Urban rail transit passenger flow short-term OD passenger flow prediction method

Abstract

The present application relates to a kind of urban rail transit passenger flow short-term OD passenger flow prediction method, comprising: step one) based on the influence factor set influencing multi-scene short-term OD passenger flow prediction, the label of scene feature is constructed, and scene feature is embedded in the time series of historical OD passenger flow;Step two) according to the time series of historical OD passenger flow embedded scene feature, construct subway network space-time graph, construct sparse graph convolution network for capturing spatial features;Step three) according to the time series of historical OD passenger flow embedded scene feature, construct cross attention mechanism for capturing time features;Step four) based on the sparse graph convolution network and cross attention mechanism, construct multi-scene short-term passenger flow prediction combination model based on Transformer, output the final predicted OD passenger flow matrix.

Description

Technical Field

[0001] This invention relates to the field of passenger flow forecasting technology, specifically to a method for short-term origin-destination (OD) passenger flow forecasting in urban rail transit. Background Technology

[0002] With the development of informatization in various industries, most cities' rail transit systems have entered the network operation stage. The ever-expanding network size and increasing passenger flow have made operation management and transportation organization strategy formulation more complex. To form a rational and scientific transportation organization, it is necessary to conduct in-depth research on the dynamic changes in passenger flow during daily network operations and accurately predict the network's origin-destination (OD) passenger flow for future operating periods. This provides real-time passenger flow data support for dynamic operation management and control, assisting the operation department in making network operation decisions, and is of great significance for optimizing the allocation of public transportation resources. The vast network size makes passenger flow distribution complex and dynamic, therefore, how to predict network OD passenger flow in a timely and accurate manner has become a challenging problem to be solved.

[0003] Obtaining OD (Original Department) passenger flow data is challenging, often relying on large-scale manual surveys. This approach is not only extremely time-consuming and labor-intensive but also results in significant data bias. Traditional prediction models for OD passenger flow forecasting struggle to fully leverage the nonlinear relationships between OD data, while machine learning methods, mostly based on shallow network structures, have limited ability to extract the complex nonlinear spatiotemporal relationships between ODs in large-scale road networks.

[0004] Meanwhile, the various factors influencing OD (Original Departure Point) passenger flow changes are complex and ever-changing. Collecting information on the corresponding influencing factors while acquiring OD passenger flow data also poses a challenge to OD passenger flow research, resulting in relatively few related predictive studies. Short-term OD passenger flow matrices under multiple scenarios still face the problems of large data dimensionality and severe data sparsity, causing the integrated scenario features to be diluted and the learning results to be insignificant.

[0005] Currently, AFC (Automatic Fare Collection) card-swiping systems are widely deployed in urban rail transit systems across my country, enabling comprehensive and accurate collection of historical passenger flow data. The rise of deep learning has provided technical support for addressing the low prediction accuracy of traditional OD (Original Departure / Origin) passenger flow prediction models. Summary of the Invention

[0006] This invention relates to the field of urban rail transit organization theory and technology, specifically providing a short-term OD passenger flow prediction method applicable to multiple scenarios. Based on historical OD passenger flow, spatial distribution of stations on the network, weather, special events, and time factors, a feature fusion prediction model for multiple scenarios is constructed using a deep learning framework to obtain predicted OD passenger flow, providing OD passenger flow information for operation organization management and route planning design.

[0007] A method for short-term origin-destination (OD) passenger flow prediction in urban rail transit includes:

[0008] Step 1) Based on the set of influencing factors affecting short-term OD passenger flow prediction in multiple scenarios, construct labels for scenario features and embed scenario features into the time series of historical OD passenger flow;

[0009] Step 2) Construct a spatiotemporal map of the subway network based on the time series of historical OD passenger flows embedded with scene features, and construct a sparse graph convolutional network to capture spatial features;

[0010] Step 3) Based on the time series of historical OD passenger flow embedded with scene features, construct a cross-attention mechanism to capture time features;

[0011] Step 4) Based on the sparse graph convolutional network and cross-attention mechanism, construct a Transformer-based multi-scenario short-term passenger flow prediction combination model and output the final predicted OD passenger flow matrix.

[0012] Optionally, the set of influencing factors affecting short-term OD passenger flow forecasting across multiple scenarios includes the following influencing factors: historical OD passenger flow, spatial distribution of road network stations, weather, special events, and time.

[0013] Optionally, the historical OD passenger flow is an OD passenger flow matrix generated using passenger card swiping data within the AFC system;

[0014] The spatial distribution of the road network stations is represented as a fully directed graph, with stations represented by nodes and connections between stations represented by edges formed by connecting nodes. Each edge contains historical OD passenger flow and road network spatial information.

[0015] The weather characteristics mentioned include the degree of sunshine and rain and air quality;

[0016] The degree of sunshine and rain includes seven levels: sunny, cloudy, overcast, light rain, moderate rain, heavy rain, and rainstorm. The air quality includes six levels: excellent, good, lightly polluted, moderately polluted, heavily polluted, and severely polluted.

[0017] The special events mentioned mainly include urban rail transit operation events and large-scale event events;

[0018] The time mentioned includes the exact moment, weekday, and non-working day information.

[0019] Optionally, the geostationary network spatiotemporal map includes the following variables: nodes, edges, spatiotemporal points, number of days, stations, and sequences;

[0020] The geostationary-temporal map of the subway network includes the following sets: the OD matrix of the input sequence, the feature set, the study time range, the station set, the set of edges with stations as points, and the number of matrices in the matrix sequence.

[0021] The subway network structure includes the following components:

[0022] A fully directed graph G(V,E,A) represents a graph network at time t;

[0023] Where V t V is the OD passenger flow value matrix of the railway network at time t. t ∈R n×n Let n be the number of stations, E be the set of edges in the subway network with stations as vertices, representing the connections between stations, and A∈R. n×n Let A be the adjacency matrix of graph G, where A0 represents the physical connection between stations and A1 represents the travel time between stations.

[0024] Optionally, the method of capturing spatial features using sparse convolutional networks is as follows:

[0025] The spatial characteristics of the subway network are captured by combining sparse graph convolution (SGCN) and graph convolution (GCN), and upsampling convolution and downsampling convolution are added to the sparse graph convolution module.

[0026] Optionally, step three includes the following steps:

[0027] Step 1) Set up a Transformer latent space with three dimensions m: ST, IS, and IT. Then, use 1×1 convolutions to transform the original OD passenger flow data X. t ∈R 1×n×n Mapping onto different latent spaces yields The calculation formula is:

[0028]

[0029] In formula (1) Let represent the OD passenger flow statistics of stations i to j that have completed their journeys at time interval t on day d; where i,j∈[1,n], and n represents the number of stations in the road network.

[0030]

[0031] In formula (2), Spare∈{ST,IS,IT} represents different latent spaces, ReLU(·) represents the activation function, and Conv 1×1 (·) represents a convolution operation with a 1×1 kernel.

[0032] Step 2) Divide the input matrix sequence with embedded features into three vector spaces, and perform differentiated feature extraction operations in different spaces to capture features of the input matrix sequence in the long-term, short-term, and medium-to-long-term time dimensions:

[0033] The ST space is the long-term feature space, and in this model branch:

[0034] α-ST = Cross Attention{SGCN ST [Pool ST (input ST (3)

[0035] The IS space is the short-term feature space, and in this model branch:

[0036] α-IS = Cross Attention{SGCN IS [Pool IS (Slipt(input IS (4)

[0037] The IT space is a medium- to long-term feature space, and in this model branch:

[0038] α-IT = Cross Attention{SGCN IT [Pool IT (GCN IT (input IT (5)

[0039] In the formula, SGCN represents sparse graph convolutional network, GCN represents graph convolutional network, Pool represents pooling operation, and Slipt represents separation operation in formula (3). Both the network and the operation layer contain weight parameters that can be used for training.

[0040] Step 3) The output Chead is mapped through a Cross-Attention mechanism. t The corresponding query spaces are mapped to the ST, IS, and IT spaces respectively. Key space With value space V t i For the Cross-Attention mechanism, the query space in different spaces is obtained by mapping information from the space itself, and the key space and value space in different spaces are obtained by mapping information from other spaces. The calculation formula is as follows:

[0041] α-IT = Cross Attention{SGCN IT [Pool IT(GCN IT (input IT (5)

[0042]

[0043]

[0044] In formula (6), (χ∈{q,k,v},i∈{1,2,…,h}) correspond to the feature extraction functions on the ST, IS, and IT branch spaces, respectively; Q t STi This represents the query vector of the i-th element in the ST space at time t, defined by the function φ. i q according to Calculated; This represents the query vector of the i-th element in the IS space at time t, defined by the function... according to Calculated; This represents the query vector of the i-th element in the IT space at time t, defined by the function... according to Calculated; V represents the key vectors in the ST, IS, and IT spaces, respectively; t STi V t ISi V t ITi Let W represent the value vectors in the ST, IS, and IT spaces, respectively; in formula (8) m The trainable parameters are represented by d = m / h, where d represents days, m represents the dimension, and h represents the length of the input sequence.

[0045] Optionally, step four includes the following steps: combining the sparse graph convolutional network and the cross-attention mechanism, and performing residual connection and normalization processing to generate a combined prediction model based on the Transformer network Encoder-Decoder architecture.

[0046] Optionally, the model combination process is as follows:

[0047] Step 1: Set up a Transformer latent space with three dimensions m: ST, IS, and IT. Then, use 1×1 convolutions to transform the original OD passenger flow data X. t ∈R 1×n×n Mapping onto different latent spaces yields

[0048] Step 2: The output Chead is mapped through a Cross-Attention mechanism. t .

[0049] Step 3: Through residual connections and normalization operations, the generalization ability of the model network is enhanced, and the effectiveness of training deep learning networks is improved.

[0050] Step 4: Obtain the output Y through a feedforward neural network (FCs) consisting of two fully connected layers. t n×n .

[0051] Step 5: Repeat step 4 to obtain OD passenger flow information Y t od .

[0052] Step 6: Let X t =Y t od Repeat steps 2 through 5 k+2 times to extract feature information at depth.

[0053] Optionally, the residual connection and normalization operations are as follows:

[0054]

[0055] In formula (9), Norm(·) represents the normalization function and GCN(·) represents the graph convolution function.

[0056] The output Y is obtained through a feedforward neural network (FCs) consisting of two fully connected layers. t n×n .

[0057]

[0058] Compared with the prior art, the present invention has the following advantages and effects:

[0059] The spatiotemporal attention mechanism model combines the cross-attention mechanism with spatiotemporal feature extraction to form a spatiotemporal attention mechanism, which improves the relevance of spatiotemporal features and the quality of feature extraction.

[0060] By embedding external environmental features into historical OD passenger flow sequences and combining them with the Transformer framework to form a prediction model, the information interaction between external environmental features and corresponding OD passenger flow in multiple scenarios can be enhanced, thus meeting the prediction needs of short-term OD passenger flow in multiple scenarios. Attached Figure Description

[0061] Figure 1 This is a flowchart of the present invention.

[0062] Figure 2This is a schematic diagram of the subway network structure.

[0063] Figure 3 This is a schematic diagram of the SGCN module.

[0064] Figure 4 This is a schematic diagram of feature capture in the input matrix sequence across three time dimensions: long-term, short-term, and medium-to-long-term. Detailed Implementation

[0065] To make the objectives and technical solutions of this invention clearer, the invention will be further described in detail. Unless otherwise specified, the experimental methods described in the following embodiments are conventional methods; where specific techniques or conditions are not specified in the embodiments, they are performed according to the techniques or conditions described in the literature in this field or according to the product instructions; unless otherwise specified, the reagents and materials are all commercially available.

[0066] Example 1

[0067] A method for short-term origin-destination (OD) passenger flow prediction in large-scale networked urban rail transit based on spatiotemporal multimodal methods includes:

[0068] Step 1) Identify the set of factors influencing short-term OD (Original Demand) passenger flow forecasts across multiple scenarios;

[0069] Factors influencing short-term OD passenger flow forecasts include historical OD passenger flow, spatial distribution of road network stations, weather, special events, and time.

[0070] The historical OD passenger flow is generated using passenger card swiping data within the AFC system to create an OD passenger flow matrix with certain historical intervals.

[0071] The spatial distribution of the road network stations can be represented as a fully directed graph, where stations are represented by nodes, and connections between stations are represented by edges formed by connecting nodes. Each edge contains historical OD passenger flow and road network spatial information.

[0072] The weather characteristics described mainly examine the degree of sunshine and rain, which is divided into 7 levels: sunny, cloudy, overcast, light rain, moderate rain, heavy rain, and rainstorm. The Air Quality Index (AQI) is divided into 6 levels: excellent, good, lightly polluted, moderately polluted, heavily polluted, and severely polluted.

[0073] The special events mentioned mainly include urban rail transit operation events and large-scale event events;

[0074] The time mentioned includes specific times, weekdays, and non-working days.

[0075] Step 2) Based on the set of influencing factors, construct labels for scene features and embed the scene features into the time series of historical OD passenger flow;

[0076] 2. Constructing scene feature labels based on a set of influencing factors and embedding them into the time series of historical OD (Original Departure) passenger flow, characterized by including label-related parameters:

[0077] The tag-related parameters include the following indexes:

[0078]

[0079] The time series data embedded in the historical OD passenger flow includes the following feature vectors for the scenario at time t:

[0080]

[0081] in These represent the categories corresponding to each feature in the table.

[0082] Step 3) Construct a spatiotemporal map of the subway network and use a sparse graph convolutional network to capture spatial features;

[0083] The aforementioned geostationary network spatiotemporal diagram includes the following variables:

[0084]

[0085] The aforementioned geostationary network spatiotemporal map includes the following sets:

[0086]

[0087] The subway network structure includes the following components:

[0088] A fully directed graph G(V,E,A) represents a graph network at time t, where V t V is the OD passenger flow value matrix of the railway network at time t. t ∈R n×n Let n be the number of stations, E be the set of edges in the subway network with stations as vertices, representing the connections between stations, and A∈R. n×n Let G be the adjacency matrix of graph G, where A0 represents the physical connection between stations and A1 represents the travel time between stations.

[0089] The method of capturing spatial features using sparse convolutional networks is as follows:

[0090] The main approach is to combine sparse graph convolution (SGCN) with GCN to capture the spatial characteristics of the subway network, and to add upsampling convolution (UPConv) and downsampling convolution (DownConv) to the SGCN module.

[0091] Step 4) Capture temporal features using the cross-attention mechanism;

[0092] The spatiotemporal sequence of the historical OD passenger flow is used to generate the query in the attention mechanism using a nonlinear function approach. key Sum The convolution and pooling operations, along with SGCN and GCN networks, are integrated into the Cross-Attention mechanism to capture temporal features. Its key feature is:

[0093] (1) Set up a Transformer latent space with three dimensions of m: ST, IS, and IT. Then, use 1×1 convolution to transform the original OD passenger flow data X. t ∈R 1×n×n Mapping onto different latent spaces yields X t ST ∈R m×n×n , The calculation formula is:

[0094]

[0095] In formula (1) Let represent the OD passenger flow statistics for stations i to j that have completed their journeys at time interval t on day d. Here, i,j∈[1,n], and n represents the number of stations within the road network.

[0096]

[0097] In formula (2), Spare∈{ST,IS,IT} represents different latent spaces, ReLU(·) represents the activation function, and Conv 1×1 (·) represents a convolution operation with a 1×1 kernel.

[0098] (2) The input matrix sequence with embedded features is divided into three vector spaces, and differentiated feature extraction operations are used in different spaces to achieve feature capture of the input matrix sequence in the three time dimensions of long-term, short-term, and medium-to-long-term:

[0099] 1) The ST space is the long-term feature space, and in this model branch:

[0100] α-ST = Cross Attention{SGCN ST [Pool ST (input ST (3)

[0101] 2) The IS space is the short-term feature space, and in this model branch:

[0102] α-IS = Cross Attention{SGCN IS [Pool IS(Slipt(input IS (4)

[0103] 3) The IT space is a medium- to long-term feature space. In this model branch:

[0104] α-IT = Cross Attention{SGCN IT [Pool IT (GCN IT (input IT (5)

[0105] In the formula, SGCN represents sparse graph convolutional network, GCN represents graph convolutional network, Pool represents pooling operation, and Slipt represents separation operation in formula (3). Both the network and the operation layer contain weight parameters that can be used for training.

[0106] (3) The output Chead is mapped through a Cross-Attention mechanism. t The corresponding query spaces are mapped to the ST, IS, and IT spaces respectively. Key space With value space V t i For the Cross-Attention mechanism, the query space in different spaces is obtained by mapping information from the space itself, and the key space and value space in different spaces are obtained by mapping information from other spaces. The calculation formula is as follows:

[0107] α-IT = Cross Attention{SGCN IT [Pool IT (GCN IT (input IT (5)

[0108]

[0109] In formula (6), (χ∈{q,k,v},i∈{1,2,…,h}) correspond to the feature extraction functions on the ST, IS, and IT branch spaces, respectively; Q t STi This represents the query vector of the i-th element in the ST space at time t, defined by the function φ. i q According to X t ST Calculated; Q t ISi This represents the query vector of the i-th element in the IS space at time t, defined by the function... according to Calculated; This represents the query vector of the i-th element in the IT space at time t, defined by the function... according to Calculated; V represents the key vectors in the ST, IS, and IT spaces, respectively; t STi V t ISi V t ITi Let W represent the value vectors in the ST, IS, and IT spaces, respectively; in formula (8) m The trainable parameter is d = m / h, where d represents days; m represents the dimension; and h represents the length of the input sequence, which is 5 in this invention.

[0110] Step 5) Construct a multi-scenario short-term passenger flow forecasting combination model based on Transformer, and output the final predicted OD passenger flow matrix. Specific steps include:

[0111] Based on the aforementioned cross-attention mechanism and feedforward neural network, residual connections and normalization are performed to generate a combined prediction model based on the Transformer network Encoder-Decoder architecture:

[0112] The residual join and normalization operations are as follows:

[0113] Y t m =Norm[X t +GCN(Chead t (9)

[0114] In formula (9), Norm(·) represents the normalization function and GCN(·) represents the graph convolution function.

[0115] The output Y is obtained through a feedforward neural network (FCs) consisting of two fully connected layers. t n×n .

[0116] Y t n×n =FCs(Y t m (10)

[0117] The model combination process is as follows:

[0118] Step 1: Set up a Transformer latent space with three dimensions m: ST, IS, and IT. Then, use 1×1 convolutions to transform the original OD passenger flow data X. t ∈R1×n×n Mapping onto different latent spaces yields

[0119] Step 2: The output Chead is mapped through a Cross-Attention mechanism. t .

[0120] Step 3: Through residual connections and normalization operations, the generalization ability of the model network is enhanced, and the effectiveness of training deep learning networks is improved.

[0121] Step 4: Obtain the output Y through a feedforward neural network (FCs) consisting of two fully connected layers. t n×n .

[0122] Step 5: Repeat step 4 to obtain OD passenger flow information Y t od .

[0123] Step 6: Let X t =Y t od Repeat steps 2 through 5 k+2 times to extract feature information at depth.

[0124] Furthermore, it should be noted that the above are merely preferred embodiments of the present invention and are not limited to the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for short-term origin-destination (OD) passenger flow prediction in urban rail transit, characterized in that, include: Step 1) Based on the set of influencing factors affecting short-term OD passenger flow forecasting in multiple scenarios, construct labels for scenario features and embed scenario features into the time series of historical OD passenger flow; Step 2) Construct a spatiotemporal map of the subway network based on the time series of historical OD passenger flows embedded with scene features, and construct a sparse graph convolutional network to capture spatial features; Step 3) Based on the time series of historical OD passenger flow embedded with scene features, construct a cross-attention mechanism to capture time features; Step 4) Construct a Transformer-based multi-scenario short-term passenger flow prediction combined model based on the sparse graph convolutional network and cross-attention mechanism, and output the final predicted OD passenger flow matrix; Step three includes the following steps: Step 1) Set up a Transformer latent space with three dimensions m: ST, IS, and IT. Convolution transforms the raw OD passenger flow data Mapping onto different latent spaces yields , , The calculation formula is: In formula (1) Indicates the first Heavenly Stations where passengers have completed their journeys within a given time interval To the station OD passenger flow statistics; among which , This represents the number of stations within the road network. In formula (2), Representing different potential spaces, Represents the activation function. The representative convolution kernel is Convolution operation; Step 2) Divide the input matrix sequence with embedded features into three vector spaces, and perform differentiated feature extraction operations in different spaces to capture features of the input matrix sequence in the long-term, short-term, and medium-to-long-term time dimensions: The ST space is the long-term feature space, and in this model branch: The IS space is the short-term feature space, and in this model branch: The IT space is a medium- to long-term feature space, and in this model branch: In the formula, SGCN represents sparse graph convolutional network, GCN represents graph convolutional network, Pool represents pooling operation, and Slipt represents separation operation in formula (3). Both the network and the operation layer contain weight parameters that can be used for training. Step 3) , , Mapped to the output via a cross-attention mechanism. The corresponding query spaces are mapped to the ST, IS, and IT spaces respectively. Key space AND value space For the Cross-Attention mechanism, the query space in different spaces is obtained by mapping information from the space itself, and the key space and value space in different spaces are obtained by mapping information from other spaces. The calculation formula is as follows: In formula (6), ( , These correspond to feature extraction functions in the ST, IS, and IT branch spaces, respectively. This represents the query vector of the i-th element in the ST space at time t, defined by the function... according to Calculated; This represents the query vector of the i-th element in the IS space at time t, defined by the function... according to Calculated; This represents the query vector of the i-th element in the IT space at time t, defined by the function... according to Calculated; , , These represent the key vectors in the ST, IS, and IT spaces, respectively. , , These represent the value vectors in the ST, IS, and IT spaces, respectively; in formula (8) Represents trainable parameters, Where d represents days; m represents dimensions; and h represents the length of the input sequence. Step four includes the following steps: combining the sparse graph convolutional network and the cross attention mechanism, and performing residual connection and normalization processing to generate a combined prediction model based on the Transformer network Encoder-Decoder architecture; The model combination process is as follows: Step 1: Set up a Transformer latent space with three dimensions m: ST, IS, and IT. Convolution transforms the raw OD passenger flow data Mapping onto different latent spaces yields , , ; Step 2: , , Mapped to the output via a cross-attention mechanism. ; Step 3: Through residual connection and normalization operations, the generalization ability of the model network is enhanced, and the effectiveness of training deep learning networks is improved. Step 4: Pass through a feedforward neural network consisting of two fully connected layers Get output ; Step 5: Repeat step 4 to obtain OD passenger flow information. ; Step 6: Order ,repeat Steps 2 through 5 involve deep feature extraction. The residual join and normalization operations are as follows: In formula (9) Represents the normalization function. Represents the graph convolution function; The feedforward neural network consists of two fully connected layers. Get output ; 。 2. The method for short-term origin-destination (OD) passenger flow prediction in urban rail transit according to claim 1, characterized in that, The set of factors influencing short-term OD passenger flow forecasting across multiple scenarios includes the following factors: historical OD passenger flow, spatial distribution of road network stations, weather, special events, and time.

3. The method for short-term OD (Original Departure) passenger flow prediction in urban rail transit according to claim 2, characterized in that, The historical OD passenger flow is an OD passenger flow matrix generated using passenger card swiping data within the AFC system; The spatial distribution of the road network stations is represented as a fully directed graph, with stations represented by nodes and connections between stations represented by edges formed by connecting nodes. Each edge contains historical OD passenger flow and road network spatial information. The weather characteristics mentioned include the degree of sunshine and rain and air quality; The degree of sunshine and rain includes seven levels: sunny, cloudy, overcast, light rain, moderate rain, heavy rain, and rainstorm. The air quality includes six levels: excellent, good, lightly polluted, moderately polluted, heavily polluted, and severely polluted. The special events mentioned mainly include urban rail transit operation events and large-scale event events; The time mentioned includes the exact moment, weekday, and non-working day information.

4. The method for short-term OD (Original Departure) passenger flow prediction in urban rail transit according to claim 1, characterized in that, The geostationary-temporal diagram of the subway network includes the following variables: nodes, edges, spatiotemporal points, number of days, stations, and sequences; The geostationary network spatiotemporal map includes the following sets: the OD matrix of the input sequence, the feature set, the study time range, the station set, the set of edges with stations as points, and the number of matrices in the matrix sequence; The subway network structure includes the following components: Completely directed graph G( V t ,E,A ),express Graph networks in real time; in yes Time-based OD passenger flow value matrix of the railway network , For the number of stations, It is the set of edges in a subway network with stations as nodes, representing the connections between stations. It is a fully directed graph The adjacency matrix.

5. The method for short-term origin-destination (OD) passenger flow prediction in urban rail transit according to claim 1, characterized in that, Using sparse convolutional networks to capture spatial features is as follows: The spatial characteristics of the subway network are captured by combining sparse graph convolution (SGCN) and graph convolution (GCN), and upsampling convolution and downsampling convolution are added to the sparse graph convolution module.

Images