A housing area plate division method based on spatiotemporal big data

Through spatiotemporal big data and artificial intelligence technology, the value index of housing characteristic attributes is calculated and clustered, which solves the problem of poor division of housing regional sectors in the existing technology, and achieves more accurate housing market supervision and policy evaluation.

CN114331084BActive Publication Date: 2024-11-19WUHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111586734.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-23
Publication Date
2024-11-19
Estimated Expiration
2041-12-23

AI Technical Summary

Technical Problem

The existing housing regional sector division technology is based on prior knowledge, and it is difficult to ensure the homogeneity of housing quality and service functions within the sector, and it cannot effectively reflect the differences in housing consumers' willingness to pay.

Method used

Using the housing characteristic value matrix calculation method based on spatiotemporal big data and artificial intelligence, the housing characteristic attribute value index of each housing unit is calculated through spatiotemporal geographic weighted regression modeling and affinity propagation clustering algorithm, and clustering from bottom to top Dividing the housing sector.

Benefits of technology

It improves the effectiveness and robustness of the division of housing regional sectors, can better reflect the differences in housing attribute payment intentions, and is suitable for housing market supervision and policy evaluation in the context of large-scale urban development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114331084B_ABST
    Figure CN114331084B_ABST
Patent Text Reader

Abstract

The present invention provides a housing regional plate division method based on spatiotemporal big data, and establishes a new method framework by introducing new indicators and new technologies: first, multi-source high-granularity spatiotemporal big data is processed and matched, and then new variables are introduced on the basis of the classic housing price model to indicate new housing consumption concepts, and then the spatiotemporal geographic weighted regression method is used to calculate the housing characteristic value matrix while considering the functional proximity and spatiotemporal proximity of housing characteristic values. Based on this, the affinity propagation algorithm in the field of artificial intelligence clustering is used to cluster housing in the potential and robustness of mining multidimensional large-scale data to obtain housing plates. The present invention helps enterprises and governments to monitor and manage the housing market in a timely, efficient and accurate manner under the background of large-scale and rapid urban development, and promotes the "digital intelligence" development of China's real estate technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of regional plate division, and in particular to a housing regional plate division method based on spatiotemporal big data. Background Art

[0002] Housing sectorization involves dividing a city into economically meaningful, decomposable units and housing regions into groups of interchangeable housing units, characterized by high homogeneity within each unit and high heterogeneity between units. For individuals, this helps clarify the supply structure of the existing housing market, broaden the differences in housing demand across sectors, and reduce search costs and transaction friction for consumers in matching supply and demand. For governments and businesses, this sectorization is crucial for deepening existing housing market monitoring, gaining insights into changes in housing market demand, improving the accuracy of housing price estimates, and accurately evaluating the effectiveness of housing policy implementation.

[0003] Existing housing sector demarcation techniques are mostly based on prior knowledge, using existing geographic boundaries, such as administrative district boundaries and major urban roads, to segment the entire urban area. This demarcation method is simple and easy to describe, but when choosing a home, consumers not only consider locational characteristics, such as the administrative district, but also consider housing quality and neighborhood environment, such as the living environment of the residential complex and the quality of nearby schools. Therefore, simply demarcating housing sectors based on prior knowledge cannot guarantee functional homogeneity within the sector, such as housing quality and available housing services. Summary of the Invention

[0004] The present invention proposes a housing area plate division method based on spatiotemporal big data, adopts a housing characteristic value matrix calculation method based on spatiotemporal big data and spatiotemporal geographic weighted regression modeling, and a housing characteristic value clustering method based on artificial intelligence, which solves the technical problem of poor division effect existing in the methods of the prior art.

[0005] In order to solve the above technical problems, the present invention provides a housing area plate division method based on spatiotemporal big data, comprising:

[0006] S1: Preprocessing and matching housing transaction data and micro-geographic big data before importing them into spatiotemporal big data. Micro-geographic big data includes housing attribute data, road network data, and point of interest data. Spatiotemporal big data includes the matched housing transaction data and road network data.

[0007] S2: Introducing new variables into the pre-set global housing price regression model, based on the spatiotemporal database, quantifying the potential influencing factors of housing prices for each housing unit, and constructing an independent variable matrix;

[0008] S3: Based on the matched housing transaction data and road network data in step S1, calculate the spatiotemporal distance and construct a spatiotemporal distance matrix. Use the spatiotemporal geographic weighted regression algorithm to consider the spatiotemporal dependence of housing prices. Embed the spatiotemporal distance matrix into the preset global housing price regression model in step S2. Then use the weighted least squares method to perform point-by-point parameter estimation of housing units to obtain the housing characteristic attribute value index of each housing unit, and construct the housing characteristic attribute value index matrix.

[0009] S4: Based on the value matrix of housing influencing factors of each housing unit obtained in step S3, an artificial intelligence clustering algorithm is used to aggregate from bottom to top starting from each residential unit point to obtain a housing unit group, which is a housing sector.

[0010] In one embodiment, in step S1, the housing transaction data includes housing transaction price, transaction time, transaction area, housing name, and housing address;

[0011] The original format of housing attribute data includes housing name, housing address, housing construction year, property management, volume ratio, and greening ratio;

[0012] The attributes of road network data include road names and road attributes;

[0013] The attributes of the POI data include POI name, POI address, and POI category.

[0014] In one embodiment, step S1 pre-processes and matches housing transaction data and micro-geographic big data, including:

[0015] The housing transaction data is parsed for geographical names and addresses to obtain geographic coordinates, and the average transaction price at the community level is calculated to obtain the average transaction price data;

[0016] Perform place name and address resolution on housing attribute data to obtain geographic coordinates, and match the geographic coordinates with housing transaction data;

[0017] Filter the road network data, remove non-walkable roads such as highways and expressways based on road attributes, build an initial road network, perform topology checks, and obtain a road network dataset;

[0018] Perform place name and address parsing on the POI data to obtain the geographic coordinates of the POI and classify them based on the POI name and category;

[0019] The average transaction price data, road network dataset and classified point of interest data are spatially matched and imported into a spatiotemporal database for storage and management.

[0020] In one embodiment, S2 includes:

[0021] S2.1: Obtain a pre-set global housing price regression model, with influencing factors including community attributes, transportation convenience, location, education and medical care, open space, commercial facilities, and job opportunities;

[0022] S2.2: Based on the existing influencing factors of the pre-set global housing price regression model, new variables are introduced, including the quality of the primary school district, the quality of the secondary school district, the distance to the road network of tertiary hospitals, the density of high-tech companies, and the Shannon diversity index of companies;

[0023] S2.3: Based on the spatiotemporal database in step 1, quantify the selected variables. The quantification methods include quantification of nominal variables and road network analysis.

[0024] S2.4: Perform z-score standardization on all influencing factors and housing transaction prices;

[0025] S2.5: Perform significance tests on all potential influencing factors processed in step S2.4;

[0026] S2.6: Perform collinearity diagnosis on all potential influencing factors processed in step S2.5;

[0027] S2.7: Establish an independent variable matrix based on all potential influencing factors processed in step S2.6.

[0028] In one embodiment, S3 includes:

[0029] S3.1: Based on the housing transaction data and road network data matched in step S1, calculate the spatiotemporal distance using the following formula:

[0030]

[0031] in, is the space-time distance between point i and point j; λ and μ are weight parameters for balancing space-time units; u, v, t are longitude, latitude and time respectively; u i and u j Represents the longitude coordinates of point i and point j respectively, v i and v j Represents the latitude coordinates of point i and point j respectively, t i and t j Respectively represent the time of point i and point j;

[0032] S3.2: Based on the spatiotemporal distance, construct a spatiotemporal weight matrix W between two observation points. W is a diagonal matrix. The diagonal elements represent the weights of the mutual influence between the two observation points, and the other elements are 0. Use cross-validation to select the optimal bandwidth.

[0033] S3.3. Based on the independent variable matrix obtained in step S2 and the matched housing transaction data in step S1, a spatiotemporal geographic weighted regression model is established. The formula is as follows:

[0034]

[0035] Among them, y i is the i-th observation point (u i ,v i ,t i ) is the explained variable, namely housing price; x ik is the independent variable matrix, i.e., the potential influencing factor k at observation point i; I and K are the number of observation points and the number of potential factors, respectively; β k (u i ,v i ,t i ) is the spatiotemporal geographic weighted regression standardization coefficient, that is, the unknown coefficient at the i-th observation point; β0(u i , v i , t i ) is a constant term, i.e., a constant term at the i-th observation point; ε i is the error term;

[0036] S3.4, use the weighted least squares estimation method to estimate the parameters, specifically: at the i-th observation point (u i , v i , t i ) when the independent variable x k Corresponding regression coefficient Using the weighted least squares estimation method, the regression coefficients are estimated by minimizing the sum of squared residuals:

[0037]

[0038] The parameter estimates are expressed as a matrix, and the regression coefficients are:

[0039]

[0040] in, represents the weight of the i-th observation point, X represents the independent variable matrix, Y represents the explained variable matrix, X T represents the transpose of the independent variable matrix;

[0041] S3.5: Based on the spatiotemporal geographical weighted regression coefficient β obtained in step S3.4 k (u i , v i , t i ), eliminate the residential unit entries whose coefficients do not meet the preset conditions, and obtain the housing characteristic attribute value index matrix, which is the i-th observation point after processing (ui , v i , t i ) is the regression coefficient of the kth housing influencing factor.

[0042] In one embodiment, step S4 includes:

[0043] S4.1: Based on the housing attribute value index matrix obtained in step S3, calculate the similarity matrix between the two points The negative Euclidean distance between any two sample points is used to measure the ability of point j to be the cluster center of point i, x ik represents the value of the k-th housing characteristic attribute value index of point i, x jk represents the value of the k-th housing characteristic attribute value index at point j;

[0044] S4.2: Initialize the reference degree to the median of the similarities, where the reference degree is a real number s(i, i), which refers to the reference degree of point i as the cluster center. The value is the value on the diagonal of the S matrix. The larger the value of s(i, i), the more likely i is the cluster center.

[0045] S4.3: Calculate the attraction matrix r(i, j) and the attribution matrix a(i, j). The attraction matrix r(i, j) refers to the degree to which point j is suitable as the cluster center of data point i. The attribution matrix a(i, j) refers to the degree to which point i chooses point j as its cluster center. The larger the sum of r(i, j) and a(i, j), the greater the probability that point j will become the final cluster center, and the greater the possibility that point i belongs to the cluster with point j as the cluster center. Initialize the attraction matrix and the attribution matrix to 0.

[0046] S4.4: Update the attraction and belongingness of each point iteratively. The termination condition of the iteration is that the cluster center is no longer updated or the maximum number of iterations is reached. The iteration formula is as follows:

[0047]

[0048] Where i′ and j′ represent the updated points i and j; s(i, j′) represents the similarity matrix after updating point j; a(i, j′) represents the attribution matrix after updating point j; r(i′, j) represents the attraction matrix after updating point i; a(j, j) and r(j, j) represent the attribution and attraction of point j as the cluster center, respectively.

[0049] S4.5. Based on the cluster centers obtained in step S4.4, the sample points are classified and finally clusters are obtained, namely the housing sector.

[0050] The above one or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:

[0051] (1) Using spatiotemporal geographic weighted regression, a value matrix of a series of housing influencing factors for each housing unit is calculated based on both functional proximity (functional homogeneity) and spatiotemporal proximity. This matrix serves as the basis for the division of housing regions. Compared with directly using the statistical characteristics of the influencing factors, it can better reflect the differences in residents' willingness to pay for different housing attributes.

[0052] (2) The use of artificial intelligence clustering algorithms reduces manual intervention and improves the robustness of clustering algorithms. Specifically, compared to traditional K-means clustering algorithms, the affinity propagation algorithm does not require the pre-specified number of clusters and can quickly process large-scale sparse data sets with fewer parameter settings. The squared error of the clustering results is also smaller. Therefore, the method of the present invention improves the partitioning effect.

[0053] Furthermore, the affinity propagation algorithm starts at each point, treating all points as potential sample sets and aggregating data from the bottom up. This aligns more closely with the logic used by offline real estate agents when selecting sites and by real estate experts when categorizing real estate sectors. This algorithm considers the attributes of each community before clustering, rather than first examining the overall landscape before segmenting. Therefore, within the context of large-scale, rapid urban development, this invention can efficiently monitor housing market dynamics, evaluate the capitalization effects of public service facilities, and assess housing policies. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is a flow chart of a housing area segmentation method according to an embodiment of the present invention;

[0056] Figure 2 Schematic diagram of spatiotemporal geographic weighted regression coefficients in an embodiment of the present invention;

[0057] Figure 3 Schematic diagram of a housing sector in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] Through extensive research and practice, the inventors of this application have discovered that in order to divide a housing area into multiple housing blocks with homogeneous housing characteristics, the following two steps are essentially required: the first step is to reveal the different preferences and willingness-to-pay differences of housing consumers for a series of housing characteristic attributes, score the importance of a series of housing characteristic attributes, and obtain a series of housing characteristic attribute value index matrices for each house; the second step is to cluster all houses based on the value index matrix to obtain a cluster of housing unit groups with high homogeneity within the group. In addition, two adjacent houses are more likely to be classified into the same group than two houses that are far apart. At the same time, housing has durability, and the value of housing characteristics will be affected by changes over time. Therefore, how to simultaneously consider functional homogeneity and spatiotemporal proximity is the key technical difficulty in achieving scientific and effective subdivision of housing areas.

[0059] To address the difficulties of existing technologies, this paper establishes a novel methodological framework by introducing new indicators and technical approaches, proposing a method for delineating housing regional sectors. This method effectively leverages the advantages of multi-source, multi-modal, and highly granular spatiotemporal big data in accurately and precisely indicating urban development trends and housing market changes. It introduces new independent variables into existing housing price models to capture new housing consumption trends in the context of rapid urbanization. It then scientifically and rationally utilizes spatiotemporal geographically weighted regression to establish the housing price model, deriving a matrix of housing attribute value indices for each house while also considering the spatiotemporal proximity between houses. More importantly, this method leverages the potential of the affinity propagation algorithm, a popular AI clustering algorithm, in mining multidimensional, large-scale data, along with its inherent robustness and generalization capabilities, to intelligently and efficiently cluster all houses based on the housing attribute value matrix, ultimately achieving a scientific and effective subdivision of housing regional sectors. This invention helps businesses and governments monitor and manage the housing market in a timely, efficient, and accurate manner within the context of large-scale and rapid urban development, thereby promoting the digitalization and intelligence of China's real estate technology sector.

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0061] An embodiment of the present invention provides a method for dividing housing area blocks based on spatiotemporal big data, including:

[0062] S1: Pre-processing and matching housing transaction data and micro-geographic big data, and then importing them into the spatio-temporal big database. The micro-geographic big data includes housing attribute data, road network data, and point of interest data. The spatio-temporal big database includes the matched housing transaction data and road network data.

[0063] S2: Introducing new variables into the pre-set global housing price regression model, based on the spatiotemporal database, quantifying the potential influencing factors of housing prices for each housing unit, and constructing an independent variable matrix;

[0064] S3: Based on the matched housing transaction data and road network data in step S1, calculate the spatiotemporal distance and construct a spatiotemporal distance matrix. Use the spatiotemporal geographic weighted regression algorithm to consider the spatiotemporal dependence of housing prices. Embed the spatiotemporal distance matrix into the preset global housing price regression model in step S2. Then use the weighted least squares method to perform point-by-point parameter estimation of housing units to obtain the housing characteristic attribute value index of each housing unit, and construct the housing characteristic attribute value index matrix.

[0065] S4: Based on the value matrix of housing influencing factors of each housing unit obtained in step S3, an artificial intelligence clustering algorithm is used to aggregate from bottom to top starting from each residential unit point to obtain a housing unit group, which is a housing sector.

[0066] Step S1 involves spatiotemporal big data preprocessing, including preprocessing and matching housing transaction data and micro-geographic big data. Step S2 involves the selection and quantification of potential influencing factors. The independent variable matrix includes a list of potential influencing factors on housing prices. Step S3 involves spatiotemporal geographic weighted regression modeling. The housing attribute value index matrix includes a list of housing attribute value indices. Step S4 involves clustering using the affinity propagation algorithm.

[0067] In general, the present invention discloses a method for dividing housing regions based on spatiotemporal big data. A new method framework is established by introducing new indicators and new technologies: first, multi-source high-granularity spatiotemporal big data is processed and matched, and then new variables are introduced on the basis of the classic housing price model (global housing price regression model) to indicate new housing consumption concepts, and then the spatiotemporal geographical weighted regression method is used to calculate the housing characteristic value matrix while considering the functional proximity and spatiotemporal proximity of the housing characteristic value. Based on this, the affinity propagation algorithm in the field of artificial intelligence clustering is used to cluster housing in terms of its potential and robustness in mining multidimensional large-scale data to obtain housing sub-markets. The present invention helps enterprises and governments to monitor and manage the housing market in a timely, efficient and accurate manner under the background of large-scale and rapid urban development, and promotes the "digital intelligence" development of China's real estate technology.

[0068] In one embodiment, in step S1, the housing transaction data includes housing transaction price, transaction time, transaction area, housing name, and housing address;

[0069] The original format of housing attribute data includes housing name, housing address, housing construction year, property management, volume ratio, and greening ratio;

[0070] The attributes of road network data include road names and road attributes;

[0071] The attributes of the POI data include POI name, POI address, and POI category.

[0072] During the specific implementation process, the original format of road network data is Shapefile.

[0073] In one embodiment, step S1 pre-processes and matches housing transaction data and micro-geographic big data, including:

[0074] The housing transaction data is parsed for geographical names and addresses to obtain geographic coordinates, and the average transaction price at the community level is calculated to obtain the average transaction price data;

[0075] Perform place name and address resolution on housing attribute data to obtain geographic coordinates, and match the geographic coordinates with housing transaction data;

[0076] Filter the road network data, remove non-walkable roads such as highways and expressways based on road attributes, build an initial road network, perform topology checks, and obtain a road network dataset;

[0077] Perform place name and address parsing on the POI data to obtain the geographic coordinates of the POI and classify them based on the POI name and category;

[0078] The average transaction price data, road network dataset and classified point of interest data are spatially matched and imported into a spatiotemporal database for storage and management.

[0079] During implementation, housing attribute data was parsed for geographical names and addresses to obtain geographic coordinates, which were then matched with housing transaction price data. Missing and extreme values ​​were processed for building year, floor area ratio, and greening ratio. For missing value data, entries with more than half of the missing attributes were deleted, and the remaining missing attributes were supplemented using regression interpolation with housing prices. For extreme value data, the 3σ method was used to remove data entries that fell outside the range (μ-3σ, μ+3σ), where μ is the mean and σ is the standard deviation.

[0080] For example, data preprocessing and data matching can be:

[0081] The repeated transaction price data of second-hand houses in Shanghai for three consecutive years were parsed to obtain geographical coordinates. Taking residential communities as the most basic analysis unit, the average transaction price of second-hand houses in each community was calculated.

[0082] Residential attribute data was parsed for geographical names and addresses to obtain geographic coordinates, which were then matched with residential second-hand housing transaction price data. Missing and extreme value processing was performed for building year, volume ratio, greening ratio, etc. For missing value data, entries with more than half of the missing attributes were deleted, and regression interpolation was used to complete the remaining missing attributes. For extreme value data, the 3σ method was used to remove data entries distributed outside the range (μ-3σ, μ+3σ), where μ is the mean and σ is the standard deviation.

[0083] The Shanghai road network data was screened, and non-walkable roads such as highways and expressways were removed based on road attributes. An initial road network was constructed and topological checks were performed to obtain a road network dataset.

[0084] Perform place name and address parsing on all points of interest data in Shanghai, obtain the geographic coordinates of the points of interest, and classify them based on their names and categories;

[0085] The pre-processed time series data, i.e., the average transaction price data of residential communities for three consecutive years, the road network dataset, and the point of interest data, are spatially matched and imported into a spatiotemporal database for storage and management.

[0086] In one embodiment, S2 includes:

[0087] S2.1: Obtain a pre-set global housing price regression model, with influencing factors including community attributes, transportation convenience, location, education and medical care, open space, commercial facilities, and job opportunities;

[0088] S2.2: Based on the existing influencing factors of the pre-set global housing price regression model, new variables are introduced, including the quality of the primary school district, the quality of the secondary school district, the distance to the road network of tertiary hospitals, the density of high-tech companies, and the Shannon diversity index of companies;

[0089] S2.3: Based on the spatiotemporal database in step 1, quantify the selected variables. The quantification methods include quantification of nominal variables and road network analysis.

[0090] S2.4: Perform z-score standardization on all influencing factors and housing transaction prices;

[0091] S2.5: Perform significance tests on all potential influencing factors processed in step S2.4;

[0092] S2.6: Perform collinearity diagnosis on all potential influencing factors processed in step S2.5;

[0093] S2.7: Establish an independent variable matrix based on all potential influencing factors processed in step S2.6.

[0094] In specific implementation, in step S2.1, housing prices are often considered the sum of the values ​​of a series of housing-influencing factors, which depend on housing structure, location, and neighborhood characteristics. Based on existing housing price models, potential influencing factors are selected from three perspectives: housing structure, location, and neighborhood characteristics, including community attributes, transportation accessibility, location, education and healthcare, open space, commercial facilities, and job opportunities. Specifically, based on existing research, potential influencing factors can be described and quantified as follows: (1) Residential properties, ① construction age, ② building type, nominal variables, need to be classified and quantified, ③ property fees, ④ residential management quality, nominal variables, need to be quantified, ⑤ floor area ratio, ⑥ greening rate; (2) Transportation convenience, ① number of buses within a 15-minute walk, ② road network distance to bus stops within a 15-minute walk, ① number of subway stations within a 15-minute walk, ② road network distance to subway stations within a 15-minute walk; (3) Location, ① district, nominal variable, ② ring line, nominal variable, ③ distance to CBD; (4) Education and medical care, ① road network distance to primary schools within a 15-minute walk, ② road network distance to middle schools within a 15-minute walk, ③ road network distance to primary schools within a 15-minute walk, The distance to the high school road network within a 15-minute walk, ④ the distance to the university road network within a 15-minute walk; ⑤ the distance to the hospital road network; (5) open space, ① the number of parks within a 15-minute walk, ② the road network distance to parks within a 15-minute walk, ③ the number of leisure squares within a 15-minute walk, ④ the road network distance to leisure squares within a 15-minute walk, ⑤ the number of open scenic spots within a 15-minute walk, ⑥ the road network distance to open scenic spots within a 15-minute walk; (6) commercial facilities, ① the number of large shopping malls within a 15-minute walk, ② the road network distance to large shopping malls within a 15-minute walk, ③ the number of small supermarkets within a 15-minute walk, ④ the road network distance to small supermarkets within a 15-minute walk; (7) job opportunities, ① the number of high-tech companies within a 15-minute walk.

[0095] Step S2.2: Based on the existing commonly used influencing factors, new variables are introduced to establish a new housing price influencing factor matrix. The new variables introduced are the quality of the primary school district, the quality of the secondary school district, the road network distance to the tertiary hospital, the density of high-tech companies, and the Shannon diversity index of companies. Specifically:

[0096] ①Quality of the primary school district and the secondary school district

[0097] Existing housing price models often use distance to primary and secondary schools as an indicator of their impact on housing prices, which is not realistic. Under the school district system, the uneven distribution of high-quality educational resources and parents' desire for quality education make the quality of the schools in the district, rather than proximity, the primary factor influencing housing prices. This invention quantifies the quality of primary and secondary schools based on parental ratings and rankings of schools on social media.

[0098] Taking Shanghai as an example, due to the correspondence between key middle schools and key primary schools in Shanghai, only the key primary school districts were considered. Based on the "Shanghai Primary School Echelon" published on social media and parents' scores and rankings of each school, the quality of primary schools was divided into two categories: ordinary primary schools and key primary schools. Based on this, the quality of the school districts in the residential areas was divided into ordinary school districts and key school districts.

[0099] ② Distance to the road network of tertiary hospitals

[0100] Existing housing price models do not distinguish between hospital grades. This invention divides hospitals into tertiary hospitals and ordinary hospitals, and considers their impact on housing prices separately.

[0101] For example, based on the list of Shanghai's top three hospitals, calculate the road network distance to the top three hospitals within 15 minutes.

[0102] ③ High-tech company core density and company Shannon diversity index

[0103] Existing housing price models only consider the distance or number of companies. This paper introduces the kernel density of high-tech companies and the Shannon diversity index of companies. Specifically:

[0104] The formula for calculating the kernel density of high-tech companies is as follows:

[0105]

[0106] For d i <r

[0107]

[0108] Where i=1,…,n is the input point, r is the search radius, Dm is the median distance to the mean center, SD is the standard distance, d i is the distance from point i to the center point.

[0109] The formula for calculating the company's Shannon Diversity Index is as follows:

[0110]

[0111] Among them, p i is the proportion of company type i among all companies, and m is the number of company types within the 15-minute reach.

[0112] Taking Shanghai as an example, based on the company POI data for Shanghai, we categorized companies into high-tech and general companies, and calculated the kernel density of high-tech companies. We also categorized companies into construction, transportation, electronics, information, biology, pharmaceuticals, materials, energy, securities and finance, services, and manufacturing, and calculated the Shannon Diversity Index for companies within a 15-minute reach.

[0113] In step S2.3, the selected variables are quantified based on the housing attribute data, road network data, and point of interest data from step S1. Specific quantification methods include quantification of nominal variables and road network analysis. For example, if the calculated road network distance within a 15-minute walking distance is null, indicating that it is unreachable, the null value is assigned the maximum value plus the average value.

[0114] In step S2.4, all potential influencing factors and housing transaction prices are z-score standardized, that is, x = (x-μ) / σ.

[0115] In step 2.5, a significance test is performed on all potential influencing factors processed in step 2.4, and the p-value is less than the significance level of 0.05.

[0116] In step 2.6, perform collinearity diagnosis on all potential influencing factors processed in step 2.5 to ensure that the variance inflation factor of all independent variables is less than 7.

[0117] Based on steps 2.1 to 2.6 of the embodiment of the present invention, 31 potential influencing factors were finally selected as shown in the following table:

[0118] Table 1 Quantification methods of potential factors affecting housing prices

[0119]

[0120] In one embodiment, S3 includes:

[0121] S3.1: Based on the housing transaction data and road network data matched in step S1, calculate the spatiotemporal distance using the following formula:

[0122]

[0123] in, is the space-time distance between point i and point j; λ and μ are weight parameters for balancing space-time units; u, v, t are longitude, latitude and time respectively; u i and uj represent the longitude coordinates of point i and point j respectively, v i and v j Represents the latitude coordinates of point i and point j respectively, t i and t j Respectively represent the time of point i and point j;

[0124] S3.2: Based on the spatiotemporal distance, construct a spatiotemporal weight matrix W between two observation points. W is a diagonal matrix. The diagonal elements represent the weights of the mutual influence between the two observation points, and the other elements are 0. Use cross-validation to select the optimal bandwidth.

[0125] S3.3. Based on the independent variable matrix obtained in step S2 and the matched housing transaction data in step S1, a spatiotemporal geographic weighted regression model is established. The formula is as follows:

[0126]

[0127] Among them, y i is the i-th observation point (u i , v i , t i ) is the explained variable, namely housing price; x ik is the independent variable matrix, i.e., the potential influencing factor k at observation point i; I and K are the number of observation points and the number of potential factors, respectively; β k (u i , v i , t i ) is the spatiotemporal geographic weighted regression standardization coefficient, that is, the unknown coefficient β at the i-th observation point k β0(u i , v i , t i ) is a constant term, that is, the constant term β0 at the i-th observation point; ε i is the error term;

[0128] S3.4, use the weighted least squares estimation method to estimate the parameters, specifically: at the i-th observation point (u i , v i , t i ) when the independent variable x k Corresponding regression coefficient Using the weighted least squares estimation method, the regression coefficients are estimated by minimizing the sum of squared residuals:

[0129]

[0130] The parameter estimates are expressed as a matrix, and the regression coefficients are:

[0131]

[0132] in, represents the weight of the i-th observation point, X represents the independent variable matrix, Y represents the explained variable matrix, X T represents the transpose of the independent variable matrix;

[0133] S3.5: Based on the spatiotemporal geographical weighted regression coefficient β obtained in step S3.4 k (u i , v i , t i ), eliminate the residential unit entries whose coefficients do not meet the preset conditions, and obtain the housing characteristic attribute value index matrix, which is the i-th observation point after processing (u i , v i , t i ) is the regression coefficient of the kth housing influencing factor.

[0134] In the specific implementation process, the cross validation method (Cross Validation) or the Akaike information criterion method (Akaike information criterion) is used to calculate the residual sum of squares to determine the optimal bandwidth parameter value: a spatial weight matrix is ​​constructed. Common spatial weight functions include the Gauss function (Gauss function), the bi-square function (Bi-square function), etc.

[0135] In the spatiotemporal geographic weighted regression model, the spatiotemporal influence between points is measured by the spatiotemporal weight matrix. The closer the spatiotemporal distance between two points, the greater the weight, and vice versa. In this implementation, the spatiotemporal weight function uses a Gaussian function:

[0136]

[0137] Among them, W ij is the weight of point i and point j, is the space-time distance between point i and point j, h ST is the bandwidth, which represents the ratio of weight decay with time and space distance. The optimal bandwidth is selected by calculating the residual sum of squares to minimize it using the cross validation method. The cross validation method expression is as follows:

[0138]

[0139] When cv(h ST ) is minimum, h ST is the optimal space-time bandwidth. i is the true value of observation point i, When the bandwidth is h ST The regression value of observation point i at time .

[0140] In the specific example, the residential unit entries with coefficients not significantly exceeding 10% are eliminated to obtain the housing characteristic attribute value index matrix, which is the i-th observation point (u i , v i , ti ) is the regression coefficient of the kth housing influencing factor.

[0141] Spatiotemporal Geographically Weighted Regression Adjustment R 2 =0.76, and the value matrix of 31 influencing factors of Shanghai residential areas is obtained. Taking school quality as an example, the spatial distribution of spatiotemporal geographic weighted regression coefficients for three consecutive years is shown in the attached figure. Figure 2 As shown in the figure, the regression coefficient values ​​are divided into three layers using the layered coloring method, namely ≤0.02, 0.02 to 0.04, and >0.04, and presented with three different grayscale values ​​to show the spatial distribution of the coefficient values ​​in different layers.

[0142] In one embodiment, step S4 includes:

[0143] S4.1: Based on the housing attribute value index matrix obtained in step S3, calculate the similarity matrix between the two points The negative Euclidean distance between any two sample points is used to measure the ability of point j to be the cluster center of point i, x ik represents the value of the k-th housing characteristic attribute value index of point i, x jk represents the value of the k-th housing characteristic attribute value index at point j;

[0144] S4.2: Initialize the reference degree to the median of the similarities, where the reference degree is a real number s(i, i), which refers to the reference degree of point i as the cluster center. The value is the value on the diagonal of the S matrix. The larger the value of s(i, i), the more likely i is the cluster center.

[0145] S4.3: Calculate the attraction matrix r(i, j) and the attribution matrix a(i, j). The attraction matrix r(i, j) refers to the degree to which point j is suitable as the cluster center of data point i. The attribution matrix a(i, j) refers to the degree to which point i chooses point j as its cluster center. The larger the sum of r(i, j) and a(i, j), the greater the probability that point j will become the final cluster center, and the greater the possibility that point i belongs to the cluster with point j as the cluster center. Initialize the attraction matrix and the attribution matrix to 0.

[0146] S4.4: Update the attraction and belongingness of each point iteratively. The termination condition of the iteration is that the cluster center is no longer updated or the maximum number of iterations is reached. The iteration formula is as follows:

[0147]

[0148] Where i′ and j′ represent the updated points i and j; s(i, j′) represents the similarity matrix after updating point j; a(i, j′) represents the attribution matrix after updating point j; r(i′, j) represents the attraction matrix after updating point i; a(j, j) and r(j, j) represent the attribution and attraction of point j as the cluster center, respectively.

[0149] S4.5. Based on the cluster centers obtained in step S4.4, the sample points are classified and finally clusters are obtained, namely the housing sector.

[0150] By using the method of the present invention, the results of the housing area division in Shanghai for three consecutive years can be obtained, as shown in the attached figure. Figure 3 As shown. The grayscale in the figure is only used to distinguish the boundaries of sub-markets (plates). Two identical grayscales that are far apart do not represent the same category. In order to verify the rationality of the housing plates divided by the present invention, the housing plates divided by Shanghai real estate companies were plotted for comparison, and descriptive statistics were performed on each housing plate. The results are shown in Table 2. Overall, there is not much difference in spatial distribution between the plates divided by the present invention and the plates divided by real estate companies, and there is no significant difference in the number of plates and the number of communities included. In order to show more clearly Figure 3 For the housing sector, please refer to the actual examination reference materials. The essential technical solution of this invention is already in the attached Figure 1 The specific implementation method is described in detail and demonstrated. Figure 2 and Figure 3 It is a schematic diagram of the effect of spatiotemporal geographic weighted regression coefficient and housing area plate division, which does not affect the essential technical solution of the present invention.

[0151] Table 2 Descriptive statistics comparison of the number of blocks and the number of included communities

[0152]

[0153] To further test the quality of the housing sector divided by the present invention, a multi-layer regression model was established with the real estate enterprise sector and the housing sector divided by the present invention as the second layer. The model performance was compared. The results are shown in the following table:

[0154] Table 3 compares the results of the multi-level regression model based on the sectors divided by the present invention and the real estate sector

[0155]

[0156] Among them, the intra-class correlation (ICC) reflects the necessity of considering sub-markets, R 2 L1 、R 2 L2 Represent the explanation rate of independent variables at the community and sub-market levels, respectively.

[0157] The results show that the ICC of all models is greater than 0.63, indicating that it is necessary to divide the housing area into sections. The multi-layer regression model established based on the sections divided by the present invention, R 2 L1 Between 0.64 and 0.69, R 2 L2 The R value of the multi-level regression model based on the real estate sector is between 0.30 and 0.34. 2 L1 All are less than 0.55, R 2 L2 The values ​​of the two groups are all less than 0.19, indicating that the sectors divided by the present invention have a stronger ability to consistently explain and predict housing prices than the sectors divided by real estate companies.

[0158] Compared with the existing technology, the advantages of the method of the present invention are: (1) the value matrix of a series of housing influencing factors of each housing unit is calculated based on the consideration of functional proximity and spatiotemporal proximity by using spatiotemporal geographic weighted regression as the basis for the division of housing regional sectors. Compared with directly using the statistical characteristics of the influencing factors, it can better reflect the differences in residents' willingness to pay for different housing attributes; (2) the use of artificial intelligence clustering algorithm reduces manual intervention and improves the robustness of the clustering algorithm. Specifically, compared with the traditional K-means clustering algorithm, the affinity propagation algorithm does not require the pre-specification of the number of clusters and can quickly process large-scale sparse data sets with fewer parameter settings, and the square error of the clustering results is small. More importantly, the algorithm starts from each point, treats all points as potential sample sets, and aggregates the data from bottom to top. It is more in line with the logic of offline real estate agents in site selection and real estate experts in dividing real estate sectors, that is, first consider the attribute characteristics of each community and then cluster, rather than first check the overall market pattern and then divide. Therefore, the present invention can efficiently carry out dynamic supervision of the housing market, evaluation of the capitalization effect of public service facilities, and evaluation of housing policies in the context of large-scale and rapid urban development.

[0159] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of protection of the patent of the present invention. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

Claims

1. A housing area plate division method based on spatiotemporal big data, It is characterized in that include: S1: pre-processing and data matching of housing transaction data and micro-geographic big data, and then importing them into the spatio-temporal big database. The micro-geographic big data includes housing attribute data, road network data, and point of interest data. The spatio-temporal big database includes the matched housing transaction data and road network data. S2: Introduce new variables into the preset global housing price regression model, quantify the potential influencing factors of housing prices for each housing unit based on the spatiotemporal database, and construct an independent variable matrix; S3: Based on the matched housing transaction data and road network data in step S1, calculate the spatiotemporal distance and construct a spatiotemporal distance matrix. Use the spatiotemporal geographic weighted regression algorithm to consider the spatiotemporal dependence between housing prices. Embed the spatiotemporal distance matrix into the preset global housing price regression model in step S2. Then use the weighted least squares method to estimate the point-by-point parameters of the housing unit to obtain the housing characteristic attribute value index of each housing unit, and construct the housing characteristic attribute value index matrix. S4: Based on the value matrix of housing influencing factors of each housing unit obtained in step S3, using an artificial intelligence clustering algorithm, starting from each residential unit point, aggregating from bottom to top to obtain a housing unit group, namely, a housing sector; Among them, S3 includes: S3.1: Based on the matched housing transaction data and road network data in step S1, calculate the spatiotemporal distance. The calculation formula is as follows: in, is the space-time distance between point i and point j; λ and μ are weight parameters for balancing space-time units; u, v, t are longitude, latitude and time respectively; u i and u j Respectively represent the longitude coordinates of point i and point j, v i and v j Represent the latitude coordinates of point i and point j respectively, t i and t j Respectively represent the time of point i and point j; S3.2: Based on the spatiotemporal distance, a spatiotemporal weight matrix W is constructed between two observation points. W is a diagonal matrix. The diagonal elements represent the weight values ​​of the mutual influence between the two observation points, and the other elements are 0. The cross-validation method is used to select the optimal bandwidth. S3.3: Based on the independent variable matrix obtained in step S2 and the matched housing transaction data in step S1, a spatiotemporal geographic weighted regression model is established, and the formula is as follows: Among them, y i is the i-th observation point (u i ,v i ,t i ) is the explained variable, i.e., housing price; x ik is the independent variable matrix, i.e., the potential influencing factor k at observation point i; I and K are the number of observation points and the number of potential factors, respectively; β k (u i ,v i ,t i ) is the spatiotemporal geographic weighted regression standardization coefficient, that is, the unknown coefficient at the i-th observation point; β 0 (u i ,v i ,t i ) is a constant term, that is, a constant term at the i-th observation point; ε i is the error term; S3.4, use the weighted least squares estimation method to estimate parameters, specifically: at the i-th observation point (u i ,v i ,t i ) when the independent variable x k Corresponding regression coefficient Using the weighted least squares estimation method, the regression coefficients are estimated by minimizing the sum of squared residuals: The parameter estimates are expressed as a matrix, and the regression coefficients are: in, represents the weight of the i-th observation point, X represents the independent variable matrix, Y represents the explained variable matrix, X T represents the transpose of the independent variable matrix; S3.5: Based on the spatiotemporal geographic weighted regression coefficient β obtained in step S3.4 k (u i ,v i ,t i ), remove the residential unit entries whose coefficients do not meet the preset conditions, and obtain the housing characteristic attribute value index matrix, which is the i-th observation point after processing (u i ,v i ,t i ) is the regression coefficient of the kth housing influencing factor; Step S4 includes: S4.1: Based on the housing attribute value index matrix obtained in step S3, calculate the similarity matrix between the two points It is measured by the negative Euclidean distance between any two sample points, which characterizes the ability of point j to serve as the cluster center of point i. ik represents the value of the k-th housing characteristic attribute value index of point i, x jk represents the value of the k-th housing characteristic attribute value index of point j; S4.2: Initialize the reference degree to the median of the similarity, where the reference degree is a real number s(i,i), which refers to the reference degree of point i as the cluster center. The value is the value on the diagonal of the S matrix. The larger the s(i,i) value is, the greater the possibility that i is the cluster center. S4.3: Calculate the attraction matrix r(i,j) and the attribution matrix a(i,j). The attraction matrix r(i,j) refers to the degree to which point j is suitable as the cluster center of data point i, and the attribution matrix a(i,j) refers to the degree to which point i chooses point j as its cluster center. The larger the sum of r(i,j) plus a(i,j), the greater the probability that point j will become the final cluster center, and the greater the possibility that point i belongs to the cluster with point j as the cluster center; initialize the attraction matrix and the attribution matrix to 0; S4.4: Update and iterate the attraction and belongingness of each point. The termination condition of the iteration is that the cluster center is no longer updated or the maximum number of iterations is reached. The iteration formula is as follows: Among them, i′, j′ represent the updated points i and j; s(i, j′) represents the similarity matrix after updating point j; a(i, j′) represents the attribution matrix after updating point j; r(i′, j) represents the attraction matrix after updating point i; a(j, j) and r(j, j) represent the attribution and attraction of point j as the cluster center respectively; S4.5: Based on the cluster centers obtained in step S4.4, the sample points are classified and finally the clusters are obtained, which are the housing sectors.

2. The housing area division method according to claim 1, It is characterized in that In step S1, the housing transaction data includes housing transaction price, transaction time, transaction area, housing name, and housing address; The original format of housing attribute data includes housing name, housing address, housing construction year, property management, volume ratio, and greening ratio; The attributes of road network data include road names and road attributes; The attributes of the POI data include the POI name, POI address, and POI category.

3. The housing area division method according to claim 1, It is characterized in that Step S1 preprocesses and matches housing transaction data and micro-geographic big data, including: The housing transaction data is parsed for geographical names and addresses to obtain geographical coordinates, and the average transaction price at the community scale is calculated to obtain the average transaction price data; Perform place name and address resolution on housing attribute data to obtain geographic coordinates, and match the geographic coordinates with housing transaction data; Filter the road network data, remove non-walkable roads such as highways and expressways based on road attributes, build an initial road network, perform topology checks, and obtain a road network dataset; Perform place name and address parsing on the POI data to obtain the geographic coordinates of the POI and classify them based on the POI name and category; The average transaction price data, road network data set and classified point of interest data are spatially matched and imported into the spatiotemporal database for storage and management.

4. The housing area division method according to claim 1, It is characterized in that S2 include: S2.1: Obtain a preset global housing price regression model, where the influencing factors include community attributes, transportation convenience, location, education and medical care, open space, commercial facilities, and job opportunities; S2.2: Based on the existing influencing factors of the preset global housing price regression model, new variables are introduced, including the quality of the primary school district, the quality of the secondary school district, the distance to the road network of tertiary hospitals, the core density of high-tech companies, and the Shannon diversity index of companies; S2.3: Based on the spatiotemporal database in step 1, quantify the selected variables, where the quantification methods include quantification of nominal variables and road network analysis; S2.4: Perform z-score standardization on all influencing factors and housing transaction prices; S2.5: Perform significance test on all potential influencing factors processed in step S2.4; S2.6: Perform collinearity diagnosis on all potential influencing factors processed in step S2.5; S2.7: Establish an independent variable matrix based on all potential influencing factors processed in step S2.6.