Urban park physical examination evaluation system construction method based on machine learning algorithm

Abstract

The invention relates to an urban park physical examination evaluation system construction method based on a machine learning algorithm. The method comprises the following steps: S1) selecting an evaluation index set; s2) taking a park system, an urban comprehensive park, an urban community park and an urban micro-greenbelt collaborative group as evaluation scales, and selecting evaluation indexes from the evaluation index set; s3) acquiring specific data of evaluation indexes to form an index data set; s4) screening repeated indexes with complex collinear features in all index data sets by using a variance expansion coefficient; s5) clustering by using a K-medoids clustering algorithm, and S6) based on an SHAP machine learning algorithm, performing full sample analysis and outputting the importance degree of the index factors, and constructing an urban park physical examination assessment system based on the importance degree of the index factors.According to the method, subjectivity and uncertainty of a traditional assessment method can be avoided, the assessment efficiency is improved, and rapid updating can be performed.

Description

Technical Field

[0001] The present application relates to the field of municipal planning technology, and in particular to a method for constructing an urban park physical examination and evaluation system based on a machine learning algorithm. Background Art

[0002] As an important carrier for optimizing healthy living environment and improving urban quality, urban parks are key projects in the urban physical examination framework. However, traditional urban park evaluation methods often rely on manual inspections and expert scoring, which leads to the problems of strong subjectivity and low efficiency in evaluation results. Summary of the invention

[0003] The present invention provides a method for constructing an urban park physical examination evaluation system based on a machine learning algorithm, which can evaluate urban parks from an objective perspective and improve evaluation efficiency.

[0004] The technical solution of this application is as follows: A method for constructing an urban park physical examination and evaluation system based on a machine learning algorithm comprises the following steps: S1) Selecting the evaluation indicators of urban park physical examination to form an evaluation indicator set; S2) taking the park system, urban comprehensive park, urban community park and urban micro green space coordination group as evaluation scales, selecting evaluation indicators for each of the evaluation scales in the evaluation indicator set; S3) Collect the specific data of the evaluation indicators under each evaluation scale to form an indicator data set; S4) Preprocess the data in each indicator dataset and use the variance inflation factor to screen repeated indicators with complex collinearity characteristics in all indicator datasets; S5) Use K-medoids clustering algorithm to cluster the indicators in all indicator data sets, and divide the evaluation indicators into corresponding categories based on the final clustering results and the minimum distance principle; S6) Based on the SHAP machine learning algorithm, a full sample analysis is conducted to output the importance of the indicator factors, and an urban park physical examination and evaluation system is constructed based on the importance of the indicator factors.

[0005] Furthermore, the evaluation indicators for the physical examination of urban parks included in the evaluation indicator set in step S1) include: citizen participation in public projects of urban parks, citizen participation in the construction of urban park green spaces, road greening compliance rate, park distribution uniformity, park accessibility, free opening rate of park green spaces, percentage of open area in parks, number of park volunteers (persons / 10,000 people), green space network connectivity, per capita park security, per capita community park area, per capita recreational green space, distance from the nearest residential area, road network density in the park, volunteer working hours, protection rate of ancient trees and subsequent resources of ancient trees, compliance rate of safety and risk avoidance facilities, number of safety incidents reported in the park, noise reduction, average illumination of the square, monitoring coverage, street green perception, protection rate of parks with historical value, human comfort, tree dust retention, average health of urban green spaces, winter greening Viewing rate, per capita area during high-frequency periods, all-age-friendly service level, three-dimensional green volume, self-sufficiency rate of native plants and local suitable plants in ecological garden projects, biomass density, species richness level, comprehensive species index, vitality value of the park surrounding area, number of jobs created, intelligent monitoring and management rate of park pipelines, three-dimensional data coverage rate of park information model basic platform construction, implementation rate of park system planning within the municipal area, perception of characteristic environment, marketing and promotion level, percentage of park managers with qualifications in ecology, environmental management, landscape architecture or other park management-related disciplines, certification rate of gardeners, application degree of smart facilities, level of natural science education, plant community composition, density of recreational facilities, number of cultural facilities, sanitation quality, greening management level, construction rate of road barrier-free facilities, and facility maintenance level.

[0006] Furthermore, in step S2), the evaluation indicators at the park system scale include: road greening compliance rate, park distribution uniformity, park accessibility, park green space free opening rate, per capita recreational green space, per capita park security, green space network connectivity, ancient trees and ancient trees and subsequent resource protection rate, street green perception, protection rate of parks with historical value, average health of urban green space, self-sufficiency rate of native plants and local suitable plants in ecological garden projects, species richness level, all-age friendly service level, park surrounding vitality value, park system planning implementation rate within the city, characteristic environment perception, marketing and promotion level.

[0007] The evaluation indicators at the scale of urban comprehensive parks include: citizen participation in public projects in urban parks, percentage of open area in the park, number of park volunteers, distance from the nearest residential area, road network density in the park, compliance rate of safety and risk avoidance facilities, number of safety incidents reported in the park, protection rate of ancient trees and subsequent resources of ancient trees, monitoring coverage, human comfort, average illumination of the square, noise reduction, green view rate in winter, per capita area occupied during high-frequency periods, comprehensive species index, all-age friendly service level, three-dimensional green volume, species richness level, number of jobs created, intelligent monitoring and management rate of park pipelines, three-dimensional data coverage of park information model basic platform construction, application degree of smart facilities, percentage of park managers with qualifications in ecology, environmental management, landscape architecture or other park management-related disciplines, certification rate of landscaping workers, and vitality value of the park surrounding area.

[0008] The evaluation indicators at the scale of urban community parks include: citizen participation in urban park green space construction, per capita community park area, volunteer working hours, monitoring coverage, human comfort, tree dust retention, average square illumination, noise reduction, per capita area during high-frequency periods, all-age friendly service level, three-dimensional green volume, species richness level, winter green view rate, comprehensive species index, park network pipeline intelligent monitoring and management rate, and the degree of application of smart facilities.

[0009] The evaluation indicators at the scale of urban micro-green space coordination group include: green space network connectivity, distance to the nearest residential area, road greening compliance rate, park distribution uniformity, street green perception, monitoring coverage, average square illumination, human comfort, average health of urban green space, all-age friendly service level, three-dimensional green volume, biomass density, species richness level, winter green view rate, degree of application of smart facilities, and vitality value of park surroundings.

[0010] Furthermore, step S5) specifically includes the following steps: S51) Take the specific data of the evaluation indicator as the Y-axis and the goal orientation as the X-axis, and randomly select k points as the initial center set, randomly select a point in the initial center set as the original center point, and calculate the distance between each remaining object and the original center point. The specific formula is as follows: ; In the formula, p For a given object, Cluster The center point of E is the sum of squared deviations of all objects in the data set; S52) assigning each remaining object to a cluster with the smallest distance therefrom to generate an initial clustering result; S53) Based on the initial clustering result, randomly select a non-original center point in the initial center set to replace the original center point and recalculate to generate an alternative clustering result; S54) determining the distance between the objects in each category of the alternative clustering result and the distance between the categories of the initial clustering result; if the distance between the objects in each category is the smallest and the distance to other categories is the largest, retaining the alternative clustering result and using it as a new initial clustering result for iteration; if not, restoring the original center point and continuing the iteration; S55) obtaining the clustering result with the greatest similarity of data set objects within each category and the greatest difference of data set objects between categories through a clustering evaluation function; after iterating to a maximum number of iterations, outputting a final clustering result, the final clustering result characterizing the indicator features contained in the classification.

[0011] Further, step S6) specifically includes: S61) Calculate the marginal contribution value of the indicator factor , the marginal contribution value The sum of the absolute values ​​of the SHAP values ​​of the indicator features represented by the final clustering results, and the marginal contribution value The calculation formula is as follows: In the formula, S is the subset of features used in the model, N is the set of all features in the training set, and its dimension is M ( =4); For feature combination S The model output value under ; S62) Calculate the actual SHAP value of the indicator factor to further explain the correlation between the indicator factor and its characteristics; if the SHAP value is positive, the factor has a positive impact on the evaluation result, and if the SHAP value is negative, the factor has a negative impact on the evaluation result. The specific formula is as follows: In the formula, Represents the predicted value of the sample in the decision tree, represents the explanatory model, Φ0 is the attributed value of each feature; Z i ′ indicates whether the corresponding feature of the indicator exists, Z i ′∈{0,1}; Indicates the features contained in the decision path where the sample is located; S63) Use summary_plot to characterize the importance of indicator features and perform a visual analysis of the actual SHAP values ​​of indicator factors.

[0012] Due to the adoption of the above technical solution, the beneficial effects of this application are as follows: 1. This application utilizes machine learning algorithms, combined with selected urban park physical examination and evaluation indicators, to achieve an objective evaluation of the quality of urban park construction and management, thereby avoiding the subjectivity and uncertainty of traditional evaluation methods.

[0013] 2. This application clusters indicators through the cluster center algorithm, and performs full sample analysis through the SHAP machine learning algorithm to solve key indicators. Based on the key indicators, this application can efficiently further evaluate the quality of urban park construction and management, and can significantly improve the evaluation efficiency.

[0014] 3. The evaluation system formed by a series of steps in this application has the ability to be continuously updated and optimized, can adapt to changes in urban park construction and management decisions, and maintain the advancement and accuracy of the evaluation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a tree diagram of the key indicators for the physical examination and evaluation of the park system in this application; Figure 2 This is a tree diagram of the key indicators for the physical examination and evaluation of urban comprehensive parks in this application; Figure 3 This is a tree diagram of the key indicators for the physical examination and evaluation of urban community parks in this application; Figure 4 This is a tree diagram of the key indicators for the physical examination and evaluation of the urban micro-green space collaborative group in this application; Figure 5 A Sankey diagram representing the indicators included in each evaluation scale in this application. DETAILED DESCRIPTION

[0016] Based on the background technology, in order to change the shortcomings of the traditional park physical examination and evaluation method, which is highly subjective and inefficient, the present application provides a method for constructing an urban park physical examination and evaluation system based on a machine learning algorithm, comprising the following steps: S1) Select urban park physical examination evaluation indicators to form an evaluation indicator set.

[0017] In this embodiment, the city park physical examination evaluation indicators can be obtained by manually screening, combing, and normalizing the four evaluation standards of landscape performance, park city, and green space index. The specific standards include the "Park City Evaluation Standards" issued by the Chinese Society of Landscape Architecture, the "National Garden City Evaluation Standards" and "2022 City Physical Examination Index System" issued by the Ministry of Housing and Urban-Rural Development, the "Park City Index 2022" jointly issued by the China Urban Planning Society, the Landscape Performance Series (LPS) issued by the Landscape Design Foundation, and the US City ParkScore Index, the European Global Green Space Index, the UK Green Flag Award Park Evaluation System and the Hong Kong Sustainable Urban Park Management Index, a total of 360 city park physical examination evaluation indicators. When selecting, you can set a number of goal orientations. As shown in the attached Figure 1 As shown, in this embodiment, with ecology, livability, health, comfort, safety, characteristics, vitality, sharing and harmony as the goal orientation, after indicator cleaning and normalization, 46 urban park physical examination evaluation indicators were selected to form an evaluation indicator set.

[0018] S2) Taking the park system, urban comprehensive park, urban community park and urban micro green space coordination group as evaluation scales, selecting evaluation indicators for each of the evaluation scales in the evaluation indicator set.

[0019] Taking the park system, urban comprehensive park, urban community park and urban micro green space coordination group as the evaluation scale, each indicator in the evaluation indicator set is filled into each evaluation scale, and different evaluation scales can have the same evaluation indicators. Please refer to the attached Figure 5 .

[0020] S3) Collect specific data of evaluation indicators under each evaluation scale to form an indicator data set.

[0021] This step is to collect various data of the park system urban comprehensive parks, urban community parks and micro green space coordination groups on site or through simulation calculation. In this embodiment, when collecting, the data can be roughly divided into five categories, including park space morphological pattern data, ecological regulation data, health and comfort data, social and cultural data and crowd activity data.

[0022] For spatial morphological pattern data, this embodiment uses the DJI Mavic 3 RTK multispectral aerial survey drone, multi-eye fisheye camera photography and GOSLAM RS100S handheld lidar scanning system, with Context Capture, Point Cloud Cube and other software to generate digital surface models (DSM), digital elevation models (DEM), orthophotos, point cloud SLAM models and other refined and quantifiable three-dimensional morphological data of urban parks to accurately identify the two-dimensional and three-dimensional scale and pattern of urban parks.

[0023] For ecological regulation data, this embodiment uses remote sensing images, airborne Zenmuse L1 lidar, street view pictures and meteorological micro-station data to measure the realization of ecological benefits such as vegetation diversity, cooling capacity, and carbon fixation capacity.

[0024] For health and comfort data, this embodiment uses a decibel meter, a wind and humidity meter, a panoramic camera, road network data, etc. to measure the data.

[0025] For social and cultural data, this embodiment obtains open source comment data by crawling social media data, and measures social and cultural vitality data through machine learning semantic segmentation and text analysis.

[0026] For crowd activity data, this embodiment is based on the activity-mobility system analysis method, combining mobile phone signaling data, network open source data, and urban Internet of Things systems to accurately measure residents' behavior paths and perceived preferences for urban parks.

[0027] The specific calculation methods of each indicator are described below. The evaluation indicators at each scale are shown in Tables 1 to 4.

[0028] 1. Road greening compliance rate = (L 达标 / L)×100%; Where, L 达标 It represents the length of the road that meets the greening standards, and L represents the total length of the road; The standards for road greening are: for roads with a red line width greater than 40m, the greening rate shall not be less than 25%; for roads with a red line width of 30-40m, the greening rate shall not be less than 20%; for roads with a red line width less than 30m, the greening rate shall not be less than 15%.

[0029] 2. Park distribution uniformity = l 综合 +l 社区 +l 口袋 , In the formula, l 综合 represents the comprehensive park service coverage, l 社区 represents the coverage rate of community park services, l 口袋 represents the service coverage of micro green space; The park service coverage rate refers to the ratio of the residential area covered by the park service radius to the residential area.

[0030] 3. Park accessibility: Using the network analysis method of GIS software, the accessibility of parks and green spaces of different levels is quantitatively analyzed based on three modes of transportation: walking, non-motorized vehicles and motor vehicles.

[0031] 4. Park green space free opening rate = S 免费开放 / S*100%; In the formula, S 免费开放 Indicates the area of ​​park green space open to the public free of charge.

[0032] 5. Per capita recreational green space = S 游憩 / R; In the formula, S 游憩 represents the area of ​​recreational green space, and R represents the population of built-up areas.

[0033] 6. Park security per capita = (S (A>5) / S 建成区 )×100%; Where A is the per capita park area = Park area per capita of a single type, S (A>5) is the area where the per capita park area is greater than 5 square meters, S 建成区 is the total built-up area.

[0034] 7. Citizen participation in public projects in urban parks = N 参与 / N*100%; Where N 参与 Indicates the number of public projects in urban parks in which citizens participate.

[0035] 8. Percentage of open area in the park = S 开放 / S*100%; In the formula, S 开放 Indicates the area of ​​the park open to tourists.

[0036] 9. Number of park volunteers: This reflects the citizens’ initiative to invest time and energy in helping to maintain the cleanliness, beauty and safety of the park environment. It reflects the citizens’ cohesion and social responsibility, and shows the public’s concern for social and environmental issues, as well as their support and participation in community co-construction and sharing.

[0037] 10. Distance from the nearest residential area: reflects the convenience of the park green space for citizens.

[0038] 11. Density of road network in the park = (L 一级 +L 二级 +L 三级 ) / S 公园 ; Where, L （一级 / 二级 / 三级）分别 It indicates the length of the first, second and third level paths in the park, and Spark indicates the area of ​​the park.

[0039] 12. Citizen participation in urban park green space construction = S 参与 / S*100%; In the formula, S 参与It represents the green area where citizens participate in the design, construction and management of urban park green space, and S represents the total green area.

[0040] 13. Community park area per capita = S 社区 / N; In the formula, S 社区 represents the area of ​​the community park; N represents the visitor capacity of the community park.

[0041] 14. Volunteer working hours: Statistics on the volunteer working hours of citizens in urban parks reflect the degree and willingness of citizens to contribute to society, and also reflect the importance that park management organizations attach to volunteer services.

[0042] 15. Green space network connectivity: The Network Analyst tool in the GIS software is used to analyze and calculate the connection strength and closeness between nodes (green spaces) in the green space network.

[0043] 16. Protection rate of ancient and famous trees and their subsequent resources: the percentage of ancient and famous trees and their subsequent resources protected in the built-up area (trees) to the total number of ancient and famous trees and their subsequent resources (trees) in the built-up area.

[0044] 17. Street green perception: refers to the proportion of green vegetation in the sight line of citizens walking in the city park, quantified by the green view rate, L=S 绿 / S*100%; When collecting data, a handheld GeoSLAM 3D laser scanner was selected to complete the collection of the park's 3D point cloud data. Then, the Trimble Realworks software was used to perform geographic coordinate registration, data cleaning and resampling on the original point cloud data. The point cloud data was automatically classified through the software algorithm, and then manually calibrated and classified to divide the complete point cloud data model into vegetation and other two types of point cloud data layers. Determine the normal human eye's visual height and field of vision, use the line of sight analysis function in the ArcGIS 3D analysis module to calculate the line of sight in the line of sight, clarify the specific spatial elements observed by each line of sight, and calculate the proportion of plant elements.

[0045] 18. Protection rate of parks with historical value: A protection list of parks with historical value is established. The number of parks with historical value protected in accordance with the list and protection requirements accounts for the percentage of the total number of parks with historical value included in the list.

[0046] 19. Compliance rate of safety and hazard avoidance facilities: The percentage of disaster prevention and hazard avoidance facilities in parks that meet the requirements of the "Guidelines for the Design of Disaster Prevention and Hazard Avoidance in Urban Green Spaces" to the total number of disaster prevention and hazard avoidance green spaces in the urban park disaster prevention and hazard avoidance system.

[0047] 20. Number of safety incidents reported in the park: Statistics on the number of various safety incidents reported in the park.

[0048] 21. Monitoring coverage: the ratio of the monitoring coverage area within the park to the park area.

[0049] 22. Human comfort level = (1.818b + 18.18)(0.88 + 0.002c) + (b- 32) / (45 -b)- 3.2d + 18.2 Wherein: b represents the average temperature, c represents the relative humidity, and d represents the wind speed.

[0050] 23. Average illumination of the square == lumen of all lamps * 85% / total lighting area 24. Noise reduction: Two locations were selected within the park within the green environment and the third location was selected at the edge of the green environment; the average ambient noise level was recorded at 9 a.m., noon and 5 p.m. each day, with each monitoring lasting 3 minutes; the test results at the two locations within the green environment area were averaged and compared with the monitoring results at the location at the edge of the green environment.

[0051] 25. Dust retention by trees = VW; Where V is the green amount of various plants, and W is the dust retention capacity coefficient.

[0052] 26. Winter green view rate: green area in the photo / viewing area of ​​the photo) × 100% 27. Average health of urban green space: Use NDVI to measure vegetation health, and the NDVI value of vegetation is between 0 and 1; 28. Self-sufficiency rate of native plants and local suitable plants in ecological garden projects = S 本 / S; In the formula, S 本 It indicates the number of native plants and local suitable plants seedlings in ecological garden projects.

[0053] 29. Species richness level == S / lnA; Where A is the park area and S is the number of species in the park.

[0054] 30. All-age friendly service level: It is characterized by the mixed degree of the activity population, taking into account the number of activity types and the number of participants in different types of activities. The calculation formula is: , where: m i is the activity mixing index of public space i, q j for j Number of participants in each activity type occupies public space i The total number of people participating in the activity is kFor public space i The total number of activity types within.

[0055] The activity type mixing index ranges from 0 to 1. The larger the value, the more balanced the distribution of various activities in the space and the higher the activity mixing degree. The smaller the value, the more uniform the distribution of activity types and the lower the mixing degree.

[0056] 31. Area per capita during high-frequency periods: ratio of park area to number of visitors in urban parks during 9:00-11:30 in the morning and 14:00-17:30 in the afternoon.

[0057] 32. Comprehensive species index = 1 / 3 (a1 / b1+a2 / b2+a3 / b3), where: a1 represents the number of bird species in urban parks, b1 represents the number of bird species in built-up areas, a2 represents the number of fish species in urban parks, b2 represents the number of fish species in built-up areas, a3 represents the number of plant species in urban parks, and b3 represents the number of plant species in built-up areas.

[0058] 33. Three-dimensional green volume: refers to the spatial volume occupied by the stems and leaves of all growing plants. The handheld GeoSLAM 3D laser scanner was used in conjunction with the Phantom 4 RTK drone to complete the collection of 3D point cloud data of the city park. The original point cloud data was then georeferenced, cleaned, and resampled using Trimble Realworks software. The point cloud data was automatically classified using software algorithms, and then manually calibrated and classified to divide the complete point cloud data model into two types of point cloud data layers: vegetation and others. Context Capture software was used for modeling and 3D volume was calculated.

[0059] 34. Biomass density S = (6*a1+4*a2+3*a3+2*a4+1.5*a5+1.5*a6+3*a7+2*a8+2*a9+1.5*a 10 +6*a 11 ) / b; In the formula: S represents the biomass density index (g / ㎡), a1 represents the forest area with understory vegetation, a2 represents the forest area without understory vegetation, a3 represents the shrub area, a4 represents the perennial plant area, a5 represents the sand plant area, a6 represents the biennial plant area, a7 represents the artificial moss area with a height greater than 7 cm, a8 represents the artificial moss area with a height less than 7 cm, a9 represents the natural grassland area with a height greater than 21 cm, a 10 Indicates the area of ​​natural grass with a height less than 21 cm, a 11 represents the wetland area, and b represents the total park area.

[0060] 35. Park surrounding vitality value = The density is centered on the city park, and the density of nine types of POIs in the surrounding areas, including catering and accommodation, commercial retail, life services, sports and leisure, science and education, culture, finance, medical care, public facilities, and transportation, are evaluated; 36. Implementation rate of park system plans formulated within the city: the ratio of the number of park system plans formulated and implemented within the city to the total number; 37. Characteristic environmental perception: With the help of street view pictures with location information, semantic information from social media, etc., combined with machine learning data sets such as intelligent recognition and image segmentation formed by perception crowdsourcing, a large-scale perception quantification of space is formed; The open source comments on urban parks on social platforms such as Qunar.com, Ctrip.com, and Dianping.com were crawled using machine learning web crawlers. For text information, the initial text database was cleaned using word frequency analysis, and the text content was coded and analyzed using the qualitative analysis tool NVivo 11. For image information, the R-based qualitative data package RQDA was used to batch parse and code the image content, and then the R-based statistical program was used to perform descriptive mathematical statistics on the code.

[0061] 38. Marketing and promotion level: Use machine learning web crawlers to crawl the park’s search volume and comment volume on social media for evaluation; 39. Number of jobs created: refers to the number of jobs created in the process of ecological landscape construction and operation. Temporary workers: the number of temporary workers set up for construction sites, seasonal operations or other temporary needs; regular workers: permanent positions directly created for operating sites.

[0062] 40. Intelligent monitoring and management rate of park pipe network = length of park pipe network that is intelligently monitored and managed by technologies such as the Internet of Things / total length of park pipe network. "Intelligent monitoring" means monitoring the operation status of water and electricity, safety warning, water quality monitoring, etc. The weight of each of the three assessment contents is 1 / 2. 41. 3D data coverage of the park information model basic platform construction: the digital ratio of the physical and functional characteristics of the park's underground pipe network and ground elements; 42. Application degree of smart facilities: Application rate of smart facilities = number of smart control devices installed in the park / total number of public equipment in the park * 100% 43. Percentage of park managers with relevant qualifications in ecology, environmental management, landscape architecture or other park management related disciplines: The percentage of managers with relevant qualifications to the total number of park managers (people).

[0063] 44. Rate of certified landscaping workers: The percentage of certified personnel in landscaping projects to the total number of technical workers in the project.

[0064] 45. Comprehensive function of park green space: expressed by the sum of the density of landscape facilities, service facilities, rest facilities and entertainment facilities.

[0065] 46. ​​Degree of openness of park boundary: Degree of openness of park boundary = length of open park boundary / length of park boundary*100%.

[0066] S4) Preprocess the data in each indicator dataset and use the variance inflation factor to screen out repeated indicators with complex collinearity characteristics in all indicator datasets.

[0067] This step can be calculated by calling the variance_inflation_factor in the machine learning Statsmodels library.

[0068] S5) Use the K-medoids clustering algorithm to cluster the indicators in all indicator data sets, and divide the evaluation indicators into corresponding categories based on the clustering results and the minimum distance principle.

[0069] Step S5) specifically includes the following steps: S51) With the specific data of the evaluation index as the Y-axis and the goal orientation as the X-axis, k points are randomly selected as the initial center set, and a point is randomly selected from the initial center set as the original center point, and the distance between each remaining object and the original center point is calculated. The specific formula is as follows: ; In the formula, p For a given object, Cluster The center point of E is the sum of squared deviations of all objects in the data set; In this embodiment, the number of goal orientations is set to 3 to 6. Too few goal orientations will result in unclear indicator classification, while too many goal orientations will result in weak boundaries of indicator classification.

[0070] Finally, after iteration of this embodiment, the clustering result is the best when k = 4. According to the meaning of the indicators contained in the clustering results, this embodiment names the four clusters as: open sharing, safe and comfortable, healthy and energetic, and diversified innovation.

[0071] S52) assigning each remaining object to a cluster with the smallest distance therefrom to generate an initial clustering result; S53) Based on the initial clustering result, randomly select a non-original center point in the initial center set to replace the original center point and recalculate to generate an alternative clustering result; S54) determining the distance between the objects in each category of the alternative clustering result and the distance between the categories of the initial clustering result; if the distance between the objects in each category is the smallest and the distance to other categories is the largest, retaining the alternative clustering result and using it as a new initial clustering result for iteration; if not, restoring the original center point and continuing the iteration; S55) obtaining the clustering result with the greatest similarity of data set objects within each category and the greatest difference of data set objects between categories through a clustering evaluation function; after iterating to a maximum number of iterations, outputting a final clustering result, the final clustering result characterizing the indicator features contained in the classification.

[0072] After clustering, scale division is performed. In this embodiment, four scales are divided into park system, urban comprehensive park, urban community park scale, and urban micro green space coordination group. The scale is used as the criterion layer, the clustering result is used as the target layer, and the specific indicator type is used as the indicator layer. The details are as follows: The evaluation indicators at the park system scale include: road greening compliance rate, park distribution uniformity, park accessibility, park green space free opening rate, per capita recreational green space, per capita park security, green space network connectivity, ancient trees and famous trees and subsequent resource protection rate of ancient trees, street green perception, protection rate of parks with historical value, average health of urban green space, self-sufficiency rate of native plants and local adaptive plants in ecological garden projects, species richness level, all-age friendly service level, comprehensive functions of park green space, vitality value of park surroundings, implementation rate of park system planning within the urban area, perception of characteristic environment, marketing and promotion level, a total of 20 indicators.

[0073] The evaluation indicators at the scale of urban comprehensive parks include: citizen participation in public projects in urban parks, percentage of open area in the park, number of park volunteers, distance from the nearest residential area, road network density in the park, degree of openness of park boundaries, compliance rate of safety and risk avoidance facilities, number of safety incidents reported in the park, protection rate of ancient trees and subsequent resources of ancient trees, monitoring coverage, human comfort, average illumination of the square, noise reduction, green view rate in winter, per capita area during high-frequency periods, comprehensive functions of park green space, comprehensive species index, all-age friendly service level, three-dimensional green volume, species richness level, number of jobs created, intelligent monitoring and management rate of park pipelines, three-dimensional data coverage of park information model (IIM) basic platform construction, application degree of smart facilities, percentage of park managers with qualifications in ecology, environmental management, landscape architecture or other park management-related disciplines, certification rate of landscaping workers, and vitality value of the park surrounding area, a total of 27 indicators.

[0074] The evaluation indicators at the scale of urban community parks include: citizen participation in urban park green space construction, per capita community park area, volunteer working hours, park boundary openness, monitoring coverage, human comfort, tree dust retention, average square illumination, noise reduction, per capita occupied area during high-frequency periods, all-age friendly service level, three-dimensional green volume, species richness level, comprehensive functions of park green space, winter green view rate, comprehensive species index, intelligent monitoring and management rate of park pipelines, and the degree of application of smart facilities, a total of 18 indicators.

[0075] The evaluation indicators at the scale of urban micro green space coordination group include: green space network connectivity, distance to the nearest residential area, road greening compliance rate, park distribution uniformity, citizen participation in urban park green space construction, street green perception, monitoring coverage, average square illumination, human comfort, average health of urban green space, all-age friendly service level, three-dimensional green volume, biomass density, species richness level, winter green view rate, degree of application of smart facilities, and vitality value of park surroundings, a total of 17 indicators.

[0076] S6) Based on the SHAP machine learning algorithm, a full sample analysis is conducted to output the importance of the indicator factors, and an urban park physical examination and evaluation system is constructed based on the importance of the indicator factors.

[0077] Step S6) specifically includes: S61) Calculate the marginal contribution value of the indicator factor , the marginal contribution value The sum of the absolute values ​​of the SHAP values ​​of the indicator features represented by the final clustering results, and the marginal contribution value The calculation formula is as follows: In the formula, S is the subset of features used in the model, N is the set of all features in the training set, and its dimension is M; It is the model output value under the feature combination S.

[0078] In this embodiment, M is 4. Considering the order, these four features have a total of 4 ! Combination of fixed features i , then the remaining Combination of.

[0079] S62) Calculate the actual SHAP value of the indicator factor to further explain the correlation between the indicator factor and its characteristics; if the SHAP value is positive, the factor has a positive impact on the evaluation result, and if the SHAP value is negative, the factor has a negative impact on the evaluation result. The specific formula is as follows: In the formula, Represents the predicted value of the sample in the decision tree, represents the explanation model, Φ0 is the attribution value of each feature; Zi′ indicates whether the corresponding feature of the indicator exists, Zi′∈{0,1}; It represents the characteristics contained in the decision path of the sample; in this embodiment, the decision path is health and vitality, safety and comfort, open sharing, and diversified innovation.

[0080] S63) Use summary_plot to characterize the importance of indicator features and perform a visual analysis of the actual SHAP values ​​of indicator factors.

[0081] In this embodiment, after the importance of the indicator characteristics is output, the indicators with high importance are further screened out according to the number of indicators as the final specific indicators for the evaluation of urban parks. In this embodiment, the physical examination evaluation indicator system of the urban park system removes the two indicators of human comfort and marketing and promotion level on the basis of the 20 indicators after clustering, and finally retains 18 indicators with high importance, such as park accessibility, park distribution uniformity, and per capita park security. The physical examination evaluation indicator system of urban comprehensive parks removes the three indicators of the percentage of open area in the park, the comprehensive species index, and the number of jobs created on the basis of the 27 indicators after clustering, and finally retains 24 indicators with high importance, such as the comprehensive function of park green space, monitoring coverage, and per capita occupied area during high-frequency periods. The physical examination evaluation indicator system of urban community parks removes the two indicators of volunteer working hours and comprehensive species index on the basis of the 18 indicators after clustering, and finally retains 16 indicators with high importance, such as tree dust retention, all-age friendly service level, and per capita community park area. The physical examination and evaluation index system of the urban micro-green space coordination group removed one index of human comfort from the 17 indexes after clustering, and finally retained 16 highly important indexes such as green space network connectivity, street green perception, and road greening compliance rate. The physical examination and evaluation index system of the urban park system, urban comprehensive park, urban community park, and urban micro-green space coordination group was constructed, and finally the urban park physical examination and evaluation system was constructed.

[0082] Anything not described in this application can be achieved by adopting or drawing on existing technologies. The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for constructing an urban park physical examination evaluation system based on a machine learning algorithm, characterized in that: The following steps are involved: S1) Selecting the evaluation indicators of urban park physical examination to form an evaluation indicator set; S2) taking the park system, urban comprehensive park, urban community park and urban micro green space coordination group as evaluation scales, selecting evaluation indicators for each of the evaluation scales in the evaluation indicator set; S3) Collect the specific data of the evaluation indicators under each evaluation scale to form an indicator data set; S4) Preprocess the data in each indicator dataset and use the variance inflation factor to screen repeated indicators with complex collinearity characteristics in all indicator datasets; S5) Use K-medoids clustering algorithm to cluster the indicators in all indicator data sets, and divide the evaluation indicators into corresponding categories based on the final clustering results and the minimum distance principle; S6) Based on the SHAP machine learning algorithm, a full sample analysis is conducted to output the importance of the indicator factors, and an urban park physical examination and evaluation system is constructed based on the importance of the indicator factors.

2. According to the method for constructing a city park physical examination evaluation system based on a machine learning algorithm according to claim 1, it is characterized in that: The urban park physical examination evaluation indicators included in the evaluation indicator set in step S1) include: citizen participation in urban park public projects, citizen participation in urban park green space construction, road greening compliance rate, park distribution uniformity, park accessibility, park green space free opening rate, percentage of open area in the park, number of park volunteers (person / 10,000 people), green space network connectivity, per capita park protection, per capita community park area, per capita recreational green space, distance from the nearest residential area, road network density in the park, volunteer working hours, ancient trees and ancient trees and subsequent resource protection rate, safety and risk avoidance facility compliance rate, number of safety incidents reported in the park, noise reduction, average square illumination, monitoring coverage, street green perception, protection rate of parks with historical value, human comfort, tree dust retention, average health of urban green space, winter green view rate, Per capita area during high-frequency periods, all-age-friendly service level, three-dimensional green volume, self-sufficiency rate of native plants and local suitable plants in ecological garden projects, biomass density, species richness level, comprehensive species index, vitality value of the park surrounding area, number of jobs created, intelligent monitoring and management rate of park pipelines, three-dimensional data coverage rate of park information model basic platform construction, implementation rate of park system planning within the municipal area, perception of characteristic environment, marketing and promotion level, percentage of park managers with qualifications in ecology, environmental management, landscape architecture or other park management-related disciplines, certification rate of gardeners, degree of application of smart facilities, level of natural science education, plant community composition, density of recreational facilities, number of cultural facilities, sanitation quality, greening management level, construction rate of road barrier-free facilities, and facility maintenance level.

3. The method for constructing an urban park physical examination evaluation system based on a machine learning algorithm according to claim 2 is characterized in that: In step S2), the evaluation indicators at the park system scale include: road greening compliance rate, park distribution uniformity, park accessibility, free opening rate of park green space, per capita recreational green space, per capita park security, green space network connectivity, ancient trees and subsequent resource protection rate of ancient trees, street green perception, protection rate of parks with historical value, average health of urban green space, self-sufficiency rate of native plants and local suitable plants in ecological garden projects, species richness level, all-age friendly service level, park surrounding vitality value, city-wide park system planning implementation rate, characteristic environment perception, marketing and promotion level; The evaluation indicators at the scale of urban comprehensive parks include: citizen participation in public projects in urban parks, percentage of open area in the park, number of park volunteers, distance from the nearest residential area, road network density in the park, compliance rate of safety and risk avoidance facilities, number of safety incidents reported in the park, protection rate of ancient trees and subsequent resources of ancient trees, monitoring coverage, human comfort, average illumination of the square, noise reduction, winter green view rate, per capita area during high-frequency periods, comprehensive species index, all-age friendly service level, three-dimensional green volume, species richness level, number of jobs created, intelligent monitoring and management rate of park pipe network pipelines, three-dimensional data coverage rate of park information model basic platform construction, application degree of smart facilities, percentage of park managers with relevant qualifications in ecology, environmental management, landscape architecture or other park management-related disciplines, certification rate of landscape workers, and vitality value of the park surroundings; The evaluation indicators at the scale of urban community parks include: citizen participation in the construction of urban park green spaces, per capita community park area, volunteer working hours, monitoring coverage, human comfort, tree dust retention, average square illumination, noise reduction, per capita area during high-frequency periods, all-age friendly service level, three-dimensional green volume, species richness level, winter green view rate, comprehensive species index, park network pipeline intelligent monitoring and management rate, and application level of smart facilities; The evaluation indicators at the scale of urban micro-green space coordination group include: green space network connectivity, distance to the nearest residential area, road greening compliance rate, park distribution uniformity, street green perception, monitoring coverage, average square illumination, human comfort, average health of urban green space, all-age friendly service level, three-dimensional green volume, biomass density, species richness level, winter green view rate, degree of application of smart facilities, and vitality value of park surroundings.

4. The method for constructing a city park physical examination evaluation system based on a machine learning algorithm according to claim 3 is characterized in that: Step S5) specifically includes the following steps: S51) Take the specific data of the evaluation indicator as the Y-axis and the goal orientation as the X-axis, and randomly select k points as the initial center set, randomly select a point in the initial center set as the original center point, and calculate the distance between each remaining object and the original center point. The specific formula is as follows: ； In the formula, p For a given object, Cluster The center point of E is the sum of squared deviations of all objects in the data set; S52) assigning each remaining object to a cluster with the smallest distance therefrom to generate an initial clustering result; S53) Based on the initial clustering result, randomly select a non-original center point in the initial center set to replace the original center point and recalculate to generate an alternative clustering result; S54) determining the distance between the objects in each category of the alternative clustering result and the distance between the categories of the initial clustering result; if the distance between the objects in each category is the smallest and the distance to other categories is the largest, retaining the alternative clustering result and using it as a new initial clustering result for iteration; if not, restoring the original center point and continuing the iteration; S55) obtaining the clustering result with the greatest similarity of data set objects within each category and the greatest difference of data set objects between categories through a clustering evaluation function; after iterating to a maximum number of iterations, outputting a final clustering result, the final clustering result characterizing the indicator features contained in the classification.

5. The method for constructing a city park physical examination evaluation system based on a machine learning algorithm according to claim 4 is characterized in that: Step S6) specifically includes: S61) Calculate the marginal contribution value of the indicator factor , the marginal contribution value The sum of the absolute values ​​of the SHAP values ​​of the indicator features represented by the final clustering results, and the marginal contribution value The calculation formula is as follows: In the formula, S is the subset of features used in the model, N is the set of all features in the training set, and its dimension is M ( =4); For feature combination S The model output value under ; S62) Calculate the actual SHAP value of the indicator factor to further explain the correlation between the indicator factor and its characteristics; if the SHAP value is positive, the factor has a positive impact on the evaluation result, and if the SHAP value is negative, the factor has a negative impact on the evaluation result. The specific formula is as follows: In the formula, Represents the predicted value of the sample in the decision tree, represents the explanatory model, Φ0 is the attributed value of each feature; Z i ′ indicates whether the corresponding feature of the indicator exists, Z i ′∈{0,1}; Indicates the features contained in the decision path where the sample is located; S63) Use summary_plot to characterize the importance of indicator features and perform a visual analysis of the actual SHAP values ​​of indicator factors.

Images