Short-term passenger flow prediction method and system for rail transit stations based on spatiotemporal characteristics

Abstract

The present invention discloses a method and system for predicting short-term passenger flow at rail transit stations based on spatiotemporal characteristics, including: representing the topology of the subway network by graph, preprocessing historical passenger flow data to construct a passenger flow feature matrix; constructing a short-term passenger flow prediction model, using the passenger flow feature matrix and adjacency matrix of historical passenger flow data as model input; simultaneously processing the adjacency matrix and passenger flow data of each historical time interval through a Chebyshev network, obtaining the short-range and long-range spatial dependencies of station passenger flow, and generating passenger flow spatial features; importing passenger flow spatial features into a downsampled convolutional interactive learning neural network to learn passenger flow time features, forming a residual connection between passenger flow time features and passenger flow space features, and generating the final short-term passenger flow prediction output through a fully connected layer. The present invention provides a new and effective solution for short-term passenger flow prediction, and accurate short-term passenger flow prediction helps stations make adjustments and warnings in advance, reducing accidents in operation.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term passenger flow prediction for rail transit, and more specifically, to a method and system for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics. Background Art

[0002] At present, subway short-term passenger flow prediction faces many challenges, such as high dimensionality, nonlinearity and non-stationarity of data. Traditional statistical methods often cannot meet the requirements of prediction accuracy. The subway system is a highly interconnected network. The passenger flow changes at a station may affect other stations, thereby increasing the difficulty of prediction. In addition, emergencies such as large-scale events also add uncertainty to subway passenger flow prediction. Therefore, how to train a model with high accuracy and low model complexity based on various limiting factors and that can be applied to real subway operation scenarios faces a great challenge. In general, subway passenger flow shows complex spatiotemporal dependence. Due to the complexity of the dynamic mechanism of passenger flow, the correlation between subway inbound and outbound passenger flow is relatively complex. Therefore, designing an effective spatiotemporal passenger flow prediction model is an urgent problem to be solved. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method and system for predicting short-term passenger flow at rail transit stations based on spatiotemporal characteristics. According to the data of the urban rail transit automatic ticket vending and checking system and the topological structure of the subway network, the characteristics of the station passenger flow and the mutual influence relationship between the station passenger flows are mined, the dynamic changes of the network passenger flow distribution are grasped, and the future short-term passenger flow changes are predicted.

[0004] The first aspect of the present invention provides a method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics, comprising:

[0005] Based on the subway automatic ticket vending and checking system, historical passenger flow data is extracted, the topology of the subway network is represented by a graph, and the historical passenger flow data is preprocessed according to the graph structure data of the subway network to construct a passenger flow feature matrix;

[0006] A short-term passenger flow prediction model is constructed based on the interactive learning neural network of Chebyshev graph convolution and down-sampling convolution, and the passenger flow feature matrix of historical passenger flow data and the adjacency matrix representing the node adjacency relationship are used as model inputs.

[0007] The Chebyshev network is used to simultaneously process the adjacency matrix and the passenger flow data of each historical time interval, obtain the short-distance and long-distance spatial dependencies of the station passenger flow, and generate the passenger flow spatial characteristics;

[0008] The passenger flow spatial features are introduced into the down-sampling convolutional interactive learning neural network to learn the passenger flow time features, all passenger flow time features are merged and rearranged to form a residual connection with the passenger flow spatial features, and the final short-term passenger flow prediction output is generated through the fully connected layer.

[0009] In this scheme, the topological information of the subway network is converted into a graph. ,in Represents a collection of nodes, represents the number of nodes, Represents a slave node To Node The set of edges of , Indicates the adjacency relationship between nodes, including 0 and 1, 1 represents adjacent nodes, and 0 represents non-adjacent nodes;

[0010] Preprocess the historical passenger flow data according to the graph structure data of the subway network to obtain the input passenger flow feature matrix , express All nodes at the moment and their corresponding inbound and outbound passenger flows.

[0011] In this scheme, the spatial feature extraction module is used in the short-term passenger flow prediction model to obtain the passenger flow spatial features. Specifically, the spatial feature extraction module is constructed by using the Chebyshev network, and the convolution kernel of the Chebyshev polynomial approximation graph is used. The Chebyshev polynomial Defined as , , …, , the convolution kernel of the spatial feature extraction module It is expressed as: in is the normalized Laplacian matrix, To initialize the Laplacian matrix, is the identity matrix, is the characteristic value, is the vector of Fourier coefficients, is the approximate order;

[0012] Introducing activation function , get the Chebyshev network representation , the adjacency matrix is ​​processed by the spatial feature extraction module The passenger flow data at each historical time interval are used to generate the passenger flow spatial characteristics through the short-range and long-range spatial dependencies of the passenger flow at the two-layer Chebyshev network sites.

[0013] In this solution, the time feature extraction module is used in the short-term passenger flow prediction model to obtain the passenger flow time features, specifically:

[0014] The temporal feature extraction module is constructed by adopting the downsampling-convolution-interaction architecture, and the sampling convolution and interaction modules are organized into a binary tree to form a hierarchical convolution network. The input of the temporal feature extraction module is generated according to the passenger flow spatial features. ;

[0015] The layered convolutional network downsamples the input Extract subsequences at odd and even positions respectively and , use a set of convolution kernels to process each subsequence and obtain temporal feature representation;

[0016] The introduction of interactive learning allows information exchange between two sequences, obtaining and , as the temporal feature of passenger flow.

[0017] In this program, in interactive learning, and Through two different one-dimensional convolution modules and Map to a high-dimensional hidden vector representation and perform feature amplification through the exp function;

[0018] Multiply the enlarged features by element and To interact, it is represented as: in Represents element-wise multiplication;

[0019] The two scale features and Further through two more one-dimensional convolution modules and Mapped to two other hidden vectors and then with and Perform addition or subtraction operations to obtain the output of the time feature extraction module and , expressed as: .

[0020] In this scheme, the passenger flow temporal features output by the temporal feature extraction module are merged and rearranged to generate a residual connection with the passenger flow spatial features as the input of the fully connected layer;

[0021] Use the fully connected layer to reconstruct the features and obtain the final short-term passenger flow prediction output .

[0022] The second aspect of the present invention provides a rail transit station short-term passenger flow prediction system based on spatiotemporal characteristics, including a data preprocessing module, a spatial feature extraction module, a temporal feature extraction module and a prediction output module;

[0023] The data preprocessing module represents the topology of the subway network in a graph, preprocesses the historical passenger flow data according to the graph structure data of the subway network, and constructs a passenger flow feature matrix;

[0024] The spatial feature extraction module obtains the short-range and long-range spatial dependencies of the passenger flow of the station through the Chebyshev network based on the passenger flow feature matrix and the adjacency matrix, and generates the passenger flow spatial features;

[0025] The time feature extraction module captures the dynamic changes of historical passenger flow at different time scales through downsampling convolution interactive learning neural network learning, and generates passenger flow time features;

[0026] The prediction output module forms a residual connection between the passenger flow spatial features and the passenger flow temporal features, reconstructs the features through a fully connected layer, and generates a short-term passenger flow prediction output.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention explores the value and characteristics of subway passenger flow data through deep learning and model fusion, designs a deep learning spatiotemporal neural network based on subway passenger flow, and provides a new and effective solution for short-term passenger flow prediction. Accurate short-term passenger flow prediction facilitates stations to make adjustments and early warnings in advance to reduce or even avoid accidents in operation.

[0029] The present invention is based on the downsampling-convolution-interaction architecture. By downsampling long sequences into subsequences, the convolution and interaction learning are used to extract time dependencies and output high-precision multi-step prediction results. The proposed architecture can also handle changes in passenger flow patterns during different operation periods.

[0030] The Chebyshev polynomial graph convolution in the model proposed in the present invention has the ability to aggregate high-order node information, model the spatial dependency between distant sites, and solve the spatial dependency between close and distant sites in the graph structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0032] Figure 1A flow chart showing a method for predicting short-term passenger flow at rail transit stations based on spatiotemporal characteristics is shown;

[0033] Figure 2 The framework diagram of the short-term passenger flow prediction model is shown;

[0034] Figure 3 shows the structural diagram of the sampling convolution and interaction modules;

[0035] Figure 4 The convolution kernel is shown , , and Detailed structure diagram of

[0036] Figure 5 The schematic diagram of the short-term passenger flow prediction system for rail transit stations based on spatiotemporal characteristics is shown. DETAILED DESCRIPTION

[0037] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0038] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0039] Figure 1 A flow chart of a method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics is shown.

[0040] like Figure 1 As shown, this embodiment provides a method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics, including:

[0041] S102, extracting historical passenger flow data based on the subway automatic ticket vending and checking system, representing the topology of the subway network in a graph, preprocessing the historical passenger flow data according to the graph structure data of the subway network, and constructing a passenger flow feature matrix;

[0042] S104, constructing a short-term passenger flow prediction model based on a Chebyshev graph convolution and downsampling convolution interactive learning neural network, and taking the passenger flow feature matrix of historical passenger flow data and the adjacency matrix representing the node adjacency relationship as model input;

[0043] S106, simultaneously processing the adjacency matrix and the passenger flow data of each historical time interval through a Chebyshev network, obtaining the short-distance and long-distance spatial dependencies of the passenger flow of the station, and generating the passenger flow spatial features;

[0044] S108, importing the passenger flow spatial features into the downsampling convolutional interactive learning neural network to learn the passenger flow time features, merging and rearranging all the passenger flow time features to form a residual connection with the passenger flow spatial features, and generating the final short-term passenger flow prediction output through the fully connected layer.

[0045] It should be noted that based on the historical passenger flow data extracted from the subway automatic fare collection (AFC) system, the passenger flow in and out of each station in the subway network in the near future is predicted. The passenger flow data of each station at 10-minute intervals is extracted from the original AFC data to predict multi-step passenger flow changes. The topological information of the subway network is converted into a graph. ,picture Graph structure data representing subway network: The subway network can be modeled as a graph, where nodes represent platforms and edges represent tracks. Represents a collection of nodes, represents the number of nodes, Represents a slave node To Node The set of edges of , Indicates the adjacency relationship between nodes, including 0 and 1, 1 represents adjacent nodes, and 0 represents non-adjacent nodes;

[0046] The in-and-out passenger flow of a subway station changes over time, so it can be used as two different features for each node. The historical passenger flow data is preprocessed according to the graph structure data of the subway network to obtain the input passenger flow feature matrix , express All nodes and their corresponding inbound and outbound passenger flows at the moment. and Respectively represent The inbound and outbound passenger flow of all sites at the moment. Similarly, the predicted future The passenger flow at a certain time can also be expressed as .

[0047] According to an embodiment of the present invention, a spatial feature extraction module is used in the short-term passenger flow prediction model to obtain passenger flow spatial features. Figure 2As shown in the figure, the Chebyshev network (ChebNet) in the graph neural network is used to perform the corresponding convolution operation to extract the spatial features of the passenger flow data. The Chebyshev network is used to build a spatial feature extraction module. The convolution kernel of the graph is approximated by Chebyshev polynomials, thereby improving the efficiency of using high-order neighbor information between nodes. By increasing the order of the convolution kernel, ChebNet can aggregate node features in a larger range, thereby enhancing the model's ability to represent graph structures.

[0048] The graph Laplacian for spectral graph analysis is defined as ,in represents the adjacency matrix, represents the degree matrix, The initialization Laplacian matrix is ​​defined as ,in is the identity matrix. From the above formula, we can see that the Laplacian matrix combines the structure of the graph and the information of neighbor nodes. Since the Laplacian matrix is ​​semi-positive definite, it can be characteristically decomposed into ,in is the Fourier basis, is the feature value, which can be regarded as the parameter that needs to be learned during the convolution training process.

[0049] The graph-based Fourier transform can be expressed as , the inverse Fourier transform is In the frequency domain, the convolution process can be expressed as a multiplication, that is, ,in is the frequency domain convolution kernel, which can be expressed as . is a vector of Fourier coefficients. The convolution kernel is a diagonal matrix whose diagonal elements represent the response of the convolution kernel to each eigenvalue. Subsequently, the convolved features are transformed from the frequency domain back to the time domain through the inverse Fourier transform, expressed as Combining these steps, the frequency domain graph convolution is defined as ,in, is the feature after graph convolution, convolution kernel For one with Related functions, by parameter definition, and are the eigenvectors and eigenvalues ​​of the Laplacian matrix, respectively.

[0050] The Chebyshev network (ChebNet) uses Chebyshev polynomials to approximate the convolution kernel function, avoiding the need for eigenvalue decomposition and computation. In addition, the use of polynomial approximation allows the selection of low-order fits, effectively shifting the computation from global to local scope. In order to approximate the convolution kernel , introducing the Chebyshev polynomial , the Chebyshev polynomial Defined as , , ..., , the convolution kernel of the spatial feature extraction module It is expressed as: in is the normalized Laplacian matrix, To initialize the Laplacian matrix, is the identity matrix, is the characteristic value, , is the approximate order;

[0051] Introducing activation function , get the Chebyshev network representation , the adjacency matrix is ​​processed by the spatial feature extraction module The passenger flow data at each historical time interval are used to generate the passenger flow spatial characteristics through the short-range and long-range spatial dependencies of the passenger flow at the two-layer Chebyshev network sites.

[0052] It should be noted that the spatial feature extraction module uses Chebyshev polynomials to approximate the Laplacian matrix of the graph, thereby achieving efficient and effective graph convolution operations, enabling the network to learn complex high-order relationships between nodes in the graph while avoiding expensive matrix operations and eigenvalue decomposition. After the topological structure of the subway network is converted into graph data, passenger flow information can be input into the spatial module to extract spatial dependencies. By adjusting the value of the approximation order K, the degree to which a node obtains information from its neighbors can be changed. In addition, increasing the number of network layers can enhance the ability to aggregate information across nodes, thereby effectively capturing the spatial dependencies of each station.

[0053] According to an embodiment of the present invention, a time feature extraction module is used in a short-term passenger flow prediction model to obtain passenger flow time features. The time feature extraction module is constructed using a downsampling-convolution-interaction architecture. Downsampling decomposes time series information and uses convolution kernels to extract time dependencies from subsequences. The sampling convolution and interaction modules (SCI-Block) of the time feature extraction module are as follows: Figure 3 The temporal feature extraction module adopts an encoder-decoder architecture, as shown in Figure 2 As shown, the sampling convolution and interaction modules are organized into a binary tree to form a hierarchical convolutional network, which uses a rich set of convolutional kernels to capture dynamic temporal dependencies at different resolutions.

[0054] Generate input of temporal feature extraction module according to the passenger flow spatial feature The sampling convolution and interaction modules of the layered convolutional network extract subsequences located at odd and even positions of the passenger flow sequence by downsampling. and , use a set of convolution kernels to process each subsequence and obtain temporal feature representation;

[0055] In order to compensate for the loss of information during downsampling, the interactive learning module allows information exchange between the two sequences. In interactive learning, and Through two different one-dimensional convolution modules and Map to a high-dimensional hidden vector representation and perform feature amplification through the exp function;

[0056] Multiply the enlarged features by element and To interact, it is represented as: in Represents element-wise multiplication;

[0057] The two scale features and Further through two more one-dimensional convolution modules and Mapped to two other hidden vectors and then with and Perform addition or subtraction operations to obtain the output of the time feature extraction module and , as the passenger flow time feature, is expressed as: . Convolutional module , , and They all have the same convolutional structure, but different parameters. The specific model structure is as follows Figure 4 shown.

[0058] It should be noted that the temporal feature extraction module effectively captures dynamic changes at different time scales through downsampling and convolution processing, which is particularly useful for predicting sharp fluctuations in subway passenger flow in the short term. It allows the model to integrate and learn from a longer time window, effectively capturing the long-term dependencies of passenger flow data. Its unique segmentation and interactive learning strategy enables the model to extract rich local and global information from the data, enhancing the understanding of the deep characteristics of passenger flow sequences.

[0059] It should be noted that the passenger flow time features output by the time feature extraction module are merged and rearranged to generate a residual connection with the passenger flow spatial features as the input of the fully connected layer; the fully connected layer is used to reconstruct the features and obtain the final short-term passenger flow prediction output ,in Represents the time step for predicting future passenger flow. The short-term passenger flow prediction model is trained by minimizing the mean absolute error between the predicted value and the true value.

[0060] Figure 5 The schematic diagram of the short-term passenger flow prediction system for rail transit stations based on spatiotemporal characteristics is shown.

[0061] The second aspect of the present invention provides a rail transit station short-term passenger flow prediction system 5 based on spatiotemporal characteristics, including a data preprocessing module 51, a spatial feature extraction module 52, a temporal feature extraction module 53 and a prediction output module 54;

[0062] The data preprocessing module 51 represents the topology of the subway network in a graph, preprocesses the historical passenger flow data according to the graph structure data of the subway network, and constructs a passenger flow feature matrix;

[0063] The spatial feature extraction module 52 obtains the short-range and long-range spatial dependencies of the passenger flow of the station through the Chebyshev network based on the passenger flow feature matrix and the adjacency matrix, and generates the passenger flow spatial features;

[0064] The time feature extraction module 53 captures the dynamic changes of historical passenger flow at different time scales through downsampling convolution interactive learning neural network learning, and generates passenger flow time features;

[0065] The prediction output module 54 forms a residual connection between the passenger flow spatial features and the passenger flow temporal features, reconstructs the features through a fully connected layer, and generates a short-term passenger flow prediction output.

[0066] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0067] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0068] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0069] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0070] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0071] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting short-term passenger flow at rail transit stations based on spatiotemporal characteristics, characterized in that: The following steps are involved: Based on the subway automatic ticket vending and checking system, historical passenger flow data is extracted, the topology of the subway network is represented by a graph, and the historical passenger flow data is preprocessed according to the graph structure data of the subway network to construct a passenger flow feature matrix; A short-term passenger flow prediction model is constructed based on the interactive learning neural network of Chebyshev graph convolution and down-sampling convolution, and the passenger flow feature matrix of historical passenger flow data and the adjacency matrix representing the node adjacency relationship are used as model inputs. The Chebyshev network is used to simultaneously process the adjacency matrix and the passenger flow data of each historical time interval, obtain the short-distance and long-distance spatial dependencies of the station passenger flow, and generate the passenger flow spatial characteristics; The passenger flow spatial features are introduced into the down-sampling convolutional interactive learning neural network to learn the passenger flow time features, all passenger flow time features are merged and rearranged to form a residual connection with the passenger flow spatial features, and the final short-term passenger flow prediction output is generated through a fully connected layer; In the short-term passenger flow prediction model, the spatial feature extraction module is used to obtain the passenger flow spatial features, specifically: The spatial feature extraction module is constructed by using the Chebyshev network, and the convolution kernel of the graph is approximated by the Chebyshev polynomial. Defined as , , …, , the convolution kernel of the spatial feature extraction module It is expressed as: in is the normalized Laplacian matrix, To initialize the Laplacian matrix, is the identity matrix, is the characteristic value, is the vector of Fourier coefficients, , is the approximate order; Introducing activation function , get the Chebyshev network representation , Represents the passenger flow feature matrix, and the adjacency matrix is ​​processed by the spatial feature extraction module The passenger flow data at each historical time interval are used to generate the passenger flow spatial characteristics through the short-range and long-range spatial dependencies of the passenger flow at the two-layer Chebyshev network sites.

2. According to claim 1, a method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics is characterized in that: Convert the topological information of the subway network into a graph ,in Represents a collection of nodes, represents the number of nodes, Represents a slave node To Node The set of edges of , Indicates the adjacency relationship between nodes, including 0 and 1, 1 represents adjacent nodes, and 0 represents non-adjacent nodes; Preprocess the historical passenger flow data according to the graph structure data of the subway network to obtain the input passenger flow feature matrix , express All nodes at the moment and their corresponding inbound and outbound passenger flows.

3. The method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics according to claim 1, characterized in that: In the short-term passenger flow prediction model, the time feature extraction module is used to obtain the passenger flow time features, specifically: The temporal feature extraction module is constructed by adopting the downsampling-convolution-interaction architecture, and the sampling convolution and interaction modules are organized into a binary tree to form a hierarchical convolution network. The input of the temporal feature extraction module is generated according to the passenger flow spatial features. ; The layered convolutional network downsamples the input Extract subsequences at odd and even positions respectively and , use a set of convolution kernels to process each subsequence and obtain temporal feature representation; The introduction of interactive learning allows information exchange between two sequences, obtaining and , as the temporal feature of passenger flow.

4. The method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics according to claim 3 is characterized in that: In interactive learning, and Through two different one-dimensional convolution modules and Map to a high-dimensional hidden vector representation and perform feature amplification through the exp function; Multiply the enlarged features by element and To interact, it is represented as: in Represents element-wise multiplication; The two scale features and Further through two more one-dimensional convolution modules and Mapped to two other hidden vectors and then with and Perform addition or subtraction operations to obtain the output of the time feature extraction module and , expressed as: .

5. The method for predicting short-term passenger flow at a rail transit station based on spatiotemporal characteristics according to claim 1, characterized in that: The passenger flow temporal features output by the temporal feature extraction module are merged and rearranged to generate a residual connection with the passenger flow spatial features as the input of the fully connected layer; Use the fully connected layer to reconstruct the features and obtain the final short-term passenger flow prediction output .

6. A short-term passenger flow prediction system for rail transit stations based on spatiotemporal characteristics, characterized in that: Implementing the short-term passenger flow prediction method for rail transit stations based on spatiotemporal characteristics as described in any one of claims 1 to 5, comprising a data preprocessing module, a spatial feature extraction module, a temporal feature extraction module and a prediction output module; The data preprocessing module represents the topology of the subway network in a graph, preprocesses the historical passenger flow data according to the graph structure data of the subway network, and constructs a passenger flow feature matrix; The spatial feature extraction module obtains the short-range and long-range spatial dependencies of the passenger flow of the station through the Chebyshev network based on the passenger flow feature matrix and the adjacency matrix, and generates the passenger flow spatial features; The time feature extraction module captures the dynamic changes of historical passenger flow at different time scales through downsampling convolution interactive learning neural network learning, and generates passenger flow time features; The prediction output module forms a residual connection between the passenger flow spatial features and the passenger flow temporal features, reconstructs the features through a fully connected layer, and generates a short-term passenger flow prediction output.

Images