Method for traffic prediction by using time domain decomposition adaptive graph neural network model and related equipment
Adaptive learning of graph structures is solved through the time domain decomposition adaptive graph neural network model, which solves the problem that predefined graph structures in the prior art are difficult to express complex traffic information, and achieves more accurate traffic prediction results.
Patent Information
- Application Number
- CN202411264716.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2025-01-03
AI Technical Summary
Existing graph neural networks use predefined graph structures in traffic prediction to accurately express complex traffic information, resulting in large errors in prediction results.
The adaptive graph neural network model of time domain decomposition is used to learn the graph structure from the macro and micro perspectives. The optimal graph structure is obtained through the time domain decomposition matrix and learnable parameters optimization, and the space-time graph neural network is input for training to improve prediction. accuracy.
Through the learning and fusion of adaptive graph structures, the accuracy of traffic prediction is significantly improved. The experimental results perform well on multiple data sets, especially when traffic flow changes dramatically, more accurate predictions can be made.
Smart Images
Figure CN119250264A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a method and related equipment for traffic prediction using a time-domain decomposition adaptive graph neural network model. Background Art
[0002] Traditional traffic prediction methods usually rely on statistical methods and machine learning models, but these methods often perform poorly when dealing with complex traffic networks and dynamic traffic conditions because traffic data exhibits complex patterns and high nonlinearity. Graph Neural Network (GNN) is a type of deep learning model specifically designed to process graph structured data. In traffic prediction, traffic networks can be naturally represented as graph structures, and the use of graph neural networks for traffic prediction has emerged. GNN captures the complex relationships between nodes and edges by passing messages on the graph structure, which can more effectively model the spatial dependencies and temporal dynamics in traffic networks.
[0003] Although graph neural networks have made great progress in the field of traffic prediction, there are still many problems. Most existing graph neural networks use a predefined graph structure determined by distance measurements. However, the predefined graph structure is not sufficient to contain all the useful information in complex traffic prediction scenarios. Therefore, it is difficult to accurately predict traffic data that is always changing dynamically using this unchanging graph structure. Summary of the invention
[0004] In order to solve the problem that the predefined graph structure is difficult to express complete traffic information during traffic prediction, resulting in large errors in prediction results, the present invention provides a time-domain decomposition adaptive spatiotemporal graph neural network for traffic prediction, attempts to learn the graph structure from both macro and micro perspectives, and uses an adaptive graph learning method to improve the graph structure so that it can more completely express the implicit relationship in the traffic pattern, thereby obtaining accurate prediction results.
[0005] In a first aspect, the present invention provides a method for traffic prediction using a time-domain decomposition adaptive graph neural network model, comprising:
[0006] Step 1: Obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range;
[0007] Step 2: Decompose the historical data in the time domain to construct a time domain decomposition matrix;
[0008] Step 3: Using the historical data to learn the current predefined graph structure to obtain a macro graph structure;
[0009] Step 4: Select the time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters and obtain the micrograph structure;
[0010] Step 5: Fusing the macro graph structure and the micro graph structure to obtain an optimal graph structure;
[0011] Step 6: Input the optimal graph structure into the spatiotemporal graph neural network for training;
[0012] Step 7: Repeat steps 4 to 6 until the stopping condition is met to obtain a trained spatiotemporal graph neural network model;
[0013] Step 8: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction results for the future time period to be predicted.
[0014] Furthermore, step 2 specifically includes:
[0015] For each traffic node, the traffic speed data within the specified historical time range is first divided according to the set time interval to obtain the traffic speed data of several time periods; the time period is set, and then the traffic speed data of the same time period in each time period are stacked; then, the super empirical mode decomposition is applied to the stacked data of each specific time period to obtain the intrinsic mode function of the traffic node in each specific time period;
[0016] For each specific time period, a node relationship matrix of each specific time period is constructed according to the intrinsic mode functions corresponding to each traffic node; wherein each element in the node relationship matrix represents the similarity between two traffic nodes at corresponding positions;
[0017] The node relationship matrices of all specific time periods together constitute the time domain decomposition matrix.
[0018] Furthermore, for each specific time period, a node relationship matrix for each specific time period is constructed according to the intrinsic mode functions corresponding to each traffic node, specifically including:
[0019] For each specific time period, the Pearson correlation coefficients of the two traffic nodes under the same eigenmode are first calculated; then the Pearson correlation coefficients of the two traffic nodes under all eigenmodes are integrated to finally obtain the similarity of the two traffic nodes.
[0020] Furthermore, the Pearson correlation coefficients of the two traffic nodes under all eigenmodes are integrated, specifically including: averaging the Pearson correlation coefficients under all eigenmodes using the following formula;
[0021]
[0022] in, represents the similarity between traffic node A and traffic node B in the characteristic time period t, represents the Pearson correlation coefficient between traffic node A and traffic node B under the i-th eigenmode, and n represents the number of eigenmodes contained in the eigenmode function corresponding to each traffic node.
[0023] Furthermore, step 4 specifically includes:
[0024] A Mi =W t ×Norm(X t )
[0025] Among them, W t represents the current learnable parameters corresponding to a specific time period t, X t represents the time domain decomposition matrix corresponding to a specific time period t, Norm represents the normalization operation, A Mi Represents the microscopic graph structure obtained after optimization.
[0026] Furthermore, step 5 specifically includes:
[0027] A * =Norm(ReLU(A Ma +W t *A Mi ))
[0028] Among them, ReLU represents the activation function, A Ma represents the macrograph structure, A * Represents the optimal graph structure after fusion.
[0029] Furthermore, the spatiotemporal graph neural network includes an input layer, L spatiotemporal blocks connected sequentially by residuals, and an output layer; wherein each of the spatiotemporal blocks includes a temporal convolution layer and a graph convolution layer connected sequentially;
[0030] Correspondingly, the optimal graph structure is input into the spatiotemporal graph neural network for training, specifically including: inputting the optimal graph structure into each of the graph convolution layers in the spatiotemporal graph neural network respectively.
[0031] In a second aspect, the present invention provides a system for traffic prediction using a time-domain decomposition adaptive graph neural network, comprising:
[0032] A data collection module, used to obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range;
[0033] A time domain decomposition module, used for performing time domain decomposition on the historical data to construct a time domain decomposition matrix;
[0034] A graph structure learning module is used to learn the current predefined graph structure using the historical data to obtain a macro graph structure; select a time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain a micro graph structure; fuse the macro graph structure and the micro graph structure to obtain an optimal graph structure; and input the optimal graph structure into a spatiotemporal graph neural network for training;
[0035] The prediction module is used to input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction results for the future time period to be predicted.
[0036] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.
[0037] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.
[0038] Beneficial effects of the present invention:
[0039] The present invention uses a time-domain decomposition adaptive graph neural network model for traffic prediction, and adaptively learns the graph structure from both macro and micro perspectives. The macro perspective is directly optimized through data, and the micro perspective is learned using the time-domain decomposition method. The graph structures learned from the two perspectives are then fused to obtain a graph structure that can better express real road network information, thereby improving the accuracy of traffic prediction. The present invention is meticulously designed, fully making up for the shortcomings of predefined graph structures in traditional traffic prediction methods, and the prediction results are accurate. Experiments have shown that the present invention performs well on both data sets. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 A framework diagram of a method for traffic prediction using a time-domain decomposition adaptive graph neural network model provided by an embodiment of the present invention;
[0041] Figure 2 Details of constructing the time domain decomposition matrix provided in the embodiment of the present invention;
[0042] Figure 3 A network framework for graph structure learning provided by an embodiment of the present invention;
[0043] Figure 4 Comparison of the prediction results of the time-domain decomposition adaptive graph neural network traffic prediction model and other algorithms provided in the embodiment of the present invention with the accuracy of the actual environmental road conditions;
[0044] Figure 5 A system framework diagram for traffic prediction using a time-domain decomposition adaptive graph neural network model provided by an embodiment of the present invention;
[0045] Figure 6 A framework diagram of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0047] like Figure 1 As shown, a method for traffic prediction using a time-domain decomposition adaptive graph neural network model provided by an embodiment of the present invention includes the following steps:
[0048] S101: Acquire historical data of a target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range, and may also include location information of each traffic node.
[0049] Specifically, this embodiment downloaded two real traffic data sets, PEMS-BAY and METR-LA, from the official website of the Caltrans Performance Measurement System as historical data of the Bay Area road network and historical data of the Los Angeles County road network, respectively. Among them, PEMS-BAY contains 6 months of traffic speed data on 325 sensors in the Bay Area; METR-LA records 4 months of traffic speed data on 207 sensors on Los Angeles County highways. These traffic data are aggregated into 5-minute intervals, with 12 time steps per hour and 288 time steps per day.
[0050] It is understandable that in order to facilitate the subsequent use of historical data for model training, it is generally necessary to preprocess the historical data, including: adding time labels and week labels to each piece of data; and then dividing the data set into 70% training set, 20% test set, and 10% validation set ratios.
[0051] S102: Decomposing the historical data in the time domain to construct a time domain decomposition matrix;
[0052] Specifically, taking the PEMS-BAY data set as an example, this embodiment sets 1 hour as the time interval, that is, for each traffic node, the traffic speed data of 6 months is first divided into hours, and the traffic speed data of several time periods (it can be understood that the length of a specific time period is 1 hour) is obtained; one day is set as a time cycle, and then the traffic speed data of the same time period in each day are stacked; then, super empirical mode decomposition (SMD) is applied to the stacked data of each specific time period to obtain the intrinsic mode function of the traffic node in each specific time period.
[0053] For example, Figure 2 As shown, the traffic speed data x of node x in a specific time period t can be t The corresponding set of intrinsic mode functions (IMFs) is denoted as: IMFs = {IMF1, IMF2, …IMF n}, n represents the number of eigenmodes. It can be understood that different eigenmode functions represent the characteristics of different frequencies of the data. The traffic speed data of each specific time period reflects the traffic characteristics of this time period, and by stacking the data of the same time period every day, the periodic characteristics with a daily cycle can be captured.
[0054] For each specific time period, a node relationship matrix of each specific time period is constructed according to the intrinsic mode functions corresponding to each traffic node; wherein each element in the node relationship matrix represents the similarity between two traffic nodes at corresponding positions;
[0055] As an example, in this embodiment, the node relationship matrix is constructed based on the Pearson correlation coefficient between the intrinsic mode functions of the nodes. Specifically, for each specific time period, the Pearson correlation coefficient of two traffic nodes under the same intrinsic mode is calculated (as shown below): ); then the Pearson correlation coefficients of the two traffic nodes under all eigenmodes are integrated, and finally the similarity of the two traffic nodes is obtained (as shown below ).
[0056] Take node A and node B as an example. Furthermore, if the Pearson correlation coefficients of all eigenmodes are averaged, the similarity between node A and node B can be obtained:
[0057] Finally, the similarity between all nodes is calculated in the above way. For a specific time period t, the corresponding node relationship matrix X t It can be expressed as follows:
[0058]
[0059] Among them, S m,1 and S 1,m Both represent the similarity between traffic node m and traffic node 1.
[0060] By applying the above method to the data of each specific time period, the node relationship matrix of all specific time periods can be obtained, which ultimately constitutes the time domain decomposition matrix of the data set.
[0061] S103: Using the historical data to learn the current predefined graph structure to obtain a macro graph structure;
[0062] Specifically, macro-graph structure learning refers to directly learning implicit factors in historical data based on a predefined graph structure using learnable parameters as a supplement to the predefined graph structure.
[0063] Specifically, the learned macro graph structure A can be Ma It is expressed as:
[0064] A Ma =A+E
[0065] in, is a learnable parameter (also called a learnable graph adjacency matrix) used to obtain implicit relationships in the graph structure, and A represents a predefined graph structure.
[0066] S104: Selecting a time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain a micrograph structure;
[0067] As an example, suppose the time domain decomposition matrix corresponding to the time period to be predicted t is X t , the optimization process of the micrograph structure can be expressed as follows:
[0068] A Mi =W t ×Norm(X t )
[0069] Among them, W t represents the current learnable parameter corresponding to a specific time period t, X t represents the time domain decomposition matrix corresponding to a specific time period t (in this case, it is the node relationship matrix mentioned above), Norm represents the normalization operation, A Mi Represents the microscopic graph structure obtained after optimization.
[0070] S105: Fusing the macro graph structure and the micro graph structure to obtain an optimal graph structure;
[0071] As an example, through steps S103 and S104, the macro-level adjacency matrix A has been obtained. Ma and the micro-level graph adjacency matrix AMi When making traffic predictions, the A at the corresponding time will be adaptively selected according to the prediction task. Mi With A Ma , fused through ReLU activation function and normalization, the fusion process can be expressed as follows:
[0072] A * =Norm(ReLU(A Ma +W t *A Mi ))
[0073] In this way, we obtain the optimal graph adjacency matrix A for graph structure learning from both macro and micro levels. * .
[0074] S106: Inputting the optimal graph structure into the spatiotemporal graph neural network for training;
[0075] like Figure 3 As shown, the spatiotemporal graph neural network in the embodiment of the present invention includes an input layer, L spatiotemporal blocks STBlocks connected in sequence by residual connection, and an output layer; wherein each of the spatiotemporal blocks STBlock includes a temporal convolution layer and a graph convolution layer connected in sequence. Both the input layer and the output layer adopt a fully connected network. Correspondingly, the optimal graph structure is respectively input into each of the graph convolution layers in the spatiotemporal graph neural network.
[0076] The data processing process of the spatiotemporal graph neural network includes: the traffic speed data is first expanded in latitude through the fully connected network of the input layer, and then sent to the STBlock. The STblock has multiple time convolution layers and graph convolution layers, which are used to extract time dependency and space dependency respectively. In the time convolution layer, this embodiment uses dilated causal convolution and gating mechanism. In the graph convolution layer, this embodiment uses the fused optimal graph structure for graph convolution. At the same time, in order to enhance the performance of the model and avoid degradation, this embodiment also adds residual connections between the spatiotemporal blocks. Finally, the data is sent to the fully connected network to be converted into prediction results.
[0077] S107: Repeat steps S104 to S106 until a stop condition is met (for example, a set loss function converges or a set number of training times is reached), and a trained spatiotemporal graph neural network model is obtained;
[0078] S108: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction result of the future time period to be predicted.
[0079] The traffic prediction method provided by the embodiment of the present invention, when designing the prediction model, studies the graph structure from both macro and micro perspectives, and applies adaptive graph learning technology to effectively express complete traffic information. The learned optimal graph structure is then input into the graph neural network for prediction, and then the prediction result is obtained, thereby improving the accuracy of the prediction result.
[0080] In order to verify the effectiveness of the method of the present invention, the mean absolute error, root mean square error, and mean absolute percentage error, three commonly used traffic prediction evaluation indicators, are used to calculate the error of the method of the present invention and compare it with other methods. The experimental settings are as follows: Adam with a learning rate of 0.001 is used as the optimizer method, and an early stopping strategy is used to determine whether the stopping criteria are met. The batch size is 64 and the number of training rounds is 100. The trained model is tested using the test set, and the prediction results are shown in Table 1. Compared with other methods, the method of using the time domain decomposition adaptive graph neural network model TDAGNN for traffic prediction has achieved more accurate prediction results.
[0081] Table 1 Comparison of traffic condition prediction performance
[0082]
[0083] In addition, in order to further verify the effectiveness of the method, the experimental results of the proposed model TDAGNN and other two models (GraphWaveNet and AdaSTNet) on the PEMS-BAY dataset are visualized and compared with the actual road conditions. Figure 4 As shown. It can be seen that when the traffic flow is stable, the three models have good prediction results. However, when the traffic flow changes dramatically and the data fluctuates greatly, the method of the present invention can better respond to the changes and make more accurate predictions.
[0084] Corresponding to the above method, an embodiment of the present invention provides a system for traffic prediction using a time-domain decomposition adaptive graph neural network, such as Figure 5 As shown, it includes: data collection module, time domain decomposition module, graph structure learning module and prediction module.
[0085] Among them, the data collection module is used to obtain the historical data of the target road network; wherein the historical data includes the traffic speed data of each traffic node within the specified historical time range; the time domain decomposition module is used to perform time domain decomposition on the historical data and construct a time domain decomposition matrix; the graph structure learning module is used to use the historical data to learn the current predefined graph structure to obtain a macro graph structure; the time domain decomposition matrix corresponding to the time period to be predicted is selected to optimize the current learnable parameters to obtain a micro graph structure; the macro graph structure and the micro graph structure are merged to obtain the optimal graph structure; the optimal graph structure is input into the space-time graph neural network for training; the prediction module is used to input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained space-time graph neural network model to obtain the prediction result for the future time period to be predicted.
[0086] It should be noted that the system for traffic prediction using time-domain decomposition adaptive graph neural network provided in the embodiment of the present invention is for the purpose of implementing the above method. Its specific functions can be referred to the above method embodiments and will not be repeated here.
[0087] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor (processor) 601, a communication interface (Communications Interface) 602, a memory (memory) 603 and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604. The processor 601 can call the logic instructions in the memory 603 to execute the traffic prediction method, which includes: Step 1: Obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range; Step 2: Decompose the historical data in the time domain to construct a time domain decomposition matrix; Step 3: Use the historical data to learn the current predefined graph structure to obtain a macro graph structure; Step 4: Select the time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain a micro graph structure; Step 5: Fusion the macro graph structure and the micro graph structure to obtain an optimal graph structure; Step 6: Input the optimal graph structure into the spatiotemporal graph neural network for training; Step 7: Repeat steps 4 to 6 until the stop condition is met to obtain a trained spatiotemporal graph neural network model; Step 8: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain a prediction result for the future time period to be predicted.
[0088] In addition, when the logic instructions in the above-mentioned memory 603 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0089] The embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: Step 1: Acquire historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range; Step 2: Decompose the historical data in the time domain to construct a time domain decomposition matrix; Step 3: Use the historical data to learn the current predefined graph structure , get the macro graph structure; Step 4: Select the time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain the micro graph structure; Step 5: Fuse the macro graph structure and the micro graph structure to obtain the optimal graph structure; Step 6: Input the optimal graph structure into the spatiotemporal graph neural network for training; Step 7: Repeat steps 4 to 6 until the stop condition is met to obtain a trained spatiotemporal graph neural network model; Step 8: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction result of the future time period to be predicted.
[0090] An embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the methods provided by the above-mentioned method embodiments are implemented, for example, including: Step 1: Obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range; Step 2: Decompose the historical data in the time domain to construct a time domain decomposition matrix; Step 3: Use the historical data to learn the current predefined graph structure to obtain a macro graph structure; Step 4: Select the time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain a micro graph structure; Step 5: Fusion the macro graph structure and the micro graph structure to obtain an optimal graph structure; Step 6: Input the optimal graph structure into a spatiotemporal graph neural network for training; Step 7: Repeat steps 4 to 6 until the stop condition is met to obtain a trained spatiotemporal graph neural network model; Step 8: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain a prediction result for the future time period to be predicted.
[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for traffic prediction using a time-domain decomposition adaptive graph neural network model, characterized in that: include: Step 1: Obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range; Step 2: Decompose the historical data in the time domain to construct a time domain decomposition matrix; Step 3: Using the historical data to learn the current predefined graph structure to obtain a macro graph structure; Step 4: Select the time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters and obtain the micrograph structure; Step 5: Fusing the macro graph structure and the micro graph structure to obtain an optimal graph structure; Step 6: Input the optimal graph structure into the spatiotemporal graph neural network for training; Step 7: Repeat steps 4 to 6 until the stopping condition is met to obtain a trained spatiotemporal graph neural network model; Step 8: Input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction results for the future time period to be predicted.
2. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to claim 1 is characterized in that: Step 2 specifically includes: For each traffic node, the traffic speed data within the specified historical time range is first divided according to the set time interval to obtain the traffic speed data of several time periods; the time period is set, and then the traffic speed data of the same time period in each time period are stacked; then, the super empirical mode decomposition is applied to the stacked data of each specific time period to obtain the intrinsic mode function of the traffic node in each specific time period; For each specific time period, a node relationship matrix of each specific time period is constructed according to the intrinsic mode functions corresponding to each traffic node; wherein each element in the node relationship matrix represents the similarity between two traffic nodes at corresponding positions; The node relationship matrices of all specific time periods together constitute the time domain decomposition matrix.
3. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to claim 2 is characterized in that: For each specific time period, the node relationship matrix of each specific time period is constructed according to the intrinsic mode functions corresponding to each traffic node, including: For each specific time period, the Pearson correlation coefficients of the two traffic nodes under the same eigenmode are first calculated; then the Pearson correlation coefficients of the two traffic nodes under all eigenmodes are integrated to finally obtain the similarity of the two traffic nodes.
4. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to claim 3 is characterized in that: The Pearson correlation coefficients of the two traffic nodes under all eigenmodes are integrated, specifically including: using the following formula to average the Pearson correlation coefficients under all eigenmodes; in, represents the similarity between traffic node A and traffic node B in the characteristic time period t, represents the Pearson correlation coefficient between traffic node A and traffic node B under the i-th eigenmode, and n represents the number of eigenmodes contained in the eigenmode function corresponding to each traffic node.
5. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to claim 1 is characterized in that: Step 4 specifically includes: A Mi =W t ×Norm(X t ) Among them, W t represents the current learnable parameters corresponding to a specific time period t, X t represents the time domain decomposition matrix corresponding to a specific time period t, Norm represents the normalization operation, A Mi Represents the microscopic graph structure obtained after optimization.
6. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to claim 5 is characterized in that: Step 5 specifically includes: A * =Norm(ReLU(A Ma +W t *A Mi )) Among them, ReLU represents the activation function, A Ma represents the macrograph structure, A * Represents the optimal graph structure after fusion.
7. The method for traffic prediction using a time-domain decomposition adaptive graph neural network model according to any one of claims 1 to 6, characterized in that: The spatiotemporal graph neural network includes an input layer, L spatiotemporal blocks connected sequentially by residuals, and an output layer; wherein each of the spatiotemporal blocks includes a temporal convolution layer and a graph convolution layer connected sequentially; Correspondingly, the optimal graph structure is input into the spatiotemporal graph neural network for training, specifically including: inputting the optimal graph structure into each of the graph convolution layers in the spatiotemporal graph neural network respectively.
8. A system for traffic prediction using time-domain decomposition adaptive graph neural network, characterized in that: include: A data collection module, used to obtain historical data of the target road network; wherein the historical data includes traffic speed data of each traffic node within a specified historical time range; A time domain decomposition module, used for performing time domain decomposition on the historical data to construct a time domain decomposition matrix; A graph structure learning module is used to learn the current predefined graph structure using the historical data to obtain a macro graph structure; select a time domain decomposition matrix corresponding to the time period to be predicted to optimize the current learnable parameters to obtain a micro graph structure; fuse the macro graph structure and the micro graph structure to obtain an optimal graph structure; and input the optimal graph structure into a spatiotemporal graph neural network for training; The prediction module is used to input the time domain decomposition matrix corresponding to the time period to be predicted in the real-time data of the target road network into the trained spatiotemporal graph neural network model to obtain the prediction results for the future time period to be predicted.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.