Group division method and system based on individual moving mode dynamic hypergraph clustering

Abstract

The invention relates to the field of traffic transportation information engineering, and discloses a group division method and system based on individual movement mode dynamic hypergraph clustering, and the method comprises the steps: constructing an individual movement mode feature matrix under each time slice; constructing an individual movement mode hypergraph under each time slice, and modeling a'many-to-many 'complex high-order association relationship among individual movement modes under the whole time period; proposing a dynamic hypergraph clustering model based on a moving mode, and performing category division of individual associated groups; the provided model is applied to a series of empirical data sets for training and testing, and it is proved that the model can obtain good group division performance superior to that of a baseline model.

Description

Technical Field

[0001] The present invention relates to the field of transportation information engineering technology, and in particular to a group division method and system based on dynamic hypergraph clustering of individual mobility patterns. Background Art

[0002] Smart travel, also known as intelligent transportation, refers to the effective penetration and integration of traditional transportation industry and the Internet with the help of technologies and concepts such as mobile Internet, cloud computing, big data and the Internet of Things. On this basis, real-time sensing and analysis technology of driving behavior can also be integrated to realize multi-mode and multi-standard dynamic navigation of public travel and improve travel efficiency.

[0003] Categorizing travel groups is an important strategy for achieving smart travel, which helps to better understand the needs and preferences of different user groups, thereby providing more personalized and customized services, as well as more refined proactive management.

[0004] The relevant technologies generally divide the groups from the perspectives of time and space, and then further infer the group categories based on time and space. However, the classification methods in the relevant technologies have a high error rate. For example, due to chance or other reasons, the travel time and travel space of an individual meet the requirements of a certain group and are classified into the corresponding group, but the actual situation is that the individual belongs to another group. This will lead to inaccurate group classification, which may further affect the formulation of corresponding policy opinions and the development of marketing activities. Summary of the invention

[0005] The present invention provides a group classification method and system based on dynamic hypergraph clustering of individual mobility patterns, which is used to solve the defects of inaccurate group classification and high misjudgment rate in related technologies. In the scheme of the present application, a hypergraph is constructed to cluster the travel records of individuals, thereby improving the accuracy of group classification.

[0006] The present invention provides a group division method based on dynamic hypergraph clustering of individual mobility patterns, comprising: obtaining a set of historical travel records of N individuals, cutting the set in time dimension by day, obtaining a set of historical travel records of N individuals under T continuous time slices, extracting D-dimensional travel features for each individual, and constructing a mobility pattern feature matrix of the N individuals;

[0007] For any time slice, each individual is regarded as a node, and N nodes are obtained; the neighborhood group of the individual is regarded as the hyperedge of the hypergraph, and N hyperedge sets are obtained to construct the hypergraph of the individual movement pattern under each time slice;

[0008] Construct a dynamic hypergraph clustering model based on mobility patterns, that is, relying on the Transformer "encoder-decoder" structure to learn the deep expression of individual mobility patterns; relying on the hypergraph convolutional network model, learn the deep expression of neighborhood group mobility patterns in local space; use the attention mechanism to integrate the deep expressions of the above individual and neighborhood group mobility patterns, and use the bidirectional hypergraph convolutional network to capture the change value of the hypergraph topological structure under the cross-action of forward time propagation and backward time propagation, and dynamically update the hypergraph structure; complete the training and testing of the model;

[0009] The historical travel record set of M individuals is input into the dynamic hypergraph clustering model of mobility patterns to conduct an instance verification of the classification of individual associated groups.

[0010] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, the D-dimensional travel features are extracted for each individual to construct a mobility pattern feature matrix of N individuals, including:

[0011] At any time slice τ, D-dimensional travel features are extracted for each individual, and the mobility pattern feature matrix X of N individuals is constructed: τ ∈R N×D , The individual mobility pattern feature matrix X = {X1,…,X τ ,…,X T}, where x τ,i (1≤i≤N) is the characteristic expression of the mobility pattern of the i-th individual in the τ-th time slice.

[0012] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, the construction of the individual mobility pattern hypergraph under each time slice includes:

[0013] For any time slice τ, the N individuals are regarded as nodes of the hypergraph to obtain N nodes; the similarity of the movement patterns of any two individuals is measured using the Euclidean distance; two nodes of the nearest neighbor are connected, and the inverse function of the Euclidean distance is used as the corresponding edge weight; according to the principle of "the smaller the Euclidean distance, the more similar the movement patterns", the k individuals with the most similar movement patterns are determined to form the i-th neighborhood group under the τ-th time slice.

[0014] The neighborhood group is used as a hyperedge to obtain a set of N hyperedges, which constitute the initial division result of the individual association group, and the individual mobility pattern hypergraph G under the τth time slice is obtained. τ =(V τ ,E τ ,X τ ,W τ ,Hτ ), where V τ is the node set that expresses all individuals, E τ is a set of hyperedges that express the movement pattern of the neighborhood group, X τ is a set of node attributes that express individual mobility patterns, W τ is a set of hyperedge weights that express the importance of the neighborhood group, H τ is the association matrix expressing the affiliation between individuals and groups, H τ =[h(v,e)] τ ∈R N×N , h(v,e) represents the affiliation relationship between node v and hyperedge e;

[0015] The affiliation relationship between the node v and the hyperedge e conforms to the following formula:

[0016]

[0017] In formula (1), v∈e indicates that the node is on the hyperedge, D τ,v =[d(v)] is the degree matrix of node v at time slice τ, which describes the current state of node v in the hypergraph G τ The importance of ; d(v) is the sum of the weights of all hyperedges connected to vertex v, calculated by formula (2):

[0018] d(v)=∑ e∈E ω(e)h(v,e) (2)

[0019] D τ,e =[d(e)] is the degree matrix of the hyperedge e at the time slice τ, which describes the hyperedge e in the hypergraph G at the time slice τ. τ The importance of d(e) is the sum of the number of nodes connected by hyperedge e, which is calculated by formula (3):

[0020] d(e)=∑ v∈V h(v,e) (3)

[0021] According to the group division method based on individual mobility pattern dynamic hypergraph clustering provided by the present invention, a dynamic hypergraph clustering model based on mobility pattern is constructed, including:

[0022] Using the Transformer model, we learn deep representations of individual movement patterns.

[0023] Using the hypergraph convolutional network model, we learn the deep expression of the neighborhood group movement pattern in the local space;

[0024] Fusion of the individual and neighborhood group movement patterns to perform dynamic hypergraph deep clustering;

[0025] The training and testing process of the dynamic hypergraph clustering model of mobility patterns is carried out to complete the dynamic division of group categories.

[0026] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, the Transformer model is used to learn the deep expression of the individual mobility pattern, including:

[0027] Through the working principle of the multi-head attention mechanism of the built-in encoder of the Transformer model, the deep expression C of the individual movement pattern in the lth (1≤l≤L) convolutional layer of the individual movement pattern feature matrix X is learned autonomously (l) , C (l) Calculated by formula (4):

[0028]

[0029] Among them, σ ReLU is a linear activation function. When l = 1, C (l-1) =C 0 is the individual mobility pattern feature matrix X; LN is the convolutional layer regularization operation; dropout is the random pruning operation of the fully connected layer; FeedForward is the fully connected network, Res (l) is the residual unit of the Transformer model decoder; U1 and U2 are the weight parameters of the model to be trained, b1 and b2 are bias values; MultiHead is a multi-head attention mechanism;

[0030] Through the built-in decoder of the Transformer model, after L convolutional layers, the reconstructed Complete the decoding operation;

[0031] Minimize the average reconstruction error To obtain the deep expression of individual mobility patterns that is closest to the actual situation, where ||·|| Fro is the F norm, where the average reconstruction error is generated by the encoding and decoding process, the average reconstruction error The expression of is shown in formula (5):

[0032]

[0033] For any time slice τ, according to the feature matrix X τ The distribution of and the real group category to which each individual belongs, calculate the real probability distribution of individual i belonging to category j, and then obtain the probability distribution curve GT of individual mobility mode category in the real scene under time slice τ;

[0034] According to the deep expression C of the individual movement pattern of the Lth convolutional layer(L) The distribution of individual mobility mode categories predicted by Transformer under time slice τ is obtained.

[0035] Determine the cumulative error of the probability distribution curve GT and the probability distribution curve TR in all time slices Determine the group classification error after learning the Transformer time series features, where Calculated by formula (6),

[0036]

[0037] Among them, GT τij and TR τij They respectively refer to the probability that individual i belongs to category j in curve GT and curve TR under time slice τ, and KL is the "Kulbeck-Leibler divergence" distribution curve.

[0038] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, the hypergraph convolutional network model is used to learn the deep expression of the neighborhood group mobility pattern in the local space, including:

[0039] Based on the hypergraph G τ As well as the initial division results of individual associated groups, the hypergraph convolutional network operation is performed to capture the local spatial topological changes of the neighborhood group. For the lth convolutional layer, the deep expression Z of the neighborhood group movement pattern (l) Calculated by formula (7):

[0040]

[0041] Where Θ is the convolution hyperparameter; when l = 1, Z (l-1) =Z 0 is the feature matrix X; H τ Transponse Indicates H τ The transposed matrix of .

[0042] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, the individual and neighborhood group mobility patterns are integrated to perform dynamic hypergraph deep clustering, including:

[0043] The individual movement patterns are expressed in C (l) Deep expression Z with neighborhood group movement pattern (l) Through formula (8), we can get C under the lth convolutional layer. (l) and Z (l) The splicing result CZ (l) , CZ (l) n For CZ(l) The nth vector sequence of ;

[0044] By introducing the attention mechanism, we can get the new expression of the neighborhood group movement pattern under the lth convolutional layer through formula (9):

[0045]

[0046]

[0047] Among them, q is the query vector, α n is the attention weight of the nth element, s(CZ(l) n ,q) is the attention scoring function used to measure CZ(l) n The similarity between the query vector q is quantified using the dot product model;

[0048] based on The distribution of the hypergraph topology under the interaction of forward time propagation and backward time propagation is captured by ΔG. τ , that is, the change in the topological structure of the hypergraph that belongs to the first or second hypergraph but not both, is calculated as shown in formula (10):

[0049] ΔG τ =difference(G 1→τ ,G T→τ )=(G 1→τ ∪G T→τ ) - (G 1→τ ∩

[0050] G T→τ ) (10)

[0051] Based on G τ and ΔG τ , update the corresponding dynamic hypergraph G τ , as the input of the hypergraph convolutional network at the convolutional layer l+1, and the output result C corresponding to the Transformer at the l+1 layer (l+1) After attention fusion, the hypergraph convolutional network operation is performed to obtain the probability distribution results of the group categories under the l+1th to Lth convolutional layers.

[0052]

[0053] Among them, W′ τ , H′ τ , D′ τ,v , D′ τ,e Dynamic Hypergraph The hyperedge weight set, incidence matrix, node degree matrix and hyperedge degree matrix of ;

[0054] For any time slice τ, according to the deep expression Z of the individual movement pattern of the lth convolutional layer (L) The distribution of individual mobility mode categories is obtained by hypergraph clustering under time slice τ, and the cumulative error between the probability distribution curve HC and the curve GT in all time slices is measured. Measuring the group classification error of deep clustering of dynamic hypergraphs that integrate the mobility patterns of individuals and neighborhood groups, Calculated by formula (12):

[0055]

[0056] Among them, HC τij is the probability that individual i belongs to category j in curve HC under time slice τ.

[0057] According to the group division method based on individual mobility pattern dynamic hypergraph clustering provided by the present invention, the training and testing process of the mobility pattern dynamic hypergraph clustering model is performed to complete the dynamic division of group categories, including:

[0058] The overall error is calculated using the following formula:

[0059]

[0060] Among them, γ1, γ2 and γ3 are as well as The weight coefficient of

[0061] Relying on the N-bit individual mobility pattern feature matrix X under T time slices, the dynamic hypergraph deep clustering model is trained to minimize the cumulative error Until The value no longer changes, the model converges, and the model training process is completed;

[0062] Input the N-bit individual movement pattern feature matrix X under T+1 time slices to the trained model T+1 , determine the individual association groups in the test data set.

[0063] According to the group division method based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, after completing the training and testing of the model, it also includes an example verification of the classification of individual associated groups, specifically including:

[0064] Input the M-bit individual mobility pattern feature matrix X′ under T time slices T, using the method established by the present invention, the individual association groups in the validation data set are found. The proposed method is applied to the public transportation passenger card swiping data set, and the partitioning performance of the proposed method and its internal modules is verified through multiple group comparisons and ablation experiments.

[0065] The present invention also provides a group division system based on individual mobility pattern dynamic hypergraph clustering, comprising:

[0066] A data acquisition module is used to obtain historical travel record data of N individuals;

[0067] Individual mobility pattern feature matrix extraction module, used to extract D-dimensional features from historical travel records and quantitatively express the mobility pattern of each individual;

[0068] An individual mobility pattern hypergraph set construction module is used to construct a hypergraph set for the current period according to the current individual mobility pattern feature matrix;

[0069] The deep expression learning module for individual movement patterns is used to learn the deep expression of individual movement patterns based on the Transformer "encoder-decoder" structure;

[0070] Neighborhood group movement pattern deep expression learning module, which is used to learn the deep expression of neighborhood group movement pattern in local space based on the hypergraph convolutional network model;

[0071] The dynamic hypergraph deep clustering module is used to use the attention mechanism to integrate the deep expression of the above-mentioned individual and neighborhood group movement patterns, and use the bidirectional hypergraph convolutional network to capture the change value of the hypergraph topological structure under the cross-action of forward time propagation and backward time propagation, and update its dynamic hypergraph structure;

[0072] The training and testing module is used to input the historical travel record set of N individuals into the dynamic hypergraph clustering model of the mobility pattern to perform training and testing on the classification of individual associated groups.

[0073] The verification module is used to input the historical travel record set of M individuals into the dynamic hypergraph clustering model of mobile patterns to perform instance verification of the classification of individual associated groups.

[0074] In the group division method and system based on dynamic hypergraph clustering of individual mobility patterns provided by the present invention, a mobility pattern matrix and a mobility pattern hypergraph can be constructed based on a set of individual travel records, wherein the constructed mobility pattern matrix can divide individuals into different categories from the time plus space perspective of "one side connecting two points", and the mobility pattern hypergraph can consider the implicit association relationship between individuals from a "many-to-many" perspective on the basis of the time plus space classification standard, and characterize groups with intrinsic association relationships through neighborhood groups in the hypergraph. In the scheme of the present application, accurate identification of the "many-to-many" association relationship in individual travel patterns is achieved to achieve accurate division of group categories. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the present invention or 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 creative work.

[0076] Figure 1 This is one of the flow diagrams of a group division method based on dynamic hypergraph clustering of individual mobility patterns provided by an embodiment of the present invention;

[0077] Figure 2 This is a second flow chart of a method for group division based on dynamic hypergraph clustering of individual mobility patterns provided by an embodiment of the present invention;

[0078] Figure 3 It is a structural schematic diagram of a group division system based on dynamic hypergraph clustering of individual mobility patterns provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0080] Figure 1 It is one of the flow charts of the method and system for group division based on dynamic hypergraph clustering of individual mobility patterns provided in an embodiment of the present invention.

[0081] Figure 2 This is the second flow chart of the method and system for group division based on dynamic hypergraph clustering of individual mobility patterns provided in an embodiment of the present invention.

[0082] like Figure 1 and Figure 2 As shown, this embodiment provides a group division method based on dynamic hypergraph clustering of individual mobility patterns, including:

[0083] Step 101, obtaining a set of historical travel records of N individuals, cutting the set in time dimension by day, obtaining a set of historical travel records of N individuals under T consecutive time slices, extracting D-dimensional travel features for each individual, and constructing a mobility pattern feature matrix of the N individuals;

[0084] Step 102, for any time slice, each individual is regarded as a node to obtain N nodes; the neighborhood group of the individual is regarded as a hyperedge of the hypergraph to obtain N hyperedge sets, and a hypergraph of the individual movement pattern under each time slice is constructed;

[0085] Step 103, construct a dynamic hypergraph clustering model based on mobility patterns, that is, relying on the Transformer "encoding-decoding" structure, learn the deep expression of individual mobility patterns; relying on the hypergraph convolutional network model, learn the deep expression of neighborhood group mobility patterns in local space; using the attention mechanism, integrate the deep expressions of the above individual and neighborhood group mobility patterns, and use the bidirectional hypergraph convolutional network to capture the hypergraph topological structure change value under the cross-action of forward time propagation and backward time propagation, dynamically update the hypergraph structure, and complete the training and testing of the model;

[0086] Step 104 , inputting the historical travel record set of M individuals into the dynamic hypergraph clustering model of the mobility pattern to perform instance verification of the classification of the individual associated groups.

[0087] The individuals may be people. Furthermore, the travel record set may include the travel time and travel location of several people who traveled. Furthermore, it may include different means of transportation, as well as the time and place of boarding the means of transportation, the time and place of leaving the means of transportation, etc.

[0088] In practical applications, the above travel record set can be obtained from a municipal transportation system, for example, from a Beijing public transportation database.

[0089] In practical applications, we can construct a matrix with N rows and D columns, X = [x1, x2, ..., x N ]∈R N×D Among them, x i(i=1,2,…,N) represents the ith individual, and column D refers to the features in the travel record set corresponding to the individual, where the features in the travel record set can include degree, degree centrality, betweenness centrality, closeness centrality, eigenvector centrality, and clustering coefficient. Degree is the number of edges directly connected to a node (individual), which measures the basic indicator of the number of direct connections of a node in the network and reflects the basic connectivity of the node; betweenness centrality measures the number of times a node appears in all shortest paths, reflects the ability of the node to control information flow and resource flow, and is an important indicator for evaluating the importance of the node; closeness centrality measures the average shortest path length from a node to all other nodes in the network, reflects the accessibility of the node in the network, and is an indicator for evaluating the superiority of the node's position in the network; eigenvector centrality not only considers the number of neighbors directly connected to the node, but also considers the importance of neighboring nodes, comprehensively reflects the influence of the node in the network, and is suitable for evaluating the long-term importance of the node in the network; clustering coefficient measures the tightness of the connection between the neighboring nodes of a node, reflects the local connectivity of the network, and is of great significance for understanding the community structure of the network and the social environment of the node.

[0090] Hypergraph is a generalized graph structure. Compared with the traditional graph structure where an edge can only connect two vertices, in the hypergraph structure, an edge can connect any number of vertices. The advantage of the hypergraph structure is that it can distinguish between the hierarchical and sub-hierarchical structures of information, and any complex information structure can be established by these two types of information.

[0091] In practical applications, the nodes and edge weights of the hypergraph can be determined based on a number of selected individuals to construct an individual mobility pattern hypergraph G m ={V,E,W}. V, E, and W represent the vertex set, hyperedge set, and edge weight set of the hypergraph, respectively. Its incidence matrix is Its internal element h(v,e) represents the affiliation between node v and hyperedge e, and v∈e indicates that the node is on the hyperedge. v and D e are the degree matrices of v and e, respectively, describing the importance of any node or hyperedge in the hypergraph. For a node v∈V in the hypergraph, its degree d(v) is the sum of the weights of all hyperedges connected to the vertex v. The set of d(v) constitutes the node degree matrix D of the hypergraph. v =[d(v)]. For a hyperedge e∈E in a hypergraph, its degree d(e) is the sum of the number of nodes in the hypergraph connected by the edge e. The set of d(e)s constitutes the hyperedge degree matrix D of the hypergraph. e =[d(e)].

[0092] When multiple individuals complete a social activity together, their mobility patterns show similarities. Multiple individuals with similar mobility patterns may influence each other. The more similar the mobility patterns are, the higher the degree of mutual influence between individuals. Compared with the traditional graph structure that only describes the correlation between pairs of nodes, the hypergraph aggregates the interactive correlation information of multiple adjacent nodes, which is more conducive to discovering potential modules in the graph and then modeling the correlation characteristics between multiple individuals. Based on this, the concept of hypergraph is introduced in the scheme of this embodiment. By constructing a hypergraph of individual mobility patterns, a complex high-order "many-to-many" correlation relationship between individual mobility patterns can be modeled.

[0093] In practical applications, the deep expression of the mobility pattern matrix can be performed based on transformer temporal feature learning. Specifically, the transformer model can be introduced, and the deep expression of the individual mobility patterns of the mobility pattern matrix in the convolutional layer can be autonomously learned through the working principle of the multi-head attention mechanism of the encoder built into the transformer model.

[0094] A neighborhood is a topological structure on a set. A neighborhood is a special curve. If a point is the center point, any open interval is called the neighborhood of that point.

[0095] In this embodiment, a hypergraph convolutional network (HGCN) operation can be performed to capture the local spatial topological changes of the neighborhood group. Furthermore, for each convolutional layer of the hypergraph convolutional network, the deep expression of the neighborhood group, that is, the second expression, can be determined.

[0096] Clustering is the process of dividing a collection of physical or abstract objects into classes of similar objects. A cluster is a collection of data objects that are similar to objects in the same cluster but different from objects in other clusters.

[0097] In the group category classification method provided in this embodiment, a mobile pattern matrix and a mobile pattern hypergraph can be constructed based on the individual travel record set, wherein the constructed mobile pattern matrix can classify individuals into different categories from the time plus space perspective of "one side connecting two points". Furthermore, the mobile pattern hypergraph can consider the implicit association relationship between individuals from a "many-to-many" perspective on the basis of the time plus space classification standard, and represent the groups with intrinsic association relationships through the neighborhood groups in the hypergraph. The solution of the present application realizes the accurate classification of group categories by considering the "many-to-many" association relationship in the individual travel patterns.

[0098] In an exemplary embodiment, extracting D-dimensional travel features for each individual and constructing a mobility pattern feature matrix of N individuals includes:

[0099] At any time slice τ, D-dimensional travel features are extracted for each individual, and the mobility pattern feature matrix X of N individuals is constructed: τ ∈R N×D , The individual mobility pattern feature matrix X = {X1,…,X τ ,…,X T}, where x τ,i (1≤i≤N) is the characteristic expression of the mobility pattern of the i-th individual in the τ-th time slice.

[0100] In practical applications, the travel record set obtained is definitely a travel record within a long period of time. In the process of group classification, this large-scale travel record set can be divided according to time to obtain several time slices. In implementation, the time slice can be in days. For example, the travel record set obtained can be a complete set of a certain year or a certain month. For example, all travel records of these individuals in August are obtained and divided according to days, such as the travel record set of August 1, the travel record set of August 2, the travel record set of August 3, etc. By cutting the travel record set in the time dimension, a relatively long time span can be divided into a small time span. In this way, the group classification in the time dimension is achieved in this embodiment.

[0101] After the individual travel record set is segmented in the time dimension, the historical travel record set of the individual under several continuous time slices can be obtained. Furthermore, for each time slice, D-dimensional travel features can be extracted for each individual, and the mobile mode feature matrix X of N individuals can be constructed. τ ∈R N×D , Thus, the individual mobility pattern feature matrix sequence X={X1,…,X τ ,…,X T}. Among them, x τ,i (1≤i≤N) is the characteristic expression of the mobility pattern of the i-th individual in the τ-th time slice. The association network between individuals is modeled based on graph theory. Specifically, for any time slice τ, each individual is regarded as a node on the graph; the connectivity between nodes is connected by edges; the inverse function of the Euclidean distance between the attributes of each pair of nodes is used as the corresponding edge weight.

[0102] Furthermore, for any time slice τ, according to the principle that “the smaller the Euclidean distance, the more similar the mobility pattern”, the k individuals with the most similar mobility patterns are circled to form the i-th neighborhood group under the τ-th time slice To express the complex "many-to-many" relationship of k individuals in a specific time slice. The neighborhood groups obtained above are analogized to hyperedges, and N hyperedge sets are obtained, which constitute the initial division result of the individual association group.

[0103] In an exemplary embodiment, constructing an individual mobility pattern hypergraph under each time slice includes:

[0104] For any time slice τ, the N individuals are regarded as nodes of the hypergraph to obtain N nodes; the similarity of the movement patterns of any two individuals is measured using the Euclidean distance; two nodes of the nearest neighbor are connected, and the inverse function of the Euclidean distance is used as the corresponding edge weight; according to the principle of "the smaller the Euclidean distance, the more similar the movement patterns", the k individuals with the most similar movement patterns are determined to form the i-th neighborhood group under the τ-th time slice.

[0105] The neighborhood group is used as a hyperedge to obtain a set of N hyperedges, which constitute the initial division result of the individual association group, and the individual mobility pattern hypergraph G under the τth time slice is obtained. τ =(V τ ,E τ ,X τ ,W τ ,H τ ), where V τ is the node set that expresses all individuals, E τ is a set of hyperedges that express the movement pattern of the neighborhood group, X τ is a set of node attributes that express individual mobility patterns, W τ is a set of hyperedge weights that express the importance of the neighborhood group, H τ is the association matrix expressing the affiliation between individuals and groups, H τ =[h(v,e)] τ ∈R N×N , h((v,e) represents the affiliation relationship between node v and hyperedge e;

[0106] The affiliation relationship between the node v and the hyperedge e conforms to the following formula:

[0107]

[0108] In formula (1), v∈e indicates that the node is on the hyperedge, D τ,v =[d(v)] is the degree matrix of node v at time slice τ, which describes the current state of node v in the hypergraph G τ The importance of ; d(v) is the sum of the weights of all hyperedges connected to vertex v, calculated by formula (2):

[0109] d(v)=∑ e∈Eω(e)h(v,e) (2)

[0110] D τ,e =[d(e)] is the degree matrix of the hyperedge e at the time slice τ, which describes the hyperedge e in the hypergraph G at the time slice τ. τ The importance of d(e) is the sum of the number of nodes connected by hyperedge e, which is calculated by formula (3):

[0111] d(e)=∑ v∈V h(v,e) (3)

[0112] In an exemplary embodiment, the construction of a dynamic hypergraph clustering model based on mobility patterns includes:

[0113] Using the Transformer model, we learn deep representations of individual movement patterns.

[0114] Using the hypergraph convolutional network model, we learn the deep expression of the neighborhood group movement pattern in the local space;

[0115] Fusion of the individual and neighborhood group movement patterns to perform dynamic hypergraph deep clustering;

[0116] Conduct the training and testing process of the dynamic hypergraph clustering model of mobility patterns.

[0117] In an exemplary embodiment, the method of using a Transformer model to learn a deep representation of an individual movement pattern includes:

[0118] Through the working principle of the multi-head attention mechanism of the built-in encoder of the Transformer model, the deep expression C of the individual movement pattern in the lth (1≤l≤L) convolutional layer of the individual movement pattern feature matrix X is learned autonomously (l) , C (l) Calculated by formula (4):

[0119]

[0120] Among them, σ ReLU is a linear activation function. When l = 1, C (l-1) =C 0 is the individual mobility pattern feature matrix X; LN is the convolutional layer regularization operation; dropout is the random pruning operation of the fully connected layer; FeedForward is the fully connected network, Res (l) is the residual unit of the Transformer model decoder; U1 and U2 are the weight parameters of the model to be trained, b1 and b2 are bias values; MultiHead is a multi-head attention mechanism;

[0121] Through the built-in decoder of the Transformer model, after L convolutional layers, the reconstructed Complete the decoding operation;

[0122] Minimize the average reconstruction error To obtain the deep expression of individual mobility patterns that is closest to the actual situation, where ||·|| Fro is the F norm, where the average reconstruction error is generated by the encoding and decoding process, the average reconstruction error The expression of is shown in formula (5):

[0123]

[0124] For any time slice τ, according to the feature matrix X τ The distribution of and the real group category to which each individual belongs, calculate the real probability distribution of individual i belonging to category j, and then obtain the probability distribution curve GT of individual mobility mode category in the real scene under time slice τ;

[0125] According to the deep expression C of the individual movement pattern of the Lth convolutional layer (L) The distribution of individual mobility mode categories predicted by Transformer under time slice τ is obtained.

[0126] Determine the cumulative error of the probability distribution curve GT and the probability distribution curve TR in all time slices Determine the group classification error after learning the Transformer time series features, where Calculated by formula (6),

[0127]

[0128] Among them, GT τij and TR τij They respectively refer to the probability that individual i belongs to category j in curve GT and curve TR under time slice τ, and KL is the "Kulbeck-Leibler divergence" distribution curve.

[0129] In an exemplary embodiment, the method of using a hypergraph convolutional network model to learn a deep expression of a neighborhood group movement pattern in a local space includes:

[0130] Based on the hypergraph G τ As well as the initial division results of individual associated groups, the hypergraph convolutional network operation is performed to capture the local spatial topological changes of the neighborhood group. For the lth convolutional layer, the deep expression Z of the neighborhood group movement pattern (l) Calculated by formula (7):

[0131]

[0132] Where Θ is the convolution hyperparameter; when l = 1, Z (l-1) =Z 0 is the feature matrix X; H τ Transponse Indicates H τ The transposed matrix of .

[0133] In an exemplary embodiment, the individual and neighborhood group movement patterns are integrated to perform dynamic hypergraph deep clustering, including:

[0134] The deep expression of individual mobility pattern C(l) and the deep expression of neighborhood group mobility pattern Z(l) are concatenated by formula (8) to obtain the concatenation result CZ(l) of C(l) and Z(l) under the lth convolutional layer, CZ (l) n For CZ (l) The nth vector sequence of ;

[0135] By introducing the attention mechanism, we can get the new expression of the neighborhood group movement pattern under the lth convolutional layer through formula (9):

[0136]

[0137]

[0138] Among them, q is the query vector, α n is the attention weight of the nth element, s(CZ(l) n ,q) is the attention scoring function used to measure CZ(l) n The similarity between the query vector q is quantified using the dot product model;

[0139] based on The distribution of the hypergraph topology under the interaction of forward time propagation and backward time propagation is captured by ΔG. τ , that is, the change in the topological structure of the hypergraph that belongs to the first or second hypergraph but not both, is calculated as shown in formula (10):

[0140] ΔG τ =difference(G 1→τ ,G T→τ )=(G 1→τ ∪G T→τ ) - (G 1→τ ∩

[0141] G T→τ ) (10)

[0142] Based on G τ and ΔG τ , dynamically update the corresponding dynamic hypergraph G τ , as the input of the hypergraph convolutional network at the convolutional layer l+1, and the output result C corresponding to the Transformer at the l+1 layer (l+1) After attention fusion, the hypergraph convolutional network operation is performed to obtain the probability distribution results of the group categories under the l+1th to Lth convolutional layers.

[0143]

[0144] Among them, W′ τ , H′ τ , D′ τ,v , D′ τ,e Dynamic Hypergraph The hyperedge weight set, incidence matrix, node degree matrix and hyperedge degree matrix of ;

[0145] For any time slice τ, according to the deep expression Z of the individual movement pattern of the lth convolutional layer (L) The distribution of individual mobility mode categories is obtained by hypergraph clustering under time slice τ, and the cumulative error between the probability distribution curve HC and the curve GT in all time slices is measured. Measuring the group classification error of deep clustering of dynamic hypergraphs that integrate the mobility patterns of individuals and neighborhood groups, Calculated by formula (12):

[0146]

[0147] Among them, HC τij is the probability that individual i belongs to category j in curve HC under time slice τ.

[0148] In an exemplary embodiment, the training and testing process of the mobile pattern dynamic hypergraph clustering model includes:

[0149] The overall error is calculated using the following formula:

[0150]

[0151] Among them, γ1, γ2 and γ3 are as well as The weight coefficient of

[0152] Relying on the N-bit individual mobility pattern feature matrix X under T time slices, the mobility pattern dynamic hypergraph clustering model is trained to minimize the cumulative error Until The value no longer changes, the model converges, and the model training process is completed;

[0153] Input the N-bit individual movement pattern feature matrix X under T+1 time slices to the trained model T+1 , determine the individual association groups in the test data set.

[0154] In an exemplary embodiment, after completing the training and testing of the model, the example verification of the classification of individual associated groups is also included, which specifically includes:

[0155] Input the M-bit individual mobility pattern feature matrix X′ under T time slices T , using the method established by the present invention, the individual association groups in the validation data set are found. The proposed method is applied to the public transportation passenger card swiping data set, and the partitioning performance of the proposed method and its internal modules is verified through multiple group comparisons and ablation experiments.

[0156] The following is a specific example to illustrate the group classification method provided by the solution of the present application.

[0157] (1) Working environment

[0158] The PyTorch framework is used to write the model code. All experimental codes are compiled and executed on the Linux software environment and run on a hardware workstation equipped with an "Intel(R) Xeon(R) CPU E5-2620v4@2.10GHz" CPU and an "NVIDIAP10016G Tesla" GPU graphics card.

[0159] (2) Introduction to experimental dataset

[0160] The experimental data set of the present invention is a data set of public transportation passengers swiping cards in a certain place. The transaction period is August 2019, which is 790,000 travel records generated by 13,962 individuals. The present invention has obtained the group categories of 2,792 (20%) individuals in advance. Among them, the first category contains 848 individuals, whose travel time and space are very regular, mainly composed of commuting office workers; the second category contains 690 individuals, whose travel time is regular, but whose travel space is irregular, mainly composed of office workers with irregular work places; the third category contains 724 individuals, whose travel time is irregular, but whose travel space is regular, mainly composed of office workers with irregular work hours; the fourth category contains 530 individuals, whose travel time and space are irregular, and are often composed of delivery personnel, domestic workers, repairmen, etc. The travel characteristics of these groups constitute a label matrix to verify the performance of the group classification results.

[0161] (3) Benchmark model selection

[0162] In the comparison and ablation experiments, 7 baseline models were selected to compare the performance with the proposed model on a given dataset.

[0163] k-means clustering model: an unsupervised learning method that discovers hidden structures in data by dividing observations in a data set into different groups or clusters;

[0164] AE (Autoencoder): The built-in encoder maps the input data to a low-dimensional latent space, and then performs the clustering task in this space;

[0165] IDEC (Improved deep embedded clustering): Improved from deep embedded clustering (DEC), it uses pre-trained autoencoder weights to map the original high-dimensional features to a low-dimensional space and perform clustering tasks in this space while maintaining the local structure;

[0166] SDCN (Structured deep clustering network): constructs a k-nearest neighbor graph based on feature similarity, performs autoencoder pre-training, and captures the aggregation structure relationship between data;

[0167] AGCN (Attention-driven graph clustering network): extracts multi-scale node attributes and graph topology feature information, dynamically fuses them, and performs data clustering tasks;

[0168] CaEGCN (Cross-attention Fusion based Enhanced Graph Convolutional Network): It uses a cross-attention deep clustering framework to fuse the auto-encoding module and the graph convolution module to aggregate heterogeneous data.

[0169] RGCC (Robust Graph Convolutional Clustering): A clustering algorithm that combines graph convolutional neural networks. It does not require knowing the number of clusters in advance. Instead, it uses group convolutional neural networks to globally optimize continuous objectives, reduce dimensionality, and achieve clustering.

[0170] (4) Experimental parameter setting

[0171] The "AdamW" optimizer was used to perform 200 iterations of training on the given data set to carry out group segmentation performance research experiments. The learning rate of all models is set to 10-4. In order to ensure fair comparison of all methods, the present invention selects the recommended network configuration parameters according to each baseline model for optimization settings. For K-means, we repeated the algorithm 11 times to obtain the best clustering results. The network size of AE and IEDC is set to "500-500-2000-10-2000-500-500". The network size of the encoder and GCN module built into SDCN is set to "500-500-2000-10". The network size of the CAE and GAE modules built into CaEGCN are both configured to "500-10-500-500". For the method proposed in the present invention, the weight coefficients γ1, γ2, and γ3 in Formula 11 are set to 1, 0.1, and 0.01, respectively. The built-in AE and HGCN of the proposed method both have a 4-layer network structure, which are set to 500, 500, 2000 and 10 respectively. The total number of iterative training is 200 times.

[0172] (5) Selection of evaluation indicators

[0173] The clustering performance of each model is evaluated by four common indicators, namely Accuracy (ACC), Normalized Virtual Information (NMI), Average Rand Index (ARI) and F1-score (F1). Generally speaking, the larger the evaluation index, the better the effect.

[0174] (6) Experimental results and analysis

[0175] Table 1 shows the experimental results of group segmentation comparison of various models on a data set of public transportation passengers swiping cards. In general, the proposed method achieves the best clustering performance (see the bold data in Table 1). Compared with the baseline model RGCC with the second best group segmentation performance, the proposed method has achieved rapid growth in various indicators, namely: ACC increased by 4.59%, NMI increased by 0.57%, ARI increased by 2.49%, and F1 increased by 4.82%, which proves that the present invention has good group segmentation performance. This is attributed to the fact that the method proposed in the present invention has a hypergraph module and a feature fusion module. The former can fully learn the similarity of deep mobility patterns between individuals and domain groups; the latter can effectively combine the mobility pattern characteristics of the individual itself and its neighborhood groups, better describe the implicit clustering characteristics of the target individual, and thus improve the performance of group segmentation.

[0176] Table 1 Comparative experimental results of various models on a public transportation passenger card swiping dataset

[0177]

[0178] Table 2 reflects the improvement of the group division performance by the hypergraph module and feature fusion module of the proposed method. The second row of data in the table is the distribution of the group division performance when the proposed method only carries the first module; the third row of data in the table is the distribution of the group division performance when the proposed method carries the above two modules. It can be seen from the table that adding the hypergraph module can increase the ACC of the proposed model by 1.93%, NMI by 2.08%, ARI by 4.11%, and F1 by 1.90%, which confirms that the hypergraph module can effectively explain the complex many-to-many association characteristics between individuals; after adding the feature fusion module, the proposed model can increase the ACC by 2.57%, NMI by 4.81%, ARI by 6.99%, and F1 by 1.48% on the basis of the hypergraph module. This fully demonstrates that the design of the hypergraph module and the feature fusion module of the present invention are both helpful to improve the accurate division performance of the group.

[0179] Table 2 Ablation test results of the built-in module of the proposed method on a public transportation passenger card swiping dataset in a certain place

[0180]

[0181] Combined with the above examples, it can be determined that the group classification method provided by this application has at least the following beneficial effects:

[0182] 1) Capture the deep expression characteristics of the mobility patterns of individuals and their neighborhood groups, use hypergraph convolutional networks for deep association learning and dynamic clustering, achieve accurate classification of group categories, and improve the accuracy and robustness of group classification;

[0183] 2) Effectively serve the field of traffic safety supervision in public places, provide important theoretical references for reasonably adjusting the contradiction between traffic supply and demand, effectively allocating urban passenger transport resources, accurately managing crowded passenger flow in places, effectively ensuring the safe operation of urban transportation systems, and vigorously improving public travel efficiency. It is expected to provide more refined group segmentation solutions for urban traffic planning, social media analysis, disease transmission prediction and other fields, and promote the development and progress of related fields.

[0184] The group category classification system provided by the present invention is described below. The group category classification system described below and the group category classification method described above can be referenced to each other.

[0185] Figure 3 It is a structural diagram of a group division system based on deep clustering of dynamic hypergraph of mobility patterns provided by an embodiment of the present invention.

[0186] like Figure 3 As shown, the group division system based on deep clustering of dynamic hypergraph of mobile patterns provided in this embodiment includes:

[0187] A data acquisition module 301 is used to acquire historical travel record data of N individuals;

[0188] An individual mobility pattern feature matrix extraction module 302 is used to extract D-dimensional features from historical travel records to quantitatively express the mobility pattern of each individual;

[0189] An individual mobility pattern hypergraph set construction module 303 is used to construct a hypergraph set of the current period according to the current individual mobility pattern feature matrix;

[0190] Individual movement pattern deep expression learning module 304, used to learn the deep expression of individual movement pattern based on Transformer "encoding-decoding" structure;

[0191] A neighborhood group movement pattern deep expression learning module 305 is used to learn the deep expression of the neighborhood group movement pattern in the local space based on the hypergraph convolutional network model;

[0192] The dynamic hypergraph deep clustering module 306 is used to utilize the attention mechanism to fuse the deep expression of the individual and neighborhood group movement patterns, and utilize the bidirectional hypergraph convolutional network to capture the change value of the hypergraph topological structure under the cross-action of forward time propagation and backward time propagation, and update its dynamic hypergraph structure;

[0193] The training and testing module 307 is used to input the historical travel record set of N individuals into the dynamic hypergraph clustering model of the mobility pattern to perform training and testing on the classification of individual associated groups.

[0194] The verification module 308 is used to input the historical travel record set of M individuals into the dynamic hypergraph clustering model of the mobility pattern to perform instance verification of the classification of the individual associated groups.

[0195] The specific implementation method of the group classification system provided in this embodiment can be implemented with reference to the above embodiments, and will not be described in detail here.

[0196] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0197] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A group division method based on dynamic hypergraph clustering of individual mobility patterns, characterized in that: include: Obtain a set of historical travel records of N individuals, cut the set in time dimension by day, obtain a set of historical travel records of N individuals under T consecutive time slices, extract D-dimensional travel features for each individual, and construct a mobile mode feature matrix of N individuals; For any time slice, each individual is regarded as a node, and N nodes are obtained; the neighborhood group of the individual is regarded as the hyperedge of the hypergraph, and N hyperedge sets are obtained to construct the hypergraph of the individual movement pattern under each time slice; Construct a dynamic hypergraph clustering model based on mobility patterns, that is, relying on the Transformer "encoder-decoder" structure to learn the deep expression of individual mobility patterns; relying on the hypergraph convolutional network model, learn the deep expression of neighborhood group mobility patterns in local space; use the attention mechanism to integrate the deep expressions of the above individual and neighborhood group mobility patterns, and use the bidirectional hypergraph convolutional network to capture the change value of the hypergraph topological structure under the cross-action of forward time propagation and backward time propagation, dynamically update the hypergraph structure, and complete the training and testing of the model; The historical travel record set of M individuals is input into the dynamic hypergraph clustering model of mobility patterns to conduct an instance verification of the classification of individual associated groups.

2. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 1 is characterized in that: The method extracts D-dimensional travel features for each individual and constructs a mobility pattern feature matrix of N individuals, including: At any time slice τ, D-dimensional travel features are extracted for each individual, and the mobility pattern feature matrix X of N individuals is constructed: τ ∈R N×D , The individual mobility pattern feature matrix X = {X1,…,X τ ,…,X T }, where x τ,i (1≤i≤N) is the characteristic expression of the mobility pattern of the i-th individual in the τ-th time slice.

3. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 1 is characterized in that: The constructing of the individual mobility pattern hypergraph under each time slice includes: For any time slice τ, the N individuals are regarded as nodes of the hypergraph to obtain N nodes; the similarity of the movement patterns of any two individuals is measured using the Euclidean distance; two nodes of the nearest neighbors are connected, and the inverse function of the Euclidean distance is used as the corresponding edge weight; according to the principle of "the smaller the Euclidean distance, the more similar the movement patterns", the k individuals with the most similar movement patterns are determined to form the i-th neighborhood group under the τ-th time slice. The neighborhood group is used as a hyperedge to obtain a set of N hyperedges, which constitute the initial division result of the individual association group, and the individual mobility pattern hypergraph G under the τth time slice is obtained. τ =(V τ ,E τ ,X τ ,W τ ,H τ ), where V τ is the set of nodes that express all individuals, E τ is a set of hyperedges that express the movement pattern of the neighborhood group, X τ is a set of node attributes that express individual mobility patterns, W τ is a set of hyperedge weights that express the importance of the neighborhood group, H τ is the association matrix expressing the affiliation between individuals and groups, H τ =[h(v,e)] τ ∈R N×N , h(v,e) represents the affiliation relationship between node v and hyperedge e; The affiliation relationship between the node v and the hyperedge e conforms to the following formula: In formula (1), v∈e indicates that the node is on the hyperedge, D τ,v =[d(v)] is the degree matrix of node v at time slice τ, which describes the current state of node v in the hypergraph G τ The importance of ; d(v) is the sum of the weights of all hyperedges connected to vertex v, calculated by formula (2): d(v)=∑ e∈E ω(e)h(v,e) (2) D τ,e =[d(e)] is the degree matrix of the hyperedge e at the time slice τ, which describes the hyperedge e in the hypergraph G at the time slice τ. τ The importance of d(e) is the sum of the number of nodes connected by hyperedge e, which is calculated by formula (3): d(e)=∑ v∈V h(v,e) (3).

4. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 1 is characterized in that: Construct a dynamic hypergraph clustering model based on mobility patterns, including: Using the Transformer model, we learn deep representations of individual movement patterns. Using the hypergraph convolutional network model, we learn the deep expression of the neighborhood group movement pattern in the local space; Fusion of the individual and neighborhood group movement patterns to perform dynamic hypergraph deep clustering; Conduct the training and testing process of the dynamic hypergraph clustering model of mobility patterns.

5. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 4 is characterized in that: The Transformer model is used to learn the deep representation of individual movement patterns, including: Through the working principle of the multi-head attention mechanism of the built-in encoder of the Transformer model, the deep expression C of the individual movement pattern in the lth (1≤l≤L) convolutional layer of the individual movement pattern feature matrix X is learned autonomously (l) , C (l) Calculated by formula (4): Among them, σ ReLU is a linear activation function. When l = 1, C (l-1) =C 0 is the individual mobility pattern feature matrix X; LN is the convolutional layer regularization operation; dropout is the random pruning operation of the fully connected layer; FeedForward is the fully connected network, Res (l) is the residual unit of the Transformer model decoder; U1 and U2 are the weight parameters of the model to be trained, b1 and b2 are bias values; MultiHead is a multi-head attention mechanism; Through the built-in decoder of the Transformer model, after L convolutional layers, the reconstructed Complete the decoding operation; Minimize the average reconstruction error To obtain the deep expression of individual mobility patterns that is closest to the actual situation, where ||·|| Fro is the F norm, where the average reconstruction error is generated by the encoding and decoding process, the average reconstruction error The expression of is shown in formula (5): For any time slice τ, according to the feature matrix X τ The distribution of and the real group category to which each individual belongs, calculate the real probability distribution of individual i belonging to category j, and then obtain the probability distribution curve GT of individual mobility mode category in the real scene under time slice τ; According to the deep expression C of the individual movement pattern of the Lth convolutional layer (L) The distribution of individual mobility mode categories predicted by Transformer under time slice τ is obtained. Determine the cumulative error of the probability distribution curve GT and the probability distribution curve TR in all time slices Determine the group classification error after learning the Transformer time series features, where Calculated by formula (6), Among them, GT τij and TR τij They respectively refer to the probability that individual i belongs to category j in curve GT and curve TR under time slice τ, and KL is the "Kulbeck-Leibler divergence" distribution curve.

6. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 4 is characterized in that: The hypergraph convolutional network model is used to learn the deep expression of the neighborhood group movement pattern in the local space, including: Based on the hypergraph G τ As well as the initial division results of individual associated groups, the hypergraph convolutional network operation is performed to capture the local spatial topological changes of the neighborhood group. For the lth convolutional layer, the deep expression Z of the neighborhood group movement pattern (l) Calculated by formula (7): Where Θ is the convolution hyperparameter; when l = 1, Z (l-1) =Z 0 is the feature matrix X; H τ Transponse Indicates H τ The transposed matrix of .

7. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 4 is characterized in that: The individual and neighborhood group movement patterns are integrated to perform dynamic hypergraph deep clustering, including: The individual movement patterns are expressed in C (l) Deep expression Z with neighborhood group movement pattern (l) Through formula (8), we can get C under the lth convolutional layer. (l) and Z (l) The splicing result CZ (l) , CZ (l) n For CZ (l) The nth vector sequence of ; By introducing the attention mechanism, we can get the new expression of the neighborhood group movement pattern under the lth convolutional layer through formula (9): Among them, q is the query vector, α n is the attention weight of the nth element, is the attention scoring function used to measure CZ (l) n The similarity between the query vector q is quantified using the dot product model; based on The distribution of the hypergraph topology under the interaction of forward time propagation and backward time propagation is captured by ΔG. τ , that is, the change in the topological structure of the hypergraph that belongs to the first or second hypergraph but not both, is calculated as shown in formula (10): ΔG τ =difference(G 1→τ ,G T→τ )=(G 1→τ ∪G T→τ )-(G 1→τ ∩G T→τ ) (10) Based on G τ and ΔG τ , dynamically update the corresponding hypergraph G τ , as the input of the hypergraph convolutional network at the convolutional layer l+1, and the output result C corresponding to the Transformer at the l+1 layer (l+1) After attention fusion, the hypergraph convolutional network operation is performed to obtain the probability distribution results of the group categories under the l+1th to Lth convolutional layers. Among them, W′ τ , H′ τ , D′ τ,v , D′ τ,e Dynamic Hypergraph The hyperedge weight set, incidence matrix, node degree matrix and hyperedge degree matrix of ; For any time slice τ, according to the deep expression Z of the individual movement pattern of the lth convolutional layer (L) The distribution of individual mobility mode categories is obtained by hypergraph clustering under time slice τ, and the cumulative error between the probability distribution curve HC and the curve GT in all time slices is measured. Measuring the group classification error of deep clustering of dynamic hypergraphs that integrate the mobility patterns of individuals and neighborhood groups, Calculated by formula (12): Among them, HC τij is the probability that individual i belongs to category j in curve HC under time slice τ.

8. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 4 is characterized in that: The training and testing process of the mobile mode dynamic hypergraph clustering model includes: The overall error is calculated using the following formula: Among them, γ1, γ2 and γ3 are as well as The weight coefficient of Relying on the N-bit individual mobility pattern feature matrix X under T time slices, the mobility pattern dynamic hypergraph clustering model is trained to minimize the cumulative error Until The value no longer changes, the model converges, and the model training process is completed; Input the N-bit individual movement pattern feature matrix X under T+1 time slices to the trained model T+1 , determine the individual association groups in the test data set.

9. The group division method based on dynamic hypergraph clustering of individual mobility patterns according to claim 1 is characterized in that: After completing the training and testing of the model, the example verification of the classification of individual associated groups is also included, which specifically includes: Input the M-bit individual mobility pattern feature matrix X′ under T time slices T , using the method established by the present invention, the individual association groups in the validation data set are found. The proposed method is applied to the public transportation passenger card swiping data set, and the partitioning performance of the proposed method and its internal modules is verified through multiple group comparisons and ablation experiments.

10. A group partitioning system based on dynamic hypergraph clustering of individual mobility patterns, characterized in that: include: A data acquisition module is used to obtain historical travel record data of N individuals; Individual mobility pattern feature matrix extraction module, used to extract D-dimensional features from historical travel records and quantitatively express the mobility pattern of each individual; An individual mobility pattern hypergraph set construction module is used to construct a hypergraph set for the current period according to the current individual mobility pattern feature matrix; The deep expression learning module of individual movement patterns is used to learn the deep expression of individual movement patterns based on the Transformer "encoder-decoder" structure; Neighborhood group movement pattern deep expression learning module, which is used to learn the deep expression of neighborhood group movement pattern in local space based on the hypergraph convolutional network model; The dynamic hypergraph deep clustering module is used to use the attention mechanism to integrate the deep expression of the above-mentioned individual and neighborhood group movement patterns, and use the bidirectional hypergraph convolutional network to capture the change value of the hypergraph topological structure under the cross-action of forward time propagation and backward time propagation, and update its dynamic hypergraph structure; The training and testing module is used to input the historical travel record set of N individuals into the dynamic hypergraph clustering model of the mobility pattern to perform training and testing on the classification of individual associated groups; The verification module is used to input the historical travel record set of M individuals into the dynamic hypergraph clustering model of mobile patterns to perform instance verification of the classification of individual associated groups.

Images