A temporal multi-layer geographic flow clustering identification method considering topological data analysis

Abstract

The present invention belongs to the field of data mining technology, and specifically relates to a method for identifying temporal multi-layer geographic flow clusters that takes into account topological data analysis, comprising the following steps: S1, constructing temporal geographic flows; S2, reducing geographic flow dimensionality; and S3, identifying geographic flow clusters. The present invention proposes a method for identifying temporal multi-layer geographic flow clusters that takes into account topological data analysis. This novel geographic spatiotemporal analysis method can identify multi-layer geographic flow clusters. To cluster multi-layer temporal geographic flows, the inventors introduce a multi-lens tool for topological data analysis into the method persistence graph, calculate the Wasserstein distance between each persistence graph, and thus cluster multi-layer geographic flows of different temporal sequences, vividly depicting their dynamic interactions and enriching research on the dynamic organization of urban space. The experimental results can provide decision support for sustainable urban management.

Description

Technical Field

[0001] The present invention belongs to the technical field of data mining, and in particular relates to a temporal multi-layer geographic flow clustering identification method taking into account topological data analysis. Background Art

[0002] Most current multi-layer network clustering methods are based on embedding the graph into Euclidean space via graph spectral decomposition, and therefore, do not explicitly consider the local underlying graph geometry and topology.

[0003] In view of this, the present application hopes to propose a temporal multi-layer geographic flow clustering identification method that takes into account topological data analysis, and can identify multi-layer geographic flow clusters from a new geographic spatiotemporal analysis method. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above-mentioned problems existing in traditional technologies and to provide a temporal multi-layer geographic flow clustering identification method taking into account topological data analysis.

[0005] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0006] A temporal multi-layer geographic flow clustering identification method taking into account topological data analysis includes the following steps:

[0007] S1. Construction of time series geographic flow

[0008] Download the trajectory data of the target city and use the TransBigData library in Python to convert the downloaded trajectory data into the OD flow data required for the study. Build a multi-layer traffic time series network. Then use ArcGIS to create grids to segment the OD flows. The number of flows with OD points falling in the same two grids is used as the weight value between the grids to construct a geographic flow weight network with grids as nodes.

[0009] S2. Geographic Stream Dimensionality Reduction

[0010] Using the DeepWalk algorithm in network embedding, the obtained geographic flow weight network is embedded into point cloud data using a random walk model. On this basis, the sequence of network nodes in each layer is calculated and dynamic urban mobile communities are detected from the time-dependent multi-layer network.

[0011] S3. Geographic flow cluster identification

[0012] The obtained point cloud data is used to generate the corresponding persistence graph using the gudhi package in Python, and the Wasserstein distance of each layer of the network is calculated, and then this distance is used to perform geographic flow clustering identification.

[0013] Further, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, in step S1, the time scale of the traffic time-series multi-layer network is 2 hours.

[0014] Further, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, in step S2, the deepwalk algorithm mainly consists of two parts: random walk and generating representation vectors; first, use the random walk algorithm to extract some vertex sequences from the graph, and then, borrowing the idea of natural language processing, regard the generated fixed-point sequences as sentences composed of words, and all sequences can be regarded as a large corpus. Finally, use the natural language processing tool word2vec to represent each vertex as a vector with a dimension of d.

[0015] Further, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, in step S2, the deepwalk algorithm specifically includes the following steps:

[0016] 1) Number the grids created by ArcGIS, regard each grid as a network node, and generate an algorithm for the random walk sequence. The algorithm can be understood as inputting the starting point and path length, generating a random walk node sequence, then summarizing the adjacent nodes, and randomly selecting the next node from the adjacent nodes;

[0017] 2) Generate a random walk sequence with each node as the starting point, train the word2vec model in the deepwalk algorithm to embed each layer of the network into point cloud data, and then use principal component analysis for dimensionality reduction visualization and save the point cloud data generated by the embedding.

[0018] Further, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, the specific algorithm of the random walk in step 1) is as follows:

[0019] Let f(x) be a multivariate function containing n variables, and x = (x1, x2,..., xn) be an n-dimensional vector;

[0020] Given the initial iteration point x, the initial walking step length λ, and the control precision ∈;

[0021] Given the iteration control number N, and k is the current iteration number;

[0022] When k < N, randomly generate an n-dimensional vector u = (u1, u2,…, un) between (-1, 1), (-1 < ui < 1, i = 1, 2,…, n), and standardize it to get Let x1 = x + λu′ to complete the first step of the walk;

[0023] Calculate the function value. If f(x1) < f(x), that is, a point better than the initial value is found, then k is reset to 1, x1 is changed to x, and go back to step 2; otherwise, k = k + 1, and go back to step 3;

[0024] If a better value cannot be found for N consecutive times, it is considered that the optimal solution is within an N-dimensional sphere centered at the current optimal solution with the current step size as the radius; at this time, if λ < ∈, the algorithm ends; otherwise, let λ = λ2, go back to step 1, and start a new round of游走.

[0025] Furthermore, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, ∈ is a very small positive number used to control the end of the algorithm.

[0026] Furthermore, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, in this step, the point cloud data obtained by embedding each layer of the network is subjected to principal component analysis for visualization dimensionality reduction to explore their different life courses of occurrence, expansion, stability, contraction, and disappearance in different time periods.

[0027] Furthermore, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, the specific algorithm steps of the Skip-Gram model in step 2) are as follows:

[0028] First, select a point in the point cloud network as the input point;

[0029] After having the input point, define a parameter called skip_window, which represents the number of points selected from one side of the current input point; define another parameter called num_skips, which represents how many different points are selected from the entire window as output points;

[0030] The neural network will output a probability distribution based on these training data, and this probability represents the output possibility of each point in the dictionary.

[0031] Furthermore, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, in step S3, the basic principle of the clustering method is: if the local neighborhoods of two points are similar in shape at all resolution scales, then they are close enough to be grouped into a cluster.

[0032] Furthermore, in the above-mentioned time-series multi-layer geographic flow clustering recognition method considering topological data analysis, to compare the shapes of the clusters, the following steps are performed:

[0033] Consider Xn = (x1,..., Xn) in some metric space (X, D);

[0034] Set the resolution thresholds V1 < V2... < V K, and construct a VR filter

[0035] Compute the local topological summary of xi in the form of a persistence graph PD(i), i = 1, ..., n;

[0036] For all local neighborhoods N(i) of xi and x j For N(j), i, j = 1, 2, ..., N, we compute the pairwise topological or data shape dissimilarity as the Wasserstein distance between their respective persistence graphs PD(i) and PD(j):

[0037]

[0038] In (1), Δ = {(x, x) | x∈R}, γ is bi-mapped from PD(i)∪Δ to PD(j)∪Δ. The Wasserstein distance allows systematically quantifying the similar shapes of the neighborhoods of two nodes;

[0039] Form a distance graph G on W2(N(i), N(j)), i, j = 1, 2, ..., N, with an adjacency matrix a, where

[0040]

[0041] Define the cut-off point κ by elbow plot or cross-validation;

[0042] The connected components of G are the resulting clusters.

[0043] The beneficial effects of the present invention are:

[0044] This paper proposes a method for identifying temporal multi-layer geographic flow clusters that incorporates topological data analysis. This novel geographic spatiotemporal analysis method can identify multi-layer geographic flow clusters. The goal is to cluster multi-layer networks in an unsupervised environment based on the shape similarity of data recorded at multiple resolutions. Furthermore, compared to supervised community detection and classification, unsupervised learning of multi-layer networks remains significantly underdeveloped. To cluster multi-layer temporal geographic flows, the inventors incorporate multi-lens tools from topological data analysis into the method's persistence graph. The Wasserstein distance between each persistence graph is calculated to cluster multi-layer geographic flows of different temporal sequences, vividly depicting their dynamic interactions and enriching research on the dynamic organization of urban space. The experimental results can provide decision support for sustainable urban management.

[0045] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is an experimental flow chart of an embodiment;

[0048] Figure 2 Schematic diagram of OD flow at different time sequences in the embodiment;

[0049] Figure 3 Schematic diagram of weighted network data with grids as nodes in the embodiment;

[0050] Figure 4 Schematic diagram of the Skip-Gram algorithm in the embodiment;

[0051] Figure 5 Schematic diagram of dimensionality reduction of 6-point to 8-point geographic flow in the embodiment;

[0052] Figure 6 Schematic diagram of different time sequence persistence in the embodiment;

[0053] Figure 7 Schematic diagram of Wasserstein matrix at different time sequences in the embodiment;

[0054] Figure 8 Schematic diagram of the result of time series multi-layer geographic flow clustering identification in the embodiment. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 any creative efforts shall fall within the scope of protection of the present invention.

[0056] In order to fully utilize the advantages of the model, the present invention relates to a time series multi-layer geographic flow clustering identification method that takes into account topological data analysis. The Wasserstein distance is used to cluster the multi-layer geographic flows of different time series. The overall experimental flow chart is as follows: Figure 1 shown.

[0057] This method can be used to explore the topological similarity of each layer of the network to reveal their different life processes of occurrence, expansion, stability, contraction and disappearance in different time periods, and depict the dynamic interactions between them.

[0058] The technical solution adopted by the present invention is: a temporal multi-layer geographic flow clustering identification method taking into account topological data analysis, comprising the following steps:

[0059] (1) Construction of temporal geographic flow

[0060] The trajectory data of Chengdu from 6:00 am to midnight from August 3 to 30, 2016 were downloaded. The TransBigData library in Python was used to convert the downloaded trajectory data into the OD flow data required for the study. A multi-layer traffic time series network with a time scale of 2 hours was established (a total of 9 time series, numbered 0, 1, 2, 3, 4, 5, 6, 7, 8). Then, ArcGIS was used to create a 500m*500m grid to segment the OD flow. The number of flows with OD points falling in the same two grids was used as the weight value between the grids to construct a weighted network with grids as nodes.

[0061] (2) Geographic flow dimensionality reduction

[0062] The algorithm consists of two main parts: random walk and generating representation vectors. First, a random walk algorithm is used to extract vertex sequences from the graph. Then, drawing on the principles of natural language processing, the generated vertex sequences are treated as sentences composed of words. All sequences can be considered a large corpus. Finally, the natural language processing tool word2vec is used to represent each vertex as a vector of dimension d.

[0063] This part mainly applies the deepwalk algorithm in network embedding. The specific algorithm steps are as follows:

[0064] 1) Use ArcGIS to number the created grids, treat each grid as a network node, and generate a random walk sequence algorithm. The algorithm can be understood as inputting the starting point and path length, generating a random walk node sequence, then summarizing the adjacent nodes and randomly selecting the next node from the adjacent nodes. The specific algorithm of the random walk is as follows:

[0065] Suppose f(x) is a multivariate function containing n variables, and x=(x1,x2,...,xn) is an n-dimensional vector.

[0066] 1. Given the initial iteration point x, the initial walking step size λ, and the control accuracy ∈ (∈ is a very small positive number used to control the end of the algorithm).

[0067] 2. Given the number of iterations N, k is the current number of iterations.

[0068] 3. When k < N, randomly generate an n-dimensional vector u = (u1, u2, …, un) between (-1, 1) (-1 < ui < 1, i = 1, 2, …, n), and normalize it to obtain Let x1 = x + λu′, and complete the first step of the random walk.

[0069] 4. Calculate the function value. If f(x1) < f(x), that is, a point better than the initial value is found, then k is reset to 1, x1 is changed to x, and go back to step 2; otherwise, k = k + 1, and go back to step 3.

[0070] 5. If a better value cannot be found continuously for N times, it is considered that the optimal solution is within an N-dimensional sphere centered at the current optimal solution with the current step size as the radius (if it is three-dimensional, it is just a sphere in space). At this time, if λ < ∈, the algorithm ends; otherwise, let λ = λ2, go back to step 1, and start a new round of random walk.

[0071] 2) Generate a random walk sequence with each node as the starting point. Train the word2vec model in the deepwalk algorithm to embed each layer of the network into point cloud data. The word2vec algorithm is divided into Skip-Gram and CBOW. In this invention, the Skip-Gram model is used. Then use principal component analysis (PCA) for dimensionality reduction visualization and save the point cloud data generated by the embedding. The algorithm of Skip-Gram in word2vec can be understood as predicting the upper and lower points based on the center point. The details of the Skip-Gram model are as follows:

[0072] 1. First, select a point in the point cloud network as the input point.

[0073] 2. After having the input point, the inventor defines a parameter called skip_window, which represents the number of points the inventor selects from one side (left or right) of the current input point. In this invention, the parameter of skip_window selected by the inventor is 4, which means selecting 4 words on the left side and 4 words on the right side of the input point to enter the inventor's window. So the size of the entire window is 4 * 4 = 16. Another parameter is called num_skips, which represents how many different points the inventor selects from the entire window as the inventor's output points. When skip_window = 4 and num_skips = 4, the inventor will obtain two sets of training data in the form of (input point, output point).

[0074] 3. The neural network will output a probability distribution based on these training data, and this probability represents the output possibility of each point in the inventor's dictionary.

[0075] (3) Geographic flow clustering recognition

[0076] To cluster multi-layer temporal geographic flows, the inventors incorporated multi-lens tools from topological data analysis into persistence graphs. The Wasserstein distance between each persistence graph was calculated to cluster multi-layered geographic flows across different temporal sequences, vividly depicting their dynamic interactions and enriching research on the dynamic organization of urban space. The experimental results can provide decision support for sustainable urban management.

[0077] The basic principle behind the clustering method is as follows: if the local neighborhoods of two points have similar shapes at all resolution scales, then they are close enough to be grouped into a cluster. To compare shapes, the inventors performed the following steps:

[0078] 1) Consider Xn=(x1,…,Xn) in some metric space (X,D).

[0079] 2) Set the resolution threshold V1 <V2…<V K , and construct a VR filter

[0080] 3) Calculate x in the form of persistence diagram PD(i), i=1,…,n i A local topological summary of .

[0081] 4) For all local neighborhoods N(i) of xi and x j For N(j), i,j = 1, 2, ..., N, we compute the pairwise topological or data shape dissimilarity as the Wasserstein distance between their respective persistence graphs PD(i) and PD(j):

[0082] Here, Δ = {(x, x) | x∈R}, and γ is bi-mapped into PD(i)∪Δ to PD(j)∪Δ. The Wasserstein distance allows us to systematically quantify the similar shapes of two node neighborhoods. That is, we compute and compare all loops, holes, and other topological features in each node neighborhood.

[0083] 5) Form a distance graph G on W2(N(i), N(j)), i, j = 1, 2, ..., N, with an adjacency matrix a, where

[0084]

[0085] The cut-off point κ is defined by elbow plot or cross-validation.

[0086] 6) The connected components of G are the clusters obtained by the inventors. Therefore, persistence graph clustering utilizes distance functions and local geometric information around points to cluster multi-layered geographic flows of different time series using the Wasserstein distance. This fully considers the topological structure of the data, effectively improving clustering results, vividly depicting their dynamic interactions, and enriching research on the dynamic organization of urban space. The experimental results can provide decision support for sustainable urban management.

[0087] The specific embodiments of the present invention are:

[0088] Example 1

[0089] To cluster multi-layer temporal geographic flows, the inventors incorporated multi-lens tools from topological data analysis into persistence graphs. By calculating the Wasserstein distance between each persistence graph, they clustered multi-layered geographic flows across different temporal sequences, vividly depicting their dynamic interactions and enriching research on the dynamic organization of urban space. The experimental results can provide decision support for sustainable urban management. The specific implementation is as follows:

[0090] (1) Convert trajectory data into OD data. Download the trajectory data of Chengdu from 6:00 to 24:00 on August 3rd to 30th, 2016. Use the TransBigData library in Python to convert the downloaded trajectory data into the OD flow data required for the study.

[0091] (2) Construct a multi-layer temporal geographic network. From 6:00 am, a multi-layer traffic temporal network with a time scale of 2 hours is established, with a total of 9 layers of networks. Figure 2 As shown, ArcGIS was used to create a 500m*500m grid to segment the OD flow. The number of flows with OD points falling in the same two grids was used as the weight value between the grids to construct a weighted network with grids as nodes. The strength data is shown in Figure 3 shown.

[0092] (3) Use ArcGIS to number the created grids, regard each grid as a network node, and generate a random walk sequence algorithm. The algorithm can be understood as inputting the starting point and path length, generating a random walk node sequence, and then summarizing the adjacent nodes, randomly selecting the next node from the adjacent nodes, and generating a set of walk sequences for all the generated grids in turn.

[0093] (4) The word2vec model in the deepwalk algorithm is trained to reduce the dimension of each layer of the network into point cloud data. The word2vec algorithm is divided into Skip-Gram and CBOW. The Skip-Gram model is used in the present invention. The algorithm process of the model is as follows: Figure 4 As shown, principal component analysis (PCA) is used for dimensionality reduction, and finally TSNE is used for visualization and saving the embedded point cloud data. A total of 9 groups of point cloud data are generated, as shown Figure 5 As shown in the figure, the Skip-Gram algorithm in word2vec can be understood as predicting the upper and lower points based on the center point.

[0094] (5) In order to cluster multi-layer temporal geographic flows, the inventors introduced the multi-lens tool of topological data analysis into the persistence graph method. By calling the Gudhi library in Python, the point cloud data obtained by embedding the traffic network in different time periods is generated into a persistence graph. The persistence graph corresponding to each time period is as follows: Figure 6 shown.

[0095] (6) Calculate the Wasserstein distance between each persistence graph, cluster the multi-layer geographic flows of different time series according to the Wasserstein distance, and calculate the common value of the Wasserstein distance:

[0096]

[0097] Here, Δ = {(x, x) | x∈R}, and γ is bi-mapped from PD(i)∪Δ to PD(j)∪Δ. The Wasserstein distance allows the inventors to systematically quantify the similar shapes of two networks. By clustering, we can explore the topological similarity of each layer of the network to reveal their different life cycles of emergence, expansion, stability, contraction, and disappearance in different time periods, and depict the dynamic interactions between them. On this basis, we obtain a Wasserstein matrix for multi-layer temporal geographic flows, as follows: Figure 7 shown.

[0098] (7) Multi-layer temporal geographic flow clustering identification

[0099] The Wasserstein matrix generated by (6) is hierarchically clustered to explore the time series multi-layer geographic flow cluster identification, such as Figure 8 As shown in the figure, it can be clearly seen that the geographical flow clustering patterns from 6 to 8 in the morning, 8 to 10 in the morning, and from 16 to 18 in the afternoon are similar. Figure 2 It can be clearly seen that the number of flows from residential areas to various transportation stations increased significantly during this time period. The reason for this is probably that residents are commuting between various transportation stations during this time period, so the related geographical flow clustering patterns are similar. Figure 8 It can also be clearly seen that the geographical flow clustering patterns from 18:00 to midnight are similar. The reason for this may be that it is related to the similarity of residents' travel patterns. Figure 2 It can be clearly seen that the number of flows to various shopping malls increased significantly from 18:00 to 24:00. Figure 8 It can also be seen that the clustering patterns of the three time periods between 10 am and 4 pm are similar. Figure 2 It can be seen that the number of flows in these three time periods is relatively small. The reason for this is probably that during these three time periods, residents are commuting to get off work and not going out, so the clustering pattern of geographical flows in the same period is similar.

[0100] This example explores the cluster identification of temporal multi-layer geographic flows from the perspective of topological data analysis. It can reveal their different life cycles of occurrence, expansion, stability, contraction, and disappearance in different time periods, depict the dynamic interactions between them, enrich the study of residents' activity mobility, and provide decision support for the management of geographic flow networks such as sustainable urban transportation.

[0101] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A temporal multi-layer geographic flow clustering identification method taking into account topological data analysis, characterized in that: It includes the following steps: S1. Construction of time-series geographical flow Download the trajectory data of the target city, use the TransBigData library in Python to convert the downloaded trajectory data into the OD flow data required for research, establish a traffic time-series multi-layer network, and then use ArcGIS to create grids to segment the OD flow. Take the number of flows with OD points falling in the same two grids as the weight value between the grids, and construct a geographical flow weight network with grids as nodes; S2. Dimensionality reduction of geographical flow Use the deepwalk algorithm in network embedding. Apply the random walk model to the obtained geographical flow weight network to embed the weight networks in different time periods into point cloud data. On this basis, calculate the sequence of nodes in each layer of the network and detect dynamic urban moving communities from the time-related multi-layer network; The deepwalk algorithm is divided into two parts: random walk and generation of representation vectors; first, use the random walk algorithm to extract some vertex sequences from the graph, then, borrowing the idea of natural language processing, regard the generated vertex sequences as sentences composed of words, and all the sequences as a large corpus. Finally, use the natural language processing tool word2vec to represent each vertex as a vector with dimension d; The deepwalk algorithm specifically includes the following steps: 1) Number the grids created by ArcGIS, regard each grid as a network node, and generate an algorithm for random walk sequences. The algorithm is understood as inputting the starting point and the path length, generating a random walk node sequence, then aggregating adjacent nodes, and randomly selecting the next node from the adjacent nodes; 2) Generate random walk sequences with each node as the starting point, train the word2vec model in the deepwalk algorithm to embed each layer of the network into point cloud data, and then use principal component analysis for dimensionality reduction visualization and save the point cloud data generated by the embedding; S3. Clustering and identification of geographical flow Use the gudhi package in Python to generate the corresponding persistence diagram for the obtained point cloud data, calculate the Wasserstein distance of each layer of the network, and then use this distance for clustering and identification of geographical flow.

2. The temporal multi-layer geographic flow clustering identification method taking into account topological data analysis according to claim 1 is characterized in that: In step S1, the time scale of the traffic time-series multi-layer network is 2 hours.

3. The temporal multi-layer geographic flow clustering identification method considering topological data analysis according to claim 1 is characterized in that: The specific random walk algorithm in step 1) is as follows: Let f(x) be a multivariate function with n variables, and x=(x1,x2,...,xn) be an n-dimensional vector; Given the initial iteration point x, the initial walking step length λ, and the control precision ∈; Given the iteration control number N, and k is the current iteration number; When k < N, randomly generate an n-dimensional vector u = (u1, u2, …, un) between (-1, 1) (-1 < ui < 1, i = 1, 2, …, n), and normalize it to obtain Let x1 = x + λu', and complete the first-step walk; Calculate the function value. If f(x1)<f(x), that is, a point better than the initial value is found, then k is reset to 1, x1 is changed to x, and go back to step 2; otherwise k=k + 1, and go back to step 3; If no better value can be found continuously for N times, it is considered that the optimal solution is within the N-dimensional sphere centered on the current optimal solution with the current step length as the radius; at this time, if λ<∈, the algorithm ends; otherwise, let λ = λ2, go back to step 1, and start a new round of walking.

4. The temporal multi-layer geographic flow clustering identification method taking into account topological data analysis according to claim 3 is characterized in that: ∈ is a very small positive number used to control the end of the algorithm.

5. The temporal multi-layer geographic flow clustering identification method taking into account topological data analysis according to claim 1 is characterized in that: In step 2), in this step, the point cloud data obtained by embedding each layer of the network is subjected to principal component analysis and visual dimensionality reduction to explore their different life processes of occurrence, expansion, stability, contraction and disappearance in different time periods.

6. The temporal multi-layer geographic flow clustering identification method taking into account topological data analysis according to claim 1 is characterized in that: The specific algorithm steps of the Skip-Gram model in step 2) are as follows: First, select a point in the point cloud network as the input point; After having the input point, define a parameter called skip_window, which represents the number of points selected from one side of the current input point; define another parameter called num_skips, which represents how many different points are selected from the entire window as output points; Based on these training data, the neural network will output a probability distribution, which represents the output possibility of each point in the dictionary.

7. The method for identifying temporal multi-layer geographic flow clusters taking into account topological data analysis according to claim 1, characterized in that: In step S3, the basic principle of the clustering method is: if the local neighborhoods of two points have similar shapes at all resolution scales, they are close enough to be grouped into one cluster.

8. The temporal multi-layer geographic flow clustering identification method taking into account topological data analysis according to claim 7 is characterized by: To compare the shapes of the clusters, perform the following steps: Consider X = (x1, ..., Xn) in some metric space (X, D); Set the resolution threshold V1 <V2…<V K , and construct a VR filter Calculate x in the form of persistence diagram PD(i), i=1,…,n i Local topological summary of ; For all local neighborhoods N(i) of xi and x j For N(j), i, j = 1, 2, ..., N, we compute the pairwise topological or data shape dissimilarity as the Wasserstein distance between their respective persistence graphs PD(i) and PD(j): In (1), Δ = {(x, x) | x∈R}, γ is bi-mapped from PD(i)∪Δ to PD(j)∪Δ. The Wasserstein distance allows systematically quantifying the similar shapes of the neighborhoods of two nodes; Form a distance graph G on W2(N(i), N(j)), i, j = 1, 2, ..., N, with an adjacency matrix a, where Define the cut-off point κ by elbow plot or cross-validation; The connected components of G are the resulting clusters.

Images