Cross-traffic-mode passenger flow joint prediction method based on data interoperation mechanism

Abstract

The invention provides a cross-traffic-mode passenger flow joint prediction method based on a data interoperation mechanism. The method comprises the following steps: constructing single traffic sub-networks based on a hypergraph theory according to network topology structure data of a passenger flow prediction area, and obtaining historical passenger flow data of all nodes on each single traffic sub-network; constructing a cross-mode traffic sub-network by using a data interoperation mechanism based on the topological structure of each single traffic mode sub-network; and constructing a multi-mode space-time hypergraph convolutional network, inputting the historical passenger flow data of all nodes on each single traffic sub-network and the traffic network topology structure of the cross-mode traffic sub-network into the multi-mode space-time hypergraph convolutional network, and outputting a passenger flow prediction result of each single traffic sub-network by the multi-mode space-time hypergraph convolutional network. According to the method, passenger flow characteristics of stations and routes are captured by building a space-time hypergraph convolutional network; based on feature aggregation and distribution, accurate prediction of passenger flow of multiple traffic modes is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of passenger flow prediction, and in particular to a method for joint prediction of passenger flow across transportation modes based on a data interoperability mechanism. Background Art

[0002] With the increase of urban population, a single mode of transportation can no longer meet people's travel needs, and residents are increasingly inclined to adopt multiple modes of transportation. Jointly predicting passenger flow of multiple modes of transportation can achieve more accurate travel planning and resource allocation, improve the efficiency and service quality of the transportation system, reduce traffic congestion, optimize the operation of public transportation, and ultimately improve passengers' travel experience. In the early days, statistical models, neural networks and other methods were mainly used, but most of them focused on traffic flow or passenger flow prediction under the spatiotemporal relationship of a single mode of transportation, ignoring the potential correlation between various modes of transportation.

[0003] Therefore, in recent years, the traditional model has been improved and optimized through grid aggregation passenger flow and graph neural network methods. However, it is still difficult to generalize to different modes of transportation with heterogeneous spatial relationships. In addition, the traditional transportation network modeling method does not adequately consider the correlation between nodes within a single mode of transportation, and fails to accurately capture the correlation characteristics across transportation modes.

[0004] Currently, there is no effective method for joint prediction of passenger flow across transportation modes in the existing technology. Summary of the invention

[0005] The embodiment of the present invention provides a method for joint prediction of passenger flow across transportation modes based on a data interoperability mechanism, so as to achieve accurate prediction of passenger flow of multiple transportation modes.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] A cross-transportation mode passenger flow joint prediction method based on data interoperability mechanism, including:

[0008] According to the network topology data of the passenger flow prediction area, a single traffic sub-network is constructed based on the hypergraph theory, and the historical passenger flow data of all nodes on each single traffic sub-network is obtained;

[0009] Based on the topological structure of each single transportation mode sub-network, a cross-modal transportation sub-network is constructed using a data interoperability mechanism;

[0010] A multi-mode spatiotemporal hypergraph convolutional network is constructed, and the historical passenger flow data of all nodes on each single transportation sub-network and the transportation network topology structure of the cross-modal transportation sub-network are input into the multi-mode spatiotemporal hypergraph convolutional network. The multi-mode spatiotemporal hypergraph convolutional network outputs the passenger flow prediction results of each single transportation sub-network.

[0011] Preferably, the method of constructing a single traffic sub-network based on the network topology data of the passenger flow prediction area based on the hypergraph theory includes:

[0012] Define the traffic network G = {G1, G2, ..., G n ,…,G N},n=1,2,…,N,G n represents a single transportation mode subnetwork, N is the number of transportation modes, including subway and bus, N = 2; according to the network topology data of the passenger flow prediction area, a single transportation subnetwork G is constructed based on the hypergraph theory n = {V, ε, H}, V represents G n The node set of q=1,2,…,Q n , Q n G n The total number of nodes, v q Represents G n The qth node in the network corresponds to the subway and bus stations. The number of nodes in the subway sub-network is set to Q1, and the number of nodes in the bus sub-network is set to Q2. ε represents G n The set of hyperedges e in , k=1,2,…,K n , K n G n The total number of hyperedges, e k Represents G n The kth hyperedge in H corresponds to the subway and bus lines. n Represents G n The correlation matrix of Represents node v q and super edge e k The subordinate relationship between

[0013]

[0014] Node v in the subway sub-network G1 and the bus sub-network G2 q The node over-degree k(v q ) all satisfy formula (2), and the super edge degree k s (e k ) all satisfy formula (3);

[0015]

[0016] The node excess k(vq) is the node v q The number of hyperedges, all nodes v q The k(vq) forms a diagonal matrix

[0017]

[0018] Super edge super k s (e k ) refers to the hyperedge e k The number of all nodes in the k K s (e k ) forms a diagonal matrix

[0019]

[0020] Preferably, the acquisition of historical passenger flow data of all nodes on each single traffic sub-network includes:

[0021] Get a single traffic subnetwork G n Historical passenger flow data of all nodes Then G n The passenger flow data of the qth node in the historical step length U period is u=1,2,…,U;

[0022] Predict a single traffic subnetwork G within a time period of step length P n The future passenger flow of Q nodes is:

[0023] The future passenger flow of the qth node is

[0024] Preferably, the topological structure based on each single transportation mode sub-network utilizes a data interoperability mechanism to construct a cross-modal transportation sub-network, including:

[0025] Based on the data interoperability mechanism, the data between the subway subnetwork G1 and the bus subnetwork G2 are interconnected and shared, and the cross-modal transportation subnetwork is defined as G 1,2 , G 1,2 The nodes of G1 and G2 are nodes v i and v j composition;

[0026] G 1,2 The hyperedge in is a cross-mode association hyperedge, that is, M. l=1,2,…,L 1,2 , L 1,2 G 1,2 The total number of hyperedges associated in the mid-span mode, m l Represents G 1,2 The lth hyperedge in G 1,2 The correlation matrix is And Q1<Q2, v i and ml One-to-one correspondence, H 1,2 Satisfy formula (4);

[0027]

[0028] In G 1,2 Medium i The node over-degree k(v i ) is fixed to 1; v j The node over-degree k(v j ) satisfies formula (5), k(v i ) and k(v j ) together form a diagonal matrix

[0029]

[0030] Super edge super k s (m l ) satisfies formula (6), all hyperedges m l K s (m l ) forms a diagonal matrix

[0031] Preferably, the construction of a multi-mode spatiotemporal hypergraph convolutional network includes: constructing a multi-mode spatiotemporal hypergraph convolutional network using an input module, an output module and the spatiotemporal convolutional module, constructing a spatiotemporal convolutional module using hypergraph convolution based on a data interoperability mechanism, the spatiotemporal convolutional module includes 2 layers of temporal convolution and 2 layers of spatial convolution, and the output module consists of 1 layer of temporal convolution, 2 layers of fully connected layers and a feature fusion layer.

[0032] Preferably, the historical passenger flow data of all nodes on each single traffic sub-network and the traffic network topology structure of the cross-modal traffic sub-network are input into the multi-mode spatiotemporal hypergraph convolutional network, and the multi-mode spatiotemporal hypergraph convolutional network outputs the passenger flow prediction results of each single traffic sub-network, including:

[0033] The traffic subnetwork G n The historical passenger flow data of all nodes on the network are converted into a passenger flow feature matrix X, and the traffic network topology structure of the cross-modal traffic subnetwork is converted into a road network association matrix H;

[0034] The passenger flow feature matrix X and the association matrix H of the road network are input into the input module of the multi-mode spatiotemporal hypergraph convolutional network. The spatiotemporal convolutional module integrates the historical passenger flow feature matrix X using a causal convolution operation, and then performs residual calculation and Sigmod activation on the integrated data to obtain output results A and B, where A represents the result of the front C0 channel and B represents the result of the rear C0 channel.

[0035] The output of the temporal convolution is calculated by equation (7), i.e., the length is UK t +1 passenger flow time series

[0036]

[0037] In the formula, represents the gated convolution kernel, * τ represents the gated convolution operator, K t is the width of the convolution kernel corresponding to the time step, ⊙ represents the Hadamard product, σ sig (·) represents the Sigmod activation function;

[0038] The spatial characteristics of traffic subnetworks G1 and G2 are extracted through formula (8);

[0039]

[0040] Where n is 1 or 2, W n is the identity matrix, Θ n G n The filter parameters of each node in X need to be learned during the training process; n G n The historical passenger flow feature matrix, D v G n The node excess matrix, H n G n The correlation matrix, D e G n The hyperedge hyperdegree matrix of ;

[0041] The cross-modal sub-network G is extracted by formula (9) 1,2 spatial characteristics;

[0042]

[0043] Where, X merge =X 1 +X 2 , X 1 is the passenger flow in G1, X 2 is the passenger flow in G2, W i,j is the identity matrix, Θ 1,2 G 1,2 The filter parameters of each node in need to be learned during the training process; G 1,2 The node excess matrix, H i,j G 1,2 The correlation matrix, D i,j G 1,2The hyperedge hyperdegree matrix of ;

[0044] Through the two-stage process of feature aggregation and then distribution, we can obtain and That is, G obtained by operating formula (8) n The future passenger flow matrix, That is, G obtained by operating formula (9) 1,2 The future passenger flow matrix,

[0045] The output module maps the output of the spatiotemporal convolution module to a single-step prediction based on the time-gated convolution. The output module obtains nonlinear features through the fully connected layer and outputs G1, G2 and G 1,2 The passenger flow prediction sequence of all nodes As shown in formula (10):

[0046]

[0047] In the formula, ω r is the weight coefficient in the fully connected layer, x r is the value of the rth neuron in the previous layer, b r is the bias of the fully connected layer; f refers to the calculation formula of the fully connected layer, and Y refers to the passenger flow prediction results of each node on G1 and G2;

[0048] Use formula (11) to fuse the passenger flow prediction results of each sub-network node;

[0049]

[0050] Specifically G i Future passenger flow matrix

[0051] In the formula, and is a learnable parameter matrix, i is 1 or 2, Corresponding to G 1,2 The result of the spatiotemporal convolution on the ith transportation mode subnetwork, ⊙ is the Hadamard product;

[0052] Define the prediction results to be optimized and updated through the following loss function;

[0053]

[0054] Where Y z is the true value, To predict the value, the model is trained end-to-end based on the error back propagation mechanism by minimizing the mean absolute error between the predicted value and the true value, and the prediction result is updated

[0055] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the embodiments of the present invention are based on data interoperability between different modes of transportation, introduce a cross-transportation mode hypergraph association matrix to characterize the association relationship between the passenger flow hypergraph networks of various modes of transportation; capture the passenger flow characteristics of sites and routes by building a spatiotemporal hypergraph convolutional network; based on the aggregation and allocation of features, accurate prediction of passenger flow of various modes of transportation is achieved. The present invention breaks data silos based on data interoperability, and mines potential correlations driven by similar passenger flow patterns and functions at sites to better reflect the local and overall characteristics in the passenger flow network and improve the accuracy of passenger flow prediction. This method can be applied to passenger flow prediction in real transportation networks, and will obtain good prediction accuracy.

[0056] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0058] Figure 1 A schematic diagram of an implementation principle of a method for joint passenger flow prediction across transportation modes based on a data interoperability mechanism provided in an embodiment of the present invention;

[0059] Figure 2 A processing flow chart of a method for joint prediction of passenger flow across transportation modes based on a data interoperability mechanism provided by an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of a passenger flow prediction area range provided by an embodiment of the present invention;

[0061] Figure 4 Schematic diagram of the temporal convolution module of the present invention;

[0062] Figure 5 Schematic diagram of the spatial hypergraph convolution module of the present invention;

[0063] Figure 6 The MAE index comparison of the examples of the present invention under different spatiotemporal convolution combinations and stacking conditions;

[0064] Figure 7 The figure shows the RMSE comparison of the examples of the present invention under different spatiotemporal convolution combinations and stacking conditions. DETAILED DESCRIPTION

[0065] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0066] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0067] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0068] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0069] In the context of increasing urban population, large cities give priority to the development of public transportation, which not only alleviates traffic congestion problems, but also actively responds to the country's low-carbon and energy-saving development goals. At present, buses and subways account for the largest proportion of passenger flow in large cities, and they are also the most common public transportation modes. If the passenger flow carried by buses and subways can be fully mined and analyzed, and accurate and efficient passenger flow forecasts can be achieved, it will be beneficial for public transportation operators to better allocate resources and adjust transportation supply.

[0070] The embodiment of the present invention first obtains the traffic network data and passenger flow history data of a certain area. The traffic network data includes the route and station location of subways and buses. The passenger flow history data includes key information such as passengers' entry and exit time, station name, and station number. The data description used in the example of the present invention is shown in Table 1.

[0071] Table 1 Data description

[0072]

[0073] The implementation principle diagram of a method for joint prediction of passenger flow across transportation modes based on a data interoperability mechanism provided by an embodiment of the present invention is as follows: Figure 1 The specific processing flow is as shown in Figure 2 As shown, the processing steps include:

[0074] Step S1: According to the network topology data of the passenger flow prediction area, a single traffic sub-network is constructed based on the hypergraph theory.

[0075] like Figure 3 As shown in the figure, the passenger flow prediction area selected in this embodiment is intuitively displayed, and the traffic network G = {G1, G2, ..., G n ,…,G N},n=1,2,…,N,G n represents a single transportation mode subnetwork, N is the number of types of transportation modes included, and in this embodiment, N=2; based on the network topology data of the passenger flow prediction area, a single transportation subnetwork G is constructed based on the hypergraph theory. n ={V,ε,H}, where V represents G n The node set of

[0076] q=1,2,…,Q n , Q n G n The total number of nodes, v q Represents G n The qth node in G is represented by the qth node in the present embodiment. In this embodiment, the nodes correspond to the subway and bus stations. The number of nodes Q1 in the subway sub-network is 13, and the number of nodes Q2 in the bus sub-network is 128. ε represents G n The set of hyperedges e in , k=1,2,…,K n , K n G n The total number of hyperedges, e k Represents G n The kth hyperedge in H , in this embodiment, the hyperedges correspond to the subway and bus lines, the number of hyperedges K1 of the subway subnetwork is 4, and the number of hyperedges K2 of the bus subnetwork is 45; nRepresents G n The correlation matrix of Represents node v q and super edge e k The subordinate relationship between

[0077]

[0078] Node v in the subway sub-network G1 and the bus sub-network G2 q The node over-degree k(v q ) all satisfy formula (2), and the super edge degree k s (e k ) all satisfy formula (3);

[0079]

[0080] The node excess k(vq) is the node v q The number of hyperedges, all nodes v q The k(vq) forms a diagonal matrix

[0081]

[0082] Super edge super k s (e k ) refers to the hyperedge e k The number of all nodes in the k K s (e k ) forms a diagonal matrix

[0083] Step S2: Based on the topological structure of each single transportation mode sub-network, a cross-modal transportation sub-network is constructed using a data interoperability mechanism.

[0084] Based on the data interoperability mechanism, the data between the subway subnetwork G1 and the bus subnetwork G2 are interconnected and shared, and the cross-modal transportation subnetwork is defined as G 1,2 , G 1,2 The nodes of G1 and G2 are nodes v i and v j In this embodiment, the number of nodes Q1 of G1 is less than the number of nodes Q2 of G2.

[0085] G 1,2 The hyperedge in is a cross-mode association hyperedge, that is, M. l=1,2,…,L 1,2 , L 1,2 G 1,2 The total number of hyperedges associated in the mid-span mode, in this embodiment, L 1,2 =13,m lRepresents G 1,2 The lth hyperedge in G 1,2 The correlation matrix is And Q1<Q2, so v i and m l One-to-one correspondence, H 1,2 Satisfies formula (4).

[0086]

[0087] In G 1,2 Medium i The node over-degree k(v i ) is fixed to 1; v j The node over-degree k(v j ) satisfies formula (5), k(v i ) and k(v j ) together form a diagonal matrix

[0088]

[0089] Super edge super k s (m l ) satisfies formula (6), all hyperedges m l K s (m l ) forms a diagonal matrix

[0090]

[0091] Step S3: Obtain the traffic subnetwork G n Historical passenger flow data of all nodes Then G n The passenger flow data of the qth node in the historical step length U period is u=1,2,…,U;

[0092] Define the traffic network G within the prediction step P period n The future passenger flow of Q nodes is The future passenger flow of the qth node is

[0093] Step S4: construct a multi-mode spatiotemporal hypergraph convolutional network, convert the historical passenger flow data of all nodes on each single traffic sub-network and the traffic network topology structure of the above cross-modal traffic sub-network into the above multi-mode spatiotemporal hypergraph convolutional network, and the multi-mode spatiotemporal hypergraph convolutional network outputs the passenger flow prediction results of each single traffic sub-network.

[0094] Multimodal Spatio-Temporal Hypergraph Convolution Network (MSTHCN) mainly consists of three parts: input module, spatiotemporal convolution module (including temporal convolution and spatial convolution) and output module.

[0095] Step S4.1: Data input

[0096] According to steps 1-3, the obtained traffic subnetwork G n The historical passenger flow data of all nodes on the network are converted into the passenger flow feature matrix X, and the obtained traffic network topology structure of the above cross-modal traffic subnetwork is converted into the association matrix H of the road network.

[0097] Step S4.2: Spatiotemporal convolution

[0098] Based on the data interoperability mechanism, hypergraph convolution is introduced to build a spatiotemporal convolution module. The spatiotemporal convolution module contains 2 layers of temporal convolution and 2 layers of spatial convolution to obtain the spatiotemporal interaction information of passenger flow data.

[0099] Step 4.2.1: Temporal Convolution

[0100] The specific structure of a temporal convolution (TCN) module provided in an embodiment of the present invention is as follows: Figure 4 As shown in the figure, the historical passenger flow feature matrix X is input, the importance of information features is differentially identified based on the gating mechanism, and the causal convolution operation (CasualConv1d) is used to effectively integrate the historical passenger flow feature matrix X.

[0101] Then perform residual calculation and Sigmod activation on the integrated data to obtain the output results A and B. A represents the result of the front C0 channel, and B represents the result of the rear C0 channel.

[0102] The output of the temporal convolution is calculated by equation (7), i.e., the length is UK t +1 passenger flow time series

[0103]

[0104] In the formula, represents the gated convolution kernel, * τ represents the gated convolution operator, K t is the width of the convolution kernel corresponding to the time step. In this embodiment, K t =3, ⊙ represents the Hadamard product, σ sig (·) represents the Sigmod activation function.

[0105] Step 4.2.2: Spatial Convolution

[0106] Based on the Hypergraph Convolution (HGCN) network, the passenger flow characteristics of the traffic network are extracted. The specific structure of a spatial convolution module provided by the embodiment of the present invention is as follows: Figure 5 That is, the spatial features of the traffic subnetworks G1 and G2 are extracted by formula (8), and the cross-mode subnetwork G is extracted by formula (9). 1,2 The spatial features of the node are first aggregated and then distributed in two stages to achieve the feature transfer and distribution of "node→hyperedge→node", and the and

[0107]

[0108] Where n is 1 or 2, W n is the identity matrix, Θ n G n The filter parameters of each node in need to be learned during the training process.

[0109]

[0110] Where, X merge =X 1 +X 2 , X 1 is the passenger flow in G1, X 2 is the passenger flow in G2, W 1,2 is the identity matrix, Θ 1,2 G 1,2 The filter parameters of each node in need to be learned during the training process;

[0111] That is, G obtained by operating formula (8) n The future passenger flow matrix, It can be regarded as the result obtained after the convolution operation according to formula (8), that is,

[0112] That is, G obtained by operating formula (9) 1,2 Future passenger flow matrix. It can be regarded as the result obtained after the convolution operation according to formula (9)

[0113] In order to focus on verifying the impact of the stacking mode of the temporal convolution module and the spatial convolution module on the performance of the prediction method proposed in the present invention, in this embodiment, the control variable method is used. Under the same training configuration, with the goal of predicting the passenger flow in the next 0.5h, 1.0h and 2.0h, three combinations are set, namely, 1 HGCN network sandwiched between 2 TCN networks, 2 HGCN networks sandwiched between 2 TCN networks, and 3 HGCN networks sandwiched between 2 TCN networks. They are regarded as a whole and then stacked 1 time, 2 times or 3 times respectively. There are a total of 18 stacking combinations, and the MAE and RMSE of their prediction results are discussed respectively, as shown in Figure 2. Figure 6 and Figure 7 As shown, the stacking method corresponding to the best performing model is finally adopted (i.e., 2 HGCN networks are sandwiched between 2 TCN networks, stacked twice);

[0114] Step 4.3: Prediction result output

[0115] The output module consists of 1 layer of temporal convolution, 2 layers of fully connected layers and a feature fusion layer.

[0116] Step 4.3.1: Output Mapping

[0117] Mapping the output of the spatiotemporal convolution module into a single-step prediction based on time-gated convolution;

[0118] Step 4.3.2: Fully connected layer

[0119] As shown in formula (10), the fully connected layer is used to obtain nonlinear features and output G1, G2 and G 1,2 The predicted sequence of all nodes

[0120]

[0121] In the formula, ω r is the weight coefficient in the fully connected layer, x r is the value of the rth neuron in the previous layer, b r is the bias of the fully connected layer;

[0122] Step 4.3.3: Spatiotemporal feature fusion

[0123] Due to the spatiotemporal convolution results of the cross-transportation mode subnetwork The influence degree of subway and bus nodes is different, so the passenger flow prediction results of each sub-network node are integrated based on formula (11);

[0124]

[0125] Specifically G i Future passenger flow matrix

[0126] In the formula, and is a learnable parameter matrix, i is 1 or 2, Corresponding to G 1,2 The spatiotemporal convolution result on the i-th transportation mode subnetwork, ⊙ is the Hadamard product; based on the above steps, the prediction results are updated.

[0127] Define the loss function to optimize.

[0128] The prediction results of BSTHCN are defined to be optimized by the following loss function;

[0129]

[0130] Where Y z is the true value, To predict the value, BSTHCN is trained end-to-end based on the error back propagation mechanism by minimizing the mean absolute error between the predicted value and the true value.

[0131] The Adam optimizer is used to train the model, the initial learning rate is set to 0.001, 20 rounds of training are performed, and the batch size is set to 32. For the actual data set, the training set, validation set and test set can be divided into a ratio of 4:1:2, and MAE (Mean Absolute Error) and RMSE (root mean square error) are used as evaluation indicators to quantify the accuracy of the method proposed in the present invention.

[0132] The passenger flow prediction method of the present invention is not limited to buses and subways. The following embodiments or drawings are only used to illustrate the present invention but are not used to limit the scope of the present invention.

[0133] In summary, the present invention reveals the complex relationships hidden in the transportation network. Compared with the traditional graph structure, the introduction of hypergraphs in convolutional networks has significant advantages. Hypergraphs can more effectively capture the internal characteristics of transportation modes, while also mining the similarities of passenger flow patterns at each node and the potential characteristics driven by the functions of each node. This not only improves the ability to extract the internal characteristics of transportation modes, but also strengthens the mining of relationships between transportation modes, further optimizing the accuracy and reliability of passenger flow prediction.

[0134] The present invention fully considers and integrates the cross-modal spatiotemporal correlation between different modes of transportation. Traditional methods only focus on passenger flow prediction within a certain mode of transportation, and fail to characterize the correlation between different modes of transportation. The present invention introduces a hypergraph association matrix, and by incorporating the spatiotemporal characteristics of passenger flow in different transportation networks into the prediction method, it strengthens the grasp of the passenger flow dynamics of each mode of transportation, realizes the capture of overall and local passenger flow characteristics, and finally accurately predicts the passenger flow of each node.

[0135] The present invention significantly enhances the recognition and understanding of spatiotemporal features in complex transportation networks. Traditional neural network methods are difficult to capture both temporal and spatial features at the same time. The present invention stacks and combines time-gated convolution and spatial hypergraph convolution to identify the periodic changes and trends of temporal features while revealing the correlation between different geographic spaces, integrating temporal and spatial features, and further improving the rationality and accuracy of passenger flow prediction.

[0136] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0137] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0138] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0139] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A cross-transportation mode passenger flow joint prediction method based on data interoperability mechanism, characterized in that: include: According to the network topology data of the passenger flow prediction area, a single traffic sub-network is constructed based on the hypergraph theory, and the historical passenger flow data of all nodes on each single traffic sub-network is obtained; Based on the topological structure of each single transportation mode sub-network, a cross-modal transportation sub-network is constructed using a data interoperability mechanism; A multi-mode spatiotemporal hypergraph convolutional network is constructed, and the historical passenger flow data of all nodes on each single transportation sub-network and the transportation network topology structure of the cross-modal transportation sub-network are input into the multi-mode spatiotemporal hypergraph convolutional network. The multi-mode spatiotemporal hypergraph convolutional network outputs the passenger flow prediction results of each single transportation sub-network.

2. The method according to claim 1, characterized in that The single traffic sub-network is constructed based on the network topology data of the passenger flow prediction area based on the hypergraph theory, including: Define the traffic network G = {G1, G2, ..., G n ,…,G N },n=1,2,…,N,G n represents a single transportation mode subnetwork, N is the number of transportation modes, including subway and bus, N = 2; according to the network topology data of the passenger flow prediction area, a single transportation subnetwork G is constructed based on the hypergraph theory n = {V, ε, H}, V represents G n The node set of q=1,2,…,Q n , Q n G n The total number of nodes, v q Represents G n The qth node in the network corresponds to the subway and bus stations. The number of nodes in the subway sub-network is set to Q1, and the number of nodes in the bus sub-network is set to Q2. ε represents G n The set of hyperedges e in , k=1,2,…,K n , K n G n The total number of hyperedges, e k Represents G n The kth hyperedge in H corresponds to the subway and bus lines. n Represents G n The correlation matrix of Represents node v q and super edge e k The subordinate relationship between them; Node v in the subway sub-network G1 and the bus sub-network G2 q The node over-degree k(v q ) all satisfy formula (2), and the super edge degree k s (e k ) all satisfy formula (3); The node excess k(vq) is the node v q The number of hyperedges, all nodes v q The k(vq) forms a diagonal matrix Super edge super k s (e k ) refers to the hyperedge e k The number of all nodes in the k K s (e k ) forms a diagonal matrix 3. The method according to claim 2, characterized in that The acquisition of historical passenger flow data of all nodes on each single traffic sub-network includes: Get a single traffic subnetwork G n Historical passenger flow data of all nodes Then G n The passenger flow data of the qth node in the historical step length U period is u=1,2,…,U; Predict a single traffic subnetwork G within a time period of step length P n The future passenger flow of Q nodes is: The future passenger flow of the qth node is 4. The method according to claim 3, characterized in that The topological structure based on each single transportation mode sub-network utilizes a data interoperability mechanism to construct a cross-modal transportation sub-network, including: Based on the data interoperability mechanism, the data between the subway subnetwork G1 and the bus subnetwork G2 are interconnected and shared, and the cross-modal transportation subnetwork is defined as G 1,2 , G 1,2 The nodes of G1 and G2 are nodes v i and v j composition; G 1,2 The hyperedge in is a cross-mode association hyperedge, that is, M. l=1,2,…,L 1,2 , L 1,2 G 1,2 The total number of hyperedges associated in the mid-span mode, m l Represents G 1,2 The lth hyperedge in G 1,2 The correlation matrix is And Q1<Q2, v i and m l One-to-one correspondence, H 1,2 Satisfy formula (4); In G 1,2 Medium i The node over-degree k(v i ) is fixed to 1; v j The node over-degree k(v j ) satisfies formula (5), k(v i ) and k(v j ) together form a diagonal matrix Super edge super k s (m l ) satisfies formula (6), all hyperedges m l K s (m l ) forms a diagonal matrix 。 5. The method according to claim 4, characterized in that The method for constructing a multi-mode spatiotemporal hypergraph convolutional network includes: constructing a multi-mode spatiotemporal hypergraph convolutional network using an input module, an output module and the spatiotemporal convolutional module, constructing a spatiotemporal convolutional module using hypergraph convolution based on a data interoperability mechanism, wherein the spatiotemporal convolutional module includes 2 layers of temporal convolution and 2 layers of spatial convolution, and the output module consists of 1 layer of temporal convolution, 2 layers of fully connected layers and a feature fusion layer.

6. The method according to claim 5, characterized in that The historical passenger flow data of all nodes on each single traffic sub-network and the traffic network topology structure of the cross-modal traffic sub-network are input into the multi-mode spatiotemporal hypergraph convolutional network, and the multi-mode spatiotemporal hypergraph convolutional network outputs the passenger flow prediction results of each single traffic sub-network, including: The traffic subnetwork G n The historical passenger flow data of all nodes on the network are converted into a passenger flow feature matrix X, and the traffic network topology structure of the cross-modal traffic subnetwork is converted into a road network association matrix H; The passenger flow feature matrix X and the association matrix H of the road network are input into the input module of the multi-mode spatiotemporal hypergraph convolutional network. The spatiotemporal convolutional module integrates the historical passenger flow feature matrix X using a causal convolution operation, and then performs residual calculation and Sigmod activation on the integrated data to obtain output results A and B, where A represents the result of the front C0 channel and B represents the result of the rear C0 channel. The output of the temporal convolution is calculated by equation (7), i.e., the length is UK t +1 passenger flow time series In the formula, represents the gated convolution kernel, * τ represents the gated convolution operator, K t is the width of the convolution kernel corresponding to the time step, ⊙ represents the Hadamard product, σ sig (·) represents the Sigmod activation function; The spatial characteristics of traffic subnetworks G1 and G2 are extracted through formula (8); Where n is 1 or 2, W n is the identity matrix, Θ n G n The filter parameters of each node in X need to be learned during the training process; n G n The historical passenger flow feature matrix, D v G n The node excess matrix, H n G n The correlation matrix, D e G n The hyperedge hyperdegree matrix of ; The cross-modal sub-network G is extracted by formula (9) 1,2 spatial characteristics; Where, X merge =X 1 +X 2 , X 1 is the passenger flow in G1, X 2 is the passenger flow in G2, W i,j is the identity matrix, Θ 1,2 G 1,2 The filter parameters of each node in need to be learned during the training process; G 1,2 The node excess matrix, H i,j G 1,2 The correlation matrix, D i,j G 1,2 The hyperedge hyperdegree matrix of ; Through the two-stage process of feature aggregation and then distribution, we can obtain and That is, G obtained by operating formula (8) n The future passenger flow matrix, That is, G obtained by operating formula (9) 1,2 The future passenger flow matrix, The output module maps the output of the spatiotemporal convolution module to a single-step prediction based on the time-gated convolution. The output module obtains nonlinear features through the fully connected layer and outputs G1, G2 and G 1,2 The passenger flow prediction sequence of all nodes As shown in formula (10): In the formula, ω r is the weight coefficient in the fully connected layer, x r is the value of the rth neuron in the previous layer, b r is the bias of the fully connected layer; f refers to the calculation formula of the fully connected layer, and Y refers to the passenger flow prediction results of each node on G1 and G2; Use formula (11) to fuse the passenger flow prediction results of each sub-network node; Specifically G i Future passenger flow matrix In the formula, and is a learnable parameter matrix, i is 1 or 2, Corresponding to G 1,2 The result of the spatiotemporal convolution on the ith transportation mode subnetwork, ⊙ is the Hadamard product; Define the prediction results to be optimized and updated through the following loss function; Where Y z is the true value, To predict the value, the model is trained end-to-end based on the error back propagation mechanism by minimizing the mean absolute error between the predicted value and the true value, and the prediction result is updated

Images