A method for identifying urban functional areas based on functional mixing degree and ensemble learning

Abstract

The present invention relates to a method for identifying urban functional areas based on functional mixing and ensemble learning, and belongs to the field of digital information technology. The method performs the following steps: 1) data collection and preprocessing; 2) constructing 10 indicator features of the identification system of the urban functional area; 3) structured indicators; using spatial statistical tools to count the 10 indicator feature data corresponding to each plot; 4) independent variable data set construction; 5) response variable labeling; 6) dividing the training data set into several sub-training sets according to the functional mixing degree; 7) ensemble learning training based on the Stacking strategy; 8) attribute table connection to complete the functional area identification of the plot. The present invention separates the training set by grading the functional mixing degree and predicts the prediction set data according to the corresponding functional mixing degree, providing a method for more accurately mining the correlation between the urban functional area type and the urban characteristics, and realizing the urban characteristic mapping and urban functional area type identification.

Description

Technical Field

[0001] The invention relates to an urban functional area identification method based on functional mixing degree and integrated learning, belonging to the field of digital information technology. Background Art

[0002] We can re-explore the urban environment. The premise of planning is to first understand our existing urban functional pattern. The previous urban functional area identification method has low identification efficiency and poor dynamics. In addition, the city is composed of "land" and "people" and the "people-land relationship". The role of "people" cannot be ignored.

[0003] In recent years, the rationality of the layout of urban functional areas has been challenged.

[0004] Urban functional areas have always been the focus of urban planning. In the past, some scholars have used remote sensing methods to obtain land use, statistical surveys, etc. to divide urban functions. Although remote sensing-based methods can capture the physical changes in urban functional areas, they cannot display social and economic information related to urban functional areas; and this traditional method has a long data acquisition cycle and is greatly affected by subjective judgment.

[0005] In some previous studies, urban POI data was combined with other spatiotemporal big data, and traditional geographic analysis methods such as cluster analysis, population heat and density analysis were used to identify functional zones. The selection of thresholds is subjective, and due to the complex structure of big data, the use of traditional geographic analysis methods cannot better reveal the internal rules, and the processing efficiency is also low. In addition, few studies have used ensemble learning methods to study the identification of urban functional zones. The multi-source data used in urban functional zone identification have differences between data, and ensemble learning methods are urgently needed to reveal the complex internal mechanisms. Summary of the invention

[0006] The technical problem to be solved by the present invention is: how to better explore the correlation between urban functional area types and urban characteristics, and realize a method for mapping urban characteristics to urban functional area types.

[0007] In order to solve the above technical problems, the technical solution proposed by the present invention is: a method for identifying urban functional areas based on functional mixing degree and ensemble learning, performing the following steps:

[0008] 1) Collect and pre-process data; obtain corresponding source data according to the urban spatial differentiation and social differentiation of the city;

[0009] The preprocessing includes the division of urban functional areas, dirty data cleaning and research unit division;

[0010] The dirty data cleaning refers to removing missing, erroneous, duplicated or non-uniformly formatted data from the source data;

[0011] The urban functional area division refers to the process of dividing the city into various urban functional areas;

[0012] The research unit division is a process of forming a number of small closed plots based on the urban road network data;

[0013] 2) constructing 10 indicator characteristics of the identification system of the urban functional zones, wherein the indicator characteristics are used to describe the urban spatial differentiation or the social differentiation;

[0014] 3) Structural indicators: Use spatial statistical tools to count the 10 indicator characteristic data corresponding to each plot;

[0015] 4) Construction of independent variable data set: Establish the attribute set of the plots with the 10 indicator characteristic data corresponding to each plot in step 3), and retain the name of the corresponding plot;

[0016] 5) Response variable labeling: select some plots as training data sets and mark the corresponding functional area numbers, and use the remaining plots as prediction sets;

[0017] 6) Divide the training data set into several sub-training sets according to the functional mixing degree; classify the plots with the same or similar functional mixing degrees into the same sub-training set, and form several sub-training sets with gradually increasing or decreasing functional mixing degrees;

[0018] 7) Integrated learning training based on Stacking strategy; using RF, GBDT, SVM and BPNN as the first level of the Stacking integrated strategy model, and using XGBoost as the second level of the Stacking integrated strategy model; training each sub-training set in step 6) separately, and predicting the prediction set with the same level of functional mixing as each sub-training set;

[0019] 8) Attribute table connection: The prediction results of step 7) are linked to the corresponding plots through the names of each plot, thereby completing the functional area identification of the plot.

[0020] The improvement of the above technical solution is: the data source includes building data, ecological source data, bus stop data, subway station data, digital elevation data, online car-hailing demand data, Weibo check-in data and mobile phone signaling data; among which building data, ecological source data, bus stop data, subway station data and digital elevation data are used to describe the urban spatial differentiation of the city; online car-hailing demand data, Weibo check-in data and mobile phone signaling data are used to describe the social differentiation of the city.

[0021] The improvement of the above technical solution is: the urban functional area division refers to the land use classification standard in GBT21010-2017 "Classification of Current Land Use" and divides the urban functional areas into residential land, industrial land, public service facility land, commercial service facility land, road traffic land and other land in the first-level category; the land use types in the second-level category are adjusted, and religious funeral land, tourism land in special land and green space and parks in public service facility land are classified as other land; mining land, rural roads, pipeline transportation land, logistics and warehousing land, military facility land and embassy and consulate land are excluded.

[0022] The improvement of the above technical solution is: the dirty data cleaning refers to removing missing, erroneous, duplicated or non-uniformly formatted data from the source data obtained from the social differentiation of the city.

[0023] The improvement of the above technical solution is: using the road data of the OpenStreetMap map open platform, retaining some grade roads, and trimming unclosed sections.

[0024] The improvement of the above technical solution is that the 10 indicator characteristics are: POI use type, slope aspect, slope, building volume ratio, ecological green area, range subway coverage, range bus coverage, examples to main roads, number of arrivals of various groups of people and urban vitality.

[0025] The beneficial effects of the present invention are as follows: the present invention separates the training set by grading the functional mixing degree and allows the prediction data set to be predicted according to the corresponding functional mixing degree, thereby effectively improving the accuracy of each prediction set and providing a relatively accurate method for identifying the urban functional area type by mapping urban characteristics that are relevant to the urban functional area type and urban characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention is a flowchart of a method for identifying urban functional areas based on functional mixing degree and integrated learning according to an embodiment of the present invention.

[0027] Figure 2 It is the comparison relationship of 10 indicator characteristics of the embodiment of the present invention.

[0028] Figure 3 It is a diagram of the plot division of the study area of ​​an embodiment of the present invention.

[0029] Figure 4 It is the result of the embodiment of the present invention identifying the urban functional areas of the study area. DETAILED DESCRIPTION

[0030] Example

[0031] This example uses the central urban area of ​​Nanjing as a research area to explain how Figure 1 The method for identifying urban functional areas based on functional mixing degree and ensemble learning is shown in FIG. 1 , and performs the following steps:

[0032] 1) Collecting and preprocessing data (not shown in the figure because it is a data preparation process); obtaining corresponding source data according to the urban spatial differentiation and social differentiation of the city;

[0033] The road data provided by the Openstreetmap map open platform is used for the division of research. Building data, ecological source data, bus station data, subway station data, digital elevation data, etc. are used to describe the urban spatial differentiation of Nanjing. Online car-hailing demand data, Weibo check-in data, and mobile phone signaling data are used to describe the social differentiation phenomenon. The data type, data size, and data source of each type of data are shown in Table 1.

[0034] Table 1 Data sources

[0035]

[0036]

[0037] The preprocessing includes the division of urban functional areas, dirty data cleaning and research unit division;

[0038] The dirty data cleaning refers to removing missing, erroneous, duplicated or non-uniformly formatted data from the source data;

[0039] Dirty data refers to data that is missing, wrong, repeated, or has inconsistent formats. The process of removing dirty data from the database is called data cleaning, which is an important step in data preprocessing. With the help of Python 3.7.9's Pandas library, dirty data cleaning of mobile phone signaling data, Weibo check-in data, online car-hailing demand data, and urban points of interest data is performed. The cleaning standards are shown in Table 2.

[0040] Table 2 Mobile phone signaling data cleaning standards

[0041]

[0042] The urban functional area division refers to the process of dividing the city into various urban functional areas;

[0043] The urban functional area division refers to the land use classification standard in GBT21010-2017 "Classification of Current Land Use". The first-level category divides urban functional areas into residential land, industrial land, public service facility land, commercial service facility land, road traffic land and other land; the land use types in the second-level category are adjusted, and religious funeral land, tourism land in special land and green space and parks in public service facility land are classified as other land; mining land, rural roads, pipeline transportation land, logistics and warehousing land, military facility land and embassy and consulate land are excluded.

[0044] The research unit division is a process of forming a number of small closed plots based on the urban road network data;

[0045] Select the road data "gis_osm_railways_free" and "gis_osm_roads_free" downloaded from the Openstreetmap map open platform. Use the "fclass" attribute in the data to select roads of different levels, retain the attributes of "primary" (main road), "primary_link" (main road-connection), "secondary" (secondary road), "secondary_link" (secondary road-connection), "tertiary" (third-level road), "tertiary_link" (third-level road-branch road), "trunk" (arterial road), "trunk_link", "cycleway" (bicycle road), "motorway" (highway), "motorway_link" (highway-connection), and trim the unclosed sections. Compare with the georeferenced Sentinel-2 remote sensing image of Nanjing City, use ArcGIS vectorization tools to supplement the roads under construction in the central urban area of ​​Nanjing and improve the urban road network of Nanjing. Finally, the river network data of Nanjing City is combined with the above road network data to divide the urban land parcels of Nanjing City. Therefore, the land parcels in the central urban area of ​​Nanjing City are surrounded by urban roads and urban river networks, and are single closed plots. The division results are as follows: Figure 3 shown.

[0046] 2) constructing 10 indicator characteristics of the identification system of the urban functional zones, wherein the indicator characteristics are used to describe the urban spatial differentiation or the social differentiation;

[0047] Specifically, it starts from the six levels of use classification, natural conditions, policy restrictions, traffic conditions, behavioral activities and urban vitality, including 10 characteristic indicators: POI use type (including grade scale and influence weight), slope aspect, slope, building volume ratio, ecological green area, subway coverage, bus coverage, distance to main roads, number of arrivals of various groups of people, and urban vitality. Figure 2The corresponding relationship.

[0048] 3) Structural indicators: Use spatial statistical tools to count the 10 indicator characteristic data corresponding to each plot;

[0049] 31) Usage indicators: According to the “China’s Current Business Classification Standard” (GB T18106-2010), the scale weight is introduced to evaluate the scale and floor area of ​​the interest point. The scale weight is the median of the area range. For example, the area range of a small supermarket in the business scale standard is 200m 2 ~1999m 2 , then the scale weight is 1100m 2 .

[0050] In order to describe the influence of different points of interest, this embodiment uses the AHP hierarchical analysis method, introduces influence weights to distinguish the influence differences of different points of interest, and uses Yaahp software to construct an AHP hierarchical analysis method structure model with decision-making goals, intermediate layer elements and alternative plans. According to the comparison of the importance between each element (type of point of interest), the judgment matrix is ​​determined and constructed, and the influence weights of different types of points of interest are summarized.

[0051] Due to the large difference in the numerical values ​​of the scale weight and influence weight measurement indicators, in order to achieve a reasonable weight reconciliation process, the scale weight and influence weight are normalized, and the normalized results of the scale weight and the influence weight are added together to obtain the total weight. The formula is as follows:

[0052]

[0053]

[0054]

[0055] Where W 1i and W 2i are the scale weight and influence weight of type i interest point, and are the normalized results of the scale weight and influence weight of type i interest points, W i is the total weight of interest points of type i.

[0056] The calculated weights of various interest points are shown in Table 3.

[0057] Table 3 Interest point weight index table

[0058]

[0059]

[0060] The weighted scores of the points of interest of each plot are calculated and the proportions of the six types of functional areas in each plot are summarized. The functional area types of the plots are determined after comparison. The statistical formula is as follows:

[0061]

[0062] Where S ij is the weight score of the jth functional type in the ith plot, Q p W is the number of points of interest of the jth functional type (first-level class) in the i-th plot. p is the total weight value of the p-type interest points (third-level category) of the j-th functional type (first-level category) in the ith plot, n is the number of the p-type interest points (third-level category) of the j-th functional type (first-level category) in the ith plot; Q k is the point of interest of the kth functional type (first-level class) in the ith plot. k is the total weight value of the kth functional type (first-level class) in the ith plot, and m is the number of all points of interest in the ith plot.

[0063] F i =max{S ij} Formula 5

[0064] Among them, F i It is the weight score of the functional area type with the highest weight score on the i-th plot.

[0065] Mark the use classification of the i-th parcel as F in the attribute table of the demarcated parcel vector file i The functional zone types represented are: residential land is assigned a value of 1, industrial land is assigned a value of 2, public service facility land is assigned a value of 3, commercial service facility land is assigned a value of 4, road traffic land is assigned a value of 5, and other land is assigned a value of 6 (see Table 4).

[0066] Table 4 Functional area marking number table

[0067]

[0068] 32) Natural conditions are of great significance for distinguishing agricultural land, construction land and unused land, and also play an auxiliary role in the classification of functional areas within construction land. For example, tourism land among other land uses is generally located near mountains, rivers and lakes with large surface undulations. Selecting slope and aspect as indicators of natural conditions can distinguish some special other land uses. The slope tool and aspect tool of ArcGIS 10.3 spatial analysis tool set were used to analyze the DEM data of Sentinel 2 to obtain slope data and aspect data. The average slope and aspect values ​​of each plot were calculated, summarized and marked.

[0069] 33) Policy restrictions: Policy guidance is an important aspect of optimizing urban functional areas. In the face of irrationality in urban functional structures, the planning guidance role of policies is becoming more and more obvious. Policies help urban development by improving the suitability of urban functional spaces, and also regulate the rational layout of urban functional spaces through restrictive conditions. This embodiment selects two factors, ecologically important areas and floor area ratio, as policy restriction indicators.

[0070] (1) Ecologically important areas

[0071] In the important ecological areas of the city, construction land is restricted and prohibited from spreading. Important ecological areas have rich resources, ecology, environment and historical and cultural values, and are excellent tourism resources. Other land uses, as well as residential land, industrial land, commercial service facilities land, public service facilities land and road traffic land can be distinguished. Strict identification labels are set for plots in important ecological areas, and a new ecological restriction attribute column is created. Except for other land uses, other functional area type values ​​are not marked in ecologically important areas. The parcels within the scope are assigned the labels of other land uses (Table 4), and the attribute column is assigned to 0 for parcels outside ecologically important areas.

[0072] (2) Floor area ratio restrictions

[0073] Another important factor to consider is the floor area ratio requirement for urban construction. The floor area ratio is the core indicator of the urban functional zoning system, which refers to the ratio of the total area of ​​the above-ground buildings to the area of ​​the plot. If 3 meters is selected as the height of a single floor, the calculation formula for the floor area ratio is as follows:

[0074]

[0075] Far is the plot ratio of a certain plot of land, S bi is the bottom area of ​​the i-th building on a certain plot of land, H i is the height of the i-th building on a certain plot of land, S p The area of ​​the land.

[0076] According to the floor area ratio index standard in the "Technical Regulations on Urban Planning Management of Jiangsu Province" and the residential land balance control index standard in the "Urban Residential Area Planning and Design Code" (GB50180-93), residential land, public service facility land, commercial service facility land and industrial land can be distinguished. The scope of new and old areas is determined in the urban master plan. According to the floor area ratio requirements, a new floor area ratio attribute column is created in the attribute table, and the corresponding functional area type value of the plot is marked, as shown in Table 5.

[0077] Table 5 Floor Area Ratio Index Table

[0078]

[0079] 34) Traffic conditions determine the flow of residents and materials within the city to a certain extent. The travel radius of residents and the transportation distance of logistics have a profound impact on the internal spatial structure and spatial texture of the city, and the internal spatial structure and spatial texture of the city also determine the spatial heterogeneity of traffic conditions (Liu et al., 2012; Gong et al., 2022). Due to the difference in carrying functions, different types of functional areas have different requirements for traffic conditions. This example selects three types of traffic data, namely urban trunk roads, subway stations and bus stops, to measure the spatial heterogeneity of traffic conditions. The three traffic factors, namely the distance to urban trunk roads, subway station coverage and bus stop coverage, also drive the evolution of urban form and urban texture.

[0080] In residential land, commercial service facilities land and important public service facilities land, the coverage rate of subway stations and bus stops is higher than that of other types of functional areas, and the demand for residents' walkable transportation is also higher. The calculation of subway station coverage rate uses the buffer analysis tool in ArcGIS 10.3 to calculate the number of subway stations covered by 500-meter walking plots in the city. The calculation of bus station coverage rate also uses the buffer analysis tool in ArcGIS 10.3 to calculate the number of bus stops covered by 350-meter walking plots in the city. Industrial land has a large demand for logistics, is close to the main roads of the city, and is convenient and fast to connect with highways, which can efficiently input production materials and efficiently output production products. The distance to the main roads of the city is calculated using the distance analysis tool in ArcGIS 10.3 to analyze the Euclidean distance from the plot to the main roads of the city.

[0081] 35) Behavioral activities. In this embodiment, the mobile phone signaling data, Weibo check-in data, and online car-hailing data that provide user desensitized information attributes are all large CSV files with large data volume and complex data structure. The user portrait part is based on the Python 3.7.9 environment, calling Pandas, Shapely, Geopandas and other software packages for data cleaning, data structuring and data integration to improve processing efficiency.

[0082] The behavior of residents is mainly measured by counting the number of times three types of user groups appear on each plot. First, the data is divided by data attributes. Each type of data represents a certain type of people, and these data contain location information; secondly, the ArcGIS spatial connection tool is used to count the number of times each type of people appears on a certain plot; the final plot spatial data contains an attribute table of the number of times each type of people appears. The specific population division process of the three types of data is as follows:

[0083] The attributes of mobile phone signaling users mainly include three aspects. The first aspect is the gender of mobile phone signaling users, which includes male and female. The second aspect is the length of stay of mobile phone signaling users. Since the length of stay can distinguish between permanent residents and short-term floating population, the length of stay of mobile phone signaling users is divided into less than 7 days and more than 7 days. The third aspect is the age of mobile phone signaling users. Age can distinguish the social status of people, so the age of mobile phone signaling users is divided into less than 25 years old, 25 to 60 years old and over 60 years old. People under 25 years old are defined as school-age people, people between 25 and 60 years old are defined as working people, and people over 60 years old are retired people. According to the above attributes of mobile phone signaling users, 12 categories of mobile phone signaling users can be obtained by cross-combination, of which 7 categories are retained, as shown in Table 6, namely, resident working men (male, stay longer than 7 days, age 25 to 60 years old), resident working women (female, stay longer than 7 days, age 25 to 60 years old), non-resident working people (stay less than 7 days, age 25 to 60 years old), resident retired men (male, stay longer than 7 days, age over 60 years old), resident retired women (female, stay longer than 7 days, age over 60 years old), and school-age people (stay longer than 7 days, age under 25 years old).

[0084] The attributes of Weibo users mainly include two aspects. The first aspect is the gender of Weibo users, which includes male and female. The second aspect is the check-in time of Weibo users, which includes 7:00-20:00 and 20:00-7:00 the next day, which are respectively used as daytime check-in data and nighttime check-in data. According to the above Weibo user attributes, cross-combination can obtain 4 types of Weibo users, as shown in Table 6, namely daytime check-in male (male gender, check-in time is 7:00-20:00), daytime check-in female (male gender, check-in time is 7:00-20:00), nighttime check-in male (male gender, check-in time is 20:00-7:00 the next day), nighttime check-in female (female gender, check-in time is 20:00-7:00 the next day).

[0085] The attributes of online ride-hailing users include the time period and date of the ride. The time period is divided into 7:00-9:00, 9:00-16:00, 16:00-20:00, 21:00-24:00 and 00:00-24:00, representing the morning rush hour, working hours, evening rush hour, night and all day respectively; the date is divided into weekends and weekdays. According to the above attributes of online car-hailing users and considering the purpose of the research, 10 categories of mobile signaling users can be obtained by cross-combination, of which 6 categories of mobile signaling users are retained, as shown in Table 6, namely, the morning peak taxi-taking group (the taxi-taking period is 7:00-9:00 and the date is weekdays), the evening peak taxi-taking group (the taxi-taking period is 16:00-20:00 and the date is weekdays), the working period taxi-taking group (the taxi-taking period is 9:00-16:00 and the date is weekdays), the night taxi-taking group (the taxi-taking period is 21:00-24:00 and the date is weekdays), the weekday taxi-taking group (the taxi-taking period is 00:00-24:00 and the date is weekdays), and the weekend taxi-taking group (the taxi-taking period is 00:00-24:00 and the date is weekends).

[0086] Table 6 User portrait table

[0087]

[0088] 36) Urban vitality. Considering the dynamic differences in urban vitality during the day and at night, in order to comprehensively consider the urban vitality during the day and at night, this study selected the NPP / VIIRS night light dataset and the WorldPop population density dataset to measure urban vitality. The average night light value and the average population density value of each plot were calculated through spatial statistics. The sum of the average night light value and the average population density value was taken as the urban vitality value of each plot.

[0089] 4) Construction of independent variable data set: Establish the attribute set of the plots with the 10 indicator characteristic data corresponding to each plot in step 3), and retain the name of the corresponding plot;

[0090] According to the scale weight and influence weight provided in Table 3, the urban interest points of Nanjing are assigned. According to Formula 4, the proportion of each urban function in each plot in the central urban area of ​​Nanjing is calculated. According to Formula 5, the functional type with the highest proportion on each plot is obtained, and the plot is judged as the functional area type corresponding to the functional type. Based on the Sentinel-2 DEM data of Nanjing in July 2020, the slope and aspect data of the central urban area of ​​Nanjing are obtained by slope analysis and aspect analysis; the scope of ecological green space in the central urban area of ​​Nanjing and the calculation results of the plot ratio are obtained according to the policy restrictions; the spatial analysis obtains the measurement results of the three factors of subway station coverage, bus station coverage and distance to the main urban roads; according to the user statistical method, based on a single plot, the various groups of mobile phone signaling users, Weibo check-in users and online car-hailing users in the central urban area of ​​Nanjing are counted; according to the content of 3.3.2, the average value of night light data and population density data in a single plot is counted.

[0091] 5) Response variable labeling: select some plots as training data sets and mark the corresponding functional area numbers, and use the remaining plots as prediction sets;

[0092] Through field investigation of land use functions, visual interpretation of remote sensing images and judgment of street view maps, the functional area types of several plots (q blocks) are marked in the 25th column of the independent variable set. The functional area type labeling numbers are shown in Table 7 Functional Area Labeling Number Table, and finally a q*25 training data set is formed. The remaining plots (kq blocks) that do not contain the functional area type in the 25th column are used as prediction data sets and are not marked.

[0093] Table 7 Functional area marking number table

[0094]

[0095] 6) Divide the training data set into several sub-training sets according to the functional mixing degree; classify the plots with the same or similar functional mixing degrees into the same sub-training set, and form several sub-training sets with gradually increasing or decreasing functional mixing degrees;

[0096] The degree of functional mixing directly affects the degree of singleness of the land's functions. If the degree of functional mixing is larger, the function identification mechanism within the land will be more complex.

[0097] The degree of mixing of residential land, public service facility land, commercial service facility land, road traffic land, industrial land and other land in a single plot represents the degree of mixed land use functions of the plot. The higher the degree of mixing, the more diverse the land use functions of the plot, and the worse the ensemble learning training effect. The land use function mixing degree can be calculated using urban POI (point of interest) data. The calculation formula for the land use function mixing degree is as follows:

[0098]

[0099] Among them, H parcel is the mixed degree of land use functions of a certain plot, n is the total number of types of points of interest in the plot, P i is the ratio of the ith type of POI to the total number of POIs in the plot. The classification of POI types is shown in the table below.

[0100] Table 8 Point of interest category mapping table

[0101]

[0102]

[0103]

[0104] 7) Integrated learning training based on Stacking strategy; using RF, GBDT, SVM and BPNN as the first level of the Stacking integrated strategy model, and using XGBoost as the second level of the Stacking integrated strategy model; training each sub-training set in step 6) separately, and predicting the prediction set with the same level of functional mixing as each sub-training set;

[0105] Formula 7 was used to calculate the land use function mixing degree of the central urban area of ​​Nanjing. The land use mixing degree of the central urban area of ​​Nanjing is between 0 and 1. The training data set is divided into 12 sub-training data sets according to the size of the land use mixing degree. The land use mixing degrees in the same sub-training set are in the same range, marked as S1~S 12 , S1 to S 12 The land use mixing degree is gradually reduced, and the prediction set is also divided into 12 according to the same land use function mixing degree range, namely prediction set P1~P 12 , from P1 to P 12 The land use mixture degree gradually decreases. Geographical zoning is performed based on the land use functional mixture degree, and independent training is performed on 12 sub-training data sets and their corresponding prediction sets. The accuracy statistics of the above 12 sub-training data sets are performed in the training data set without geographical zoning, as shown in Table 9. Since the land use mixture degree reflects the complexity of the functional zone types within the plot, the urban characteristics and resident behavior patterns within the plots with the same land use mixture degree are similar. It is necessary to divide the training data set into multiple sub-training data sets according to the land use mixture degree and conduct separate training. After regional training, the accuracy is significantly improved.

[0106] It can be seen that the development in the city is uneven and the land use mixture in the city varies greatly. For the identification of urban functional zone types in large areas, it is necessary to split the training data set by using the land use mixture. The training data set is split into sub-training data sets based on the land use mixture. The land use mixture in each sub-training data set is similar. Each sub-training set is trained separately and the prediction set with the same land use mixture is predicted.

[0107] Table 9 Comparison of ensemble learning accuracy of sub-training sets with different land use mixed degrees

[0108]

[0109] 8) Attribute table connection: The prediction results of step 7) are linked to the corresponding plots through the names of each plot, thereby completing the functional area identification of the plot.

[0110] The visualization results of the identification of urban functional areas in the central urban area of ​​Nanjing are shown in the figure below. Figure 4 As shown, 2007 residential plots were identified, accounting for 34.1%; 624 industrial plots were identified, accounting for 10.6% of all plots; 1089 public service facility plots were identified, accounting for 18.5%; 1065 commercial service facility plots were identified, accounting for 18.0%; 124 road traffic plots and 986 other plots were identified, accounting for 2.1% and 16.7% of the total plots, respectively.

[0111] The improvement of the above technical solution is that the 10 indicator characteristics are: POI use type, slope aspect, slope, building volume ratio, ecological green area, range subway coverage, range bus coverage, examples to main roads, number of arrivals of various groups of people and urban vitality.

Claims

1. A method for identifying urban functional areas based on functional mixing degree and ensemble learning, characterized in that Follow these steps: 1) Data collection and preprocessing; obtaining corresponding source data according to the urban spatial differentiation and social differentiation of the city; The preprocessing includes the division of urban functional areas, dirty data cleaning and research unit division; The dirty data cleaning refers to removing missing, erroneous, duplicated or non-uniformly formatted data from the source data; The urban functional area division refers to the process of dividing the city into various urban functional areas; The research unit division is a process of forming a number of small closed plots based on the urban road network data; 2) Constructing 10 indicator characteristics of the identification system of the urban functional zones, wherein the indicator characteristics are used to describe the urban spatial differentiation or the social differentiation; 3) Structural indicators: Use spatial statistical tools to count the 10 indicator characteristic data corresponding to each plot; 4) Construction of independent variable data set: Establish the attribute set of the plots with the 10 indicator characteristic data corresponding to each plot in step 3), and retain the name of the corresponding plot; 5) Response variable labeling: select some plots as training data sets and mark the corresponding functional area numbers, and use the remaining plots as prediction sets; 6) Divide the training data set into several sub-training sets according to the functional mixing degree; classify the plots with the same or similar functional mixing degree into the same sub-training set, and form several sub-training sets with increasing or decreasing functional mixing degree step by step; 7) Integrated learning training based on Stacking strategy; four machine learning algorithms, RF, GBDT, SVM and BPNN, are used as the first level of the Stacking integrated strategy model, and XGBoost is used as the second level of the Stacking integrated strategy model; each sub-training set in step 6) is trained separately, and predictions are made for the prediction sets with the same level of functional mixing as each sub-training set; 8) Attribute table connection: Link the prediction results of step 7) to the corresponding plots through the names of each plot, so as to complete the functional area identification of the plot.

2. The urban functional area identification method based on functional mixing degree and ensemble learning according to claim 1 is characterized by: The data sources include building data, ecological source data, bus station data, subway station data, digital elevation data, online car-hailing demand data, Weibo check-in data and mobile phone signaling data; Among them, building data, ecological source data, bus station data, subway station data and digital elevation data are used to describe the urban spatial differentiation of the city; online car-hailing demand data, Weibo check-in data and mobile phone signaling data are used to describe the social differentiation of the city.

3. The urban functional area identification method based on functional mixing degree and ensemble learning according to claim 1 is characterized by: The division of urban functional areas refers to the land use classification standard in GBT21010-2017 "Classification of Current Land Use". The first-level category divides urban functional areas into residential land, industrial land, public service facility land, commercial service facility land, road traffic land and other land; the land use types in the second-level category are adjusted, and religious funeral land, tourism land in special land and green space and parks in public service facility land are classified as other land; mining land, rural roads, pipeline transportation land, logistics and warehousing land, military facility land and embassy and consulate land are excluded.

4. The urban functional area identification method based on functional mixing degree and ensemble learning according to claim 1 is characterized by: The dirty data cleaning refers to removing missing, erroneous, duplicated or non-uniformly formatted data from the source data obtained from the social differentiation of the city.

5. The urban functional area identification method based on functional mixing degree and ensemble learning according to claim 1 is characterized by: Utilize the road data of OpenStreetMap map open platform, retain some graded roads, and trim unclosed road sections.

6. The urban functional area identification method based on functional mixing degree and ensemble learning according to claim 1 is characterized by: The 10 indicator characteristics are: POI use type, slope aspect, slope, building volume ratio, ecological green space area, subway coverage, bus coverage, distance to main roads, number of arrivals of various groups of people and urban vitality.

Images