Rail passenger flow short-term prediction method based on dynamic multi-graph fusion space-time deep learning

Abstract

The invention discloses a rail passenger flow short-term prediction method based on dynamic multi-graph fusion space-time deep learning, and the method comprises the steps: carrying out the convolution of historical passenger flow feature data, and forming input data; then, a fusion graph convolution module is constructed, the fusion graph convolution module comprises multiple predefined graph convolution and dynamic graph convolution, and causal time is introduced into the fusion graph convolution module to form a space-time learning device; processing the input data by using a space-time learning device; and finally, the output of the space-time learner is connected through a jump connection layer, the jump connection layer introduces the output of the space-time learner into an output layer, and a prediction result is obtained through the output layer. According to the method, the combination of the predefined graph convolution and the dynamic graph convolution is used as the adjacent matrix, the complex spatial features of the passenger flow can be effectively learned, a causal time multi-head attention mechanism is introduced, and multi-layer spatial-temporal feature extraction can be performed, so that the spatial-temporal feature extraction capability of the model is improved; and more accurate feature representation is provided for short-term passenger flow prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term prediction of rail passenger flow, and in particular to a short-term prediction method of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning. Background Art

[0002] With the rapid expansion of the Urban Rail Transit (URT) network and the continuous increase in passenger flow, the temporal and spatial distribution of passengers has become increasingly complex. Short-term passenger flow forecasting provides precise data, allowing operators to adjust strategies in real time to ensure smooth system operation. This forecast not only helps passengers plan their trips more effectively, thereby reducing costs, but also strengthens the management of the entire system. Therefore, accurate short-term passenger flow forecasting is indispensable for the effective management and operation of urban rail transit systems.

[0003] At present, in the field of urban rail transit passenger flow prediction, graph convolutional networks (GCN) are generally used to predict traffic. However, this prediction method requires complex professional domain knowledge and may have the problem of incomplete information utilization. Or the method of adaptively constructing a spatial adjacency matrix based on data. The principle of the above method is to construct a matrix using shared parameters of each layer and update the matrix through methods such as gradient descent. Although GCN has achieved good results in urban rail transit prediction, the following problems still exist.

[0004] 1. In the dynamic spatiotemporal characteristics of urban rail transit. Due to the rapid changes in commuting patterns, traffic congestion and other factors, the interactions between stations or nodes in the network are dynamic. For example, the traffic volume or traffic density between two stations during peak hours (such as morning commuting time) may be very different from the traffic volume or traffic density during non-peak hours (such as afternoon). Traditional models often fail to capture the dynamic and evolving nature of the spatial relationships between these stations, resulting in reduced prediction accuracy.

[0005] 2. The variables that affect urban rail transit passenger flow are multifactorial and complex. The physical structure of the urban rail transit network, historical traffic patterns, and other environmental and temporal factors such as weather, holidays, or special events are all interrelated. These factors are usually represented by predefined graphs. However, each predefined graph can only capture a fragment of the overall narrative, bypassing the interactions between these factors. For example, a graph that emphasizes the physical structure of the urban rail network may ignore the impact of historical passenger flow patterns and environmental variables. This one-sided dynamic perspective will further affect the prediction accuracy of traditional models. Summary of the invention

[0006] The purpose of the present invention is to provide a short-term prediction method for rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning to address the above-mentioned problems, and overcome the limitations of static matrices in traditional methods. In the prior art, traditional graph neural network methods usually use static adjacency matrices in urban rail transit predictions, which ignores the dynamic changes in the correlation between stations during the day. The present invention uses the Tucker decomposition method to construct a time-varying dynamic graph, which can accurately capture the spatial correlation of passenger flow in different time intervals, effectively overcoming this limitation; enriching the extraction of multivariate spatial correlation of passenger flow, and previous research methods are relatively single in extracting passenger flow spatial correlation. The present invention constructs three predefined graphs, namely the physical structure graph, the passenger flow similarity graph, and the OD correlation graph, and fuses them with the dynamic graph to construct a fused graph convolution module, which greatly enriches the multivariate spatial correlation of passenger flow and provides more comprehensive information for more accurate short-term passenger flow prediction; improves the ability to extract spatiotemporal features, and the prior art has deficiencies in spatiotemporal feature extraction. The present invention introduces a causal temporal multi-head attention mechanism, which together with the fusion graph convolution module forms a spatiotemporal learner, which can perform multi-layer spatiotemporal feature extraction, thereby improving the model's ability to extract spatiotemporal features and providing more accurate feature representation for short-term passenger flow prediction. In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning is provided, comprising the following steps: S1, convolve the historical passenger flow feature data to form input data; S2. construct a fused graph convolution module, wherein the fused graph convolution module includes multiple predefined graph convolutions and dynamic graph convolutions, and introduces a causal time multi-head attention mechanism into the fused graph convolution module to form a spatiotemporal learner; S3, connecting the plurality of spatiotemporal learners in series in sequence, and using each of the spatiotemporal learners to process the input data respectively; S4. The outputs of all the spatiotemporal learners are connected through a skip connection layer. The skip connection layer introduces the outputs of the spatiotemporal learners into an output layer, and a prediction result is obtained through the output layer.

[0007] Preferably, in step S1, the historical passenger flow characteristic data is represented by a historical passenger flow sequence, specifically:

[0008] in, represents the historical passenger flow sequence; represents the inflow data of the nth station in the tth time interval; Indicates the total number of stations.

[0009] Preferably, in step S2, the output of the fused graph convolution module is expressed by the following formula:

[0010] in, is the output of the fused graph convolution; is the gate value; is the output of multiple predefined graph convolutions; is the output of dynamic graph convolution; is the Hadamard product.

[0011] Preferably, the gate value is represented by the following formula:

[0012] in, is the output of multiple predefined graph convolutions; is the output of dynamic graph convolution; is the gate value; is the bias parameter.

[0013] Preferably, the output of the multiple predefined graph convolutions It can be expressed by the following formula:

[0014] in, It is the convolution output of the physical structure graph; is the output of the passenger flow similarity graph convolution; is the output of OD related graph convolution; For tensor connection.

[0015] Preferably, the physical structure graph convolution output , the output of the passenger flow similarity graph convolution , OD related graph convolution They are expressed by the following formulas:

[0016] in, It is the graph convolution operation; It is a physical structure diagram; is the passenger flow similarity graph; is the OD correlation diagram; is the output of the causal temporal multi-head attention mechanism.

[0017] Preferably, in step S2, the dynamic space graph can be used Indicates that at a specific moment , the dynamic space graph can be expressed as:

[0018] in, Represents a collection of sites; Indicates time spatial connectivity; Representation Node and nodes The corresponding weights between .

[0019] Preferably, in step S4, the output of the spatiotemporal learner can be expressed as:

[0020] in, is the output of the fused graph convolution, is the output of the spatiotemporal learner.

[0021] Preferably, in step S4, the prediction result obtained by the output layer can be expressed by the following formula:

[0022] in, is the number of spatiotemporal learners, is the output of the spatiotemporal learner.

[0023] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: The present invention first convolves the historical passenger flow feature data for dimensional alignment; then, a spatiotemporal learner is constructed by fusion graph convolution and causal time mechanism. In the process of fusion graph convolution, a combination of predefined graph convolution and dynamic spatial graph convolution is used as the adjacency matrix, which can effectively learn the complex spatial characteristics of passenger flow. The causal time multi-head attention mechanism is introduced, and together with the fusion graph convolution module, a spatiotemporal learner is constructed, which can perform multi-layer spatiotemporal feature extraction, thereby improving the model's ability to extract spatiotemporal features and providing more accurate feature representation for short-term passenger flow prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of the dynamic graph construction process of the present invention; Figure 3 It is a schematic diagram of the structure of the fusion graph convolution module of the present invention; Figure 4 is a visualization schematic diagram of CT-MSA of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are only for the purpose of 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 realized even without these specific details.

[0026] See also Figure 1 The present invention provides a short-term prediction method for rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning, and the technical solution is as follows: A short-term prediction method for rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning, comprising the following steps: S1. Convolve the historical passenger flow feature data to form input data.

[0027] Specifically, a historical passenger flow sequence is extracted, and the historical passenger flow sequence includes data from the automatic ticket collection (AFC) system. The data from the automatic ticket collection (AFC) system includes the passenger's card number, departure station, arrival station, and respective timestamps. After extracting the historical passenger flow sequence from the data of the automatic ticket collection (AFC) system, the historical passenger flow sequence is structured into a matrix to form passenger flow features configured at different time intervals. In this embodiment, the time intervals are set at intervals of 10, 15, and 60 minutes. The sequence can be represented by the following formula:

[0028] in, represents the historical passenger flow sequence; Indicates In the time interval Inflow data of each site; Indicates the total number of stations.

[0029] After obtaining the passenger flow sequence, 1x1 convolution is performed on the passenger flow sequences in different time periods, and the convolution results can be used as input data for subsequent processing.

[0030] S2. Construct a fused graph convolution module, wherein the fused graph convolution module includes multiple predefined graph convolutions and dynamic graph convolutions, and introduces a causal temporal multi-head attention mechanism into the fused graph convolution module to form a spatiotemporal learner.

[0031] First, construct a dynamic space graph; Since existing methods usually rely on predefined dependencies, such as distance and functional similarity, or use static graphs, they cannot explain the heterogeneity of different time intervals. Therefore, in this embodiment, a dynamic spatial graph construction method for dynamic graph convolution is used to capture the dynamic spatial dependencies in passenger flow data by tailoring the spatial representation of latent attributes to specific time intervals. Specifically, based on the concept of periodicity, it is assumed that inputs from the same time period on different days can be used to construct a shared dynamic graph. The dynamic graph can be represented as , at a specific moment , the dynamic space graph can be expressed as:

[0032] in, Represents a collection of sites; Indicates time spatial connectivity; Representation Node and nodes The corresponding weights between them. The method of building a shared dynamic graph reduces the number of separate constructions. A dynamic graph approach. Using tensors Indicates construction A dynamic graph to determine a specific time Suitable dynamic graph, use Calculate the corresponding index and select As the time Dynamic graph of . Using Tucker decomposition to reorganize the adjacency tensor can effectively reduce the number of required parameters. Allocate three matrices and one core tensor: the slot embedding matrix , source node embedding matrix , target node embedding matrix , and a core tensor . The adjacency tensor can be determined from the above elements:

[0033] in, and denote the time slot, original node, target node and embedding dimension respectively. Functions and Functions are used to normalize and activate adjacency tensors, which can improve the performance and stability of dynamic graph convolution. Figure 2 The process of constructing a dynamic graph using Tucker decomposition is shown. The core tensor and factor matrix are learned from data to capture the latent spatial correlations at different time intervals.

[0034] Then, in order to solve the problem that relying solely on dynamic graphs cannot fully explore hidden spatial dependencies and lacks sufficient interpretability (although the dynamic spatial graph generated by the dynamic graph builder can reflect the dynamic correlation of passenger flow characteristics at different stations within a day, the dynamic graph is essentially a supplement to the uncertain relationship between nodes). In order to more deeply explore these hidden spatial dependencies and enhance the explanatory power of the model, in this embodiment, Figure 3 As shown in the figure, by constructing a fusion graph convolution module and introducing a causal time multi-head attention mechanism in the fusion graph convolution module to form a spatial learner, more prior knowledge is added. Specifically, the fusion graph convolution module consists of a dynamic graph convolution module and multiple predefined graph convolution modules. The dynamic graph convolution module uses the constructed dynamic adjacency matrix To capture and integrate the dynamic spatial information in the rail passenger flow network. This process can be expressed mathematically as follows:

[0035] in, Represents the graph convolution operation, with a dynamic adjacency matrix and causal time Module Output as its input.

[0036] Among them, for , representing the urban rail network as a graph ,in is the set of nodes (stations), is an edge set, representing the abstract connection relationship between different stations. The adjacency matrix of the graph can be expressed as ,in Indicates the number of nodes. Row and Elements of a column Represents the weight coefficient, which represents the strength of the relationship between stations. By using a graph convolutional network (GCN) to operate on different adjacency matrices, different spatial dependencies between stations can be mined.

[0037] The GCN operation can be expressed as follows:

[0038] in, is the node representation matrix of the lth layer; is the weight matrix to be learned in the lth layer; is the adjacency matrix with self-connection added, where is the identity matrix; yes The degree matrix of .

[0039] In multiple predefined graph convolution modules, using expertise in fields such as topology and similarity, a variety of graphs can be designed to illustrate various spatial connections between nodes (especially stations). Stations with similar functions also have similar passenger flow change patterns. At the same time, the size of the OD (origin-destination) passenger flow between stations can also reflect the correlation between the two stations. Therefore, based on prior knowledge, three types of graphs representing different spatial relationships are constructed. These three types of graphs are the physical adjacency matrix (physical graph of URT network), the functional similarity matrix (functional similarity graph), and the OD correlation matrix (origin-destination (OD) correlation graph). They are named respectively. and .in, represents a set of stations; Indicates the number of URT stations; express Central Station With the station In order to capture the multifaceted spatial interdependencies in the entire URT network, Creating multiple weight matrices The following are descriptions of the three types of graphs: 1) Physical diagram: picture is constructed directly from the physical layout of the subway system being studied. Nodes in and In the real world, an edge is created to connect nodes. and To assign weights to these edges, we first construct a physical connection matrix , the dimension is If the node and If there is an edge between the elements Set to 1, otherwise Set to 0. Finally, the edge weight is obtained by linear normalization of each row . Specifically, the element The calculation method is as follows:

[0040] 2) Functional similarity graph: The spatial layout between stations is crucial, but it is equally critical to assess their functional consistency. Despite being geographically distant, the connections between some stations can be attributed to their similar roles, such as commuting or commercial centers. Therefore, this functional similarity must be incorporated when capturing spatial interdependencies. To further explore this, Represented as a site Historical passenger attributes, where represents the total passenger characteristics, Representation interval The time step within. and The similarity weight between can be expressed as:

[0041] According to the given similarity weight matrix , select sites with high similarity weights to build edges A threshold is set in advance , used to identify these sites, as follows:

[0042] In the modified similarity weight matrix After that, each row is normalized to facilitate training. The calculation formula of the normalized matrix is:

[0043] 3) OD correlation diagram: The OD (origin-destination) information of the target site is an important indicator to measure the closeness of the connection between the site and other sites. In order to capture the closeness between sites, the OD features between different sites are extracted from the original AFC data to form an OD association graph. To site The OD correlation coefficient is defined as follows and is expressed as :

[0044] in, Indicates from Stand The total number of passengers at the station; Indicates the total number of stations.

[0045] By comparison and preset allocation ratio To determine the stations with high allocation ratio:

[0046] In order to further improve the graph, the rows and columns are normalized to obtain the OD correlation weight matrix :

[0047] In each layer of ST-Feature Learner, the output of the causal time attention mechanism is input, and the convolution results of the three different types of graphs mentioned above are obtained. They are then concatenated and fused into tensors to finally obtain the output of multiple predefined graph convolutions, which are expressed as follows:

[0048] in, It is the convolution output of the physical structure graph; is the output of the passenger flow similarity graph convolution; is the output of OD related graph convolution; For tensor connection; It is the output of multiple predefined graph convolutions.

[0049] The high-order spatial information representations extracted from two types of multiple predefined graph convolutions and dynamic graph convolutions are respectively , and on this basis, the gate control value is further calculated as follows:

[0050] in, is the output of dynamic graph convolution; is the gate value; is the bias parameter.

[0051] Therefore, the fused graph convolution output of the l-th ST feature learner layer is calculated as follows:

[0052] in, is the output of the fused graph convolution, is the output of the ST feature learner, is the Hadamard product.

[0053] The above fused graph convolution module can derive the spatial dependency of passenger flow. In addition to spatial dependency, the passenger flow of rail stations is also affected by its historical passenger flow characteristics. Therefore, the causal temporal MSA (CT-MSA) is introduced. As shown in Figure 4, its implementation is generally similar to the standard MSA. However, the difference is that CT-MSA uses the domain knowledge of time series and makes two major modifications. They are: Local window: Since adjacent time steps usually show stronger correlation than distant time periods, MSA is implemented in non-overlapping windows to encapsulate local interactions between time steps, so that the computational cost is O(TWC), which effectively reduces TW compared to standard MSA (W is the window size). At the same time, in order to maintain the wide receptive field of standard MSA, the window size is gradually expanded in different ST feature learners.

[0054] Temporal causality: Since the passenger flow characteristics of the current step are irrelevant to its future, based on WaveNet, causality is incorporated into MSA to ensure that the model does not violate the temporal order of the input data. Causality can be achieved through the masked attention mechanism. To facilitate location-aware MSA, a learnable absolute position encoding is introduced in the input of CT-MSA.

[0055] In terms of extracting temporal features of passenger flow, compared with traditional recurrent neural networks, the multi-head attention mechanism can process information of each time step in parallel, effectively handle the dependencies of various time spans, and improve the prediction accuracy of the model.

[0056] In MSA, each element of the input sequence is mapped into three vectors, namely query (Q), key (K), and value (V), through a separate linear transformation. The attention score is then calculated by taking the dot product of Q and K, and then a soft maximum operation is performed to ensure that the scores form a valid probability distribution. These scores are then used to calculate the weighted sum of the V vector, thereby providing a time-related representation for each specific time period in the prediction model. In the multi-head design, each "head" corresponds to an independent attention calculation process, so that time information of different dimensions can be extracted simultaneously, thereby enriching the expressive power of the model.

[0057] The form of MSA operation can be expressed as follows:

[0058] The calculation formula for each head is:

[0059] The attention function is defined as:

[0060] By introducing the causal temporal MSA (CT-MSA), both historical passenger flow data and causal relationships are considered, thereby enhancing the processing of time series characteristics and being able to learn the temporal dynamic changes of rail passenger flow more accurately.

[0061] S3. Connect multiple spatiotemporal learners in series and use each spatiotemporal learner to process the input data separately.

[0062] Specifically, multiple spatiotemporal learners are connected in series in sequence, the output of the previous spatiotemporal learner is connected to the input of the next spatiotemporal learner, and the last spatiotemporal learner is connected to the output module. The causal temporal MSA module and the fusion graph convolution module of each learner process the input, and the residual connection ensures effective gradient back propagation.

[0063] S4. The outputs of all the spatiotemporal learners are connected through a jump connection module, and the outputs of the jump connection module are introduced into an output module to obtain a prediction result through the output module.

[0064] The skip connection layer (skip connection module) can effectively avoid gradient vanishing by integrating information from different layers (different spatiotemporal learners), thereby improving the prediction accuracy of the model. Specifically, the output of each spatiotemporal learner is first summed up as a tensor, and then a The standard convolution transforms the information connected to the output module into a sequence of length 1, where Represents the input sequence length of each connection layer.

[0065] The output module consists of two 1×1 standard convolutional layers in order to adjust the channel dimension of the input to achieve the desired output dimension. Specifically, if the task is to predict one step into the future, the desired output dimension is 1. However, if the prediction spans Q consecutive steps, the output dimension is Q.

[0066] By finding a function (with ), using this function to predict future passenger flow data within a specified time period. The prediction is based on a series of historical passenger flow data, expressed as , where the variable represents the historical time step. Other factors, such as the predefined URT network graph (denoted as Gp, Gs, and Gc) and the dynamic space graph (denoted as Gd), are integral to the forecast. Therefore, the function The mathematical expression can be expressed as:

[0067] in, Indicates the next time step Forecasted passenger flow.

[0068] The prediction of future passenger flow can be expressed as follows:

[0069] All in the model The combined result of the individual outputs of the layer.

[0070] The prediction process of the present invention is as follows: first, a 1x1 convolution is performed on the historical passenger flow feature data to align the dimensions. Then, a spatiotemporal learner is composed of a causal time MSA module and a fusion graph convolution module. Multiple spatiotemporal learners are connected in series to form a DMSTGCN model. The causal time MSA module and the fusion graph convolution module of each learner process the input, and the residual connection ensures effective gradient back propagation; then, the outputs of all learners are connected through jumps and input into the output layer to obtain the final result. In the process of fusion graph convolution, a combination of predefined and dynamic spatial graphs is used as the adjacency matrix, which can effectively learn the complex spatial characteristics of passenger flow.

[0071] The present invention overcomes the limitation of static matrix in traditional methods. Traditional graph neural network methods usually use static adjacency matrix in urban rail transit prediction, which ignores the dynamic change of station correlation in a day. The present invention can accurately capture the spatial correlation of passenger flow in different time intervals by using Tucker decomposition method to construct time-varying dynamic graph, effectively overcoming this limitation; and overcomes the singleness of existing methods in extracting passenger flow spatial correlation, enriching the multivariate spatial correlation extraction of passenger flow. By constructing three predefined graphs, namely physical structure graph, passenger flow similarity graph and OD correlation graph, and fusing them with dynamic graph to construct fusion graph convolution module, the multivariate spatial correlation of passenger flow is greatly enriched, providing more comprehensive information for more accurate short-term passenger flow prediction. The causal time multi-head attention mechanism is introduced, and together with the fusion graph convolution module, a spatiotemporal learner is formed. The spatiotemporal learner can be used to extract multi-layer spatiotemporal features, thereby improving the model's ability to extract spatiotemporal features and providing more accurate feature representation for short-term passenger flow prediction.

[0072] The beneficial effects of the present invention are: by constructing a time-varying dynamic graph and a predefined graph integrating multivariate spatial correlations, and introducing a causal time multi-head attention mechanism for multi-layer spatiotemporal feature extraction, the short-term passenger flow of urban rail transit can be predicted more accurately. A large number of experimental results conducted on three real-world data sets have confirmed the superior performance of this model, providing a more reliable decision-making basis for urban traffic management and crowd control; the interpretability of dynamic spatial relationships, the present invention not only provides accurate short-term passenger flow prediction results, but also shows the intuitive results of the dynamic spatial graph, further proving its effectiveness.

[0073] This interpretability enables traffic managers and relevant decision makers to better understand the dynamic changes in inter-station correlations, and thus take more targeted measures for traffic management and crowd control; the dynamic multi-graph fusion spatiotemporal graph convolutional network (DMSTGCN) proposed in this invention is a new model architecture that integrates Tucker decomposition, predefined graph fusion and causal temporal multi-head attention mechanism. This architecture provides new ideas and methods for short-term passenger flow prediction in urban rail transit, and is expected to promote technological development in this field.

[0074] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A short-term prediction method for rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning, characterized by: The following steps are involved: S1, convolve the historical passenger flow feature data to form input data; S2. construct a fused graph convolution module, wherein the fused graph convolution module includes multiple predefined graph convolutions and dynamic graph convolutions, and introduces a causal time multi-head attention mechanism into the fused graph convolution module to form a spatiotemporal learner; S3, connecting the plurality of spatiotemporal learners in series in sequence, and using each of the spatiotemporal learners to process the input data respectively; S4. The outputs of all the spatiotemporal learners are connected through a skip connection layer. The skip connection layer introduces the outputs of the spatiotemporal learners into an output layer, and a prediction result is obtained through the output layer.

2. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 1 is characterized by: In step S1, the historical passenger flow characteristic data is represented by a historical passenger flow sequence, specifically: in, represents the historical passenger flow sequence; represents the inflow data of the nth station in the tth time interval; Indicates the total number of stations.

3. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 1 is characterized by: In step S2, the output of the fused graph convolution module is expressed as follows: in, is the output of the fused graph convolution; is the gate value; is the output of multiple predefined graph convolutions; is the output of dynamic graph convolution; is the Hadamard product.

4. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 3 is characterized by: The gate value is expressed by the following formula: in, is the output of multiple predefined graph convolutions; is the output of dynamic graph convolution; is the gate value; is the bias parameter.

5. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 3 is characterized by: The output of the multiple predefined graph convolutions It can be expressed by the following formula: middle, It is the convolution output of the physical structure graph; is the output of the passenger flow similarity graph convolution; is the output of OD related graph convolution; For tensor connection.

6. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 5 is characterized by: The physical structure graph convolution output , the output of the passenger flow similarity graph convolution , OD related graph convolution They are expressed by the following formulas: in, It is the graph convolution operation; It is a physical structure diagram; is the passenger flow similarity graph; is the OD correlation diagram; is the output of the causal temporal multi-head attention mechanism.

7. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 1 is characterized by: In step S2, the dynamic space graph is available Indicates that at a specific moment , the dynamic space graph can be expressed as: in, Represents a collection of sites; Indicates time spatial connectivity; Representation Node and nodes The corresponding weights between .

8. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 3 is characterized by: In step S4, the output of the spatiotemporal learner can be expressed as: in, is the output of the fused graph convolution, is the output of the spatiotemporal learner.

9. The method for short-term prediction of rail passenger flow based on dynamic multi-graph fusion spatiotemporal deep learning according to claim 4 is characterized by: In step S4, the prediction result obtained by the output layer can be expressed by the following formula: in, is the number of spatiotemporal learners, is the output of the spatiotemporal learner.

Images