Park green land planning scheme deduction method based on multi-agent quarterly simulation

Abstract

The invention discloses a park green land planning scheme deduction method based on multi-agent quarterly simulation. The method comprises four main steps of park historical data acquisition, crowd seasonal travel feature extraction, multi-agent visiting modeling and iterative optimization of park green land, and park layout deduction and scheme output. According to the method, the relationship between the seasonal travel mode of the park green land and the urban built environment is captured, the deducable park green land multi-agent travel model is constructed, and then the deducable park green land multi-agent travel model is connected with the multi-scene park green land planning scheme data to form the park green land passenger flow deduction in quarters so as to assist the intelligent evaluation of the planning scheme. According to the method, the park green land scheme passenger flow deduction and intelligent evaluation in quarters can be realized, and the existing park green land planning scheme evaluation method is optimized.

Description

Technical Field

[0001] The present invention relates to a technology for predicting passenger flow in urban parks and green spaces, and belongs to the technical field of urban planning and green space system planning, and in particular to a multi-agent predicted passenger flow status and evaluation method for park green spaces. Background Art

[0002] Urban green space is an important part of the urban system, with ecological, economic and social value, and is of great significance to promoting sustainable urban development, improving the quality of ecological links and improving the happiness index of citizens' lives. In actual work, urban green space system planning needs to evaluate and deliberate on multiple green space layout plans based on the existing green space development indicators, under the premise of coordinating the existing landscape resources of the city, the daily needs of residents and the overall construction requirements of the city.

[0003] At present, the evaluation of green space system solutions mainly focuses on post-use evaluation, that is, after the park green space is selected and built, the existing problems are summarized based on the passenger flow data and residents' evaluation of the park green space to guide the next step of park green space construction and optimization. It is difficult to effectively predict its passenger flow before the plan is determined, and the current evaluation methods often ignore the differences in the number of passenger flows and visitor types in each quarter of the park green space, and fail to achieve an effective evaluation of the use of the park green space each quarter. In the stage of plan deliberation, the evaluation of specific plans often needs to be completed by planning and design personnel, who need to manually organize the existing landscape resource data and infer the passenger flow. This process has the problems of high labor costs, economic costs and time costs, and the evaluation results are highly subjective and lack quantitative basis.

[0004] The emergence and continuous maturity of multi-agent simulation technology provides a technical method for the simulation and evaluation of passenger flow in urban park green space schemes. Summary of the invention

[0005] The purpose of the present invention is to construct a multi-agent simulation model of park green space travel by season based on multi-agent simulation technology, to simulate the passenger flow and evaluate the schemes of each park green space under the scenarios of each season, so as to effectively deal with the problem that the design of park green space layout schemes in the original green space system planning is highly subjective and does not take into account the differences in passenger flow and tourist type composition between each season.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] The park green space planning scheme deduction method based on multi-agent quarterly simulation includes:

[0008] Step S1: Park historical data collection

[0009] According to the scope of the target city, basic spatial data, park green space climate and environmental data, and urban spatiotemporal trajectory big data are obtained; the urban spatiotemporal trajectory big data is preprocessed for data cleaning and stop point identification, and the mobile phone user trajectory is matched with the park green space through coordinate data, and the name of the park green space passed by each user is associated with the corresponding user data to generate a quarterly park green space historical usage data set.

[0010] Step S2: Extraction of seasonal travel characteristics of the crowd

[0011] S2-1 Basic characteristic variable set of multi-agents. All user IDs and their location attributes, as well as the date distribution d and time distribution t of each user's park green space trip are extracted from the quarterly park green space historical usage data set described in S1, and the user's characteristic indicators are classified according to the seasonal anchoring classification method to generate a total of 12 types of multi-agents.

[0012] S2-2 Park green space travel probability set of multi-agents. For each type of multi-agent user ID, extract its quarterly park green space historical usage data set, and use the travel frequency accumulation method to extract the park green space travel probability set {P1, P2, P3, ..., P 24}, where P = {probability of traveling to park green space, park number}, 24 is the time slice, and the slice length is one hour.

[0013] S2-3 Construction of the “Season-Multi-agent Subject” feature set of park green space. The park green space travel probability sets of various multi-agents in S2-2 are aggregated to each park green space by time and space to generate the “Season-Multi-agent Subject” feature set of each park green space. The “Season-Multi-agent Subject” feature set contains the daily multi-agent type composition and effective stay intensity feature set {H1, H2, H3, …, H u}, where H u ={effective stay probability of the u-th type of multi-agent, total stay duration of the u-th type of multi-agent}; and adopt a sampling method to verify the actual resident composition and usage status of the selected sample park green space in different seasons and time periods.

[0014] S2-4 Construction of the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of park green space. From the basic spatial data described in S1, extract the characteristic indicators of the built environment inside and around the park green space in each season and each time slice, and generate the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of the park green space through correction of the measured environmental data. The spatiotemporal environment feature set includes the measured environment, the scope and scale of the park green space, the POI density of the surrounding built environment, and the road network density index of the surrounding built environment. The surrounding built environment refers to the urban built space within a straight-line distance of 1 km from the park green space.

[0015] Step S3: Multi-agent visit modeling and iterative optimization of park green spaces

[0016] The large language model (LLM) is used as the decision-making agent. The feature set of "seasonal-multi-agent subject" and "seasonal-multi-agent spatiotemporal environment" of each park green space and the spatiotemporal trajectory big data in the seasonal park green space historical use data set are used as machine learning labels to capture the relationship between multi-agent model variables and travel trajectories, match the seasonal travel characteristics of the crowd with the learned travel model, and build a multi-agent travel plan. Through self-evaluation and recursive reasoning, the agent continuously compares the generated trajectory with the real-world situation to improve its output, and finally obtains a multi-agent simulation model of park green space visits.

[0017] Step S4: Park layout deduction and plan output

[0018] Input multi-scenario park green space layout plans for multi-agent simulation, generate the "park-multi-agent" pairing results for each quarter under different layout plans, and combine with social integration evaluation to construct a park green space plan evaluation module, and output the optimal plan for the park green space layout scenario.

[0019] Furthermore, in the quarterly park green space historical usage data collection in step S1, basic spatial data is obtained from the Open Street Map website, specifically including park green space vector data, urban road network vector data, building vector data, and business POI data; park green space climate environment data is obtained through the website of the meteorological bureau in the study area, including wind environment data, humidity data, temperature data, and solar radiation data; urban spatiotemporal trajectory big data is LBS data obtained from mobile phone information push service providers, including LBS effective stay point data and user attribute data.

[0020] The quarterly park green space historical usage data set refers to the spatiotemporal trajectory big data and park green space climate and environmental data within the boundaries of each park summarized in four quarters, forming a spatiotemporal trajectory big data set and a park green space climate and environmental data set; wherein, the summary of the spatiotemporal trajectory big data needs to be matched by quarter according to the time range and the stay threshold, and the park green space climate and environmental data selects the 24-hour meteorological data corresponding to the date with the largest number of effective stays in each quarter, and the 24-hour meteorological data includes wind environment data, humidity data, and temperature data.

[0021] Furthermore, the seasonal anchorage classification method in step S2-1 is:

[0022] Step S2-11 Quarterly park green space travel characteristics classification. The proportion of the number of park green space trips of the user in the i-th quarter to the total number of trips in the year is recorded as F i , then the distribution entropy of the user's quarterly park green space travel is:

[0023]

[0024] If H(F)≥0.7ln4, the user is marked as a four-season traveler, otherwise it is marked as a single-season traveler.

[0025] Step S2-12: Classification of daily park green space travel characteristics. Mark 6:00-18:00 (including 6:00 and excluding 18:00) as daytime, and mark the rest of the time as nighttime. Based on the daily travel characteristics of users, mark users whose number of park green space trips during the daytime in the statistical year is greater than or equal to 80% of the total number of trips in the year as daytime travelers; mark users whose number of park green space trips at night in the statistical year is greater than or equal to 80% of the total number of trips in the year as nighttime travelers; and mark the rest of the users as full-day travelers.

[0026] Step S2-13: Territorial characteristics classification: Based on the user's territorial characteristics, the users whose park green space travel trajectories are located in the same place as their territorial location are marked as locals, and the rest of the users are marked as outsiders.

[0027] Step S2-14: Classification of multi-agents: The multi-agents are classified into 12 types according to the seasonal travel characteristics, daily travel characteristics, and local travel characteristics.

[0028] Furthermore, the travel frequency accumulation method in step S2-2 is:

[0029] Step S2-21: Summarize park green space travel information. Number the park green spaces within the target city, and summarize the park numbers h where each type of multi-agent effectively stays and the total length of stay s in each park by quarter.

[0030] Step S2-22: Calculate the probability of traveling to the park green space. The effective stay probability set of each type of multi-agent in the park green space for one day in each quarter is:

[0031]

[0032] Among them, a is the time slice number, b is the park green space number where this type of multi-agent has the most effective stays in the time slice, n is the total number of users in the time slice, m is the number of users with effective stay records in park green space b in the time slice, and s is the total stay time in park green space b in the time slice.

[0033] Furthermore, the method for verifying the actual resident composition and usage status of the selected sample park green space in different seasons and time periods in step S2-3 is:

[0034] Step S2-31 determines the sampling object. Select time segment a in the qth quarter, and conduct sampling inspection on the ten parks and green spaces with the largest number of visitors in the qth quarter and time segment a within the target city, and mark the parks and green spaces as b1, b2, b3...b 10 .

[0035] Step S2-32: Construction of the field verification set. In time period a, the dynamic video data of the ten parks and green spaces are obtained by drone photography, and the actual number of effective residents k in park green space b in time segment a is obtained by video semantic recognition. ab And obtain the type of crowd through field research (the result of the number of crowd types is the same as the multi-agent category classification table in step S2-14). According to the value k, the ranking R1 of the ten parks is obtained.

[0036] Step S2-33: Sampling feature set construction. From the "quarter-multi-agent subject" feature set, extract the sampling feature set of park green space according to the park number, quarter feature q, and time segment a. The sampling feature set contains the daily multi-agent type composition and its effective stay intensity feature set H; mark each multi-agent type b in each extracted park green space as The effective stay probabilities of various multi-agents are summarized, and then the number of people in each park green space extracted is calculated; the specific formula is as follows:

[0037]

[0038] Where P i represents the effective stay probability of the i-th type of multi-agent, n i Represents the total number of samples of the i-th type of multi-agent in the quarterly park green space historical usage dataset.

[0039] Step S2-34: Data verification between the field verification set and the sampling feature set. Based on the type number and number of people sorting verification between the field verification data set and the sampling feature set, a high-quality data set with a degree of consistency of more than half is screened out. The specific rules are as follows: According to the value l, the ranking R2 of the ten parks is obtained, and R1 and R2, the type composition of the crowd and the type composition of the multi-agent are compared. If they are the same, the simulation of the park green space is considered to be consistent with the actual situation, otherwise it is considered that the simulation of the park green space is inconsistent with the actual situation. In this way, the ten park green spaces sampled and tested are all tested. If more than half of the park green spaces match the actual situation, the data set is considered to have passed the test.

[0040] Furthermore, the feature set of "Seasonal-Multi-agent Space-Time Environment" in step S2-4 includes public transportation facility coverage, job-residence ratio, urban land function information entropy and physiological equivalent temperature.

[0041] Furthermore, in step S4, the analytic hierarchy process is used to evaluate the degree of social integration, and the specific steps are as follows:

[0042] Step S4-1: Crowd diversity dimension evaluation. The total number of multi-agent categories with effective stays in the simulation results of the park green space to be evaluated is recorded as u, the total number of effective stays is recorded as n, and the number of effective stays of the i-th multi-agent category is recorded as m i , then the population diversity distribution entropy G of this layout scheme is:

[0043]

[0044] The population diversity distribution entropy G is used as the evaluation standard for the population diversity dimension. The mean value of the population diversity distribution entropy G of the park green space in four quarters is obtained. The higher the value, the better the overall performance of the scheme in terms of population diversity. The population diversity distribution entropy of the park green space in the i-th quarter is recorded as G i ,but:

[0045]

[0046] If H(G)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the population diversity dimension, and the best and worst quarters in the population diversity dimension are recorded.

[0047] Step S4-2: Evaluation of the park green space capacity suitability dimension. The "Park Design Code" (GB 51192-2016) stipulates that the average land area per visitor to a comprehensive park is 30 to 60 square meters, the average land area per visitor to a special park is 20 to 30 square meters, the average land area per visitor to a community park is 20 to 30 square meters, and the average land area per visitor to a park is 30 to 60 square meters. The total number of multi-intelligent effective stays with effective stays in the simulation results of the park green space to be evaluated is recorded as n, and the park green space area g is obtained from the basic spatial data of step S1, then the park green space capacity suitability index C is:

[0048]

[0049] The park green space capacity suitability index C is used as the park green space capacity suitability evaluation standard. The average value of the park green space capacity suitability index C in four quarters is obtained. The smaller the value, the better the overall performance of the scheme in terms of park green space capacity suitability. The c index of the park green space in the i quarter is recorded as C i ,but:

[0050]

[0051] If H(C)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded.

[0052] Step S4-3: Comprehensive environmental dimension evaluation. The public transportation facility coverage rate x1, job-housing ratio x2, urban land function information entropy x3, and physiological equivalent temperature x4 of the evaluated park green space are obtained from the step S2-4 "Season-Multi-agent Spatiotemporal Environment". The same data are normalized in all evaluation schemes to obtain y1, y2, y3, and y4. The normalization method is:

[0053]

[0054] Then the comprehensive environmental dimension index E of the layout scheme is:

[0055]

[0056] The comprehensive environmental dimension index E is used as the comprehensive environmental dimension evaluation standard. The average value of the comprehensive environmental dimension index E of the park green space in four quarters is obtained. The higher the value, the better the overall performance of the scheme in the comprehensive environmental dimension. The comprehensive environmental dimension index of the park green space in the i quarter is recorded as E i ,but:

[0057]

[0058] If H(E)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded.

[0059] Step S4-4: Scheme evaluation and output: Based on the evaluation dimension of the multi-scenario park green space layout scheme, the dimension is scored and evaluated, and the park green space layout scheme and quarterly characteristic evaluation under the dimension are output.

[0060] Beneficial effects:

[0061] 1. The present invention realizes the quarterly deduction of the passenger flow of park green spaces. Steps S2 and 3 are based on multi-agent simulation technology and seasonal anchor feature extraction, and innovatively propose a passenger flow deduction model based on the spatiotemporal characteristics of park green spaces by season, breaking through the limitation that traditional green space system planning cannot predict quarterly passenger flow characteristics before construction.

[0062] 2. The present invention takes into account the different emphases of planning schemes and the applicability of intelligent inspection in the work. In step 4, based on the prediction results achieved by the present invention and the actual work requirements, the three dimensions of population diversity, capacity suitability, and comprehensive environment are evaluated, and the evaluation results are output according to the work requirements.

[0063] 3. The present invention starts from the prediction accuracy and process scientificity of multi-scenario planning and evaluation. In terms of the time consumption of multi-scenario planning and evaluation, it changes from the traditional expert review meeting that requires a lot of manpower and financial resources to quarterly intelligent deduction and evaluation based on historical data sets. The time cost is greatly reduced, and the deduction dimension, decision-making efficiency and accuracy are also greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a flow chart of the park green space planning scheme deduction and evaluation method of the present invention. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0066] like Figure 1 As shown, the technical solution of the present invention is described in detail with the multi-scenario planning scheme of Gulou District, Nanjing City, Jiangsu Province:

[0067] Step S1: Park historical data collection

[0068] According to the scope of the target city, basic spatial data, park green space climate and environmental data, and urban spatiotemporal trajectory big data are obtained; the urban spatiotemporal trajectory big data is preprocessed for data cleaning and stop point identification, and the mobile phone user trajectory is matched with the park green space through coordinate data, and the name of the park green space passed by each user is associated with the corresponding user data to generate a quarterly park green space historical usage data set. Among them, the basic spatial data is obtained from the Open streetmap website, including park green space vector data, urban road network vector data, building vector data, and business POI data; the park green space climate and environmental data is obtained through the website of the meteorological bureau in the study area, including wind environment data, humidity data, temperature data, and solar radiation data; the urban spatiotemporal trajectory big data is LBS data obtained from mobile phone information push service providers, including LBS effective stop point data and user attribute data. The specific attributes of each type of data are described as follows:

[0069] Data types and their property descriptions

[0070]

[0071] The quarterly park green space historical use data set refers to the spatiotemporal trajectory big data and park green space climate and environmental data within the boundaries of each park summarized in four quarters, forming a spatiotemporal trajectory big data set and a park green space climate and environmental data set; wherein, the summary of the spatiotemporal trajectory big data needs to be matched by time range and stay threshold for quarterly, and the park green space climate and environmental data selects the 24-hour meteorological data corresponding to the date with the largest number of valid stays in each quarter, and the 24-hour meteorological data includes wind environment data, humidity data, and temperature data. The specific attributes of each type of data are described as follows:

[0072] Attribute description of the historical park green space use dataset by quarter

[0073]

[0074] Step S2: Extraction of seasonal travel characteristics of the crowd

[0075] S2-1 Basic characteristic variable set of multi-agents. All user IDs and their location attributes, as well as the date distribution d and time distribution t of each user's park green space trip are extracted from the quarterly park green space historical usage data set described in S1, and the user's characteristic indicators are classified according to the seasonal anchoring classification method to generate a total of 12 types of multi-agents.

[0076] The seasonal anchorage classification method in step S2-1 is:

[0077] Step S2-11 Quarterly park green space travel characteristics classification. The proportion of the number of park green space trips of the user in the i-th quarter to the total number of trips in the year is recorded as F i , then the distribution entropy of the user's quarterly park green space travel is:

[0078]

[0079] If H(F)≥0.7ln4, the user is marked as a four-season traveler, otherwise it is marked as a single-season traveler.

[0080] Step S2-12: Classification of daily park green space travel characteristics. Mark 6:00-18:00 (including 6:00 and excluding 18:00) as daytime, and mark the rest of the time as nighttime. Based on the daily travel characteristics of users, mark users whose number of park green space trips during the daytime in the statistical year is greater than or equal to 80% of the total number of trips in the year as daytime travelers; mark users whose number of park green space trips at night in the statistical year is greater than or equal to 80% of the total number of trips in the year as nighttime travelers; and mark the rest of the users as full-day travelers.

[0081] Step S2-13: Territorial characteristics classification: Based on the user's territorial characteristics, the users whose park green space travel trajectories are located in the same place as their territorial location are marked as locals, and the rest of the users are marked as outsiders.

[0082] Step S2-14: Classification of multi-agents: Based on the seasonal travel characteristics, daily travel characteristics, and local travel characteristics, the multi-agents are divided into 12 types, and the specific characteristics are as follows.

[0083] Multi-agent classification

[0084]

[0085] S2-2 Park green space travel probability set of multi-agents. For each type of multi-agent user ID, extract its quarterly park green space historical usage data set, and use the travel frequency accumulation method to extract the park green space travel probability set {P1, P2, P3, ..., P 24}, where P = {probability of traveling to park green space, park number}, 24 is the time slice, and the slice length is one hour.

[0086] The travel frequency accumulation method in step S2-2 is:

[0087] Step S2-21: Summarize park green space travel information. Number the park green spaces within the target city, and summarize the park numbers h where each type of multi-agent effectively stays and the total length of stay s in each park by quarter.

[0088] Step S2-22: Calculate the probability of traveling to the park green space. The effective stay probability set of each type of multi-agent in the park green space for one day in each quarter is:

[0089]

[0090] Among them, a is the time slice number, b is the park green space number where this type of multi-agent has the most effective stays in the time slice, n is the total number of users in the time slice, m is the number of users with effective stay records in park green space b in the time slice, and s is the total stay time in park green space b in the time slice.

[0091] S2-3 Construction of the “Season-Multi-agent Subject” feature set of park green space. The park green space travel probability sets of various multi-agents in S2-2 are aggregated to each park green space by time and space to generate the “Season-Multi-agent Subject” feature set of each park green space. The “Season-Multi-agent Subject” feature set contains the daily multi-agent type composition and effective stay intensity feature set {H1, H2, H3, …, H u}, where H u ={effective stay probability of the yth type of multi-agent, total stay duration of the uth type of multi-agent}; and adopt a sampling method to verify the actual resident composition and usage status of the selected sample park green space in different seasons and time periods.

[0092] The method for checking and verifying the actual resident composition and usage status of the selected sample park green space in different seasons and time periods in step S2-3 is:

[0093] Step S2-31 determines the sampling object. Select time segment a in the qth quarter, and conduct sampling inspection on the ten parks and green spaces with the largest number of visitors in the qth quarter and time segment a within the target city, and mark the parks and green spaces as b1, b2, b3...b 10 .

[0094] Step S2-32: Construction of the field verification set. In time period a, the dynamic video data of the ten parks and green spaces are obtained by drone photography, and the actual number of effective residents k in park green space b in time segment a is obtained by video semantic recognition. ab And obtain the type of crowd through field research (the result of the number of crowd types is the same as the multi-agent category classification table in step S2-14). According to the value k, the ranking R1 of the ten parks is obtained.

[0095] Step S2-33: Sampling feature set construction. From the "quarter-multi-agent subject" feature set, extract the sampling feature set of park green space according to the park number, quarter feature q, and time segment a. The sampling feature set contains the daily multi-agent type composition and its effective stay intensity feature set H; mark each multi-agent type b in each extracted park green space as The effective stay probabilities of various multi-agents are summarized, and then the number of people in each park green space extracted is calculated; the specific formula is as follows:

[0096]

[0097] Where P i represents the effective stay probability of the i-th type of multi-agent, n i Represents the total number of samples of the i-th type of multi-agent in the quarterly park green space historical usage dataset.

[0098] Step S2-34: Data verification between the field verification set and the sampling feature set. Based on the type number and number of people sorting verification between the field verification data set and the sampling feature set, a high-quality data set with a degree of consistency of more than half is screened out. The specific rules are as follows: According to the value l, the ranking R2 of the ten parks is obtained, and R1 and R2, the type composition of the crowd and the type composition of the multi-agent are compared. If they are the same, the simulation of the park green space is considered to be consistent with the actual situation, otherwise it is considered that the simulation of the park green space is inconsistent with the actual situation. In this way, the ten park green spaces sampled and tested are all tested. If more than half of the park green spaces match the actual situation, the data set is considered to have passed the test.

[0099] S2-4 Construction of the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of park green space. From the basic spatial data described in S1, extract the characteristic indicators of the built environment inside and around the park green space in each season and each time slice, and generate the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of the park green space through correction of the measured environmental data. The spatiotemporal environment feature set includes the measured environment, the scope and scale of the park green space, the POI density of the surrounding built environment, and the road network density index of the surrounding built environment. The surrounding built environment refers to the urban built space within a straight-line distance of 1 km from the park green space.

[0100] The feature set of "Seasonal-Multi-agent Spatiotemporal Environment" in step S2-4 includes public transportation facility coverage, job-residence ratio, urban land function information entropy and physiological equivalent temperature. The specific indicators are as follows:

[0101] "Quarterly-Multi-agent Spatiotemporal Environment" feature set indicators

[0102]

[0103]

[0104] Step S3: Multi-agent visit modeling and iterative optimization of park green spaces

[0105] The large language model (LLM) is used as the decision-making agent. The feature set of "seasonal-multi-agent subject" and "seasonal-multi-agent spatiotemporal environment" of each park green space and the spatiotemporal trajectory big data in the seasonal park green space historical use data set are used as machine learning labels to capture the relationship between multi-agent model variables and travel trajectories, match the seasonal travel characteristics of the crowd with the learned travel model, and build a multi-agent travel plan. Through self-evaluation and recursive reasoning, the agent continuously compares the generated trajectory with the real-world situation to improve its output, and finally obtains a multi-agent simulation model of park green space visits.

[0106] Step S4: Park layout deduction and plan output

[0107] Input multi-scenario park green space layout plans for multi-agent simulation, generate the "park-multi-agent" pairing results for each quarter under different layout plans, and combine with social integration evaluation to construct a park green space plan evaluation module, and output the optimal plan for the park green space layout scenario.

[0108] In step S4, the analytic hierarchy process is used to evaluate the degree of social integration, and the specific steps are as follows:

[0109] Step S4-1: Crowd diversity dimension evaluation. The total number of multi-agent categories with effective stays in the simulation results of the park green space to be evaluated is recorded as u, the total number of effective stays is recorded as n, and the number of effective stays of the i-th multi-agent category is recorded as m i , then the population diversity distribution entropy G of this layout scheme is:

[0110]

[0111] The population diversity distribution entropy G is used as the evaluation standard for the population diversity dimension. The mean value of the population diversity distribution entropy G of the park green space in four quarters is obtained. The higher the value, the better the overall performance of the scheme in terms of population diversity. The population diversity distribution entropy of the park green space in the i-th quarter is recorded as G i ,but:

[0112]

[0113] If H(G)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the population diversity dimension, and the best and worst quarters in the population diversity dimension are recorded.

[0114] Step S4-2: Evaluation of the park green space capacity suitability dimension. The "Park Design Code" (GB 51192-2016) stipulates that the average land area per visitor to a comprehensive park is 30 to 60 square meters, the average land area per visitor to a special park is 20 to 30 square meters, the average land area per visitor to a community park is 20 to 30 square meters, and the average land area per visitor to a park is 30 to 60 square meters. The total number of multi-intelligent effective stays with effective stays in the simulation results of the park green space to be evaluated is recorded as n, and the park green space area g is obtained from the basic spatial data of step S1, then the park green space capacity suitability index C is:

[0115]

[0116] The park green space capacity suitability index C is used as the park green space capacity suitability evaluation standard. The average value of the park green space capacity suitability index C in four quarters is obtained. The smaller the value, the better the overall performance of the scheme in terms of park green space capacity suitability. The c index of the park green space in the i quarter is recorded as C i ,but:

[0117]

[0118] If H(C)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded.

[0119] Step S4-3: Comprehensive environmental dimension evaluation. The public transportation facility coverage rate x1, job-housing ratio x2, urban land function information entropy x3, and physiological equivalent temperature x4 of the evaluated park green space are obtained from the step S2-4 "Season-Multi-agent Spatiotemporal Environment". The same data are normalized in all evaluation schemes to obtain y1, y2, y3, and y4. The normalization method is:

[0120]

[0121] Then the comprehensive environmental dimension index E of the layout scheme is:

[0122]

[0123] The comprehensive environmental dimension index E is used as the comprehensive environmental dimension evaluation standard. The average value of the comprehensive environmental dimension index E of the park green space in four quarters is obtained. The higher the value, the better the overall performance of the scheme in the comprehensive environmental dimension. The comprehensive environmental dimension index of the park green space in the i quarter is recorded as E i ,but:

[0124]

[0125] If H(E)<0.5ln4, it is considered that the park green space has obvious quarterly differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded.

[0126] Step S4-4: Scheme evaluation and output: Based on the evaluation dimension of the multi-scenario park green space layout scheme, the dimension is scored and evaluated, and the park green space layout scheme and quarterly characteristic evaluation under the dimension are output.

[0127] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A park green space planning scheme deduction method based on multi-agent quarterly simulation, characterized in that: The method comprises the following steps: Step S1: Park historical data collection According to the scope of the target city, basic spatial data, park green space climate and environmental data, and urban spatiotemporal trajectory big data are obtained; the urban spatiotemporal trajectory big data is preprocessed for data cleaning and stop point identification, and the mobile phone user trajectory is matched with the park green space through the coordinate data, and the name of the park green space passed by each user is associated with the corresponding user data to generate a quarterly park green space historical usage data set; Step S2: Extraction of seasonal travel characteristics of the crowd S2-1 Basic characteristic variable set of multi-agents; extract all user IDs and their location attributes from the quarterly park green space historical usage data set described in S1, as well as the date distribution d and time distribution t of each user's park green space trip, and classify the user's characteristic indicators according to the seasonal anchoring classification method to generate a total of 12 types of multi-agents; S2-2 Park green space travel probability set of multi-agents; for each type of multi-agent user ID, extract its quarterly park green space historical usage data set, and use the travel frequency accumulation method to extract the park green space travel probability set {P1, P2, P3, ..., P 24 }, where P = {probability of traveling to park green space, park number}, 24 is the time slice, and the slice length is one hour; S2-3 Construction of the "Season-Multi-agent Subject" feature set of park green space: The park green space travel probability sets of various multi-agents in S2-2 are aggregated to each park green space by time and space to generate the "Season-Multi-agent Subject" feature set of each park green space. The "Season-Multi-agent Subject" feature set contains the daily multi-agent type composition and effective stay intensity feature set {H1, H2, H3, ..., H u }, where H u = {effective stay probability of the u-th type of multi-agent, total stay time of the u-th type of multi-agent}; and adopt the sampling method to check and verify the actual resident composition and usage status of the selected sample park green space in different seasons and time periods; S2-4 Construction of the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of park green space: Extract the characteristic indicators of the built environment inside and around the park green space in each season and each time slice from the basic spatial data described in S1, and generate the "Seasonal-Multi-agent Spatiotemporal Environment" feature set of park green space through correction of the measured environmental data; Step S3: Multi-agent visit modeling and iterative optimization of park green spaces The large language model is used as the decision-making agent, and the "seasonal-multi-agent subject" feature set, "seasonal-multi-agent spatiotemporal environment" feature set of each park green space and the spatiotemporal trajectory big data in the seasonal park green space historical use data set are used as machine learning labels to capture the relationship between the multi-agent model variables and travel trajectories, match the seasonal travel characteristics of the crowd with the learned travel model, and build a multi-agent travel plan; through self-evaluation and recursive reasoning, the agent continuously compares the generated trajectory with the real-world situation to improve its output, and finally obtains a multi-agent simulation model of park green space visits; Step S4: Park layout deduction and plan output Input multi-scenario park green space layout plans for multi-agent simulation, generate the "park-multi-agent" pairing results for each quarter under different layout plans, and construct a park green space plan evaluation module in combination with social integration evaluation to output the optimal plan for displaying the park green space layout scenario.

2. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 1 is characterized in that: In the quarterly park green space historical usage data collection in step S1, basic spatial data is obtained from the Open Streetmap website, specifically including park green space vector data, urban road network vector data, building vector data, and business POI data; park green space climate environment data is obtained through the website of the meteorological bureau in the research area, including wind environment data, humidity data, temperature data, and solar radiation data; urban spatiotemporal trajectory big data is LBS data obtained from mobile phone information push service providers, including LBS effective stay point data and user attribute data; The quarterly park green space historical usage data set refers to the spatiotemporal trajectory big data and park green space climate and environmental data within the boundaries of each park summarized in four quarters, forming a spatiotemporal trajectory big data set and a park green space climate and environmental data set; wherein, the summary of the spatiotemporal trajectory big data needs to be matched by quarter according to the time range and the stay threshold, and the park green space climate and environmental data selects the 24-hour meteorological data corresponding to the date with the largest number of effective stays in each quarter, and the 24-hour meteorological data includes wind environment data, humidity data, and temperature data.

3. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 2 is characterized in that: The seasonal anchorage classification method in step S2-1 is: Step S2-11 Quarterly park green space travel characteristics classification; the proportion of the number of park green space trips of the user in the i-th quarter to the total number of trips in that year is recorded as F i , then the distribution entropy of the user's quarterly park green space travel is: If H(F)≥0.7ln4, the user is marked as a four-season traveler, otherwise it is marked as a single-season traveler; Step S2-12: daily park green space travel characteristics classification: mark 6:00-18:00 as daytime, and the rest of the time as nighttime; According to the daily travel characteristics of users, users whose number of trips to parks and green spaces during the daytime during the statistical year is greater than or equal to 80% of the total number of trips in the year are marked as daytime travelers; users whose number of trips to parks and green spaces at night during the statistical year is greater than or equal to 80% of the total number of trips in the year are marked as nighttime travelers; the rest of the users are marked as full-day travelers; Step S2-13: Territorial characteristics classification: based on the user's territorial characteristics, mark the user whose park green space travel trajectory is located in the same place as his / her territorial location as a local user; and mark the rest of the users as outsiders; Step S2-14: multi-agent classification; multi-agents are classified into 12 types based on seasonal travel characteristics, daily travel characteristics, and territorial travel characteristics.

4. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 3 is characterized in that: The travel frequency accumulation method in step S2-2 is: Step S2-21: Summarize park green space travel information; number the park green spaces within the target city, and summarize the park numbers h where each type of multi-agent effectively stays and the total length of stay s in each park by quarter; Step S2-22: Calculate the probability of traveling to the park green space; the effective stay probability set of each type of multi-agent in the park green space for one day in each quarter is: Among them, a is the time slice number, b is the park green space number where this type of multi-agent has the most effective stays in the time slice, n is the total number of users in the time slice, m is the number of users with effective stay records in park green space b in the time slice, and s is the total stay time in park green space b in the time slice.

5. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 4 is characterized in that: The method for checking and verifying the actual resident composition and usage status of the selected sample park green space in different seasons and time periods in step S2-3 is: Step S2-31 determines the sampling object; selects time segment a in the qth quarter, and performs sampling inspection on the ten parks and green spaces with the largest number of visitors in the qth quarter and time segment a within the target city, and marks the parks and green spaces as b1, b2, b3…b 10 ; Step S2-32: constructing a field verification set; obtaining dynamic video data of the ten park green spaces by drone photography during time period a, and obtaining the actual number of effective residents k in park green space b during time period a by video semantic recognition ab , and obtain the type of people through field research; obtain the ranking R1 of the ten parks according to the value k; Step S2-33: Sampling feature set construction; extracting the sampling feature set of park green space from the "quarter-multi-agent subject" feature set according to the park number, quarter feature q, and time segment a, wherein the sampling feature set includes a daily multi-agent type composition and its effective stay intensity feature set H; marking each multi-agent type b in each extracted park green space as The effective stay probabilities of various multi-agents are summarized, and then the number of people in each park green space extracted is calculated; the specific formula is as follows: Where P i represents the effective stay probability of the i-th type of multi-agent, n i Represents the total number of samples of the i-th type of multi-agent in the quarterly park green space historical use dataset; Step S2-34: Data verification between the field verification set and the sampling feature set; based on the type number and population ranking verification between the two data sets of the field verification data set and the sampling feature set, screen out high-quality data sets with a degree of consistency with the actual situation of more than half; the specific rules are as follows: obtain the ranking R2 of the ten parks according to the size of the value l, compare R1 and R2, the population type composition and the multi-agent type composition, if they are the same, it is considered that the simulation situation of the park green space is consistent with the actual situation, otherwise it is considered that the simulation situation of the park green space is inconsistent with the actual situation, and the ten park green spaces sampled and tested are all tested in this way. If the simulation situation is consistent with the actual situation for more than half of the park green spaces, the data set is deemed to have passed the test.

6. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 5 is characterized in that: The surrounding built environment described in the "Seasonal-Multi-agent Space-Time Environment" feature set in step S2-4 refers to the urban built space within a straight-line distance of 1 km from the park green space; specifically includes the coverage rate of public transportation facilities, the job-residence ratio, the information entropy of urban land function and the physiological equivalent temperature.

7. The method for deducing a park green space planning scheme based on multi-agent quarterly simulation according to claim 6 is characterized in that: In step S4, the analytic hierarchy process is used to evaluate the degree of social integration, and the specific steps are as follows: Step S4-1: Crowd diversity dimension evaluation; the total number of multi-agent categories with effective stays in the simulation results of the park green space to be evaluated is recorded as u, the total number of effective stays is recorded as n, and the number of effective stays of the i-th multi-agent is recorded as m i , then the population diversity distribution entropy G of this layout scheme is: The entropy G of the population diversity distribution is used as the evaluation standard of the population diversity dimension; the mean value of the entropy G of the population diversity distribution in the park green space in four quarters is obtained The higher the value, the better the overall performance of the scheme in terms of population diversity. The population diversity distribution entropy of the park green space in the i-th quarter is recorded as G i ,but: If H(G)<0.5ln4, it is considered that the park green space has obvious quarterly differences in population diversity, and the best and worst quarters in terms of population diversity are recorded; Step S4-2: Evaluation of the park green space capacity suitability dimension; The "Park Design Specifications" stipulate that the average land area per visitor to a comprehensive park is 30-60 m2, the average land area per visitor to a special park is 20-30 m2, the average land area per visitor to a community park is 20-30 m2, and the average land area per visitor to a park is 30-60 m2; the total number of multi-intelligent effective stays with effective stays in the simulation results of the park green space to be evaluated is recorded as n, and the park green space area g is obtained from the basic spatial data of step S1, then the park green space capacity suitability index C is: The park green space capacity suitability index C is used as the park green space capacity suitability evaluation standard; the average value of the park green space capacity suitability index C in four quarters is obtained The smaller the value, the better the annual comprehensive performance of the scheme in the dimension of park green space capacity suitability; the c index of the park green space in the i quarter is recorded as C i ,but: If H(C)<0.5ln4, the park green space is considered to have obvious seasonal differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded at the same time; Step S4-3: Comprehensive environmental dimension evaluation; the public transportation facility coverage rate x1, job-housing ratio x2, urban land function information entropy x3, and physiological equivalent temperature x4 of the evaluated park green space are obtained from the step S2-4 "quarterly-multi-agent spatiotemporal environment", and the similar data are normalized in all evaluation schemes to obtain y1, y2, y3, and y4. The normalization method is: Then the comprehensive environmental dimension index E of the layout scheme is: The comprehensive environmental dimension index E is used as the comprehensive environmental dimension evaluation standard; the average value of the comprehensive environmental dimension index E of the park green space in four quarters is obtained The higher the value, the better the overall performance of the scheme in the comprehensive environmental dimension. The comprehensive environmental dimension index of the park green space in the i-th quarter is recorded as E i ,but: If H(E)<0.5ln4, the park green space is considered to have obvious seasonal differences in the suitability of park green space capacity, and the best and worst quarters in the comprehensive environmental dimension are recorded; Step S4-4: Scheme evaluation and output; based on the evaluation dimension of the multi-scenario park green space layout scheme, the dimension is scored and evaluated, and the park green space layout scheme and quarterly characteristic evaluation under the dimension are output.

Images