i respectively. and green space ratio violations The sum of the violations of plot ratio and green space ratio for each plot is used to obtain the amount of violation of legally binding constraints. ; (3) Based on the set of existing pedestrian route segments that must be protected Verify the connectivity of each existing pedestrian path segment that must be protected in the road network under the corresponding spatial adjustment scheme of the particle, and adjust the path segments. The connectivity state in the adjusted road network is represented by a binary indicator. , where if path segment Maintain connectivity. If path segment Not connected Based on the binary indicator quantities corresponding to each path segment Calculate the adjusted path connectivity index The expression is as follows: The corresponding path segments The set of existing pedestrian path segments that must be protected is identified as a violation of connectivity constraints. The connectivity constraint violations of each path segment are summed to obtain the connectivity violation amount of the current pedestrian path segments that must be protected. ; Based on the violation of spatial accessibility hard constraints , legally binding constraints violated And the number of violations of connectivity constraints of existing pedestrian path segments that must be protected. The total penalty term is calculated as follows: in, , , The first positive value penalty coefficient, the second positive value penalty coefficient, and the third positive value penalty coefficient are preset; The weighted sum is calculated by summing the absolute values ​​of the accessibility degradation degree corresponding to the community's use of intelligent agents and the expected net profit change rate corresponding to the development and operation of intelligent agents, and then the weighted sum is multiplied by the total penalty term. The fitness values ​​of each particle are obtained by adding them together. The individual historical best position and global best position are updated based on the fitness value of each particle, and the particle velocity and position updates and fitness value calculations are repeated until the preset particle swarm iteration termination condition is met. The optimal particle is determined from the particles that meet the hard constraints of spatial accessibility, legal rigid constraints, and the connectivity constraints of existing pedestrian path segments that must be protected. The centroid coordinates, land boundary control point coordinates, and pedestrian connection node coordinates of the facilities after the bounded space replanning are determined based on the spatial control point coordinates corresponding to the optimal particle, forming an updated comprehensive conflict resolution draft.

8. The multi-agent urban renewal land use management multi-objective optimization method according to claim 1, characterized in that, The consensus-based modification operation set is transformed into structured editing operations that can be directly applied to the individual gene coding of the scheme through preset mapping rules, including: Map the target plot number and new land use category code of the land use change order to the targeted assignment operation of the corresponding land use location; Map the transfer-out plot number, receiving plot number, and transfer amount of the plot ratio transfer instruction to a decrement operation of the continuous development intensity variable of the transfer-out plot and an increment operation of the continuous development intensity variable of the receiving plot, respectively. The instructions for fine-tuning the boundaries of public facility land and adjusting the scale of supporting facilities are mapped to displacement operations of the geometric control point coordinates of the affected plots and assignment operations of the supporting facility scale parameters, respectively. The instructions for adding green space and pedestrian connecting passages are mapped to assignment operations of the land use category codes of the corresponding plots and writing operations of the connecting passage node coordinates to the spatial topology coding layer.

9. The multi-agent urban renewal land use management multi-objective optimization method according to claim 1, characterized in that, The targeted mutations and compliance corrections described in S6 include the following steps: Based on the priority weights of the manager agents corresponding to the candidate land use categories, a weighted roulette wheel is used to determine the target land use category code from the set of candidate land use categories for the corresponding plot, and the target land use category code is assigned to the land use location of the corresponding plot. The priority weights of the manager agents are then used to determine the target land use category code. The expression is as follows: in, Indicating land price potential index, , These represent the first weighting coefficient and the second weighting coefficient, respectively. The difference between the maximum achievable floor area ratio (FAR) and the current FAR for a candidate land use category is expressed as: in, This represents the maximum floor area ratio corresponding to candidate land use category j. This indicates the current average plot ratio of the corresponding land parcel; Determine the standard deviation of the adaptive decay perturbation based on the current iteration number. Wherein, the standard deviation of the disturbance corresponding to generation t. The expression is as follows: in, This represents the standard deviation of the initial disturbance. The decay coefficient represents the standard deviation of the disturbance, and t represents the current iteration number; From a mean of zero and having the aforementioned perturbation standard deviation The perturbation values ​​are extracted from the Cauchy distribution and superimposed onto the continuous development intensity variable of the corresponding plot to obtain a new scheme individual with directional variation. The compliance correction for individual new schemes includes: sequentially projecting the upper limit of floor area ratio, the lower limit of green space ratio, and the bottom line of public facility land area, and using the minimum Euclidean distance mapping to the boundary of the feasible region to map variables exceeding the legal rigid constraints to the corresponding feasible region; merging the new schemes that have completed the compliance correction with the schemes marked with the temporary development intensity freeze mark to obtain a set of candidate sub-schemes, and performing compliance correction verification on the set of candidate sub-schemes, forcibly mapping schemes that exceed the upper limit of floor area ratio, the lower limit of green space ratio, or the bottom line of public facility land area to the feasible region to obtain new schemes after compliance correction.

10. The multi-agent urban renewal land use management multi-objective optimization method according to claim 1, characterized in that, The population of multi-objective land use control schemes is iteratively calculated using a set of candidate offspring schemes, including the following steps: S7.1: Initialize an external archive independent of the multi-objective land use control scheme population, the external archive being used to store non-dominated scheme individuals obtained during the iteration process; S7.2: Merge the candidate offspring scheme set with the current multi-objective land use control scheme population to obtain an expanded population; S7.3: Perform non-dominated ranking on the individuals of the schemes in the extended population, and perform elite selection based on the non-dominated ranking results to obtain the next generation of multi-objective land use control scheme population; S7.4: Merge the non-dominant scheme individuals in the extended population with the scheme individuals in the external archive, and use the Pareto dominance relation to remove the dominated scheme individuals to obtain the updated external archive. S7.5: Calculate the hypervolume index of the Pareto front using the updated external archive, and calculate the multigeneration moving average rate of change based on the hypervolume index over several generations. S7.6: Convergence is determined based on the multi-generation moving average change rate and the current iteration number. If the preset convergence condition is not met and the current iteration number has not reached the preset maximum iteration number, the next generation of multi-objective land use control scheme population obtained in S7.3 is used as the current multi-objective land use control scheme population, and the process returns to step S2 to continue iterating. The temporary development intensity freeze flag is used as a constraint for the manager agent to generate the first type of structural modification instruction set, and the generation of adjustment instructions for irreconcilable land parcels in the scheme is prohibited. If the preset convergence condition is met or the current iteration number reaches the preset maximum iteration number, the iteration is terminated, and the set of all Pareto optimal solutions of the multi-objective land use control scheme population is obtained based on the updated external archive. The step of performing non-dominated ranking of individuals in the expanded population and selecting elites based on the non-dominated ranking results includes the following steps: On the unit hyperplane of the M-dimensional target space, reference points are generated uniformly according to the preset division of each target dimension. Based on the reference points, a fast non-dominated sort is performed on the extended population to obtain multiple non-dominated layers. For any scheme individual in the extended population, if there is no other scheme individual that is not inferior to the scheme individual in all optimization objectives and is superior to the scheme individual in at least one optimization objective, then the scheme individual is determined as a non-dominated scheme individual. For each non-dominated layer, the distance between adjacent individuals of each scheme is calculated in each target dimension. The distance between adjacent individuals is normalized based on the maximum and minimum values ​​of the current non-dominated layer in each target dimension. The normalized results corresponding to each target dimension are then aggregated to obtain the adaptive crowding distance of each scheme individual. The expression is as follows: Where M is the objective dimension and k is the scheme number. For individual solutions In each objective The distance between adjacent individuals on the surface and The current non-dominated layer is in the th... Maximum and minimum values ​​on the target To prevent division by a constant; The scheme individuals in the extended population are sorted in order of non-dominated hierarchy from low to high and adaptive congestion distance within the same non-dominated hierarchy from large to small. The scheme individuals to be selected are compared with the scheme individuals in the external archive in a joint Pareto dominance comparison. Scheme individuals that are Pareto dominated by any scheme individual in the external archive are removed. Elite individuals are selected from the remaining scheme individuals based on non-dominated hierarchy and adaptive congestion distance to form the next generation of multi-objective land use management scheme population. Calculating the multigenerational moving average rate of change of the hypervolume index using the updated external archives includes the following steps: Based on the updated external archives, the hypervolume index HV(g) of the Pareto front in the g-th generation external archives is determined. Specifically, the hypervolume reference point vector is calculated by taking the worst values ​​of each generation's population across each target dimension. The hypervolume reference point vector, along with the Lebesgue measure of the dominance region defined by the scheme individuals in the target space within the external archives, is then used to determine the hypervolume index HV(g). Based on the hypervolume index HV(g) and the preset moving window width... Calculate the moving average of the hypervolume index for generation g. The expression is as follows: The first The moving average of the hypervolume index of the generation With the The moving average of the hypervolume index of the generation Comparison, calculation of the multi-generation moving average rate of change of the hypervolume index The expression is as follows: 。

Images