Traffic flow prediction method, device, electronic device and storage medium

Abstract

The present invention discloses a traffic flow prediction method, device, electronic device, and storage medium, belonging to the technical field of traffic prediction. The traffic flow prediction method of the present invention includes obtaining a traffic spatiotemporal probability map; parameterizing the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; extracting features from the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedded feature vector and a dynamic embedded feature vector; mapping the global embedded feature vector and the dynamic embedded feature vector to a preset vector space to obtain inference parameters; constructing a spatiotemporal sequence network based on a preset recursive graph convolution module and inference parameters; and predicting the current traffic flow using the spatiotemporal sequence network to obtain predicted traffic data. This traffic flow prediction method can improve the accuracy of traffic flow prediction in time and space.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic prediction, and in particular to a traffic flow prediction method, device, electronic equipment and storage medium. Background Art

[0002] At present, neural network models are often used to predict traffic flow. However, the neural network models in related technologies often have certain limitations when predicting traffic flow in both time and space dimensions, which affects the prediction accuracy. Therefore, how to provide a traffic flow prediction method to achieve accurate prediction of traffic flow has become an urgent problem to be solved. Summary of the Invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a traffic flow prediction method that can improve the accuracy of traffic flow prediction in time and space.

[0004] The present invention also provides a traffic flow prediction device having the above traffic flow prediction method.

[0005] The present invention also provides an electronic device having the above traffic flow prediction method.

[0006] The present invention also provides a computer-readable storage medium.

[0007] A traffic flow prediction method according to a first embodiment of the present invention includes:

[0008] Obtain traffic spatiotemporal probability map;

[0009] performing parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters;

[0010] Using a preset feature extraction network to extract features from the traffic spatiotemporal parameters, and obtaining a global embedding feature vector and a dynamic embedding feature vector;

[0011] Mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters;

[0012] Constructing a spatiotemporal sequence network according to a preset recursive graph convolution module and the inference parameters;

[0013] The spatiotemporal sequence network is used to predict the current traffic flow to obtain predicted traffic data.

[0014] The traffic flow prediction method according to an embodiment of the present invention has at least the following beneficial effects: This traffic flow prediction method obtains a traffic spatiotemporal probability map; parameterizes the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; uses a preset feature extraction network to extract features from the traffic spatiotemporal parameters to obtain a global embedding feature vector and a dynamic embedding feature vector; maps the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters; constructs a spatiotemporal sequence network based on a preset recursive graph convolution module and inference parameters; and uses the spatiotemporal sequence network to predict current traffic flow to obtain predicted traffic data. This method, through the spatiotemporal sequence network, can predict traffic flow in both the temporal and spatial dimensions, meeting prediction requirements and thereby improving prediction accuracy.

[0015] According to some embodiments of the present invention, obtaining a traffic spatiotemporal probability map includes:

[0016] Obtaining spatiotemporal traffic data;

[0017] A traffic spatiotemporal probability map is generated based on the traffic spatiotemporal data.

[0018] According to some embodiments of the present invention, extracting features from the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector includes:

[0019] Performing convolution processing on the traffic spatiotemporal parameters in a preset dimension to obtain initial convolution features;

[0020] Performing dimension conversion processing on the initial convolution feature to obtain a target convolution feature;

[0021] A feature extractor is used to extract features from the target convolution features to obtain a global embedding feature vector and a dynamic embedding feature vector.

[0022] According to some embodiments of the present invention, extracting the target convolution feature using a feature extractor to obtain a global embedding feature vector and a dynamic embedding feature vector includes:

[0023] Using a first feature extractor to extract the target convolution features in a preset first sequence set to obtain a global embedding feature vector;

[0024] A second feature extractor is used to extract the target convolution features in a preset second sequence set to obtain a dynamic embedding feature vector.

[0025] According to some embodiments of the present invention, mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters includes:

[0026] Mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain a feature vector pair;

[0027] Perform edge prediction processing on the feature vector pair to obtain the inference parameter.

[0028] According to some embodiments of the present invention, constructing a spatiotemporal sequence network based on a preset recursive graph convolution module and the inference parameters includes:

[0029] The graph structure of the spatiotemporal sequence network is regularized using a graph Laplace algorithm.

[0030] According to some embodiments of the present invention, constructing a spatiotemporal sequence network based on a preset recursive graph convolution module and the inference parameters includes:

[0031] Calculating the loss value of the spatiotemporal sequence network using the Huber function;

[0032] The spatiotemporal sequence network is optimized according to the loss value to update the spatiotemporal sequence network.

[0033] According to a second embodiment of the present invention, a traffic flow prediction device includes:

[0034] Traffic space-time probability map acquisition module, used to obtain traffic space-time probability map;

[0035] A parameterization processing module, configured to perform parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters;

[0036] A feature extraction module, configured to extract features from the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector;

[0037] A mapping processing module, configured to map the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters;

[0038] A spatiotemporal sequence network construction module, configured to construct a spatiotemporal sequence network based on a preset recursive graph convolution module and the inference parameters;

[0039] The traffic flow prediction module is used to predict the current traffic flow using the spatiotemporal sequence network to obtain predicted traffic data.

[0040] The traffic flow prediction device according to an embodiment of the present invention has at least the following beneficial effects: the traffic flow prediction device obtains a traffic spatiotemporal probability map through a traffic spatiotemporal probability map acquisition module; the parameterization processing module performs parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; the feature extraction module uses a preset feature extraction network to extract features from the traffic spatiotemporal parameters to obtain a global embedding feature vector and a dynamic embedding feature vector; the mapping processing module maps the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters; the spatiotemporal sequence network construction module constructs a spatiotemporal sequence network based on a preset recursive graph convolution module and inference parameters; and the traffic flow prediction module uses the spatiotemporal sequence network to predict the current traffic flow and obtain predicted traffic data. The device can predict traffic flow in both the temporal and spatial dimensions through the spatiotemporal sequence network, meeting prediction requirements and thereby improving prediction accuracy.

[0041] An electronic device according to an embodiment of a third aspect of the present invention includes:

[0042] at least one processor, and

[0043] a memory communicatively connected to the at least one processor; wherein,

[0044] The memory stores instructions, and the instructions are executed by the at least one processor so that the at least one processor implements the traffic flow prediction method as described in the embodiment of the first aspect when executing the instructions.

[0045] The electronic device according to an embodiment of the present invention has at least the following beneficial effects: the electronic device uses the above-mentioned traffic flow prediction method to obtain a traffic spatiotemporal probability map; parameterizes the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; uses a preset feature extraction network to extract features from the traffic spatiotemporal parameters to obtain a global embedding feature vector and a dynamic embedding feature vector; maps the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters; constructs a spatiotemporal sequence network based on a preset recursive graph convolution module and inference parameters; and uses the spatiotemporal sequence network to predict the current traffic flow to obtain predicted traffic data. This method can predict traffic flow in both the temporal and spatial dimensions through the spatiotemporal sequence network, meeting prediction requirements and thereby improving prediction accuracy.

[0046] According to a computer-readable storage medium of an embodiment of the fourth aspect of the present invention, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the traffic flow prediction method as described in the embodiment of the first aspect.

[0047] The computer-readable storage medium according to an embodiment of the present invention has at least the following beneficial effects: the computer-readable storage medium executes the above-mentioned traffic flow prediction method by obtaining a traffic spatiotemporal probability map; parameterizing the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; extracting features from the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector; mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters; constructing a spatiotemporal sequence network based on a preset recursive graph convolution module and inference parameters; and predicting the current traffic flow using the spatiotemporal sequence network to obtain predicted traffic data. This method can predict traffic flow in both the temporal and spatial dimensions through the spatiotemporal sequence network, meeting prediction requirements and thereby improving prediction accuracy.

[0048] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0050] Figure 1 is a flow chart of a traffic flow prediction method according to an embodiment of the present invention;

[0051] Figure 2 for Figure 1 Flowchart of step S101 in FIG.

[0052] Figure 3 for Figure 1 Flowchart of step S103;

[0053] Figure 4 for Figure 1 Flowchart of step S104;

[0054] Figure 5 for Figure 1 Flowchart of step S105;

[0055] Figure 6 Schematic diagram of the structure of a traffic flow prediction device according to an embodiment of the present invention.

[0056] Reference numerals: 701, traffic spatiotemporal probability map acquisition module; 702, parameterization processing module; 703, feature extraction module; 704, mapping processing module; 705, spatiotemporal sequence network construction module; 706, traffic flow prediction module. DETAILED DESCRIPTION

[0057] The following describes embodiments of the present invention in detail. Examples of the embodiments 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 intended only to explain the present invention and are not to be construed as limiting the present invention.

[0058] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention.

[0059] In the description of the present invention, "several" means more than one, "plurality" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0060] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0061] In the description of the present invention, reference to terms such as "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the exemplary expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0062] First, refer to Figure 1 The traffic flow prediction method of the embodiment of the present invention includes:

[0063] S101, obtaining a traffic spatiotemporal probability map;

[0064] S102, parameterizing the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters;

[0065] S103, using a preset feature extraction network to extract features from traffic spatiotemporal parameters to obtain a global embedding feature vector and a dynamic embedding feature vector;

[0066] S104, mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters;

[0067] S105, constructing a spatiotemporal sequence network according to the preset recursive graph convolution module and inference parameters;

[0068] S106, using the spatiotemporal sequence network to predict the current traffic flow and obtain predicted traffic data.

[0069] The process of traffic flow prediction first requires obtaining a spatiotemporal traffic probability map. This map can be obtained by obtaining spatiotemporal traffic data and performing probability calculations on the spatiotemporal traffic data using, for example, a softmax function. The spatiotemporal traffic probability map is then parameterized, converting the graphical information into spatiotemporal traffic parameters. A pre-defined feature extraction network is then used to extract features from the spatiotemporal traffic parameters, generating global embedding feature vectors and dynamic embedding feature vectors. This feature extraction network, comprising an input layer, a convolutional layer, and a fully connected layer, extracts features from the spatiotemporal traffic parameters on different sequence sets, generating global embedding feature vectors and dynamic embedding feature vectors, respectively. The global embedding feature vectors and dynamic embedding feature vectors are then mapped to a pre-defined vector space to obtain feature vector pairs. Edge prediction is then performed on these feature pairs to obtain inference parameters. Using the pre-defined recursive graph convolution module and inference parameters, the spatiotemporal dependencies of traffic flow are simulated, resulting in a spatiotemporal sequence network. Finally, the constructed spatiotemporal sequence network is used to predict the current traffic flow, generating predicted traffic data. This method can predict traffic flow in both time and space dimensions through a spatiotemporal sequence network, meeting prediction needs and thus improving prediction accuracy.

[0070] Furthermore, when simulating the spatiotemporal dependency of traffic flow based on the preset recursive graph convolution module and inference parameters, the spatiotemporal dependency relationship constructed can be expressed as follows:

[0071] R t =sigmoid(W R *[X t ‖H t-1 ]+b R );

[0072] C t =tanh(W C *[X t ‖R t &H t-1 ]+b C );

[0073] U t =sigmoid(W U *[X t ‖H t-1 ]+b U );

[0074] H t =U t &H t-1 +(1-U t )&C t ;

[0075] Among them, Xt is the input data, Rt represents the reset gate signal, Ut represents the update gate signal, and the reset gate signal and update gate signal are calculated by the sigmoid function respectively. W represents the learnable parameter, b is the constant parameter, and & represents the dot product. Ct represents the activation gate signal. The parameters in the network can be scaled to the range of -1 to 1 by the tanh function. Finally, according to the update gate signal Ut and the reset gate signal Rt, as well as the hidden state input H of the previous state t-1 Combine them to get the hidden state output Ht at the current time.

[0076] Furthermore, the input data can be convolved by matrix multiplication, namely:

[0077]

[0078] Among them, Y is the input data, W Q and W k is a learnable variable, k represents the order of matrix multiplication, A is the adjacency matrix, A T is the transpose of A, D I and D O They represent the in-degree matrix and out-degree matrix derived from A respectively.

[0079] Reference Figure 2 In some embodiments, step S101 includes:

[0080] S201, obtaining traffic spatiotemporal data;

[0081] S202: Generate a traffic spatiotemporal probability map based on the traffic spatiotemporal data.

[0082] In the process of predicting traffic flow, we first obtain traffic spatiotemporal data, use the softmax function to perform probability calculation on the traffic spatiotemporal data, and use the softmax function to create a probability distribution of traffic spatiotemporal data on different classification labels, so that we can generate a traffic spatiotemporal probability map based on this probability distribution.

[0083] Reference Figure 3In some embodiments, step S103 includes:

[0084] S301, performing convolution processing on the traffic spatiotemporal parameters in a preset dimension to obtain initial convolution features;

[0085] S302, performing dimension conversion processing on the initial convolution feature to obtain the target convolution feature;

[0086] S303: Use a feature extractor to extract the target convolution features to obtain a global embedding feature vector and a dynamic embedding feature vector.

[0087] When extracting features from traffic spatiotemporal parameters using a preset feature extraction network, the feature extraction network includes an input layer, a convolutional layer, a fully connected layer, and so on. The convolutional layer can convolute the input traffic spatiotemporal parameters along the time dimension to obtain initial convolutional features. The initial convolutional features are then subjected to dimensionality reduction processing so that the initial convolutional features and the participating convolutional features have the same dimension, thereby obtaining target convolutional features. Furthermore, the feature extractors in the feature extraction network are used to extract features from the target convolutional features on different sequence sets to obtain global embedded feature vectors and dynamic embedded feature vectors. This feature extraction network makes it easier to extract features from traffic spatiotemporal parameters, eliminate parameters with low correlation, and more accurately obtain feature vectors with high correlation, thereby improving prediction accuracy.

[0088] In some embodiments, step S303 includes:

[0089] Using a first feature extractor to extract target convolution features in a preset first sequence set to obtain a global embedding feature vector;

[0090] The second feature extractor is used to extract the target convolution features in a preset second sequence set to obtain a dynamic embedding feature vector.

[0091] In order to improve the accuracy of feature extraction, the feature extraction network includes multiple feature extractors, for example, the feature extraction network includes a first feature extractor and a second feature extractor. The first feature extractor extracts the target convolution feature in a preset first sequence set to obtain a global embedding vector z i,g The preset first sequence set is the entire sequence set currently obtained. The second feature extractor extracts the target convolution feature in the preset second sequence set to obtain the dynamic embedding feature vector z i,d . Among them, the preset second sequence set is the preset input sequence set By extracting different feature vectors separately through multiple feature extractors, the accuracy of feature extraction can be improved.

[0092] Reference Figure 4 In some embodiments, step S104 includes:

[0093] S401, mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain a feature vector pair;

[0094] S402: Perform edge prediction processing on the feature vector pairs to obtain inference parameters.

[0095] When obtaining inference parameters, we can first embed the global feature vector z i,g and the dynamic embedding feature vector z i,d Mapped to the preset vector space, the eigenvector pair (z i ,z j ), and then the feature vector pair is processed by the edge prediction model to obtain the inference parameter θ ij and T; wherein the edge prediction model includes a first function P(T)=N τ (μ(z i ,z j ),σ 2 (z i ,z j )) and the second function P(T)=N τ (μ(z i ,z j ),σ 2 (z i ,z j )), where μ(.) and σ 2 (.) is the attention-based hidden layer. The two have a shared input layer and hidden layer. N τ This method can be used to easily obtain the required inference parameters based on the global embedding feature vector and the dynamic embedding feature vector, thereby improving the prediction efficiency.

[0096] In some embodiments, step S105 includes:

[0097] The graph structure of spatiotemporal sequence networks is regularized using the graph Laplace algorithm.

[0098] In order to improve the prediction performance of the spatiotemporal sequence network, the graph Laplace algorithm can also be used to regularize the graph structure of the spatiotemporal sequence network. Specifically, the graph structure can be regularized from the perspective of graph smoothness. For example, the stationarity of the spatiotemporal sequence X on G can be expressed using the quadratic form of the graph Laplace as:

[0099]

[0100] Among them, A ij∈A, i~j means traversing all combinations of i and j, A represents the adjacent matrix, X T is the transpose of the input data X, x i and xj represent the values ​​corresponding to point i and point j respectively.

[0101] Reference Figure 5 In some embodiments, step S105 includes:

[0102] S601, using Huber function to calculate the loss value of the spatiotemporal sequence network;

[0103] S602: Optimize the spatiotemporal sequence network according to the loss value to update the spatiotemporal sequence network.

[0104] In order to improve the prediction performance of the spatiotemporal sequence network, it is also necessary to optimize the spatiotemporal sequence network. Specifically, the loss value of the spatiotemporal sequence network can be calculated using the Huber function, where the Huber function can be expressed as:

[0105]

[0106] Where τ is the time interval, T is the total time, X is the input matrix, is the adjacency matrix, δ is the sensitivity;

[0107] The Huber function can be used to calculate the loss value by traversing all points N and time intervals τ, and then dividing it by the total time and total intervals. This can easily identify outliers and extreme values ​​in the spatiotemporal sequence network, filter this series of interference data, and effectively improve the performance of the spatiotemporal sequence network. At the same time, by backpropagating the loss value and comparing the loss value with the preset threshold, the spatiotemporal sequence network is continuously optimized, which can achieve updates to the spatiotemporal sequence network, improve its prediction performance, and thus improve the accuracy of traffic flow prediction.

[0108] Secondly, refer to Figure 6 , the traffic flow prediction device of an embodiment of the present invention includes:

[0109] Traffic spatiotemporal probability map acquisition module 701, used to acquire a traffic spatiotemporal probability map;

[0110] The parameterization processing module 702 is used to perform parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters;

[0111] A feature extraction module 703 is used to extract features of traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector;

[0112] A mapping processing module 704 is used to map the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain inference parameters;

[0113] A spatiotemporal sequence network construction module 705 is used to construct a spatiotemporal sequence network according to a preset recursive graph convolution module and inference parameters;

[0114] The traffic flow prediction module 706 is used to predict the current traffic flow using a spatiotemporal sequence network to obtain predicted traffic data.

[0115] The specific implementation of the traffic flow prediction device is basically the same as the specific embodiment of the above-mentioned traffic flow prediction method, and will not be repeated here.

[0116] In a third aspect, an electronic device according to an embodiment of the present invention includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions, and the instructions are executed by the at least one processor so that the at least one processor implements the traffic flow prediction method according to the first aspect embodiment when executing the instructions.

[0117] In a fourth aspect, the present invention further provides a computer-readable storage medium having computer-executable instructions stored therein, the computer-executable instructions being used to cause a computer to execute the traffic flow prediction method according to the first embodiment.

[0118] While the embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to the embodiments described above. Various modifications may be made within the scope of knowledge possessed by a person skilled in the art without departing from the spirit of the present invention. Furthermore, the embodiments of the present invention and the features thereof may be combined with one another unless there is a conflict.

Claims

1. Traffic flow prediction method, characterized in that: include: Obtain traffic spatiotemporal probability map; performing parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; Using a preset feature extraction network to extract features from the traffic spatiotemporal parameters, and obtaining a global embedding feature vector and a dynamic embedding feature vector; Mapping the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain a feature vector pair; Performing edge prediction processing on the feature vector pair to obtain inference parameters; Constructing a spatiotemporal sequence network according to a preset recursive graph convolution module and the inference parameters; Using the spatiotemporal sequence network to predict current traffic flow, and obtain predicted traffic data; The step of constructing a spatiotemporal sequence network based on the preset recursive graph convolution module and the inference parameters includes: Calculating the loss value of the spatiotemporal sequence network using the Huber function; The spatiotemporal sequence network is optimized according to the loss value to update the spatiotemporal sequence network.

2. The traffic flow prediction method according to claim 1, characterized in that: The obtaining of the traffic spatiotemporal probability map includes: Obtaining spatiotemporal traffic data; A traffic spatiotemporal probability map is generated based on the traffic spatiotemporal data.

3. The traffic flow prediction method according to claim 1, characterized in that: The method of extracting features of the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector includes: Performing convolution processing on the traffic spatiotemporal parameters in a preset dimension to obtain initial convolution features; Performing dimension conversion processing on the initial convolution feature to obtain a target convolution feature; A feature extractor is used to extract features from the target convolution features to obtain a global embedding feature vector and a dynamic embedding feature vector.

4. The traffic flow prediction method according to claim 3, characterized in that: The method of extracting the target convolution feature by using a feature extractor to obtain a global embedding feature vector and a dynamic embedding feature vector includes: Using a first feature extractor to extract the target convolution features in a preset first sequence set to obtain a global embedding feature vector; A second feature extractor is used to extract the target convolution features in a preset second sequence set to obtain a dynamic embedding feature vector.

5. The traffic flow prediction method according to any one of claims 1 to 4, characterized in that: The step of constructing a spatiotemporal sequence network based on a preset recursive graph convolution module and the inference parameters includes: The graph structure of the spatiotemporal sequence network is regularized using a graph Laplace algorithm.

6. Traffic flow prediction device, characterized in that include: Traffic space-time probability map acquisition module, used to obtain traffic space-time probability map; A parameterization processing module, configured to perform parameterization processing on the traffic spatiotemporal probability map to obtain traffic spatiotemporal parameters; A feature extraction module, configured to extract features from the traffic spatiotemporal parameters using a preset feature extraction network to obtain a global embedding feature vector and a dynamic embedding feature vector; A mapping processing module, configured to map the global embedding feature vector and the dynamic embedding feature vector to a preset vector space to obtain a feature vector pair; and perform edge prediction processing on the feature vector pair to obtain an inference parameter; A spatiotemporal sequence network construction module, configured to construct a spatiotemporal sequence network based on a preset recursive graph convolution module and the inference parameters; wherein the construction of the spatiotemporal sequence network based on the preset recursive graph convolution module and the inference parameters includes: calculating a loss value of the spatiotemporal sequence network using a Huber function; and optimizing the spatiotemporal sequence network based on the loss value to update the spatiotemporal sequence network; The traffic flow prediction module is used to predict the current traffic flow using the spatiotemporal sequence network to obtain predicted traffic data.

7. An electronic device, characterized in that include: at least one processor, and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions, and the instructions are executed by the at least one processor, so that the at least one processor implements the traffic flow prediction method according to any one of claims 1 to 5 when executing the instructions.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the traffic flow prediction method according to any one of claims 1 to 5.

Images