Urban updating area identification method based on machine learning
Through multi-time node data and Gaussian mixed model clustering algorithm, urban renewal areas are identified, which solves the problem that the time dimension cannot be included in the existing technology, and realizes the identification of urban morphological evolution process and scientific support for urban planning, which is efficient and Reusability.
Patent Information
- Application Number
- CN202410964165.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-12
AI Technical Summary
The existing technology is difficult to incorporate time dimensions into machine learning algorithms, resulting in the inability to vertically track and compare the evolution of urban morphology, focusing on the cross-sectional data of urban morphology in a single year, and it is difficult to fully understand the dynamic process of urban development.
Multi-time node data and Gaussian hybrid model (GMM) clustering algorithm are used to calculate the multi-dimensional matrix composed of architectural morphological characteristics, Bayesian information criterion optimization is carried out, urban renewal areas are identified, and multi-time node data analysis of urban morphology is realized.
It realizes the identification of the urban morphological evolution process within a large-scale spatial scope, provides a more scientific basis for urban planning, saves manpower, has strong reusability, and can store key model parameters for identification work in other regions.
Smart Images

Figure CN118941948A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an intelligent identification technology for urban renewal areas based on machine learning. Background Art
[0002] Urban morphological evolution is an important representation of regional social, economic and cultural changes. At the same time, urban morphological changes also have a reverse effect on the natural and social space of the region. In the era of stock planning, the urbanization gradient from cities to rural areas may be more diverse, and the renewal and improvement of urban and rural space has become the goal of spatial planning development at this stage. In recent years, the combination of remote sensing technology and machine learning algorithms has provided new methods and tools for identifying and analyzing urban morphological evolution, especially in the acquisition of high-precision and large-scale urban building data.
[0003] The commonly used algorithm for urban morphology identification is to conduct cluster analysis based on the morphological attributes of buildings and the demographic and economic attributes of regions. Commonly used machine learning algorithms include k-means and GMM clustering algorithms, but existing algorithms often focus on the differentiated distribution of urban morphology in different spaces under a single time section, and are unable to incorporate the time dimension into the machine learning algorithm, resulting in a focus on urban morphology cross-sectional data of a single year, making it difficult to track and compare the evolution of urban morphology longitudinally. Summary of the invention
[0004] In view of the shortcomings of the existing technology, the present invention proposes a technical method for identifying urban renewal areas using a machine learning algorithm. The method emphasizes intelligent identification of urban renewal areas in large-scale areas based on multi-time node data and machine learning algorithms, which is conducive to refined urban management.
[0005] The technical solution of the present invention is as follows:
[0006] A method for identifying urban renewal areas based on machine learning, comprising the following steps:
[0007] Step (1): Use the initial time node t1 to the end time node t i The building outline data within the study area is J space units in total. The calculation includes the sum of the building areas intersecting within the circle with a radius of k meters and the center of each space unit j as the center point. Average building perimeter Building orientation angle entropy Building shape index average Average building compactness index and number of buildings Six architectural features form a multidimensional matrix x ij , the specific formula is as follows:
[0008]
[0009] In formula 1 is the total building area of space unit j in year i, is the jth time in the i-th year in the circle building The area of the building.
[0010]
[0011] In formula 2 is the average building perimeter of space unit j in year i, P i is the jth time in the i-th year in the circle building The perimeter of the building.
[0012]
[0013] In formula 3 is the entropy value of the building orientation angle of spatial unit j in the i-th year, and Pik is the probability distribution of the building orientation angle k in the i-th year.
[0014]
[0015] In formula 4 is the average value of the building shape index of spatial unit j in year i, The number of buildings j in year i is building The minimum circumscribed circle area of the outline polygon.
[0016]
[0017] In formula 5 is the average building compactness index of space unit j in year i, For building j in year i building Building j inside the minimum circumscribed circle of the outline polygon building circumference.
[0018]
[0019] In formula 6, x ij A multidimensional matrix containing 6 architectural morphological characteristics for each spatial unit j.
[0020] Step (2): The multidimensional matrix x of architectural form characteristics of all spatial units in the first year (i=1) obtained in step (1) i Perform Gaussian mixture model (GMM) cluster analysis, the number of clusters m cluster The value is 1, 2, ..., m, and the mixing coefficient corresponding to different clustering numbers is obtained. Mean vector And the covariance matrix ∑m cluster , the formula is as follows:
[0021]
[0022] In formula 7 For space unit j belongs to m cluster The posterior probability of the class, So is the mean, ∑m cluster is the covariance matrix of the multivariate normal distribution.
[0023]
[0024] In formula 8 It is the mth cluster The mixing coefficient of clusters satisfies and is an indicator variable. If space unit j belongs to m cluster It is 1 if the class is true, otherwise it is 0.
[0025]
[0026] In formula 9 is the mean vector.
[0027]
[0028] ∑m in Equation 10 cluster is the covariance matrix for each cluster, For vector The cosquare matrix is obtained by multiplying and its transpose.
[0029] Step (3): The mixing coefficient calculated in step (2) Mean vector And the covariance matrix ∑m cluster , calculate the corresponding Bayesian Information Criterion The formula is as follows:
[0030]
[0031] In formula 11 The number of clusters is m cluster The corresponding maximum likelihood value.
[0032]
[0033] In formula 12, The number of clusters is m cluster The corresponding Bayesian Information Criterion (BIC) is
[0034] Step (4): The number of clusters m calculated according to step (3) clusterCorresponding The number of clusters corresponding to the minimum BIC value is obtained by comparison as the optimal number of clusters m opt , the formula is as follows:
[0035]
[0036] In formula 13, m opt is the optimal number of clusters, To obtain the minimum algorithm.
[0037] Step (5): The optimal number of clusters m obtained in step (4) opt , calculate M respectively opt =1,2,3,......,m opt The clustering probability density function of the corresponding Gaussian mixture model and save the mixing coefficient Mean vector and the covariance matrix ∑M opt , as the final clustering training model The core parameters of are as follows:
[0038]
[0039] In formula 14, p(x i ) is the number of clusters m opt The probability density function of the Gaussian mixture model is .
[0040] Step (6): According to the clustering probability p(x i ), classify each spatial unit j into the type m with the highest probability opt * The formula is as follows:
[0041]
[0042] In formula 15, p(M opt |x i ) is M opt =1,2,3,......,m opt The corresponding probability density.
[0043] In formula 16 So is the mean, ∑M opt , m opt is the covariance matrix of the multivariate normal distribution.
[0044] Step (7): From time node t2 to end time node t i All year space units, use Repeat steps (5) and (6) to obtain the spatial unit j at different time nodes. i The clustering results
[0045] Step (8): Clustering results of each spatial unit j obtained in step (7) Summarize different statistics separately m opt The multidimensional matrix x of the six architectural form characteristics of all spatial units in the type i get Using the between-group comparison method Sort and get the corresponding rank i , used to represent the degree of urbanization, ranking from 1 to The smaller the number, the higher the degree of urbanization. The formula is as follows:
[0046]
[0047] In formula 17, For each cluster The statistical analysis of the building form characteristic matrix includes the calculation of mean, median, variance and other statistics. The sort() function represents the sorting method based on statistics, which is used to determine the order or level of the characteristic matrix of each cluster type. The sorting method is judged according to the characteristics of the location of the study area. It is generally believed that the higher the building density, the more buildings, the higher the compactness, etc., the higher the degree of urbanization.
[0048] Step (9): The rank of each spatial unit j at different time nodes obtained in step (8) i , if rank i =1 and rank i-1 Greater than rank i The spatial unit i is then identified as an urban renewal area.
[0049] The present invention considers the urbanization gradient changes at multiple time nodes, can more comprehensively understand the dynamic process of urban development, capture the trend of urban expansion and density changes, thereby identifying urban renewal areas and providing a more scientific basis for urban planning. The specific beneficial effects are:
[0050] (1) Based on multi-time node data, the urban morphological evolution process in a large spatial scale can be identified;
[0051] (2) Based on machine learning algorithms, the previous method of identifying urban renewal areas based on manual work is changed to save manpower;
[0052] (3) It has strong reusability. The urban renewal area identification model trained by the method of the present invention can store key model parameters and be applied to the urban renewal area identification work in other regions. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 : The technical roadmap of the present invention.
[0054] Figure 2 :Seven urban form types in Shanghai in 2016.
[0055] Figure 3 :Seven urban form types in Shanghai from 2017 to 2021.
[0056] Figure 4 : Ranking by degree of urbanization.
[0057] Figure 5 : Urban renewal area identification results from 2016 to 2021. DETAILED DESCRIPTION
[0058] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.
[0059] The present invention relates to an intelligent identification technology for urban renewal areas based on machine learning.
[0060] Case selection: Shanghai
[0061] Figure 2 : 7 types of urban forms in Shanghai in 2016
[0062] Figure 3 : 7 types of urban forms in Shanghai from 2017 to 2021
[0063] Figure 1 The steps of the embodiment method are as follows:
[0064] A method for identifying urban renewal areas based on machine learning, comprising the following steps:
[0065] Step (1): Using the building outline data within the study area from the initial time node 2016 (t1) to the end time node 2021 t6, time nodes i = 1, 2, 3, 4, 5, 6, a total of 3,430,994,784 (J) space units, calculate the total building area intersecting within a circle with a radius of 250 meters and the center of each space unit j as the center point Average building perimeter Building orientation angle entropy Building shape index average Average building compactness index and number of buildings Six architectural features form a multidimensional matrix x ij , the specific formula is as follows:
[0066]
[0067] In formula 1 is the total building area of space unit j in year i, is the jbth in the i-th year in the circle uilding The area of the building.
[0068]
[0069] In formula 2 is the average building perimeter of space unit j in year i, P i is the jth time in the i-th year in the circle building The perimeter of the building.
[0070]
[0071] In formula 3 is the entropy value of the building orientation angle of spatial unit j in the i-th year, and Pik is the probability distribution of the building orientation angle k in the i-th year.
[0072]
[0073] In formula 4 is the average value of the building shape index of spatial unit j in year i, The number of buildings j in year i is building The minimum circumscribed circle area of the outline polygon.
[0074]
[0075] In formula 5 is the average building compactness index of space unit j in year i, For building j in year i building Building j inside the minimum circumscribed circle of the outline polygon building circumference.
[0076]
[0077] In formula 6, x ij A multidimensional matrix containing 6 architectural morphological characteristics for each spatial unit j.
[0078] Step (2): The multidimensional matrix x of architectural form characteristics of all spatial units in 2016 (i=1) obtained in step (1) is i Perform Gaussian mixture model (GMM) cluster analysis, the number of clusters mcluster The value is 1, 2, ..., 20, and the mixing coefficient corresponding to different clustering numbers is obtained. Mean vector And the covariance matrix ∑m cluster , the formula is as follows:
[0079]
[0080] In formula 7 For space unit j belongs to m cluster The posterior probability of the class, So is the mean, ∑m cluster is the covariance matrix of the multivariate normal distribution.
[0081]
[0082] In formula 8 It is the mth cluster The mixing coefficient of clusters satisfies and is an indicator variable. If space unit j belongs to m cluster It is 1 if the class is true, otherwise it is 0.
[0083]
[0084] In formula 9 is the mean vector.
[0085]
[0086] ∑m in Equation 10 cluster is the covariance matrix for each cluster, For vector The cosquare matrix is obtained by multiplying and its transpose.
[0087] Step (3): The mixing coefficient calculated in step (2) Mean vector And the covariance matrix ∑m cluster , calculate the corresponding Bayesian Information Criterion The formula is as follows:
[0088]
[0089] In formula 11 The number of clusters is m cluster The corresponding maximum likelihood value.
[0090]
[0091] In formula 12, The number of clusters is m cluster The corresponding Bayesian Information Criterion (BIC) is
[0092] Step (4): The number of clusters m calculated according to step (3) cluster Corresponding By comparison, the number of clusters corresponding to the minimum BIC value is taken as the optimal number of clusters, which is 7 (m opt ), the formula is as follows:
[0093]
[0094] In formula 13, m opt is the optimal number of clusters, To obtain the minimum algorithm.
[0095] Step (5): According to the optimal number of clusters 7 obtained in step (4), calculate M opt =1, 2, 3, ..., 7 (m opt ) The clustering probability density function of the Gaussian mixture model corresponding to , and save the mixing coefficient Mean vector and the covariance matrix ∑M opt , as the final clustering training model The core parameters of are as follows:
[0096]
[0097] In formula 14, p(x i ) is the number of clusters m opt The probability density function of the Gaussian mixture model is .
[0098] Step (6): According to the clustering probability p(x i ), classify each spatial unit j into the type m with the highest probability opt * , we get 7 types of urban morphology in Shanghai in 2016, see the attached Figure 2 , the formula is as follows:
[0099]
[0100] In formula 15, p(M opt |x i ) is M opt =1,2,3,......,m opt The corresponding probability density.
[0101] In formula 16 So is the mean, ∑Mopt , m opt is the covariance matrix of the multivariate normal distribution.
[0102] Step (7): For time node t 2017 To the end time node t 2021 All year space units, use Repeat steps (5) and (6) to obtain the spatial unit j at different time nodes i. i The clustering results See attached Figure 3 .
[0103] Step (8): Clustering results of each spatial unit j obtained in step (7) Summarize different statistics separately The multidimensional matrix x of the six architectural form characteristics of all spatial units in the type i get Using the between-group comparison method Sort and get the corresponding rank i , used to represent the degree of urbanization, ranking from 1 to The smaller the number, the higher the degree of urbanization. Figure 4 , the formula is as follows:
[0104]
[0105] In formula 17, For each cluster The statistical analysis of the building form characteristic matrix includes the calculation of mean, median, variance and other statistics. The sort() function represents the sorting method based on statistics, which is used to determine the order or level of the characteristic matrix of each cluster type. The sorting method is judged according to the characteristics of the location of the study area. It is generally believed that the higher the building density, the more buildings, the higher the compactness, etc., the higher the degree of urbanization.
[0106] Step (9): The rank of each spatial unit j at different time nodes obtained in step (8) i , if rank i =1 and rank i -1 is greater than rank i The spatial unit i is identified as an urban renewal area. Finally, the results of urban renewal identification in Shanghai from 2016 to 2021 are shown in the attached Figure 5 .
[0107] The above technical solution uses building outline data of multiple time nodes and applies cluster analysis algorithm to intelligently identify urban areas that need to be updated. By comparing data at different time nodes, this method can effectively track the evolution of urban spatial form and identify urban renewal areas accordingly. This method not only has high repeatability and operational efficiency, but also can retain the machine learning model parameters of the initial time node so that the trained model can be applied to cross-year data, thereby achieving continuous and efficient identification of urban renewal areas. The present invention provides strong technical support for the delineation of urban renewal areas under stock planning.
Claims
1. An intelligent identification method for urban renewal areas based on machine learning, It is characterized in that The following steps are involved: Step (1): Use the initial time node t 1 To the end time node t i The building outline data within the study area is J space units in total. The calculation includes the sum of the building areas intersecting within the circle with a radius of k meters and the center of each space unit j as the center point. Average building perimeter Building orientation angle entropy Building shape index average Average building compactness index and number of buildings Six architectural features form a multidimensional matrix x ij , the specific formula is as follows: In formula 1 is the total building area of space unit j in year i, is the jth time in the i-th year in the circle building The area of the building; In formula 2 is the average building perimeter of space unit j in year i, P i is the jth time in the i-th year in the circle building The perimeter of the building; In formula 3 is the entropy value of the building orientation angle of spatial unit j in the i-th year, and pik is the probability distribution of the building orientation angle k in the i-th year; In formula 4 is the average value of the building shape index of spatial unit j in year i, The number of buildings j in year i is building The minimum circumscribed circle area of the contour polygon; In formula 5 is the average building compactness index of space unit j in year i, For building j in year i building Building j within the minimum circumscribed circle of the outline polygon building circumference; In formula 6, x ij A multidimensional matrix containing 6 architectural morphological features for each spatial unit j; Step (2): The multidimensional matrix x of architectural form characteristics of all spatial units in the first year (i=1) obtained in step (1) i Perform Gaussian mixture model (GMM) cluster analysis, the number of clusters m cluster The value is 1, 2, ..., m, and the mixing coefficient corresponding to different clustering numbers is obtained. Mean vector And the covariance matrix ∑m cluster , the formula is as follows: In formula 7 For space unit j belongs to m cluster The posterior probability of the class, So is the mean, ∑m cluster is the covariance matrix of the multivariate normal distribution; In formula 8 It is the mth cluster The mixing coefficient of clusters satisfies and is an indicator variable. If space unit j belongs to m cluster If the class is true, it is 1, otherwise it is 0; In formula 9 is the mean vector; ∑m in Equation 10 cluster is the covariance matrix for each cluster, For vector The co-squared matrix obtained by multiplying and its transpose; Step (3): The mixing coefficient calculated in step (2) Mean vector And the covariance matrix ∑m cluster , calculate the corresponding Bayesian Information Criterion The formula is as follows: In formula 11 The number of clusters is m cluster The maximum likelihood value corresponding to ; In formula 12, The number of clusters is m cluster The corresponding Bayesian Information Criterion (BIC) when ; Step (4): The number of clusters m calculated according to step (3) cluster Corresponding The number of clusters corresponding to the minimum BIC value is obtained by comparison as the optimal number of clusters m opt , the formula is as follows: In formula 13, m opt is the optimal number of clusters, To obtain the minimum algorithm; Step (5): The optimal number of clusters m obtained in step (4) opt , calculate M respectively opt =1,2,3,…,m opt The clustering probability density function of the corresponding Gaussian mixture model and save the mixing coefficient Mean vector and the covariance matrix ∑M opt , as the final clustering training model The core parameters of are as follows: In formula 14, p(x i ) is the number of clusters m opt The probability density function of the Gaussian mixture model is: Step (6): According to the clustering probability p(x i ), classify each spatial unit j into the type m with the highest probability opt * ; The formula is as follows: In formula 15, p(M opt |x i ) is M opt =1,2,3,......,m opt The corresponding probability density; In formula 16 So is the mean, ∑M opt , m opt is the covariance matrix of the multivariate normal distribution; Step (7): For time node t 2 To the end time node t i All year space units, use Repeat steps (5) and (6) to obtain the clustering results of each spatial unit ji at different time nodes. Step (8): Clustering results of each spatial unit j obtained in step (7) Different statistics are summarized separately m opt The multidimensional matrix x of the six architectural form characteristics of all spatial units in the type i get Using the between-group comparison method Sort and get the corresponding rank i , used to represent the degree of urbanization, ranking from 1 to The smaller the number, the higher the degree of urbanization; the formula is as follows: In formula 17, For each cluster The statistical analysis of the architectural form feature matrix is performed; the sort() function represents a sorting method based on statistics, which is used to determine the order or level of the feature matrix of each cluster type. The sorting method is judged according to the characteristics of the location of the study area. The higher the building density, the more buildings, and the higher the compactness, the higher the degree of urbanization; Step (9): The rank of each spatial unit j at different time nodes obtained in step (8) i , if rank i =1 and rank i-1 Greater than rank i The spatial unit i is then identified as an urban renewal area.
Citation Information
Cited By
Urban update region identification method based on multi-source data
CN120067235A
A Method for Identifying Urban Renewal Areas Based on Multi-Source Data
CN120067235B
Urban update region identification method and device based on machine learning
CN122656306A
