Multi-agent-based translation and moving decision-making method for residents in one-storey area of old city

Abstract

The invention provides an old city bungalow area resident translation and moving decision-making method based on multiple agents. The method comprises the steps of collecting plot information of an old city bungalow area and constructing a data set, establishing a resident agent and a developer agent and setting behavior rules and action spaces of the resident agent and the developer agent, setting a learning method of the developer agent as multi-target hierarchical reinforcement learning, setting multi-target hierarchical rules respectively, and performing multi-target hierarchical reinforcement learning. And setting a reward mechanism of reinforcement learning, running all agents and performing data simulation, dynamically adjusting a relocation strategy in the simulation process until a convergence condition is met, outputting data and forming a result list. Compared with the application of the prior art, the multi-agent-based translation and moving decision-making method for residents in the one-storey area of the old city innovatively introduces the agents into a new scene of stock development, helps developers to make a more scientific moving decision, and can obtain a win-win balance result of the developers and the residents.

Description

Technical Field

[0001] The present invention relates to the field of urban renewal simulation, and in particular to a multi-agent-based decision-making method for the relocation of residents in bungalow areas in old cities. Background Art

[0002] In the process of contemporary urban renewal, old town bungalow areas are generally trying a gradual renewal method, where residents need to gradually move out and move back. However, unreasonable relocation plans will cause problems such as the original residents in the bungalow areas sharing courtyards with vacant land and mutual constraints, resulting in the spatial regeneration of old town bungalow areas falling into a situation of discrete space utilization, disordered functional layout, and difficulty in resource integration.

[0003] On the one hand, the complex scenario of gradual renewal in the bungalow areas of the old city no longer lends itself to traditional relocation solutions such as financial compensation and relocation. Instead, the remaining scattered residents must be relocated to several courtyards within the site, freeing up more intact courtyards for intensive development. However, in this scenario, developers' relocation decisions are highly subjective, making it difficult to identify win-win relocation decisions for both developers and residents among the vast number of options. It is also difficult to individually research the needs of residents in different plots. A more scientific, flexible, and efficient decision-making approach is urgently needed to address this relocation and relocation issue.

[0004] Multi-agent technology has the ability to simulate the complex behaviors and choices of the crowd in a refined manner, and is often used to solve intelligent problems in urban incremental planning (such as urban growth simulation and new functional site selection). However, existing technologies lack feasible solutions for using multi-agent technology to solve the decision-making of relocation plans for residents in old town bungalow areas under complex stock environments. Summary of the Invention

[0005] In view of the shortcomings and defects of the existing technology, the present invention provides a multi-agent-based decision-making method for the relocation of residents in the old town bungalow area, which includes the following steps:

[0006] S1 collects plot information in the old town bungalow area and constructs a plot information dataset; the plot information dataset includes the following data items: plot ID, yard ID where the plot is located, plot orientation O, plot area S, plot location characteristics, and property owners on the plot;

[0007] S2: Create a resident agent for each household in the old town bungalow area. The attributes of the resident agent include the following attributes: resident ID, location ID, and relocation probability P, and assign initial values ​​to each attribute.

[0008] S3: Create a developer agent. The attributes of the developer agent include the following: principal Q0, relocation cost W, expected total development income M, total profit value Q D , and assign initial values ​​to each attribute;

[0009] S4: Setting the behavior rules of the resident agent to relocate and maximize profit, and setting the action space of the resident agent;

[0010] S5: setting the behavior criterion of the developer agent to be profit maximization and setting the action space of the developer agent;

[0011] S6: Personalize modeling of one or more of the resident agents;

[0012] S7: Set the developer agent’s learning method to multi-objective hierarchical reinforcement learning, set multi-objective hierarchical rules, and set the reinforcement learning reward mechanism;

[0013] S8: Run all agents and perform data simulation, dynamically adjusting the relocation strategy during the simulation until convergence conditions are reached;

[0014] S9: Output the data and form a result list.

[0015] Furthermore, the relocation costs include rental compensation and communication costs.

[0016] Furthermore, the action space of the resident agent in step S4 is one of the following possibilities:

[0017] No relocation;

[0018] relocation to optimize daylighting;

[0019] Optimize the relocation of area;

[0020] Optimize relocation;

[0021] Furthermore, the action space of the developer agent in step S5 is one of all possibilities of relocating any one of the resident agents to other plots or courtyards;

[0022] Furthermore, the personalized modeling of one or more resident agents in step S6 is specifically as follows: setting the preference parameters a of different compensation factors in the calculation formula of the relocation probability P according to the preferences of the resident agent. n , n = 1, 2, 3; the calculation formula of the relocation probability P is: P = a1 * lighting score + a2 * area score + a3 * location score;

[0023] Furthermore, the total profit value Q in step S3 D = expected income from the whole hospital development M_ relocation cost W;

[0024] Furthermore, the multi-objective hierarchical rule in step S7 is:

[0025] Goal 1: Maximize total profit Q D and / or maximizing the number of vacant intact compound buildings;

[0026] Goal 2: Minimize the number of relocation rounds.

[0027] Objective 3: Maximize residents’ willingness to relocate.

[0028] Furthermore, the reward mechanism of the reinforcement learning in step S7 is: the profit obtained by the developer after each relocation is calculated as the reward function, the increase in the number of vacant intact courtyards is also quantified as profit growth, and the communication cost of each relocation is introduced, so that the relocation round affects the developer's profit growth.

[0029] Furthermore, the step S8 includes the following sub-steps:

[0030] S91 randomly selects a resident agent to relocate and uses a reinforcement learning algorithm to select the optimal relocation action based on the current state, where the relocation action is selected based on the maximization of the reward function;

[0031] S92 calculates the change in the principal Q0 of this relocation, updates the number of vacant and complete courtyards, and further calculates the total profit value Q of the developer if new vacant and complete courtyards appear. D ;

[0032] S93 updates the status of the resident agent, as well as the remaining principal Q0 and relocation round of the developer agent;

[0033] S94 repeats steps S91 to S93, trying different initial relocation options in each round of iteration, and finding the optimal relocation sequence through the reinforcement learning algorithm, comparing the results of different sequence orientations, and selecting the best solution that maximizes developer profits and increases resident satisfaction.

[0034] Furthermore, the method also includes step S10: developing a visualization tool through simulation and multi-agent system platform, allowing decision makers to adjust parameters and view simulation results in real time, superimposing the above simulation results and spatial data in the system, and completing the effect display of translation and maneuvering multi-scene simulation.

[0035] This method establishes resident and developer agents to achieve localized and scientific layout of relocation plans based on agent technology, helping developers make the most appropriate decisions, promoting the smooth implementation of this type of urban renewal and the intensive use of spatial resources, and quickly resolving the difficult problem of horizontal relocation of residents in the old city's bungalow areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the specific embodiments of the present invention, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art.

[0037] Figure 1 A flowchart of a multi-agent-based decision-making method for relocating residents in old town bungalow areas provided by an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of the structure of a plot dataset constructed in a multi-agent-based decision-making method for relocating residents in old town bungalow areas provided by an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of key steps in a multi-agent-based decision-making method for relocating residents in an old town bungalow area provided by an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of the list structure of the final output results of a multi-agent-based decision-making method for the relocation of residents in old town bungalow areas provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The structure and operation of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the accompanying drawings are provided only for a better understanding of the present invention and should not be construed as limiting the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative effort shall fall within the scope of protection of the present invention.

[0042] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer" and the like indicate positions or locations based on those shown in the accompanying drawings. These terms are intended solely to facilitate and simplify the description of the present invention and are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0043] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to direct connections, indirect connections through an intermediate medium, or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention.

[0044] Example 1

[0045] Combine Figures 1 to 4 This embodiment discloses a multi-agent-based decision-making method for the relocation of residents in the old town bungalow area. The method includes the following steps: Figure 1 As shown:

[0046] S1 collects plot information in the old town bungalow area and constructs a plot information dataset; the plot information dataset includes the following data items: plot ID, yard ID where the plot is located, plot orientation O, plot area S, plot location characteristics, and property owners on the plot;

[0047] S2: Create a resident agent for each household in the old town bungalow area. The attributes of the resident agent include the following attributes: resident ID, location ID, and relocation probability P, and assign initial values ​​to each attribute.

[0048] S3: Create a developer agent. The attributes of the developer agent include the following: principal Q0, relocation cost W, expected total development income M, total profit value Q D , and assign initial values ​​to each attribute;

[0049] S4: Setting the behavior rules of the resident agent to relocate and maximize profit, and setting the action space of the resident agent;

[0050] S5: setting the behavior criterion of the developer agent to be profit maximization and setting the action space of the developer agent;

[0051] S6: Personalize modeling of one or more of the resident agents;

[0052] S7: Set the developer agent’s learning method to multi-objective hierarchical reinforcement learning, set multi-objective hierarchical rules, and set the reinforcement learning reward mechanism;

[0053] S8: Run all agents and perform data simulation, dynamically adjusting the relocation strategy during the simulation until convergence conditions are reached;

[0054] S9: Output the data and form a result list.

[0055] Specifically, the relocation costs include rental compensation and communication costs.

[0056] In step S1, the land parcel information of the old town bungalow area can be collected through the Geographic Information System (GIS). After data collection, the following data processing steps can also be included:

[0057] Data structuring: Construct a data set of plots in the old town bungalow area, store the above information as structured data, and complete the initial assignment to facilitate subsequent analysis and call.

[0058] Preprocessing: Clean the data, handle missing values ​​and outliers, ensure the accuracy and completeness of the data, and integrate multi-dimensional information such as plots, property rights, and area into the coordinates of the courtyard space.

[0059] The orientation of the plot O affects the subsequent residents' choices by affecting the lighting environment, which includes three categories: east-facing N or south-facing S, east-facing E, and west-facing W. For the lighting environment: N=S>E>W, > means that residents are more accepting of this orientation than other orientations.

[0060] The inclusion of the attribute of plot location characteristics can increase the possible dimensions of the action plan and also increase the degree of restoration of the simulation to the actual situation. The plot location characteristics require adding a location evaluation step and assigning expert scores to the location conditions of each plot.

[0061] In this invention, there are two major types of agents for simulation decision-making: resident agents and developer agents.

[0062] Resident Agent: Each household head who is still in the site is modeled as an agent, with attributes including resident ID, location plot ID, where the location plot ID can be connected to the plot data table through this key, and relocation probability.

[0063] Developer Agent: The developer is modeled as an agent with attributes including principal Q0, relocation cost W, expected total development income M, and total profit value Q D .

[0064] Each attribute of the resident agent and the developer agent needs to be assigned an initial value, which is the original value of each attribute during the simulation. For example, according to the actual situation, the initial plot ID of the resident agent and the current relocation probability are both P0; the developer's principal Q0, and the relocation cost are divided into two parts: ① the amount of compensation the developer will make to the residents based on the average rent of individual households in the current block w1; ② the initial communication cost p for each household relocated; the expected added rent w2 (w2>>w1) for the entire courtyard development, etc., and the total profit value Q D .

[0065] Specifically, the action space of the resident agent in step S4 is one of the following possibilities: no relocation; relocation to optimize lighting; relocation to optimize area; relocation to optimize location;

[0066] Among them, non-relocation means that residents choose to stay where they are, do not participate in the relocation, and cannot obtain the relocation compensation benefits given by the developer; relocation to optimize lighting means that residents whose land faces east or west, resulting in poor lighting, choose to relocate to a target land with better lighting conditions and an area no smaller than the existing land; relocation to optimize area means that residents choose to relocate to a larger land; relocation to optimize location means that residents choose to relocate to a land with better location conditions.

[0067] The rewards residents receive for taking action are converted from compensation from developers.

[0068] Specifically, the action space of the developer agent in step S5 is one of all possibilities of relocating any one of the resident agents to other plots or courtyards;

[0069] The developer agent aims to maximize profits, i.e., to vacate as many entire courtyards as possible at the lowest relocation cost. Therefore, the developer agent needs to explore how to meet the relocation needs of as many residents as possible at the lowest economic cost. Its action space includes multiple possible solutions: which residents to relocate to which plots.

[0070] The developer's reward function is clear, that is, for each completed relocation, the developer's total profit value is Q D Changes occur when a relocation action causes Q D The developer chooses the action that results in the smallest loss or the largest gain, and finally iterates the cycle to present the overall solution.

[0071] Specifically, the personalized modeling of one or more resident agents in step S6 is specifically as follows: setting the preference parameters a of different compensation factors in the calculation formula of the relocation probability P according to the preferences of the resident agent. n , n = 1, 2, 3; the calculation formula of the relocation probability P is: P = a1 * lighting score + a2 * area score + a3 * location score;

[0072] That is, the resident agent is personalized by considering the influence of different compensation factors such as residents’ preferences on the probability of relocation. Residents’ preferences may be expressed as preferences for lighting, area, location, etc. The preference parameter a in the calculation formula is n Represents the weight of a certain preference, through a n Adjustments to the value of can reflect the importance residents place on a particular compensation factor. For example, residents with extremely poor lighting conditions may place higher demands on daylight compensation, while residents who work in remote areas may place higher demands on location and transportation.

[0073] The more the target plot (i.e., compensation plan) offered by the developer satisfies residents' preferences, the greater the probability of residents moving. When a developer's compensation plan makes the probability of residents moving P greater than 70%, residents are willing to move.

[0074] During the developer decision-making process, determining an appropriate global reward method can drive the developer agent to automatically process massive amounts of data, avoid enumeration methods, and more efficiently find the appropriate decision solution. Therefore, step S7 introduces a multi-objective hierarchical reinforcement learning method to enable developers to better hierarchize their decisions and achieve autonomous learning and decision-making.

[0075] Specifically, firstly, a quantitative calculation scheme for developer profit is given. The total profit value Q in step S3 is D = expected income from the whole hospital development M_ relocation cost W;

[0076] Where W = ∑i (communication cost per relocation p + rental loss converted from area loss)

[0077] Rental loss converted from area loss = ΔS*w1, where ΔS is the difference in area before and after residents are relocated, and w1 is the rental compensation per unit area;

[0078] The added value of the vacant yard = Syard*w2-(Syard-Si)*w1; where Syard is the area of ​​the vacant yard, w2 is the rent per unit area of ​​the vacant yard, and S i It is the area of ​​the land where the residents lived before they moved.

[0079] Specifically, the multi-objective hierarchical rule in step S7 is:

[0080] Goal 1: Maximize total profit Q D and / or maximizing the number of vacant intact compound buildings;

[0081] Goal 2: Minimize the number of relocation rounds.

[0082] Objective 3: Maximize residents’ willingness to relocate.

[0083] Specifically, the reinforcement learning reward mechanism in step S7 is as follows: the profit obtained by the developer after each relocation is calculated as the reward function, the increase in the number of vacant intact courtyards is also quantified as profit growth, and the communication cost of each relocation is introduced, so that the relocation round affects the developer's profit growth.

[0084] Specifically, step S8 includes the following sub-steps:

[0085] S91 randomly selects a resident agent to relocate, and uses a reinforcement learning algorithm to select the optimal relocation action based on the current state. The selection of the relocation action is based on maximizing the reward function. The reinforcement learning algorithm is Q-learning or other reinforcement learning algorithms.

[0086] S92 calculates the change in the principal Q0 of this relocation, updates the number of vacant and complete courtyards, and further calculates the total profit value Q of the developer if new vacant and complete courtyards appear. D ;

[0087] S93 updates the status of the resident agent, as well as the remaining principal Q0 and relocation round of the developer agent;

[0088] S94 repeats steps S91 to S93, trying different initial relocation options in each round of iteration, and finding the optimal relocation sequence through the reinforcement learning algorithm, comparing the results of different sequence orientations, and selecting the best solution that maximizes developer profits and increases resident satisfaction.

[0089] After completing this step, you can output the data to form a result list, such as Figure 4 As shown, the following content is output and a result data list is formed:

[0090] Final relocation strategy - the initial and final state of the arrangement of resident agents in the plot;

[0091] A list of resident relocation probabilities to test whether most residents of the site are compensated in a relatively satisfactory manner;

[0092] The developer's economic losses at each step of the relocation process and the total profit after the relocation is completed, monitoring whether the developer's relocation plan achieves the goal of reinforcement learning to maximize the developer's profit;

[0093] Then, the method also includes step S10: developing a visualization tool through simulation and multi-agent system platform, allowing decision makers to adjust parameters and view simulation results in real time, superimposing the above simulation results and spatial data in the system, and completing the effect display of translation and maneuvering multi-scene simulation.

[0094] The following is a specific example to illustrate the specific steps of this method.

[0095] Figure 3 This paper shows the key steps in a multi-agent-based decision-making method for relocating residents in the bungalow area of ​​the old city in a specific embodiment of the present invention. This embodiment is completed through four core processes: data preparation - agent setting - model establishment and operation - interactive display. Taking the Beijing West Headquarters update as a reference, a multi-agent-based decision-making method for relocating residents in the bungalow area of ​​the old city is provided. Figure 1The specific steps of the method are shown.

[0096] Step S1: Collect the plot information of the old town bungalow area and construct the plot information dataset.

[0097] Specifically, the geographic information of the bungalow area in the old city was initially collected through investigation and image analysis via the GIS platform, and then all information was collected by further combining on-site investigation and pre-updated data.

[0098] Because the old town's bungalows are located within overlapping spatial units, the necessary site data for this model includes the property parcel number to be relocated, the courtyard number where the parcel is located, basic attributes such as the parcel's area, orientation, and locational characteristics, and whether the parcel is inhabited (if so, it is marked as 1, and if not, it is marked as 0). The parcel location requires expert scoring during the structuring process to complete a basic evaluation. The value is set from 0 to 1 to facilitate the calculation of the subsequent preference function.

[0099] After the data is structured, Figure 2 As shown, a quantitative list of the complex spatial environment of the old town bungalow area will be formed.

[0100] Assign values ​​to the list based on actual conditions, focusing on property plots, integrating information such as site ownership, physical space, and resident distribution. Further data cleaning is performed to review and address missing and outlier values ​​to ensure data accuracy and completeness for subsequent operations.

[0101] Next, corresponding Figure 1 Steps S2 to S5 in the process complete the multi-agent modeling and action rule setting.

[0102] This step is based on the specific situation of Beijing Xizongbu. First, the modeling and initial states of the resident agent and the developer agent are set, including their attributes, action space, and reward function.

[0103] Resident agent modeling: The attributes of the resident agent include resident ID, location ID, and relocation probability P.

[0104] Resident ID: uniquely identifies each resident, such as "R1, R2".

[0105] Plot ID: Associated with the plot ID in the plot dataset, for example, "P1".

[0106] Relocation Probability: The initial value is set to P0, which represents the resident's willingness to relocate in the initial state. For example, resident R1 is located in plot P3, and the initial relocation probability P0 = 0, indicating that the resident has no probability of relocation in the initial state. Here, the initial P0 = 0 for all residents is agreed.

[0107] Developer Agent Modeling: The attributes of the developer agent include principal Q0, relocation cost W, expected total development income M, and total profit value Q D wait.

[0108] Principal Q0: the developer's initial capital, for example, the developer of Xizongbu presets Q0 = 20 million yuan.

[0109] The relocation cost W includes the developer's loss of self-interest caused by compensating residents under certain conditions, that is, the compensation cost, as well as the communication cost paid in negotiating with residents during the relocation process.

[0110] Compensation cost: calculated based on the average rent w1 of retail tenants in the current block, for example w1 = 8 yuan / square meter / day.

[0111] Communication cost: The communication cost for each household moving is initially set at p = 2,000 yuan.

[0112] Expected income from the development of the entire courtyard M: Since the rent of the entire courtyard development will be much higher than that of individual households, there will be an added value rent w2 for the development of the entire courtyard. In this embodiment, let w2 = 30 yuan / square meter / day.

[0113] Total profit value Q D : The initial value is 0, indicating the developer's initial profit.

[0114] In this example, the developer's initial capital, Q0, equals 20 million yuan. The compensation cost for each relocated household is calculated based on the individual tenant rent, w1. For example, if a 25-square-meter household is provided with a new plot of land of 28 square meters, the rent is converted to R = (28-25) square meters * 365 days * w1 * 3 years. The communication cost is 2,000 yuan, calculated based on the three-month total work hours of the relocation communication staff. If the vacated household vacates a complete courtyard with a total area of ​​200 square meters, the expected profit from the development of the entire courtyard over three years is 200 square meters * 365 days * 3 years * w2 - (200-25) square meters * 365 days * 3 years * w1 = 6,570,000 - 1,533,000 = 5,037,000 yuan.

[0115] After completing the basic value selection, the action plan criteria for the two types of agents are set. Based on the actual situation of Beijing Xizongbu Flats District, the basic rules of action plans are set for residents and developers.

[0116] Resident Agent Action Guidelines: Residents act based on relocation and maximizing profit. Their action space includes the following possibilities:

[0117] No relocation: Residents choose to stay where they are and not participate in the relocation.

[0118] Relocation to optimize lighting: For example, residents move from the west-facing plot P3 with poor lighting to the south-facing plot P with better lighting. 15 .

[0119] Optimize the relocation area: For example, residents move from the smaller plot P3 to the larger plot P 10 .

[0120] Relocation to optimize location: For example, residents move from plot P8, which has inconvenient transportation and leads to a very low location score, to plot P with convenient transportation. 16 .

[0121] Since the reward function of residents is transformed from the compensation of developers. For example, resident R1 moves from plot P3 to P 15 , the compensation is better lighting, and the relocation probability P increases to 0.7, then the residents agree to relocate. On the basis of agreeing to relocate, the residents will compare which relocation plan will give them the highest probability of relocation and make their choice.

[0122] However, the developer has a higher priority in the selection. The developer will choose the relocation plan that will maximize its own profit and make a choice based on the residents’ P value. For example, if P3 is moved to P 15 In this example, the developer only provides lighting compensation, so that the developer does not suffer any economic loss. Therefore, this solution is the best solution.

[0123] Developer Agent Action Guidelines: Developers act based on profit maximization, and their action space includes multiple possible options:

[0124] First, select the resident to relocate: for example, select resident R1 for relocation.

[0125] Then, select the target plot: for example, relocate resident R1 to plot P 15 .

[0126] The developer’s reward function is the total profit value Q D For example, the developer relocates resident R1, frees up plot P3 and courtyard Y3, and the total profit is Q D = Principal Q0 - Relocation Cost W + Expected Overall Development Return M. In this example, refer to the profit calculation process described above to complete the calculation for the actual plot of land, using a three-year profit estimate as the benchmark.

[0127] In order to be closer to the actual situation of Beijing Xizongbu Flats District, this embodiment performs personalized modeling of resident agents and designs resident preference functions, which corresponds to step S6.

[0128] Through the survey, the preference parameter a is adjusted according to the special conditions of the residents. n ,For example:

[0129] Lighting preference: Residents with extremely poor lighting environments have higher requirements for lighting compensation, a1 = 0.5.

[0130] Area preference: Residents with more family members require higher area compensation, a2 = 0.3.

[0131] Location preference: Residents who work in more distant areas have higher requirements for transportation convenience, with a3 = 0.2. This value can be adjusted based on detailed research. In this example, the evaluation is based on the residents' geographical conditions and family conditions.

[0132] The relocation probability function is designed as: P = a1×lighting score + a2×area score + a3×location score.

[0133] For example, resident R2 has a lighting score of 0.8 and an area increase of 0.3, which is equivalent to the highest condition offered by the developer, a score of 1, and a location score of 0.4. The relocation probability P = 0.5×0.8+0.3×1+0.2×0.4=0.78.

[0134] Relocation threshold setting: When the compensation plan offered by the developer makes the probability of residents relocating P greater than 70%, the residents are willing to relocate. For example, if the relocation probability of resident R2 is P = 0.78 > 0.7, then the residents are willing to relocate.

[0135] Corresponding to step S7, this embodiment introduces a multi-objective hierarchical reinforcement learning method to optimize the developer's decision-making process.

[0136] Quantitative calculation scheme for developer profit: Developer's profit Q D The calculation is as follows:

[0137] Total profit value Q D = Expected revenue from the entire development of the hospital M_Relocation cost W; where: W = ∑i (communication cost per relocation p + rental loss converted from area loss)W = i∑ (communication cost per relocation p + rental loss converted from area loss)

[0138] Multi-objective hierarchical rule optimization

[0139] Goal 1: Maximize total revenue Q D .

[0140] Goal 2: Minimize the number of relocation rounds.

[0141] Objective 3: Maximize residents’ willingness to relocate.

[0142] For example, developers relocate residents R1-R 118 , total profit Q D The increase is 5,360,000 yuan / year, the relocation round is 118, and the average willingness value of residents is greater than 85%, which is a better relocation plan.

[0143] Corresponding to step S8, all intelligent agents are run and data simulation is performed, and the relocation strategy is dynamically adjusted during the simulation process until the convergence condition is reached.

[0144] Through multiple rounds of iterative simulation, the relocation strategy is dynamically adjusted until convergence conditions are reached. This condition can be set based on actual needs: for example, the relocation plan is finalized when the developer's rental profits reach 15 times the principal over 10 years after relocation.

[0145] Single-round relocation decision-making: First, simulate a developer randomly selecting a resident to relocate. Using Q-learning or other reinforcement learning algorithms, the optimal relocation action is selected based on the current state. The relocation action is selected based on maximizing the reward function.

[0146] Then calculate the change in the principal of this relocation. Further update the number of vacant whole courtyards. If there are new vacant whole courtyards, further calculate the developer's income according to the profit calculation formula and update the developer's profit Q D .

[0147] Update agent status: Update the status of the resident agent (such as the plot after relocation), as well as the developer's remaining principal and relocation rounds.

[0148] For example, the developer selects resident R1 for relocation and relocates him from plot P3 to P 15 , total profit Q D Increase by 10,000 yuan / year.

[0149] Multiple rounds of iterative testing: Repeat a single round of relocation decisions until all residents are relocated. Since the order of relocation does affect the developer's overall interests, prioritizing the relocation of residents who have contributed most to the vacancy of entire courtyards may increase the number of vacant courtyards more quickly, thereby increasing profits.

[0150] If the relocation sequence is not reasonable, it may lead to more relocation rounds, thereby increasing communication costs and rental losses.

[0151] To optimize the relocation sequence, developers need to iterate repeatedly. They can try different initial relocation options in each round of iteration and find the optimal relocation sequence through reinforcement learning. They can compare the results of different iterative sequences and select the best solution that maximizes developer profits and increases resident satisfaction.

[0152] For example, through multiple iterations, the developer finds the optimal relocation sequence, with a total profit Q D It reached 5.36 million yuan per year, with 118 relocation rounds.

[0153] Corresponding to step S9, the data is output and a result list is formed. Result list: Figure 4 As shown, the final relocation strategy, the residents' relocation probability list, the developer's total profit and other data are output.

[0154] Finally, a visualization tool was developed using the NetLogo platform to display simulation results in real time. For example, data such as the distribution of land parcels before and after residents’ relocation and changes in developer profits were visualized to assist decision makers in adjusting parameters.

[0155] The gradual renewal of bungalow areas in old cities is a key practice in urban renewal, and the relocation of residents is a crucial step in resettling existing residents. Developers and planners should design relocation plans in a scientific, flexible, and efficient manner. Intelligent agents have the technical core to simulate decision-making in complex urban scenarios and have been widely used in urban growth scenarios. Compared to existing technologies, the multi-agent-based decision-making method for the relocation of residents in bungalow areas in old cities innovatively introduces intelligent agents into the new context of stock development, helping developers make more informed relocation decisions and achieving a win-win outcome for both developers and residents.

[0156] The above embodiments are only used to illustrate the present invention, in which the structure, connection method and manufacturing process of each component can be changed. Any equivalent transformations and improvements based on this technical solution should not be excluded from the scope of protection of the present invention.

Claims

1. A multi-agent-based decision-making method for relocation of residents in old town bungalow areas, characterized by: The method comprises the following steps: S1 collects plot information in the old town bungalow area and constructs a plot information dataset; the plot information dataset includes the following data items: plot ID, yard ID where the plot is located, plot orientation O, plot area S, plot location characteristics, and property owners on the plot; S2: Create a resident agent for each household in the old town bungalow area. The attributes of the resident agent include the following attributes: resident ID, location ID, and relocation probability P, and assign initial values ​​to each attribute. S3: Create a developer agent. The attributes of the developer agent include the following: principal Q0, relocation cost W, expected total development income M, total profit value Q D , and assign initial values ​​to each attribute; S4: Setting the behavior rules of the resident agent to relocate and maximize profit, and setting the action space of the resident agent; S5: setting the behavior criterion of the developer agent to be profit maximization and setting the action space of the developer agent; S6: Personalize modeling of one or more of the resident agents; S7: Set the developer agent’s learning method to multi-objective hierarchical reinforcement learning, set multi-objective hierarchical rules, and set the reinforcement learning reward mechanism; S8: Run all agents and perform data simulation, dynamically adjusting the relocation strategy during the simulation until convergence conditions are reached; S9: Output the data and form a result list.

2. The method according to claim 1, wherein: The relocation costs include rental compensation and communication costs.

3. The method according to claim 1, wherein: The action space of the resident agent in S4 is one of the following possibilities: No relocation; relocation to optimize daylighting; Optimize the relocation of area; Optimize relocation location.

4. The method according to claim 1, wherein: The action space of the developer agent in S5 is one of all possibilities of relocating any one of the resident agents to other plots or courtyards.

5. The method according to claim 1, wherein: The personalized modeling of one or more resident agents in S6 is specifically as follows: setting the preference parameters a of different compensation factors in the calculation formula of the relocation probability P according to the preferences of the resident agent. n , n=1,2,3; The calculation formula for the relocation probability P is: P = a1*lighting score + a2*area score + a3*location score.

6. The method according to claim 1, wherein: The total profit value Q in S3 D =Q D =Expected total hospital development income M - Relocation cost W.

7. The method according to claim 1, wherein: The multi-objective hierarchical rules in S7 are: Goal 1: Maximize total profit Q D and / or maximizing the number of vacant intact compound buildings; Goal 2: Minimize relocation rounds; Objective 3: Maximize residents’ willingness to relocate.

8. The method according to claim 1, wherein: The reinforcement learning reward mechanism in S7 is as follows: the profit obtained by the developer after each relocation is calculated as the reward function, the increase in the number of vacant intact courtyards is also quantified as profit growth, and the communication cost of each relocation is introduced, so that the relocation round affects the developer's profit growth.

9. The method according to claim 1, wherein: The S8 includes the following sub-steps: S91 randomly selects a resident agent to relocate and uses a reinforcement learning algorithm to select the optimal relocation action based on the current state, where the relocation action is selected based on the maximization of the reward function; S92 calculates the change in the principal Q0 of this relocation, updates the number of vacant and complete courtyards, and further calculates the total profit value Q of the developer if new vacant and complete courtyards appear. D ; S93 updates the status of the resident agent, as well as the remaining principal Q0 and relocation round of the developer agent; S94 repeats S91 to S93, trying different initial relocation options in each iteration, and finding the optimal relocation sequence through the reinforcement learning algorithm. The results of different sequence orientations are compared, and the best solution is selected to maximize the developer's profit and increase the residents' satisfaction.

10. The method according to claim 1, wherein: The method also includes S10: developing a visualization tool through simulation and multi-agent system platform, allowing decision makers to adjust parameters and view simulation results in real time, superimposing the above simulation results and spatial data in the system, and completing the effect display of translation and maneuvering multi-scene simulation.

Images