A shared bicycle travel network community mining method based on deep learning

Abstract

The application provides a shared bicycle travel network community mining method based on deep learning. The method comprises the following steps: analyzing and counting the road network data information, the shared bicycle data information and the travel data of users in a specified area, and constructing a traffic travel network; quantitatively describing the travel characteristics of users in the shared bicycle travel network, taking the user travel quantitative indicators as network measurement indicators of a graph neural network model, and constructing the graph neural network model; performing community clustering analysis based on the graph neural network model by using a ClusterNet algorithm, and mining community structures. According to the method, a dynamic travel network is constructed according to different time periods, and the spatiotemporal distribution of travel demands in different stages is analyzed. By using a spatial statistics and a complex network method, indicators are constructed to quantitatively describe the travel characteristics, so that the changes of the travel modes of users in different periods can be clearly understood, and the community structures in the travel network can be dynamically mined.

Description

Technical Field

[0001] This invention relates to the field of computer application technology, and in particular to a method for mining shared bicycle travel network communities based on deep learning. Background Technology

[0002] Cities are vast and complex systems, and urban planning and transportation are closely related to our daily lives. Therefore, quantifying the spatiotemporal patterns of residents' travel flows can effectively reflect the dynamics of urban components. Urban economic development leads to the continuous expansion of transportation scale, and people's travel demands are increasing daily. Frequent traffic problems, such as traffic congestion and uneven distribution of travel demand, are becoming increasingly prominent and have received widespread attention from all sectors of society. Understanding changes in users' travel needs, behaviors, and community structures can help governments and operators provide better services to passengers. Users exhibit different travel choices for different modes of transportation; emerging shared bicycles provide a new mode of transportation for short-distance travel and strengthen connections with other modes of transportation such as buses and subways.

[0003] Currently, some breakthroughs and innovations have been made in methods for studying human travel and mobility characteristics. However, some problems still exist. First, research on changes in user travel needs, travel behavior, and travel community structure is insufficient. Second, emerging deep learning graph neural networks can create more powerful representations of node attributes and community structure, taking into account both network topology and node feature learning, and possessing strong learning capabilities for community discovery in user travel networks. Summary of the Invention

[0004] Embodiments of the present invention provide a deep learning-based method for mining communities in shared bicycle travel networks, so as to effectively identify communities existing in shared bicycle travel networks and changes in community structure.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A deep learning-based method for mining shared bicycle travel network communities includes:

[0007] Analyze and statistically analyze road network data, shared bicycle data, and user travel data within a designated area to construct a transportation network;

[0008] The travel characteristics of users in the shared bicycle travel network are quantitatively described, and the quantitative indicators of user travel are used as the network metric of the graph neural network model to construct the graph neural network model.

[0009] Community clustering analysis is performed using the ClusterNet algorithm based on a graph neural network model to uncover community structure.

[0010] Preferably, the analysis and statistical analysis of road network data, shared bicycle data, and user travel data within a designated area to construct a transportation network includes:

[0011] The data on shared bicycles and road network in a designated area are analyzed, and a transportation network is constructed based on user travel data. The points in the transportation network are the origin and destination of the user's trip, and the journey from the origin to the destination is considered as the edge between the nodes.

[0012] A directed transportation network G = (V, E, W) is constructed based on shared bicycle data from the beginning, middle, and end of a specified time period, where V = {V1, ..., V...}. N Let} represent the set of points, E = {e ij |i,j=1,2,…,N,i≠j},e ij =1 indicates that there is an edge between node i and node j, e ij =0 indicates that there is no connecting edge between node i and node j, W = {w ij Let |i,j=1,2,…,N,i≠j} be the set of weights, w ij Representing edge e ij The weight of the edge is the number of trips between nodes i and j.

[0013] Preferably, the quantitative description of user travel characteristics in the shared bicycle travel network, using user travel quantification indicators as network metrics for the graph neural network model, and constructing the graph neural network model includes:

[0014] This paper uses quantitative indicators of user travel to quantitatively describe the travel characteristics of users in the shared bicycle travel network. The quantitative indicators of user travel are used as network metrics of the graph neural network model. Users in the shared bicycle travel network are treated as nodes in the graph neural network. The graph neural network model is constructed based on shared bicycle travel data using all network metrics. The network metrics include: degree, strength, clustering coefficient, PageRank value, net flow ratio and Moran index of the nodes.

[0015] In a graph neural network model, the degree d of node i is defined. i The number of nodes connected is given by formula 1. If j is a neighbor node of i, then e ij =1; otherwise, e ij =0;

[0016]

[0017] Define the strength s of node i i To describe the passenger flow intensity between nodes, as shown in Formula 2, w ijIt is the origin-destination (OD) passenger flow between node i and node j;

[0018]

[0019] The clustering coefficient C of the network is defined as the average of the clustering coefficients of all nodes, as shown in Formulas 3 and 4.

[0020]

[0021]

[0022] C is the clustering coefficient of the network, C i It is the clustering coefficient of node i, e i Let i be the number of edges between the neighboring nodes of node i.

[0023] The PageRank value is used to represent the influence score of a node. The formula for calculating the PageRank value of a node is shown in formula (5).

[0024]

[0025] c i Let c be the PageRank value of the i-th node. i ∈[0,1], p is the damping coefficient, Let a represent the out-degree of the j-th node. In a complex network, the out-degree of a node refers to the number of connections originating from that node, i.e., the number of edges pointing from that node to other nodes. ji It is the adjacency matrix of any directed network. The higher the PageRank value of a node, the more important the node is.

[0026] The net flow ratio (NFR) was used to analyze the inflow and outflow of passengers in different areas during different peak periods. The calculation method for NFR is shown in Formula 6:

[0027]

[0028] NFR i The range is between -1 and 1, O i and D i These represent the inflow and outflow of travel within region i.

[0029] The Moran index is used to represent the spatial distribution and evolution of shared bicycles. The formula for calculating the Moran index is as follows:

[0030]

[0031] Where n represents the number of spatial regions, w ij y represents the weight between site i and site j. iand y j This represents the attribute values ​​for site i and site j. This is the average of all observations.

[0032] Preferably, the community clustering analysis based on the graph neural network model using the ClusterNet algorithm to mine community structure includes:

[0033] Community clustering analysis is performed using the ClusterNet algorithm based on a graph neural network model. The input data is embedded into a graph convolutional network (GCN), and the output of the convolutional network is fed into the K-means clustering function for iterative clustering. Finally, the loss function, i.e. the optimization objective, is calculated using the output assignment matrix and modularity. The parameters are optimized through backpropagation of the error. The output of the ClusterNet algorithm is the community division label for each shared bike station.

[0034] A graph neural network model is constructed based on shared bicycle travel data to extract the travel connections between each pair of stations. The travel volume of each pair of origin and destination stations is used as the adjacency matrix, which is then used as the feature of the graph convolutional network edges. This is achieved by defining the adjacency matrix... To describe the spatial connectivity between stations, the features of each station are defined as its latitude and longitude, daily departure passenger flow, daily arrival passenger flow, hourly departure passenger flow, and hourly arrival passenger flow. t , the characteristic matrix γ t As input data for the ClusterNet algorithm;

[0035] The graph convolutional network model constructs a filter in the Fourier domain, which acts on the nodes of the graph. Based on the first-order neighborhood of the filter, it captures the spatial features between shared bicycle stations. A deep GCN model is constructed by stacking multiple convolutional layers. This modeling process is represented by Equation 8:

[0036]

[0037] in, It is an adjacency matrix, I N It is the identity matrix. It is the degree matrix of the shared bicycle station network, where H (l) It is the output of the l-th layer, θ (l) These are the training parameters of the l-th layer, σ(·) represents the activation function of the nonlinear model, using the ReLU activation function, given the feature matrix γ. t and adjacency matrix A two-layer GCN model is represented by Equation 9, where θ (1)θ is the trainable weight matrix from the input layer to the hidden layer. (2) It is the trainable weight matrix from the hidden layer to the output layer.

[0038]

[0039] The feature matrix is ​​used as the input to the clustering module. The K-means algorithm is used to divide the community. Assuming there are N nodes, each node represents an input. Based on the ClusterGCN model, the nodes of the graph are divided into k different communities according to the input. The goal of model training is to find a partitioning method r that maximizes the modularity of the k communities. The modularity is defined as the loss function, and the calculation formula of the loss function is shown in Equation 10.

[0040]

[0041] Where i and j are any two nodes in the graph, and A is defined as a graph where i and j are directly connected. i,j =1, otherwise A i,j =0. d i It is the degree of point i. δ(r) i ,r j ) is used to determine whether nodes i and j are in the same community. If they are in the same community, δ(r) i ,r j ) = 1, otherwise δ(r) i ,r j ) = 0;

[0042] The gradient is calculated using the backpropagation algorithm, and the network parameters are updated using the optimization algorithm. The steps of forward propagation, loss calculation, backpropagation, and parameter update are repeated to iteratively train the neural network. During the training process, the model is evaluated using a validation set. The model learns the shared bicycle stations and the travel characteristics between stations, and the community structure of the shared bicycle travel network is discovered.

[0043] As can be seen from the technical solutions provided by the embodiments of the present invention above, the present invention provides a deep learning method for user travel feature analysis and community structure detection based on graph neural networks. A dynamic travel network is constructed according to different time periods to analyze the spatiotemporal distribution of travel demand at different stages. Spatial statistics and complex network methods are used to construct indicators to quantify travel characteristics, thereby clearly understanding the changes in user travel patterns at different times. Then, an end-to-end deep learning model, ClusterGCN, is proposed to dynamically mine the community structure in the travel network.

[0044] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a deep learning-based method for mining shared bicycle travel network communities is provided in this embodiment of the invention.

[0047] Figure 2 A schematic diagram of the New York City shared bicycle research area;

[0048] Figure 3 The daily travel demand distribution characteristics of shared bicycles in a designated area;

[0049] Figure 4 Spatial distribution map of shared bicycle origin-destination (OD) travel in a designated area;

[0050] Figure 5 A layout diagram of the Fruchterman Reingold shared bicycle network;

[0051] Figure 6 To explore the community structure of shared bicycle travel networks at different times. Detailed Implementation

[0052] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0053] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0054] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0055] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0056] This invention proposes a deep learning method for detecting user travel characteristics and community structure based on shared bicycle travel data. A dynamic travel network is constructed according to different time periods to analyze the spatiotemporal distribution of travel demand at different stages. Spatial statistics and complex network methods are used to construct indicators to quantify travel, thereby clearly understanding the changes in user travel patterns over different periods. Then, an end-to-end deep learning model, ClusterGCN, is proposed to dynamically mine the community structure in the travel network to explore changes in user travel characteristics.

[0057] The processing flow of a deep learning-based method for mining shared bicycle travel network communities provided in this embodiment of the invention is as follows: Figure 1 As shown, the processing flow includes the following:

[0058] Step S1: Data Preparation and Analysis. Prepare and analyze statistical data on road network information and shared bicycle information within the specified area.

[0059] Step S2: Network construction for transportation and travel networks.

[0060] This paper analyzes shared bicycle data in a specified area and constructs a transportation network based on user travel data. Points in the transportation network represent the origin and destination of user trips, and the journey from the origin to the destination is considered as an edge between nodes. A directed weighted network G = (V, E, W) is constructed based on shared bicycle data from the beginning, middle, and end of a specified time period, where V = {V1, ..., V...}. N Let} represent the set of points, E = {e ij |i,j=1,2,…,N,i≠j}. Where, e ij =1 indicates that there is an edge between node i and node j; otherwise, e ij =0. W = {w ij |i,j=1,2,…,N,i≠j} is the set of weights, w ij Representing edge e ij The weight of the edge is the number of trips between nodes i and j.

[0061] Step S3: Construct quantitative indicators of user travel through spatial statistics and complex network methods, and use these quantitative indicators of user travel as network metrics for the graph neural network model.

[0062] This paper uses quantitative user travel metrics to quantify the travel characteristics of users in a shared bicycle network. These metrics are then used as network metrics for a graph neural network (Graph Neural Network) model. Users in the shared bicycle network are treated as nodes in the Graph Neural Network, and the model is constructed based on shared bicycle travel data using all available network metrics. These network metrics include: node degree, strength, clustering coefficient, PageRank, net traffic ratio, and Moran's index. A graph convolutional neural network (GNN) is a machine learning model used to process graph-structured data. It updates node representations by considering information from neighboring nodes and edges, enabling information transfer and feature aggregation on graph data. A typical GNN structure consists of an input layer, graph convolutional layers, pooling layers, and an output layer. The input layer receives the representation of the graph data, typically structural information such as node feature vectors and edge attributes. The graph convolutional layer is the core layer of the GNN, used for information transfer and feature aggregation on the graph data. The pooling layer reduces the dimensionality and complexity of the graph data, thereby reducing computational and memory consumption. The output layer transforms the final representation of the GNN into the desired output form. Graph convolutional neural networks gradually aggregate more local and global information through multiple iterations.

[0063] In a graph neural network model, the degree d of node i i Defined as the number of connected nodes, it measures the importance of node i in the network. The degree of a node is defined as in Formula 1. If j is a neighbor node of i, e ij =1; otherwise, e ij =0.

[0064]

[0065] Define the strength s of node i i To describe the passenger flow intensity between nodes, as shown in Formula 2. ij It is the OD (Origin to Destination) passenger flow between node i and node j.

[0066]

[0067] The clustering coefficient reflects the degree of aggregation in a network; a network with a larger clustering coefficient is likely to have more clusters of nodes with higher connectivity. C represents the network's clustering coefficient. i d is the clustering coefficient of node i. It is calculated as shown in Formulas 3 and 4. Where d iLet be the degree of node i. i Let be the number of edges between the neighboring nodes of node i. The clustering coefficient C of the network is the average of the clustering coefficients of all nodes.

[0068]

[0069]

[0070] The importance of nodes in a network can be determined by the PageRank metric, which uses the PageRank algorithm to calculate a node's influence score. The PageRank metric was proposed as an algorithm for calculating the importance of web pages on the internet, and its calculation method is shown in Formula 5. Where c... i Score c for the influence of the i-th node. i ∈[0,1]. p is the damping coefficient, empirically set as p = 0.85. Let a represent the out-degree of the j-th node. ji It is the adjacency matrix of any directed network. The higher the PageRank value of a node, the more important the node is.

[0071]

[0072] In complex networks, each node has in-degree and out-degree, which describe the node's connectivity within the network. In-degree refers to the number of connections pointing to a given node, i.e., the number of edges from other nodes to that node. In-degree represents how many other nodes are connected to that node, and can be understood as the number of connections that node receives. Out-degree refers to the number of connections originating from a given node, i.e., the number of edges from that node to other nodes. Out-degree represents how many other nodes can be reached through that node, and can be understood as the number of connections that node sends.

[0073] To extract the differences in shared bicycle usage characteristics among users at different stages, the net flowrate (NFR) was used to analyze the inflow and outflow passenger flow in different areas during different peak hours. The calculation method for NFR is shown in Formula 6. Wherein, NFR... i The range is between -1 and 1, O i and D i NFR represents the inflow and outflow of travel within region i. i A value less than 0 indicates that the number of vehicle pick-ups in region i is greater than the number of vehicle returns, meaning there is a large outflow of vehicles from region i, requiring more vehicles to be deployed to meet user travel demand. Conversely, NFR... i A value greater than 0 indicates that the number of car returns exceeds the number of car pick-ups, with inflow volume dominating.

[0074]

[0075] Spatial autocorrelation analysis is a statistical method used to study geospatial data, revealing regional structural information of spatial variables. This study investigates the spatial distribution and evolution of shared bicycles using the global Moran's index and the local spatial autocorrelation Moran's index (Moran's I). The calculation formula is as follows: Where n represents the number of spatial regions, w... ij y represents the weight between site i and site j. i and y j This represents the attribute values ​​for site i and site j. This is the average of all observations. The Moran's index typically ranges from -1 to 1. A positive value indicates that Moran's I is a coefficient between -1 and 1. A value greater than 0 indicates a positive correlation, and a larger value indicates a more significant spatial correlation.

[0076]

[0077] Step S4: Perform community clustering analysis using the ClusterNet algorithm based on the graph neural network model to mine the community structure.

[0078] This invention proposes an end-to-end model, ClusterGCN. First, data is embedded into a graph using a Convolutional Neural Network (GCN). Then, the output of the GCN is fed into a K-means clustering function for iterative clustering. Finally, the loss function, i.e., the optimization objective, is calculated using the output assignment matrix and modularity. Error backpropagation is then used for parameter optimization. The ClusterNet algorithm, based on shared bicycle travel data, takes the latitude and longitude of the stations as input, the number of departing passengers on the day of departure, the number of arriving passengers on the day of arrival, the hourly departure passenger flow, the hourly arrival passenger flow, and the inter-station travel volume as input. The output of the ClusterNet algorithm is the community classification label for each shared bicycle station.

[0079] A graph neural network model is constructed based on shared bicycle travel data to extract the travel connections between each pair of stations. The travel volume (OD volume) of each pair of origin and destination stations is used as the adjacency matrix, which is then used as the feature of the graph convolutional network edges. This is achieved by defining the adjacency matrix... This describes the spatial connectivity between stations. The features of a station are defined using its latitude and longitude, daily departure passenger flow, daily arrival passenger flow, hourly departure passenger flow, and hourly arrival passenger flow statistics. A feature matrix γ is then defined. t The GCN model constructs a filter in the Fourier domain, which acts on the nodes of the graph to capture the spatial features between shared bicycle stations based on their first-order neighborhoods. A deep GCN model is then built by stacking multiple convolutional layers. This modeling process is represented by Equation 8.

[0080]

[0081] in, It is an adjacency matrix, I N It is the identity matrix. It is the degree matrix of the shared bicycle station network, where H (l) It is the output of the l-th layer, θ (l) These are the training parameters of the l-th layer, and σ(·) represents the activation function of the nonlinear model. This invention uses the ReLU activation function. Given the feature matrix γ t and adjacency matrix A two-layer GCN model can be represented by Equation 9. Where θ (1) θ is the trainable weight matrix from the input layer to the hidden layer. (2) It is the trainable weight matrix from the hidden layer to the output layer.

[0082]

[0083] After feature extraction using a dual-graph neural network, the structural and attribute information in the original graph structure is effectively reduced in dimensionality. The feature matrix is ​​then used as input to a clustering module for community partitioning. In this module, the classic K-means algorithm is employed for community partitioning.

[0084] The K-means algorithm is an unsupervised clustering method that is significantly faster than typical community detection algorithms while achieving excellent clustering results, making it suitable for community detection applications. The basic idea of ​​K-means is to group nodes in the graph around k cluster centers, grouping the nodes closest to each cluster together. Iterative updates to the cluster centers are used until the optimal clustering result is achieved. The K-means algorithm first determines the number of cluster centers k. Then, it clusters the other nodes in the graph according to parameter k. In the final clustering results, nodes with high similarity are grouped together, while nodes with low similarity are not grouped. The similarity is calculated by measuring the Euclidean distance between the vectors of the nodes.

[0085] Assume there are N nodes, each representing an input. Based on the ClusterGCN model, the nodes are divided into k communities according to the input, and there are m connections between the nodes. The goal of the community partitioning task is to divide the nodes of the graph into k distinct communities, where these subgroups are densely packed internally and have few connections between them. Therefore, the objective of model training is to find a partitioning method r that maximizes the modularity, defined as Equation 10.

[0086]

[0087] The quality of network community segmentation is often measured by modularity. Q represents modularity; the higher the modularity, the more reasonable the community segmentation; the lower the modularity, the more ambiguous the network community segmentation. The Q value ranges from -0.5 to 1. Studies indicate that a Q value between 0.3 and 0.7 indicates good clustering performance. Here, i and j are any two nodes in the graph; when two nodes are directly connected, A... i,j =1, otherwise A i,j =0. d i It is the degree of point i. δ(r) i ,r j ) is used to determine whether nodes i and j are in the same community. If they are in the same community, δ(r) i ,r j ) = 1, otherwise δ(r) i ,r j ) = 0.

[0088] The training process of ClusterGCN, a community partitioning model for shared bicycle travel, includes the following key steps: First, based on shared bicycle travel data, travel features of stations and travel volume features between stations are extracted to construct a shared bicycle travel network. Then, the graph network data is input into the model, and the community partitioning results are obtained through forward propagation. During forward propagation, the ClusterGCN network sequentially performs graph convolution, pooling, and K-means clustering operations. The propagation between layers in the graph convolutional network is shown in Equations 8 and 9. Subsequently, modularity is defined as the loss function to measure the quality of the network's community partitioning, as shown in Equation 10. The gradient is calculated using the backpropagation algorithm, and the network parameters are updated using an optimization algorithm. The steps of forward propagation, loss calculation, backpropagation, and parameter update are repeated to iteratively train the neural network. During training, the model is evaluated using a validation set to understand its performance on unseen data. Through these steps, the ClusterGCN community partitioning model for shared bicycle travel can learn the travel features of shared bicycle stations and between stations, thus enabling its application to the task of community structure partitioning.

[0089] Urban transportation community segmentation based on travel data enables intelligent traffic management and resource optimization. By analyzing large-scale travel data, different travel modes and groups can be identified, such as commuters, students, and business travelers, thus providing an accurate information foundation for urban transportation planning and decision-making. Community segmentation allows for a better understanding of the differences in travel behavior and needs among different groups, enabling the provision of customized transportation services. For example, optimizing public transportation for commuters and designing safer campus transportation plans for students will improve residents' travel experience and satisfaction. Community segmentation based on travel data can guide the optimal allocation of resources and promote the popularization and promotion of sustainable travel modes, such as public transportation, walking, and bicycle sharing, which helps reduce traffic congestion and promote sustainable urban development.

[0090] Example

[0091] Figure 1 This is a schematic diagram of the shared bicycle travel network community mining method in this embodiment, referring to... Figure 1 The method includes:

[0092] Step S1: Data preparation and data analysis.

[0093] This invention requires the following data: site data and shared bicycle travel record data for a designated area in the United States. To verify the quantitative effect of the constructed indicators based on spatial statistics and complex network methods on travel, and the effectiveness of the proposed community structure mining algorithm based on the graph neural network model ClusterGCN, shared bicycle data from the early, middle, and late periods of a designated time period in the United States (2019-2022) were selected as the data source. Travel characteristics of users in different time periods were mined. The New York CitiBike dataset includes order number, user pick-up and drop-off time, pick-up and drop-off site, user ID, user gender, and rental time. (Appendix) Figure 2 Describes the research area for shared bicycles in New York City; Appendix Figure 3 The study describes the time distribution characteristics of daily travel demand for shared bicycles in a designated area in 2019, 2020, and 2021.

[0094] Step S2: Analyze the shared bicycle data in the specified area and construct a shared bicycle travel network based on the data. In this network, points are shared bicycle stations, and the origin-destination (OD) values ​​between stations are the network edge weights. (Appendix) Figure 4 This reflects the origin-destination (OD) distribution of shared bicycle trips at different times within a specified time period. Shared bicycles are primarily used for short-distance travel, therefore, trips are concentrated in downtown Manhattan. The specific time period has a relatively small impact on shared bicycle trip volume. For example... Figure 4(a) and 4(b), before the specified time period, the daily number of shared bicycle trips was 58,869, with 784 stations. During the more severe period in April 2020, the daily number of shared bicycle trips was 22,758, a year-on-year decrease of 61.34%. During the recovery phase in the specified time period, the number of shared bicycle trips increased. In particular, a large number of new stations were built, further expanding the service area of ​​shared bicycles. Specifically, in April 2022, the daily trip demand was 77,320, with 1,547 stations.

[0095] Step S3: Construct indicators to quantify travel using spatial statistics and complex network methods.

[0096] Table 1 shows the changes in travel patterns of the shared bicycle network on weekends and weekdays before, during, and after a specified time period. On weekdays, the node degree of the shared bicycle network decreases due to the specified time period, indicating poorer connectivity between areas. During the recovery period of the specified time period, the degree distribution of the shared bicycle network decreases, indicating a reduction in the connectivity of nodes in the shared bicycle network after the specified time period. The shared bicycle system added a significant number of stations after the specified time period, increasing from 784 stations in 2019, 887 stations in 2020, to 1396 stations in 2021. Regarding network strength indicators, the node strength of the shared bicycle network also decreased slightly due to the specified time period. A larger clustering coefficient indicates that the network may have more node groups with higher connectivity; the shared bicycle network has a smaller clustering coefficient, suggesting less tight connectivity between stations. Before the specified time period, the inflow of shared bicycles exceeded the outflow, while after the specified time period, the outflow exceeded the inflow. Regarding the spatial distribution of network characteristics, Moran's I was 0.6996 before the specified time period. A higher Moran's I value indicates more significant spatial correlation. The Moran's I decreased during the most severe period within the specified time period. As the specified time period recovered, the spatial correlation of the shared bicycle network weakened. Compared to weekdays, the degree and strength of nodes in the shared bicycle network decreased on rest days.

[0097] Table 1 Average metrics of shared bicycle networks

[0098]

[0099]

[0100] The Fruchterman Reingold layout is used to display the trip volume and graphical structure of the bike-sharing system; see appendix. Figure 5The graph exhibits significant multi-core, multi-cluster characteristics. Node size represents the degree of a node, which is the sum of its in-degree and out-degree. The grayscale of the line color indicates the magnitude of the origin-destination (OD) between sites; a brighter color indicates a larger OD. Results show that the graph structure exhibits different characteristics at different stages of the specified time period. Before the specified time period, there are more sites in multi-clusters, and the degree of shared single-site nodes is relatively high. During the specified time period, the number of sites in multi-clusters decreases significantly, and the connection between nodes also weakens. Later in the specified time period, the degree distribution of shared single-site nodes does not increase significantly, and the connections between sites are not as strong as before the specified time period, which is related to the large number of newly established sites.

[0101] Step S4: Mining community structure based on the Graph Neural Network model ClusterGCN

[0102] Based on shared bicycle data from a designated area from 2019 to 2022, the ClusterNet algorithm was used for community clustering analysis. New York's CitiBike is mainly concentrated in the Manhattan area. Table 2 shows the modularity index of community division; a modularity value between 0.3 and 0.7 is considered a good clustering effect. The modularity of the shared bicycle system was relatively high in January, indicating a tighter network connection, while the modularity value was lower in April. This is because shared bicycle travel is greatly affected by climate; in winter, when temperatures are low, shared bicycle travel tends to cluster within communities. As the weather gradually warms, user travel patterns become more diversified, and inter-community travel gradually increases.

[0103] The community structure based on shared bicycle data is shown in the attached figure. Figure 6 As shown, different symbols are used to represent the community structure. From Figure 6 As can be seen, the community division structure has changed to some extent at different times. Figure 6 Two neighborhoods in Mid-Manhattan and one in Queens were subsequently merged into one large neighborhood. This shift in zonal structure reflects changes in user travel patterns at different stages, as well as variations in the intensity of interaction with other areas. Designated time periods facilitated the integration of travel within each borough, with cycling connecting more localized areas, resulting in smaller, more closely aligned clusters of communities with geographic boroughs.

[0104] Table 2. ClusterNet Community Discovery Modularity Indicators

[0105]

[0106] In summary, this invention constructs a network based on shared bicycle travel data from different periods to analyze the dynamic changes in the spatiotemporal distribution of travel demand at different stages. It utilizes spatial statistics and complex network methods to construct indicators to quantify travel and reveal the inherent patterns of changes in user travel patterns.

[0107] This invention proposes an end-to-end deep learning model, ClusterGCN, which uses graph convolutional neural networks and the K-means algorithm to perform community partitioning and iterative optimization of the user's dynamic travel network. It takes into account both the learning of network topology and node features, and effectively identifies the communities existing in the shared bicycle travel network and the changes in community structure.

[0108] The method of this invention uses a graph neural network to extract multi-source features and trains the neural network through iterative optimization. It is easy to understand and calculate and has strong applicability.

[0109] The method of this invention understands changes in users' travel needs, travel behavior, and travel community structure, which can provide better services for passengers' "last mile" travel.

[0110] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0111] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0112] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0113] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for mining shared bicycle travel network communities based on deep learning, characterized in that, include: Analyze and statistically analyze road network data, shared bicycle data, and user travel data within a designated area to construct a transportation network; The travel characteristics of users in the shared bicycle travel network are quantitatively described, and the quantitative indicators of user travel are used as the network metric of the graph neural network model to construct the graph neural network model. Based on a graph neural network model, the ClusterNet algorithm is used to perform community clustering analysis and mine community structure. The method of quantitatively describing the travel characteristics of users in a shared bicycle travel network, and using the quantitative indicators of user travel as network metrics for a graph neural network model, includes the following: This paper uses quantitative indicators of user travel to quantitatively describe the travel characteristics of users in the shared bicycle travel network. The quantitative indicators of user travel are used as network metrics of the graph neural network model. Users in the shared bicycle travel network are treated as nodes in the graph neural network. The graph neural network model is constructed based on shared bicycle travel data using all network metrics. The network metrics include: degree, strength, clustering coefficient, PageRank value, net flow ratio and Moran index of the nodes. In a graph neural network model, nodes are defined. degree The number of nodes to connect is given by formula 1. yes The neighboring nodes, ;otherwise, ; Define nodes intensity To describe the passenger flow intensity between nodes, as shown in Formula 2. It is a node and nodes Od / destination passenger flow between origin and destination; Define the clustering coefficient of a network The average value of the clustering coefficients of all nodes is shown in Formulas 3 and 4. The clustering coefficients of the network. It is a node Clustering coefficient, For nodes The number of edges between neighboring nodes; The PageRank value is used to represent the influence score of a node. The formula for calculating the PageRank value of a node is shown in formula (5). For the first PageRank values ​​of each node. , The damping coefficient is... Let represent the out-degree of the j-th node. In a complex network, the out-degree of a node refers to the number of connections originating from that node, i.e., the number of edges pointing from that node to other nodes. It is the adjacency matrix of any directed network. The higher the PageRank value of a node, the more important the node is. The net flow ratio (NFR) was used to analyze the inflow and outflow of passengers in different areas during different peak periods. The calculation method is shown in Formula 6: The range is between -1 and 1. and It is a region Inflow and outflow of travel within China; The Moran index is used to represent the spatial distribution and evolution of shared bicycles. The formula for calculating the Moran index is shown in formula (7) below: in, Indicates the number of spatial regions. For the site and sites The weights between them and Indicates site and sites The attribute value, The average of all observations; The aforementioned community clustering analysis based on the graph neural network model using the ClusterNet algorithm to mine community structure includes: Community clustering analysis is performed using the ClusterNet algorithm based on a graph neural network model. The input data is embedded into a graph convolutional network (GCN), and the output of the convolutional network is fed into the K-means clustering function for iterative clustering. Finally, the loss function, i.e. the optimization objective, is calculated using the output assignment matrix and modularity. The parameters are optimized by backpropagation of the error. The output of the ClusterNet algorithm is the community division label for each shared bicycle station. A graph neural network model is constructed based on shared bicycle travel data to extract the travel connections between each pair of stations. The travel volume of each pair of origin and destination stations is used as the adjacency matrix, which is then used as the feature of the graph convolutional network edges. This is achieved by defining the adjacency matrix... To describe the spatial connectivity between stations, a feature matrix is ​​defined, using the station's latitude and longitude, daily departure passenger flow, daily arrival passenger flow, hourly departure passenger flow, and hourly arrival passenger flow as station features. , the feature matrix As input data for the ClusterNet algorithm; The graph convolutional network model constructs a filter in the Fourier domain, which acts on the nodes of the graph. Based on the first-order neighborhood of the filter, it captures the spatial features between shared bicycle stations. A deep GCN model is constructed by stacking multiple convolutional layers. This modeling process is represented by Equation 8: in, , It is an adjacency matrix. It is the identity matrix. It is the degree matrix of the shared bicycle station network, where , It is the first The output of the layer, It is the first Layer training parameters, The activation function for the nonlinear model is ReLU, given the feature matrix. and adjacency matrix A two-layer GCN model is represented by Equation 9, where It is the trainable weight matrix from the input layer to the hidden layer. It is a trainable weight matrix from the hidden layer to the output layer; The feature matrix is ​​used as input to the clustering module, and the K-means algorithm is used to divide the community. Assume there are... There are 10 nodes, each representing an input. Based on the ClusterGCN model, the nodes of the graph are divided into 10 nodes according to the input. Different communities, nodes share resources with each other. The goal of model training is to find a way to partition the data. ,make To maximize the modularity of each community, modularity is defined as a loss function, and the calculation formula of the loss function is shown in Equation 10. in, and It refers to any two nodes in the graph, when the two nodes are directly connected. ,otherwise , It is a point The degree, It is used to determine nodes and Are they in the same community? ,otherwise ; The gradient is calculated using the backpropagation algorithm, and the network parameters are updated using the optimization algorithm. The steps of forward propagation, loss calculation, backpropagation, and parameter update are repeated to iteratively train the neural network. During the training process, the model is evaluated using a validation set. The model learns the shared bicycle stations and the travel characteristics between stations, and the community structure of the shared bicycle travel network is discovered.

2. The method according to claim 1, characterized in that, The analysis and statistical analysis of road network data, shared bicycle data, and user travel data within a designated area to construct a transportation network includes: The data on shared bicycles and road network in a designated area are analyzed, and a transportation network is constructed based on user travel data. The points in the transportation network are the origin and destination of the user's trip, and the journey from the origin to the destination is considered as the edge between the nodes. A directed transportation network is constructed based on shared bicycle data from the beginning, middle, and end of a specified time period. ， in Set representing points ， , Represents a node and nodes There is an edge between them. Represents a node and nodes There are no connecting edges between them. It is a set of weights. Representing an edge The weight of the node and nodes The amount of travel between them.

Images