Urban rail transit short-time OD passenger flow prediction method based on multi-task learning

Abstract

The invention discloses an urban rail transit short-time OD passenger flow prediction method based on multi-task learning, and the method comprises the steps: firstly constructing a space-time propagation module, taking the in-out passenger flow as a node feature, and taking the OD passenger flow as a side feature; performing spatio-temporal feature learning on the node features and the edge features to obtain spatio-temporal features of the nodes and the edges; transmitting and aggregating the spatio-temporal feature information of the nodes and the edges layer by layer to obtain feature representations of the nodes and the edges; and constructing an OD passenger flow prediction module and an in-out passenger flow prediction module to obtain a passenger flow prediction result. According to the method, through combined modeling of the OD flow and the in-and-out flow, the space-time dependence in a rail transit network can be better captured, OD passenger flow prediction and in-and-out passenger flow prediction are used as a combined task for learning, the space-time characteristics of the in-and-out flow can be fully utilized, the precision of OD passenger flow prediction and the generalization ability of the model are improved, and the prediction efficiency of the rail transit network is improved. The method has high practical application value, and provides more reliable technical support for rail transit management and decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit passenger flow prediction, and particularly to a short-term OD passenger flow prediction method for urban rail transit based on multi-task learning. Background Art

[0002] With the rapid development of urban rail transit systems, passenger flow prediction, as an important means to improve operation efficiency, alleviate traffic congestion, optimize resource allocation, and enhance the passenger experience, has become a key process in rail transit management. Short-term OD (origin-destination) passenger flow prediction is of great practical significance for foreseeing the passenger flow demand of different lines and stations within a certain future time period and guiding operation dispatching and resource allocation. However, traditional prediction methods based on statistics and simple machine learning often neglect the complex spatio-temporal dependencies in the rail transit network and the dynamic correlations between stations, resulting in deficiencies when facing complex passenger flow patterns and external influencing factors.

[0003] In recent years, deep learning, especially graph neural networks (GNNs), has made remarkable progress in processing spatio-temporal data with complex structures. GNNs can effectively model the relationships between nodes, especially in systems with obvious spatial and temporal correlations such as urban rail transit, showing strong advantages. Among them, message-passing graph neural networks (MPNNs), as an innovative graph learning method, can capture spatial dependencies and temporal dynamic changes through the transmission of information between graph nodes, and have gradually become a research hotspot in the field of rail transit passenger flow prediction. However, existing research mainly focuses on the optimization of single-task models, often neglecting the interconnections and synergistic effects between multiple prediction tasks in the rail transit system. Summary of the Invention

[0004] The purpose of the present invention is to: in view of the above problems, provide a short-term OD passenger flow prediction method for urban rail transit based on multi-task learning. By combining the joint modeling of OD flow and in-out station flow, it can better capture the spatio-temporal dependencies in the rail transit network. Learning OD passenger flow prediction and in-out station passenger flow prediction as a joint task can make full use of the spatio-temporal characteristics of in-out station flow, improve the accuracy of OD passenger flow prediction and the generalization ability of the model, have strong practical application value, provide more reliable technical support for rail transit management and decision-making, provide new ideas and methods for the development of the short-term passenger flow prediction field, and have broad application prospects.

[0005] In order to achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0006] According to one aspect of the present invention, there is provided a short-term OD passenger flow prediction method for urban rail transit based on multi-task learning, including the following steps:

[0007] S1. Use inbound and outbound passenger flow as node features and OD passenger flow as edge features to construct a spatiotemporal propagation module, where the spatiotemporal propagation module is composed of multiple spatiotemporal learning units stacked with decreasing time steps layer by layer;

[0008] S2. Use the spatiotemporal learning unit to perform spatiotemporal feature learning on node features and edge features to obtain spatiotemporal features of nodes and edges;

[0009] S3, the spatiotemporal feature information of nodes and edges is transmitted and aggregated layer by layer in the spatiotemporal propagation module to obtain the feature representation of nodes and edges;

[0010] S4. Taking the feature representation of edges and nodes as input, we construct the OD passenger flow prediction module as the main task and the inbound and outbound passenger flow prediction module as the auxiliary task respectively. We use the OD passenger flow prediction module to obtain the OD matrix prediction result, and the inbound and outbound passenger flow prediction module to obtain the inbound and outbound passenger flow.

[0011] Preferably, in step S2, the spatiotemporal learning unit includes a node spatial feature learning subunit, a node temporal feature learning subunit and an edge spatiotemporal feature learning subunit;

[0012] The node space feature learning subunit is used to complete node space feature learning;

[0013] The node time feature learning subunit is used to complete node time feature learning;

[0014] The edge spatiotemporal feature learning subunit is used to complete edge spatiotemporal feature learning.

[0015] Preferably, the node space feature learning comprises the following steps:

[0016] Calculate the neighbor node attention coefficient:

[0017] The neighbor node attention coefficient is calculated by the following formula:

[0018]

[0019] in, is the set of neighbor nodes of node i; j is the neighbor node, α i,j is the neighbor node attention coefficient; H i and H j are the features of nodes i and j respectively, W is the learnable weight matrix, a is the attention vector; k is any neighbor node of node i; H k is the feature vector of neighbor node k; | is vector concatenation;

[0020] Calculate the node connection edge attention coefficient:

[0021] The attention coefficient of the node connection edge is calculated by the following formula:

[0022]

[0023] where e i,j is the edge feature between nodes i and j; β i,j is the attention coefficient of the node connection edge; W e is the edge feature transformation matrix; b is the attention vector; e i,k represents the edge feature between node i and its adjacent node k;

[0024] Message aggregation:

[0025] Based on the attention coefficient, aggregate the information of neighbor nodes and edges to obtain the spatial feature of node i:

[0026]

[0027] where H space is the spatial feature of node i.

[0028] Preferably, the edge spatio-temporal feature learning includes the following steps:

[0029] Node information aggregation:

[0030] Based on the edge e i,j , aggregate the features of the in-node i and the out-node j:

[0031]

[0032] where φ node is the node information aggregation function;

[0033] Historical edge information fusion:

[0034] Fuse the current edge feature with the edge feature at the previous moment :

[0035]

[0036] where φ temp is the historical edge information aggregation function;

[0037] Edge feature update:

[0038] Fuse the node information and the historical edge information to update the edge feature:

[0039]

[0040] where φ edge is the edge feature update function.

[0041] Preferably, in step S4, the OD passenger flow prediction module includes the following prediction steps:

[0042] Edge feature extraction:

[0043] Obtain the edge feature matrix E from the spatio-temporal propagation module i,h ∈R N×N×H , where E i,j represents the spatio-temporal implicit feature from station i to station j, and H is the hidden layer dimension.

[0044] Fully connected mapping:

[0045] Perform a non-linear transformation on each edge feature and map it to the prediction space:

[0046]

[0047] where, W OD1 ∈R H×H and b OD1 ∈R H are learnable parameters;

[0048] OD matrix prediction:

[0049] Output the predicted OD passenger flow values for the next K time steps through the second-layer fully connected network:

[0050]

[0051] where, and b OD2 ∈R are the prediction layer parameters; the final output is where each matrix dimension is N×N.

[0052] Preferably, in step S4, the in-out passenger flow prediction module includes the following prediction steps:

[0053] Node feature extraction:

[0054] Obtain the node feature matrix H from the spatio-temporal propagation module L ∈R N×H , where H L [i] represents the spatio-temporal implicit feature of station i;

[0055] Feature transformation and activation:

[0056] Perform a non-linear transformation on the node features:

[0057]

[0058] where, W IOt ∈R H×Hand b IOt ∈R H are learnable parameters; LayerNorm is a layer normalization operation;

[0059] Passenger flow prediction:

[0060] Output the inbound and outbound passenger flows for the next K time steps through a fully connected network:

[0061]

[0062] where and b IOt ∈R are the parameters of the prediction layer;

[0063] The final output is Each vector dimension is N.

[0064] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are as follows:

[0065] By jointly modeling the OD flow and the inbound and outbound flow, the present invention can better capture the spatio-temporal dependencies in the rail transit network. Learning the OD passenger flow prediction and the inbound and outbound passenger flow prediction as a joint task can make full use of the spatio-temporal characteristics of the inbound and outbound flow, improve the accuracy of the OD passenger flow prediction and the generalization ability of the model, have strong practical application value, provide more reliable technical support for rail transit management and decision-making, provide new ideas and methods for the development of the short-term passenger flow prediction field, and have broad application prospects. Brief Description of the Drawings

[0066] Figure 1 is a schematic flow chart of the present invention;

[0067] Figure 2 is a schematic structural diagram of the spatio-temporal propagation module of the present invention;

[0068] Figure 3 is a schematic structural diagram of the prediction model of the present invention. Detailed Embodiments

[0070] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the following preferred embodiments are cited with reference to the accompanying drawings to further elaborate on the present invention. However, it should be noted that many details listed in the specification are only for enabling the reader to have a thorough understanding of one or more aspects of the invention, and these aspects of the present invention can be implemented even without these specific details.

[0071] Please refer to Figure 1 , the present invention provides a method for short-term OD passenger flow prediction in urban rail transit based on multi-task learning, and the technical solution is as follows:

[0072] A short-term OD passenger flow prediction method for urban rail transit based on multi-task learning, comprising the following steps:

[0073] S1. Construct a spatio-temporal propagation module.

[0074] Construct a spatio-temporal propagation module based on a Message Passing Neural Network (MPNN). The overall structure of the spatio-temporal propagation module is an upward triangle, which is actually composed of multiple spatio-temporal learning units with gradually decreasing time steps stacked together. The spatio-temporal propagation module uses the in-out passenger flow as the node feature and the OD passenger flow as the edge feature. The in-out passenger flow and the OD passenger flow are used as the input of the spatio-temporal propagation module, and the spatio-temporal propagation module is used to capture the spatio-temporal dependence relationship of passenger flow data. MPNN models the spatio-temporal dependence relationship between stations through a graph structure, effectively captures the dynamic spatial relationship between stations and its time-varying characteristics, thereby improving the accuracy of passenger flow prediction.

[0075] S2. Use the spatio-temporal learning unit to perform spatio-temporal feature learning on the node feature and the edge feature to obtain the spatio-temporal features of the node and the edge.

[0076] After the node feature and the edge feature are input, the spatio-temporal learning unit is used to implement the spatio-temporal feature learning of the node and the edge in the message passing graph neural network. The spatio-temporal learning unit sequentially completes the node spatial feature learning, the node time feature learning, and the spatio-temporal feature learning of the edge. Specifically:

[0077] Node spatial feature learning

[0078] Node spatial feature learning is used to aggregate the information of neighbor nodes to capture the spatial dependence relationship between stations. The specific steps are as follows:

[0079] 1. Calculate the neighbor node attention coefficient:

[0080] For node i and its neighbor nodes is the set of neighbor nodes of node i, calculate the attention coefficient α i,j :

[0081]

[0082] where, H i and H j are the features of node i and j respectively, W is a learnable weight matrix, a is an attention vector; k is any neighbor node of node i; H k is the feature vector of neighbor node k; | is vector concatenation.

[0083] 2. Calculate the node connection edge attention coefficient:

[0084] For the edge e between nodes i and j i,j , calculate the edge attention coefficient β i,j :

[0085]

[0086] where e i,j is the edge feature between nodes i and j; W e is the edge feature transformation matrix; b is the attention vector; e i,k represents the edge feature between node i and its adjacent node k.

[0087] 3. Message aggregation:

[0088] Based on the attention coefficient, aggregate the information of neighbor nodes and edges to obtain the spatial feature of node i:

[0089]

[0090] Node temporal feature learning

[0091] Node temporal feature learning is used to fuse the node information at the current moment and the previous moment to capture the temporal evolution law of passenger flow data. The specific steps are as follows:

[0092] 1. Temporal information fusion:

[0093] Fuse the node feature at the current moment with the node feature at the previous moment :

[0094]

[0095] where φ temp is the temporal fusion function, usually implemented using a gating mechanism (such as GRU or LSTM).

[0096] 2. Spatiotemporal feature fusion:

[0097] Fuse the spatial feature H space with the temporal feature to update the node feature:

[0098]

[0099] where φ fuse is the feature fusion function, usually using concatenation or weighted summation.

[0100] Spatiotemporal feature learning of edges

[0101] Spatiotemporal feature learning of edges is used to aggregate the information of in-nodes and out-nodes, and combine historical edge features to update the edge feature. The specific steps are as follows:

[0102] 1. Node information aggregation:

[0103] For edge e i,j , aggregate the features of ingress node i and egress node j:

[0104]

[0105] Among them, φ node It is a node information aggregation function, which is usually implemented using a fully connected network.

[0106] 2. Historical side information fusion:

[0107] Set the current edge feature The edge feature at the previous moment To perform the fusion:

[0108]

[0109] Among them, φ temp is the historical side information aggregation function;

[0110] 3. Edge feature update:

[0111] Fusion of node information and historical edge information, update of edge features:

[0112]

[0113] Among them, φ edge It is the edge feature update function, which is usually implemented using a fully connected network.

[0114] Through the above calculation process, the spatiotemporal learning unit can simultaneously capture the spatiotemporal dependencies of nodes and edges, thereby fully capturing the spatiotemporal dependencies of passenger flow data and providing efficient feature representation for passenger flow prediction tasks. This unit has significant advantages in modeling complex spatiotemporal patterns and effectively improves prediction accuracy.

[0115] S3: The spatiotemporal feature information of nodes and edges is transmitted and aggregated layer by layer in the spatiotemporal propagation module to obtain the feature representation of nodes and edges.

[0116] Specifically, Figure 2As shown, it is assumed that the input data contains L time steps. The spatio-temporal propagation module is composed of multiple spatio-temporal learning units with gradually decreasing time steps layer by layer. Each layer in the spatio-temporal propagation module passes the time step information of the current layer to the adjacent time steps of the next layer through a message passing mechanism. For example, the information of time steps t1 and t2 in the first layer will flow to time step t_2 in the second layer, and so on. As the time steps gradually decrease layer by layer, the information propagates L times on the time and space scales and finally converges to the last layer. By passing and aggregating information layer by layer, the long-term dependencies and spatial correlations in the passenger flow data can be effectively captured, and the efficient learning of spatio-temporal features can be achieved.

[0117] The output of the spatio-temporal propagation module is used as the feature input for the subsequent passenger flow prediction task. Through the hierarchical progressive information passing mechanism, the spatio-temporal dynamic features of the passenger flow data can be learned simultaneously, thereby improving the prediction accuracy. The module has significant advantages in capturing complex spatio-temporal patterns and provides a reliable feature representation for the passenger flow prediction task.

[0118] S4. Construct the OD passenger flow prediction module and the in-out passenger flow prediction module, with the feature representations of nodes and edges as inputs, the OD passenger flow prediction module and the in-out passenger flow prediction module.

[0119] The prediction module includes the OD passenger flow prediction module and the in-out passenger flow prediction module. The OD passenger flow prediction is the main task, and the in-out passenger flow prediction is the auxiliary task. Fine-grained prediction is achieved through a task-specific network structure. Specifically:

[0120] OD passenger flow prediction module

[0121] The OD passenger flow prediction module takes the edge features (i.e., the implicit representation of the OD relationship) output by the spatio-temporal propagation module as input and generates the prediction result of the OD matrix for future time steps through a multi-layer network structure. The specific implementation steps are as follows:

[0122] 1. Edge feature extraction:

[0123] Obtain the edge feature matrix E from the spatio-temporal propagation module i,j ∈R N×N×H where E i,j represents the spatio-temporal implicit feature from station i to station j, and H is the dimension of the hidden layer.

[0124] 2. Fully connected mapping:

[0125] Perform a non-linear transformation on each edge feature and map it to the prediction space:

[0126]

[0127] where, W OD1 ∈R H×H and b OD1 ∈RH are learnable parameters.

[0128] 3. OD matrix prediction:

[0129] Output the predicted OD passenger flow values for the next K time steps through the second fully connected network:

[0130]

[0131] where and b OD2 ∈R are the parameters of the prediction layer. The final output is where each matrix dimension is N×N.

[0132] Set the OD matrix at time t as M t , where the element m i,j represents the number of passengers entering at station i and finally leaving at station j at time t. Assume there are N stations in the rail transit network, then the dimension of M t is N×N, forming the whole-network OD demand map at this time point.

[0133] In-out passenger flow prediction module

[0134] The in-out passenger flow prediction module takes the node features output by the spatio-temporal propagation module as input and predicts the in-out passenger flow of each station in the future. The specific implementation steps are as follows:

[0135] 1. Node feature extraction:

[0136] Obtain the node feature matrix H L [i]∈R N×H , where H L [i] represents the spatio-temporal implicit feature of station i.

[0137] 2. Feature transformation and activation:

[0138] Perform non-linear transformation on the node features:

[0139]

[0140] where W IOt ∈R H×H and b IOt ∈R H are learnable parameters; LayerNorm is the layer normalization operation.

[0141] 3. Passenger flow prediction:

[0142] Output the in-out passenger flow for the next K time steps through the fully connected network:

[0143]

[0144] Among them, and b IOt ∈R are the parameters of the prediction layer. The final output is The dimension of each vector is N.

[0145] Set the passenger flow vector at time t as F t , where the element represents the total in-out passenger flow at station i at time t. By performing time series modeling on the passenger flow data at different times, short-term passenger flow prediction can provide accurate demand prediction support for rail transit managers, ensuring sufficient transportation capacity to meet passenger demand in the upcoming time period.

[0146] In the present invention, short-term OD passenger flow prediction and short-term in-out passenger flow prediction are optimized as two subtasks in multi-task learning, enabling them to share the spatio-temporal features extracted in the model. This joint learning can not only obtain additional information from the in-out flow, but also improve the accuracy of OD passenger flow prediction. Especially during peak hours or special events, the changes in the in-out flow play an important auxiliary role in OD passenger flow prediction. Compared with traditional single-task prediction methods, the present invention effectively improves the accuracy of OD passenger flow prediction, especially in the face of scarce or imbalanced data, and can provide more reliable prediction results.

[0147] Through shared spatio-temporal feature learning, not only can the number of model parameters be reduced, but also the computational efficiency of the model can be improved, avoiding the overfitting problem. In the framework of multi-task learning, the joint learning of OD flow and in-out flow provides a more comprehensive spatio-temporal feature representation for the model, promotes the mutual learning between the two, and thus improves the overall prediction accuracy. Through the multi-task learning architecture, the short-term passenger flow prediction ability in the rail transit system can be effectively improved, providing more accurate decision-making support for urban traffic scheduling and operation management.

[0148] In actual tests, the OD passenger flow prediction method based on multi-task learning is significantly better than traditional single-task methods. Especially in the prediction tasks at different time periods and different stations, it shows high prediction accuracy and strong generalization ability. This method not only provides a new solution idea for short-term OD passenger flow prediction, but also provides more accurate data support for the intelligent scheduling and resource allocation of rail transit.

[0149] In summary, through the joint modeling of OD flow and in-out flow, the present invention can better capture the spatio-temporal dependence in the rail transit network, has strong practical application value, provides more reliable technical support for rail transit management and decision-making, and the multi-task learning framework provides new ideas and methods for the development of the short-term passenger flow prediction field, with broad application prospects.

[0150] The present invention also discloses a short-term OD passenger flow prediction model for urban rail transit based on multi-task learning, as Figure 3 shown. The model includes a spatio-temporal propagation module stacked by spatio-temporal learning units with gradually decreasing number of time steps layer by layer, an OD passenger flow prediction module, and an in-out station passenger flow prediction module. Based on the multi-task learning framework, the model is used to improve the accuracy of short-term OD passenger flow prediction.

[0151] The model takes short-term OD passenger flow prediction as the main task and in-out station passenger flow prediction as the auxiliary task. Through the synergistic effect between tasks, the overall prediction performance is improved. First, the model constructs a spatio-temporal propagation module using a message-passing based graph neural network (MPNN). In this module, the in-out station passenger flow volume is used as the node feature, and the OD passenger flow volume is used as the edge feature for modeling. MPNN models the spatio-temporal dependence relationship between stations through the graph structure, effectively capturing the dynamic spatial relationship between stations and its time-varying characteristics, thereby improving the accuracy of passenger flow prediction. Under the multi-task learning framework, OD passenger flow prediction and in-out station passenger flow prediction are jointly trained by sharing network parameters. The in-out station passenger flow volume, as the auxiliary task, provides a useful supplement for OD flow prediction. By sharing parameters, the model can promote knowledge sharing between tasks during the training process. Especially during peak periods or special events, the changes in the in-out station passenger flow volume help to improve the accuracy of OD passenger flow prediction. In addition, the model also introduces external features, such as date type and weather data. These external information can reflect the impact on passenger flow in different time periods. Combining these features with spatio-temporal data helps to further enhance the model's perception ability of complex spatio-temporal features and improve the prediction performance. Through the collaborative optimization between tasks and the fusion of external features, the model can simultaneously predict OD passenger flow and in-out station passenger flow, providing more accurate short-term passenger flow prediction results. It not only improves the accuracy of OD passenger flow prediction but also enhances the robustness of the model in complex environments, providing reliable decision-making support for the scheduling and management of urban rail transit.

[0152] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A short-term OD passenger flow prediction method for urban rail transit based on multi-task learning, characterized in that: The following steps are involved: S1. Use inbound and outbound passenger flow as node features and OD passenger flow as edge features to construct a spatiotemporal propagation module, where the spatiotemporal propagation module is composed of multiple spatiotemporal learning units stacked with decreasing time steps layer by layer; S2. Use the spatiotemporal learning unit to perform spatiotemporal feature learning on node features and edge features to obtain spatiotemporal features of nodes and edges; S3, the spatiotemporal feature information of nodes and edges is transmitted and aggregated layer by layer in the spatiotemporal propagation module to obtain the feature representation of nodes and edges; S4. Taking the feature representation of edges and nodes as input, we construct the OD passenger flow prediction module as the main task and the inbound and outbound passenger flow prediction module as the auxiliary task respectively. We use the OD passenger flow prediction module to obtain the OD matrix prediction result, and the inbound and outbound passenger flow prediction module to obtain the inbound and outbound passenger flow.

2. The urban rail transit short-term OD passenger flow prediction method based on multi-task learning according to claim 1 is characterized by: In step S2, the spatiotemporal learning unit includes a node spatial feature learning subunit, a node temporal feature learning subunit and an edge spatiotemporal feature learning subunit; The node space feature learning subunit is used to complete node space feature learning; The node time feature learning subunit is used to complete node time feature learning; The edge spatiotemporal feature learning subunit is used to complete edge spatiotemporal feature learning.

3. The urban rail transit short-term OD passenger flow prediction method based on multi-task learning according to claim 2 is characterized by: The node space feature learning comprises the following steps: Calculate the neighbor node attention coefficient: The neighbor node attention coefficient is calculated by the following formula: in, is the set of neighbor nodes of node i; j is the neighbor node, α i,j is the neighbor node attention coefficient; H i and H j are the features of nodes i and j respectively, W is the learnable weight matrix, a is the attention vector; k is any neighbor node of node i; H k is the feature vector of neighbor node k; | is vector concatenation; Calculate the node connection edge attention coefficient: The node connection edge attention coefficient is calculated by the following formula: Among them, e i,j is the edge feature between nodes i and j; β i,j is the node connection edge attention coefficient; W e is the edge feature transformation matrix; b is the attention vector; e i,k Represents the edge features between node i and its adjacent node k; Message aggregation: Based on the attention coefficient, the information of neighboring nodes and edges is aggregated to obtain the spatial features of node i: Among them, H space is the spatial feature of node i.

4. The urban rail transit short-term OD passenger flow prediction method based on multi-task learning according to claim 3 is characterized by: The node time feature learning comprises the following steps: Temporal Information Fusion: Fusion of the node features at the current moment with the node features at the previous moment: in, is the node feature at the current moment; is the node feature at the previous moment; φ temp is the time fusion function; Spatiotemporal feature fusion: The spatial feature H space With time characteristics Perform fusion and update node features: Among them, φ fuse is the feature fusion function.

5. The method for predicting short-term OD passenger flow of urban rail transit based on multi-task learning according to claim 4 is characterized by: The edge spatiotemporal feature learning includes the following steps: Node information aggregation: Based on edge i,j , aggregate the features of ingress node i and egress node j: Among them, φ node It is the node information aggregation function; Historical side information fusion: Set the current edge feature The edge feature at the previous moment To perform the fusion: Among them, φ temp is the historical side information aggregation function; Edge feature update: Fusion of node information and historical edge information, update of edge features: Among them, φ edge Update function for edge features.

6. The urban rail transit short-term OD passenger flow prediction method based on multi-task learning according to claim 1 is characterized by: In step S4, the OD passenger flow prediction module includes the following prediction steps: Edge feature extraction: Get the edge feature matrix E from the space-time propagation module i,j ∈R N×N×H , where E i,j It represents the temporal and spatial implicit features from site i to site j, and H is the hidden layer dimension. Fully connected mapping: Perform a nonlinear transformation on each edge feature and map it to the prediction space: Among them, W OD1 ∈R H×H and b OD1 ∈R H is a learnable parameter; OD matrix prediction: The second layer of fully connected network outputs the predicted OD passenger flow value for the next K time steps: in, and b OD2 ∈R is the prediction layer parameter; the final output is Among them, each matrix dimension is N×N.

7. The urban rail transit short-term OD passenger flow prediction method based on multi-task learning according to claim 1 is characterized by: In step S4, the inbound and outbound passenger flow prediction module includes the following prediction steps: Node feature extraction: Get the node feature matrix H from the spatiotemporal propagation module L [i]∈R N×H , where H L [i] represents the temporal and spatial implicit features of site i; Feature transformation and activation: Perform nonlinear transformation on node features: Among them, W IOt ∈R H×H and b IOt ∈R H is a learnable parameter; LayerNorm is a layer normalization operation; Passenger flow forecast: Output the inbound and outbound passenger flow in the next K time steps through the fully connected network: in, and b IOt ∈R is the prediction layer parameter; The final output is Each vector has dimension N.

Images