Traffic flow prediction model construction method based on coupling multi-factor collaborative graph network

Abstract

The invention discloses a traffic flow prediction model construction method based on a coupling multi-factor collaborative graph network, and the method comprises the steps: obtaining the traffic characteristic data of an urban road network, and carrying out the preprocessing, and obtaining the standardized traffic characteristic data; simulating the traffic network into a graph network according to the standardized traffic characteristic data; constructing a distance adjacency matrix, a node correlation matrix, an environment comfort matrix and a subjectivity matrix according to the graph network, and performing fusion to obtain a coupling multi-factor feature; according to the attention mechanism and the neural network, constructing a time-space model coupled with multi-factor feature collaboration; and training the space-time model, and performing evaluation to obtain a traffic flow prediction model. According to the traffic flow prediction model construction method based on the coupled multi-factor collaborative graph network provided by the invention, coupling spatial-temporal characteristics and complex scene factors are considered, so that the trained traffic flow prediction model can accurately predict the traffic flow in a multi-source complex scene.

Description

Technical Field

[0001] The present application relates to the technical field of traffic flow prediction, and in particular to a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network. Background Art

[0002] With the rapid growth of population and economic development, the transportation sector is facing some complex problems that are difficult to solve with traditional methods. In addition, with the continuous acceleration of urbanization, heavy traffic pressure will seriously affect traffic capacity in scenarios with large traffic volume and high driving speed. Traffic congestion brings negative impacts of high energy consumption to society, and is highly complex, time-sensitive and random. The healthy development of transportation can reduce the pressure of social and economic development, curb the occurrence of unnecessary economic burdens and the healthy development of the environment.

[0003] In order to cope with the challenges of traffic problems in social development and ensure the sustainable development of an eco-friendly society, researchers have used a variety of advanced technologies such as deep learning, intelligent perception, big data, information and communication technology, wireless communication technology, and electronic control to integrate the entire traffic management system. The traffic management system has gradually become an application hotspot in the transportation field with its intelligence, digitization, easy sharing, and high coupling, meeting people's needs for visualization, analysis, prediction, and planning of traffic spatiotemporal information.

[0004] Traffic prediction methods have evolved from traditional statistical theories to current deep learning technologies. Traditional traffic prediction methods are model-driven and use linear system theory prediction methods. These methods assume that the target data is stable. This stable state cannot effectively handle non-parametric nonlinear data, nor can it describe the spatiotemporal nature of traffic scenes. Deep learning technology is constructed through a neural network structure and exhibits superior performance in capturing time and space alone. However, these methods ignore the spatiotemporal dependencies of traffic characteristics during the prediction process. At present, the prediction of urban spatiotemporal traffic networks focuses on scene features, mainly on scene features at the time and space levels, which can be summarized as a spatiotemporal prediction model of a spatiotemporal dual-dimensional prediction model. The spatiotemporal dual-dimensional prediction model matches the sequence algorithm for extracting time series features with traffic data, extracts spatial features in combination with the spatial network structure, and predicts through aggregation and update mechanisms. It uses time series to mine time features and relies on the adjacency matrix to extract spatial features. The parameters extracted by the model are solidified. The extraction of spatial features mainly relies on the multi-source heterogeneous mode of data. The use of Euclidean distance connection features invisibly increases the possibility of the model extracting invalid information, thereby causing a loss in the prediction accuracy of the model, resulting in the prediction effect affecting the effectiveness and applicability of spatiotemporal traffic features. Therefore, this model cannot accurately describe the complex traffic conditions in traffic scenarios, and it is difficult to meet the needs of residents for urban traffic prediction results in complex scenarios.

[0005] At the same time, the spatiotemporal prediction model embedded with collaborative single-factor features embeds scene factors with strong traffic relevance into traffic features, such as POI, holidays, and weather conditions, etc., and uses convolutional neural networks and attention mechanisms to extract the characteristics of urban areas, coupling single features with traffic main factors. In recent years, there has been extensive research on graph networks in this area. Their basic idea is to consider factors such as weather and holidays in the model input, effectively predicting road speeds and traffic congestion. However, the research on spatiotemporal prediction embedded with single factors focuses on the efficiency and performance of the model itself, and lacks consideration of complex real-world scenarios, resulting in the accuracy of the prediction results in complex spatiotemporal scenarios being difficult to meet the scenario requirements, and the prediction results are poorly generalizable when reflecting multi-source complex scenarios. Summary of the invention

[0006] To this end, the present application provides a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network to solve the problem that the traffic flow prediction methods in the prior art are inaccurate in multi-source complex scenarios.

[0007] In order to achieve the above objectives, this application provides the following technical solutions:

[0008] In a first aspect, a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network comprises:

[0009] Step 1: Obtain the traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data;

[0010] Step 2: Simulate the traffic network into a graph network based on standardized traffic characteristic data;

[0011] Step 3: construct a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix, and a subjectivity matrix according to the graph network, and fuse them to obtain coupled multi-factor features;

[0012] Step 4: constructing the spatiotemporal model of the coupled multi-factor feature collaboration based on the attention mechanism and the neural network;

[0013] Step 5: Train the spatiotemporal model and evaluate it to obtain a traffic flow prediction model.

[0014] Optionally, in step 1, the preprocessing includes: unifying the format of traffic characteristic data from different sources and preliminarily cleaning them, then normalizing the traffic characteristic data using Z-value, and converting each traffic characteristic data according to the mean and standard deviation to obtain standardized traffic characteristic data.

[0015] Optionally, in step 3, constructing a distance adjacency matrix based on the graph network is specifically as follows: extracting different nodes in the graph network and calculating the distance between each pair of nodes; the connectivity between nodes is represented by distance adjacency, and the calculated distance is filled into the adjacency matrix to obtain the distance adjacency matrix.

[0016] Optionally, the distance between each pair of nodes is calculated using a Euclidean distance or geographic distance formula based on the node position coordinates.

[0017] Optionally, in step 3, constructing a node correlation matrix according to the graph network is specifically as follows: using the Pearson correlation coefficient method to describe the degree of correlation between the nodes of the graph network with the numerical characteristics of covariance to obtain a node correlation matrix.

[0018] Optionally, in step 3, constructing an environmental comfort matrix based on the graph network is specifically as follows: taking temperature, weather conditions and wind speed as the main factors affecting the traffic scene, using the outdoor weather comfort index to superimpose meteorological elements, and then using a nonlinear model to mathematically express different nodes of the graph network to obtain the environmental comfort matrix.

[0019] Optionally, in step 3, constructing the subjectivity matrix according to the graph network is specifically as follows: using statistical methods to collect subjective willingness data of a certain sample in a layered manner, and based on the graph network, calculating the subjective willingness to choose driving according to the subjectivity of different weather conditions, thereby obtaining a human subjective matrix.

[0020] Optionally, the step 4 is specifically:

[0021] Step 401: extracting temporal features in complex scenarios using a GRU neural network with an attention mechanism;

[0022] Step 402: Using graph convolutional networks and attention mechanisms to mine dynamic spatiotemporal features;

[0023] Step 403: Input the temporal features, the spatiotemporal features and the coupled multi-factor features into a fully connected layer to obtain a spatiotemporal model.

[0024] Optionally, in step 5, mean absolute error, root mean square error and mean absolute percentage error are used to evaluate the spatiotemporal model.

[0025] In a second aspect, a traffic flow prediction model construction device based on a coupled multi-factor collaborative graph network comprises:

[0026] The data acquisition module is used to acquire the traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data;

[0027] A graph network construction module, which is used to simulate the traffic network into a graph network based on standardized traffic characteristic data;

[0028] A fusion module is used to construct a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix and a subjectivity matrix according to the graph network, and fuse them to obtain coupled multi-factor features;

[0029] A spatiotemporal model building module, used to build the spatiotemporal model of coupled multi-factor feature collaboration based on the attention mechanism and the neural network;

[0030] The model training module is used to train the spatiotemporal model and perform evaluation to obtain a traffic flow prediction model.

[0031] Compared with the prior art, this application has at least the following beneficial effects:

[0032] The present application provides a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network, by obtaining the traffic characteristic data of the urban road network and preprocessing it to obtain standardized traffic characteristic data; simulating the traffic road network as a graph network based on the standardized traffic characteristic data; constructing a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix, and a subjectivity matrix based on the graph network, and fusing them to obtain coupled multi-factor features; constructing a spatiotemporal model of coupled multi-factor feature collaboration based on the attention mechanism and the neural network; training the spatiotemporal model and evaluating it to obtain a traffic flow prediction model. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network provided in the present application takes into account the coupled spatiotemporal characteristics and complex scene factors, so the trained traffic flow prediction model can accurately predict traffic flow in multi-source complex scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more intuitively illustrate the prior art and the present application, exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing the present application; for example, those skilled in the art are capable of easily making conventional adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components) based on the technical concepts and exemplary drawings disclosed in the present application.

[0034] Figure 1 A flow chart of a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network provided in Example 1 of the present application;

[0035] Figure 2 A schematic diagram of the structure of a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network provided in Example 1 of the present application;

[0036] Figure 3 A schematic diagram of the traffic flow prediction model structure provided in Example 1 of the present application;

[0037] Figure 4 A schematic diagram of the prediction effect of the traffic flow prediction model provided in Example 1 of the present application. DETAILED DESCRIPTION

[0038] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0039] In the description of this application: unless otherwise specified, the meaning of "plurality" is two or more. The terms "first", "second", "third", etc. in this application are intended to distinguish the objects referred to, and do not have special meanings in terms of technical connotations (for example, they should not be understood as emphasizing the importance or order, etc.). Expressions such as "including", "comprising", "having", etc. also mean "not limited to" (certain units, components, materials, steps, etc.).

[0040] The terms such as "upper", "lower", "left", "right", "middle", etc. cited in this application are usually used to indicate the general relative position relationship for the purpose of intuitive understanding by referring to the drawings, and are not absolute limitations on the position relationship in the actual product.

[0041] With the impact of complex multi-source factors on spatiotemporal traffic, various external conditions, such as policies, economy and weather, have direct or indirect impacts on the operating performance of the transportation system. By integrating the characteristics of the two and constructing a coupled graph spatiotemporal deep learning model, and mining the coupled spatiotemporal prediction model under complex scenarios, feasible solutions are provided to alleviate traffic congestion and other traffic-related problems, which brings new ideas for safer, more efficient and sustainable transportation development. Therefore, the research on this idea has broad application prospects.

[0042] Embodiment 1

[0043] See also Figure 1 and Figure 2 This embodiment provides a method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network, including:

[0044] S1: Obtain traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data;

[0045] This step obtains traffic characteristic data under the urban road network and preprocesses the traffic characteristic data into a basic structure, thereby obtaining a regular characteristic data set. Specifically, it includes:

[0046] S101: Use multiple data sources such as traffic sensors and GNSS (Global Navigation Satellite System) to collect traffic characteristics data such as road network, traffic flow and vehicle speed;

[0047] S102: Unify the formats of data from different sources and perform preliminary cleaning to remove redundant and invalid data;

[0048] S103: normalizing the traffic characteristic data using the Z-value, converting each traffic characteristic data according to the mean and the standard deviation to obtain standardized traffic characteristic data, so that the characteristics of the entire data set are on the same standardized scale.

[0049] S2: Simulate the traffic network as a graph network based on standardized traffic characteristic data;

[0050] This step simulates the traffic network into a graph network G = (V, A, X) based on the standardized traffic feature data. Specifically, each traffic sensor represents a node. Assume that the node set is V, the number of nodes is N, and the feature sequence in the T time period after fusion is X. The main goal of prediction is to use historical data to predict future situations. Therefore, this process requires training a model that can contain parameters, which can be abstractly expressed as:

[0051]

[0052] In formula (1), H represents the step length of historical traffic characteristic data, and F represents the step length of future prediction data. Since the data is taken as a step length every 5 minutes, the step length of historical data is taken as H, and the step length of predicted data is taken as F. In general, F = 1, which means predicting one step length, that is, predicting the next 5 minutes.

[0053] S3: Construct the distance adjacency matrix, node correlation matrix, environmental comfort matrix and subjectivity matrix according to the graph network, and fuse them to obtain the coupled multi-factor features;

[0054] The specific steps of this step are:

[0055] S301: construct a distance adjacency matrix;

[0056] S3011: Extract different nodes in the graph network and calculate the distance between each pair of nodes;

[0057] Specifically, the Euclidean distance or geographic distance formula can be used to calculate according to the node location coordinates. The distance formula can be expressed as:

[0058]

[0059] In formula (2), x i and x jIndicates the horizontal coordinate of GPS collected data, y i and j It represents the ordinate of the GNSS collected data, and d represents the distance.

[0060] S3012: The connectivity between nodes is represented by distance adjacency, and the calculated distance is filled into the adjacency matrix to obtain the distance adjacency matrix.

[0061] Specifically, connected nodes are represented as 1, and disconnected nodes are represented as 0. Therefore, the distance adjacency matrix can be expressed as:

[0062]

[0063] In formula (2), A Dis Represents the distance adjacency matrix.

[0064] S302: Constructing a node correlation matrix;

[0065] Specifically, the Pearson correlation coefficient method is used to describe the correlation between the nodes in the graph network with the numerical characteristics of covariance to obtain the N-dimensional node correlation matrix A Corr The node correlation matrix is ​​also called the correlation coefficient matrix, which is expressed as:

[0066]

[0067] S303: Construct environmental comfort matrix;

[0068] Specifically, temperature, weather conditions and wind speed are taken as the main factors affecting traffic scenes, and the outdoor weather comfort index is used to superimpose meteorological elements. Then, the different nodes of the graph network are mathematically expressed using a nonlinear model to reflect the subjective outdoor overall comfort. After the different nodes are expressed, an N-dimensional environmental comfort matrix A can be formed. Envi , expressed as:

[0069] A E =C-(K averT +K maxT +K weather +K wind )(5)

[0070] In formula (5), C represents a constant, which is composed of the sum of the average temperature term and the maximum temperature term calculated by the extremely uncomfortable threshold temperature, and K averT represents the mean temperature, K maxT Indicates the maximum temperature, K weather Indicates weather conditions, K wind Indicates wind speed.

[0071] S304: Constructing the subjectivity matrix;

[0072] Specifically, we use statistical methods to collect subjective willingness data of a certain sample in a hierarchical manner, and based on the graph network, we calculate the subjective willingness of choosing to drive according to the subjectivity of different weather conditions, thereby obtaining the human subjectivity matrix A. Sub .

[0073] S305: The distance adjacency matrix, the node correlation matrix, the environmental comfort matrix and the subjectivity matrix are integrated to obtain coupled multi-factor features.

[0074] Specifically, the matrices of four different features are aligned and weighted according to their importance in fusion, and then the weighted matrices are spliced ​​together along the column direction to finally obtain a comprehensive fusion matrix. The matrix after feature fusion can be expressed as:

[0075]

[0076] In formula (6), A Dis is the distance adjacency matrix, A Corr is the node correlation matrix, A Envi is the environmental comfort matrix, A Sub is the subjectivity matrix, M F ∈R N×N Represents the fused features, W∈R N×N represents the weight of the fusion sub-matrix, A N represents the fused adjacency matrix, b N Represents the bias vector of the fused feature, W N Represents the weight parameter.

[0077] S4: Construct a spatiotemporal model of coupled multi-factor feature collaboration based on attention mechanism and neural network;

[0078] The specific steps of this step are:

[0079] S401: Extracting temporal features in complex scenarios using the GRU neural network with attention mechanism;

[0080] S4011: Smoothing of time series data;

[0081] Specifically, each convolution kernel is slid on a local window of the time series, and the local pattern is extracted by weighted summation.

[0082] S4012: Constrain the GRU and attention mechanisms to output feature X H Mapped into query (Q), key (K) and value (V) vectors, and then get the attention weights, which are expressed as;

[0083] Q=X H WQ ,K=X H W K ,V=X H W V (7)

[0084]

[0085] In formula (7) and (8), W Q ,W K ,W V represents the learnable weight matrix, d k represents the dimension of the key vector, K T represents the transpose of parameter K, U represents the attention weight, and is a probability distribution after the attention score is converted.

[0086] S4013: By calculating the similarity between different time steps in the sequence, the dynamic weight of the feature of each time step is assigned to better handle complex traffic flow time series data. The final time series feature is expressed as X T :

[0087] X T =Temporal(UV)(9)

[0088] S402: Mining dynamic spatiotemporal features using graph convolutional networks and attention mechanisms;

[0089] S4021: Represent spatial data as a graph structure, assign features to each node, and aggregate neighborhood information of each node through convolution operations;

[0090] S4022: Use GCN to extract high-level features of nodes in the graph, expressed as:

[0091]

[0092] In formula (10) and (11), I N represents the identity matrix, represents the road network degree matrix, H l represents the lth layer network features, θ (l) is the corresponding hyperparameter, and σ(·) refers to the ReLU activation function.

[0093] S4023: Introduce an attention mechanism to dynamically adjust the weights between nodes.

[0094] Specifically, the attention mechanism is applied to perform weighted combination of these initially extracted features according to the importance of different nodes. Nodes i and j can be expressed as:

[0095] e ij =LeakyReLU(aT [Wx i ||Wx j ])(12)

[0096]

[0097] S4024: Weight and sum the features of adjacent nodes to generate updated node representations, and finally obtain the spatiotemporal feature X S It is expressed as:

[0098]

[0099] S403: Input the temporal features, spatiotemporal features and coupled multi-factor features into the fully connected layer to obtain a spatiotemporal model.

[0100] See also Figure 3 ,This step uses the fully connected layer, and the spatiotemporal model can be effectively integrated to capture the complex interactive relationships in traffic data. The specific process is as follows:

[0101] S4031: Construct the weights of temporal and spatial features, and then perform weighted processing on them;

[0102] S4032: The coupled multi-factor features are input into the fully connected layer, and the fused features are predicted and expressed through the regression layer (i.e., the spatiotemporal model). The prediction result can be expressed as:

[0103]

[0104] In formula (15), W1, W2 represent the spatiotemporal feature weights, and f represents the feature collaborative fusion function.

[0105] S5: Train the spatiotemporal model and evaluate it to obtain the traffic flow prediction model.

[0106] This step is used to train the spatiotemporal model obtained previously, including:

[0107] S501: Divide the data set into training set, validation set and test set in a ratio of 6:2:2. Select Adam as the optimizer, set the training step size to 0.01, the batch size to 10, and the iteration limit to 200. When the number of training times reaches the set k value, the model stops training.

[0108] S502: Predict the trained feature parameters in the test set to obtain the predicted value, and then compare it with the true value to obtain the prediction accuracy;

[0109] S503: The performance of different methods is evaluated using three commonly used traffic prediction indicators, namely, mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). The calculation formula is as follows:

[0110]

[0111] In formulas (16), (17) and (18), X R (k) represents the real data, X P (k) represents the predicted data, and θ represents the number of sample points.

[0112] The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network provided in this embodiment takes into account coupled spatiotemporal characteristics and complex scenario factors, so the trained traffic flow prediction model can accurately predict traffic flow in multi-source complex scenarios.

[0113] The method provided in this embodiment can be used to extract the spatiotemporal correlation of complex scenes. Specifically, the time series features in complex scenes are extracted through a collaborative mechanism, and dynamic spatiotemporal features are mined, so that the direct coupling correlation between spatiotemporal features and multiple external factors can be effectively mined through a collaborative mechanism, thereby achieving accurate and effective traffic flow prediction.

[0114] Compared with the current single-factor spatiotemporal prediction model that lacks consideration of complex real-world scenarios, this embodiment proposes a coupled multi-factor collaborative graph process network (CM-SGN), which uses a new fusion mechanism to couple traffic features with coupled scene features, and uses the synergy of spatiotemporal networks and attention to predict traffic flow. This method has better short- and medium-term prediction performance and more effective mining of data features, so that it can maintain high accuracy and stability in prediction tasks at multiple time scales, and improves the accuracy of short- and medium-term predictions. For specific results, see Figure 4 .

[0115] Embodiment 2

[0116] This embodiment provides a traffic flow prediction model construction device based on a coupled multi-factor collaborative graph network, including:

[0117] The data acquisition module is used to acquire the traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data;

[0118] A graph network construction module, which is used to simulate the traffic network into a graph network based on standardized traffic characteristic data;

[0119] A fusion module is used to construct a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix and a subjectivity matrix according to the graph network, and fuse them to obtain coupled multi-factor features;

[0120] A spatiotemporal model building module, used to build the spatiotemporal model of coupled multi-factor feature collaboration based on the attention mechanism and the neural network;

[0121] The model training module is used to train the spatiotemporal model and perform evaluation to obtain a traffic flow prediction model.

[0122] Regarding the specific implementation contents of each module in a traffic flow prediction model construction device based on a coupled multi-factor collaborative graph network, please refer to the above definition of a traffic flow prediction model construction method based on a coupled multi-factor collaborative graph network, which will not be repeated here.

[0123] The technical features of the above embodiments may be arbitrarily combined (as long as there is no contradiction in the combination of these technical features). To make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

Claims

1. A method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network, characterized in that: include: Step 1: Obtain the traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data; Step 2: Simulate the traffic network into a graph network based on standardized traffic characteristic data; Step 3: construct a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix, and a subjectivity matrix according to the graph network, and fuse them to obtain coupled multi-factor features; Step 4: constructing the spatiotemporal model of the coupled multi-factor feature collaboration based on the attention mechanism and the neural network; Step 5: Train the spatiotemporal model and evaluate it to obtain a traffic flow prediction model.

2. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1 is characterized in that: In the step 1, the preprocessing includes: unifying the format of traffic characteristic data from different sources and preliminarily cleaning them, then normalizing the traffic characteristic data using the Z-value, and converting each traffic characteristic data according to the mean and standard deviation to obtain standardized traffic characteristic data.

3. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1 is characterized in that: In the step 3, constructing a distance adjacency matrix according to the graph network is specifically as follows: extracting different nodes in the graph network and calculating the distance between each pair of nodes; using the distance adjacency to represent the connectivity between the nodes, the calculated distance is filled into the adjacency matrix to obtain the distance adjacency matrix.

4. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 3 is characterized in that: When calculating the distance between each pair of nodes, the Euclidean distance or geographic distance formula is used for calculation according to the node position coordinates.

5. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1 is characterized in that: In the step 3, constructing a node correlation matrix according to the graph network is specifically as follows: using the Pearson correlation coefficient method to describe the correlation between the nodes of the graph network with the numerical characteristics of covariance to obtain a node correlation matrix.

6. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1, characterized in that: In the step 3, the environmental comfort matrix is ​​constructed according to the graph network as follows: temperature, weather conditions and wind speed are taken as the main factors affecting the traffic scene, the meteorological elements are superimposed using the outdoor weather comfort index, and then the different nodes of the graph network are mathematically expressed using a nonlinear model to obtain the environmental comfort matrix.

7. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1, characterized in that: In the step 3, constructing the subjectivity matrix according to the graph network is specifically as follows: using statistical methods to collect subjective intention data of a certain sample in a layered manner, and based on the graph network, calculating the subjective intention of choosing to travel by car according to the subjectivity of different weather conditions, thereby obtaining a human subjective matrix.

8. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1 is characterized in that: The step 4 is specifically as follows: Step 401: extracting temporal features in complex scenarios using a GRU neural network with an attention mechanism; Step 402: Using graph convolutional networks and attention mechanisms to mine dynamic spatiotemporal features; Step 403: Input the temporal features, the spatiotemporal features and the coupled multi-factor features into a fully connected layer to obtain a spatiotemporal model.

9. The method for constructing a traffic flow prediction model based on a coupled multi-factor collaborative graph network according to claim 1, characterized in that: In step 5, mean absolute error, root mean square error and mean absolute percentage error are used to evaluate the spatiotemporal model.

10. A traffic flow prediction model construction device based on a coupled multi-factor collaborative graph network, characterized in that: include: The data acquisition module is used to acquire the traffic characteristic data of the urban road network and perform preprocessing to obtain standardized traffic characteristic data; A graph network construction module, which is used to simulate the traffic network into a graph network based on standardized traffic characteristic data; A fusion module is used to construct a distance adjacency matrix, a node correlation matrix, an environmental comfort matrix and a subjectivity matrix according to the graph network, and fuse them to obtain coupled multi-factor features; A spatiotemporal model building module, used to build the spatiotemporal model of coupled multi-factor feature collaboration based on the attention mechanism and the neural network; The model training module is used to train the spatiotemporal model and perform evaluation to obtain a traffic flow prediction model.

Images