Traffic flow prediction method fusing air pollution factors

Abstract

The application provides a traffic flow prediction method fusing air pollution factors, which comprises the following steps: acquiring traffic topology structure, historical traffic flow and air pollutant concentration data, constructing a traffic feature matrix, an adjacency matrix, a related pollutant concentration attribute matrix and a mapping function, and fusing the air pollutant concentration data through an attribute enhancement unit; adopting K-means clustering to screen the pollutants with the highest correlation with traffic flow, and using the attribute enhancement unit to integrate the pollutants into traffic flow features; and building and training a deep learning model to predict traffic flow. The method overcomes the limitation of the prior art which only depends on traffic flow data by combining the correlation between traffic flow spatiotemporal features and air pollutants, and effectively improves the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic flow prediction, in particular to a traffic flow prediction method fusing air pollution factors. BACKGROUND

[0002] Traffic flow prediction refers to a technology that uses historical and / or real-time traffic data (such as vehicle flow, speed, event information, etc.) to estimate the traffic operating state (such as flow, congestion level) of a specific period, road section or area in the future through mathematical models or algorithms, aiming to provide support for traffic management and travel decision-making.

[0003] The current common prediction methods include time series models based on historical statistical rules (such as ARIMA), traditional machine learning methods relying on artificial feature extraction (such as support vector machine, random forest), and deep learning methods capable of automatically learning complex spatio-temporal dependencies (such as RNN, LSTM, GNN and its variants STGCN, DCRNN), as well as hybrid methods fusing multi-source data (such as weather, events).

[0004] However, the existing methods generally have the following shortcomings: first, they fail to fully utilize the spatio-temporal correlation characteristics of traffic flow data, and second, they ignore the influence mechanism of air pollutant concentration data on traffic flow. SUMMARY

[0005] The purpose of the present application is to provide a traffic flow prediction method fusing air pollution factors, which integrates the spatio-temporal correlation characteristics of road traffic flow and environmental air quality information to at least solve one problem existing in the prior art.

[0006] The traffic flow prediction method fusing air pollution factors provided by the present application comprises the following steps:

[0007] Obtaining the road network topology information, historical traffic flow time series data set and air pollutant concentration data set of a target area;

[0008] Establishing an adjacency matrix A representing the connectivity of the road network according to the road network topology information, establishing a traffic feature matrix X according to the historical traffic flow time series data set, and establishing a concentration matrix Y of a plurality of target pollutants according to the air pollutant concentration data set, and representing the concentration matrix Y of the target pollutant with the highest correlation with the traffic feature matrix X as a related pollutant concentration attribute matrix K;

[0009] Building a neural network model, training the neural network model using the adjacency matrix A, the traffic feature matrix X and the related pollutant concentration attribute matrix K; wherein an augmented matrix P is established according to the traffic feature matrix X and the related pollutant concentration attribute matrix K to fuse air pollutant concentration data;

[0010] According to the loss function, the parameters of the neural network model are optimized by a back propagation algorithm to obtain a traffic flow prediction model.

[0011] Traffic flow is predicted using the traffic flow prediction model.

[0012] The method uses air pollutant concentration data for traffic flow prediction, improving the reliability of traffic flow prediction. At the same time, according to the historical traffic flow time series data set and the air pollutant concentration data set, a relevant pollutant concentration attribute matrix is established to associate historical traffic flow with air pollutant concentration, improving the accuracy of traffic flow prediction.

[0013] Based on the traffic flow prediction method fusing air pollution factors, the application provides a traffic flow prediction device, which comprises one or more processors and a memory; the memory is coupled with the one or more processors, and the memory is used to store computer program codes, the computer program codes comprising computer instructions, and the one or more processors invoke the computer instructions to realize the traffic flow prediction method fusing air pollution factors.

[0014] In addition, the application further provides a computer readable storage medium, which stores instructions, and the instructions are executed by a processor to realize the traffic flow prediction method fusing air pollution factors.

[0015] The beneficial technical effects of the application are that: by constructing a multi-modal fusion traffic flow prediction model of traffic flow and air pollutants, a time-space and environmental double-dimensional feature extraction mechanism is innovatively established, the time-space evolution law of traffic flow data is fully mined, the influence of air quality fluctuation on traffic mode is accurately captured, and the problem of insufficient modeling of traffic environment interaction in traditional models is effectively solved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 The flowchart of the traffic flow prediction method fusing air pollution factors in the embodiments of the application.

[0018] Figure 2 The schematic diagram of the road network (street) of the target area in the embodiments of the application.

[0019] Figure 3Fig. 1 is a schematic diagram of a road network topology and an adjacency matrix in an embodiment of the present application, wherein (a) is a road network topology, and (b) is an adjacency matrix.

[0020] Figure 4 Fig. 2 is a schematic diagram of traffic feature matrix integrated with air pollutant concentration data in an embodiment of the present application.

[0021] Figure 5 Fig. 3 is a schematic diagram of a neural network model in an embodiment of the present application.

[0022] Figure 6 Fig. 4 is a comparison diagram of prediction results of the traffic flow prediction method integrated with air pollution factors in an embodiment of the present application, wherein (a) is a prediction result diagram, and (b) is a real result diagram. DETAILED DESCRIPTION

[0023] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the present application is not limited to the specific embodiments disclosed below.

[0024] Traffic flow prediction plays an important role in urban planning. Traffic flow prediction combined with air quality data not only helps planners to more comprehensively assess the impact of traffic on the environment, thereby developing more sustainable traffic management strategies, but also improves prediction accuracy and optimizes traffic control measures through the correlation between air quality and traffic flow. However, due to the complex interaction mechanism between urban traffic system and air pollution, accurately modeling the relationship between the two is still a challenging task.

[0025] Traffic flow prediction and air quality have complex time dependence. Traffic flow and pollutant concentration both change dynamically over time and exhibit periodic characteristics. For example, traffic congestion during peak hours is often accompanied by rising pollutant concentrations, while air quality may improve during low-flow hours at night. Existing traffic flow prediction methods (such as Kalman filtering, support vector regression, recurrent neural networks, etc.) can capture time dependence, but most do not fully integrate the influence mechanism of air quality time series data on traffic flow.

[0026] In addition, there is a complex spatial dependence between traffic flow and air quality. From the spatial distribution, the diffusion of traffic pollutants is affected by road network structure and meteorological conditions, and the feedback of air quality to traffic flow varies in different regions. For example, the high concentration of pollutants around the main road may inhibit travel demand, while the low pollution level in the green area may attract more traffic. Although existing research attempts to use convolutional neural networks or graph neural networks to model spatial correlation, it is difficult to accurately depict the heterogeneous correlation between traffic patterns and air quality in different regions, as the coupling relationship between road network topology, traffic flow and pollutant diffusion is not considered. Therefore, how to effectively integrate multi-source spatio-temporal data (such as traffic flow, road network structure, and air quality monitoring data) and establish a more accurate prediction model is still a difficult point in current research.

[0027] To solve the above at least one technical problem, the application provides a traffic flow prediction method fusing air pollution factors. The method innovatively introduces air quality monitoring data as auxiliary features into the prediction model, and by establishing a traffic-environment coupling analysis framework, the accuracy of the prediction results and the robustness of the model in complex environments are significantly improved.

[0028] As shown in Figure 1 , a traffic flow prediction method fusing air pollution factors includes the following steps:

[0029] S1: Obtain the road network topology structure information, historical traffic flow time series data set and air pollutant concentration data set of the target region;

[0030] S2: According to the road network topology structure information, an adjacency matrix A representing the connectivity of the road network is established, according to the historical traffic flow time series data set, a traffic feature matrix X is established, and according to the air pollutant concentration data set, a concentration matrix Y of a plurality of target pollutants is established. The concentration matrix Y of the target pollutant with the highest correlation with the traffic feature matrix X is represented as a related pollutant concentration attribute matrix K;

[0031] S3: Build a neural network model, and train the neural network model using the adjacency matrix A, the traffic feature matrix X and the related pollutant concentration attribute matrix K; wherein an augmented matrix P is established according to the traffic feature matrix X and the related pollutant concentration attribute matrix K, so as to fuse the air pollutant concentration data;

[0032] S4: According to the loss function, the parameters of the neural network model are optimized by the back propagation algorithm, and a traffic flow prediction model is obtained;

[0033] S5: Use the traffic flow prediction model to predict traffic flow.

[0034] The road network topology information can be obtained by collecting traffic road network information of the target area, the historical traffic flow time series data set can be obtained by measuring in the target area or based on traffic monitoring data, and the air pollutant concentration data set can be obtained by measuring in the target area or based on meteorological monitoring data. The data can be preprocessed, for example, the historical traffic flow time series data set is normalized by using a MinMax scaler, and the missing part of the data is completed by using a linear interpolation method. The data set is divided into a training set and a test set for training the neural network model. The target pollutant can include one of carbon monoxide, particulate matter, sulfur dioxide, nitrogen dioxide and ozone, and thus a concentration matrix Y of five target pollutants can be established.

[0035] Further, in step S2, establishing an adjacency matrix A representing road network connectivity according to the road network topology information comprises: modeling the road network topology information to obtain a road network topology graph G = (V, E) representing the connection relationship between roads, an adjacency matrix A representing road network connectivity, and a mapping function ; wherein V represents a set of road segments, and E represents a set of edges representing connectivity between two road segments.

[0036] Specifically, as shown in Figure 2 and Figure 3 , the set of road segments is represented as V, V = v1, v2, …, v n , where n represents the number of road segments, also known as the number of nodes; the set of edges is represented as E, E = e1, e2, …, e m , and m represents the number of edges; if two road segments are connected, the connection relationship is marked as an undirected edge; if two road segments are not connected, there is no edge relationship; the set of road segments V and the set of edges E are drawn as a point-line graph, denoted as a road network topology graph G;

[0037] Without considering the self-connection of road segments, an adjacency matrix A that is not self-connected is created, which is an n x n matrix. Each element of the adjacency matrix is initialized to 0, and the element in the adjacency matrix A is 0, indicating that there is no edge connection between two road segments;

[0038] All edges in the road network topology graph are traversed, and for an undirected edge connecting road segment i and road segment j, the elements of the i-th row and the j-th column and the j-th row and the i-th column in the adjacency matrix A are set to 1; after the traversal is completed, the adjacency matrix A corresponding to the road network topology graph G is obtained, and the mapping function .

[0039] Further, in step S2, establishing a traffic feature matrix X according to the historical traffic flow time series data set comprises: establishing a traffic feature matrix X from historical traffic flow time series data from time t-T to time t , , ;in, express Traffic flow on n road segments at a given time. Indicates that the i-th road segment is in Traffic flow at any given moment.

[0040] Further, in step S2, establishing a concentration matrix Y of multiple target pollutants based on the air pollutant concentration dataset includes: establishing a concentration matrix of target pollutants from the target pollutant concentration data from time tT to time t. , , ;in, express The pollutant concentrations at n road segments at given time. Indicates that the i-th road segment is in The concentration of pollutants at any given time.

[0041] Furthermore, representing the concentration matrix Y of the target pollutant most correlated with the traffic feature matrix X as the related pollutant concentration attribute matrix K includes: representing the traffic feature matrix... Concentration matrix of a certain type of target pollutant Traffic flow data and pollutant concentration data were clustered separately to obtain two sets of data with cluster labels. A matching calculation was then performed on these two sets of data to obtain a cluster similarity index. Cluster similarity index The concentration matrix Y of the target pollutant with the highest value is represented as the related pollutant concentration attribute matrix K.

[0042] Specifically, the clustering method can be a two-step K-means clustering method. Cluster similarity index. The larger the value, the higher the correlation. By adjusting the set threshold, the pollutant concentration data with the highest correlation can be used as the data to construct the augmented matrix P.

[0043] For traffic flow datasets A dataset of pollutant concentrations First, traffic flow data and air pollutant concentration data are clustered to obtain data with cluster labels. Then, a matching matrix is ​​defined. ,in, This represents the number of identical sample points in the i-th cluster after clustering the traffic flow dataset X and the j-th cluster after clustering the pollutant concentration dataset Y, where k and q represent the maximum number of rows and columns of the matrix, respectively. Finally, a clustering similarity index is defined. The calculation formula is:

[0044]

[0045] wherein, , , is specifically defined as follows:

[0046]

[0047]

[0048]

[0049] By the calculation formula of , it can be found that when the class number changes in two times of clustering, the value of does not change, that is, the size of has nothing to do with the label of clustering. At the same time, by the calculation formula of and the calculation formula of , it can be known that , so ; when , it means that the results of two times of clustering are completely similar; when , it means that the results of two times of clustering are completely dissimilar.

[0050] represents the total number of samples of the traffic flow clustering as the ith class, that is, the total distribution of the class in all pollutant concentration clustering classes; represents the total number of samples of the pollutant concentration clustering as the jth class, that is, the total distribution of the class in all traffic flow clustering classes; represents the total sum of all elements in the matching matrix M, that is, the total number of sample points intersection of all class combinations in the results of two times of clustering. By comparing the similarity results of the traffic flow and air pollutant concentration clustering results of each grid, the correlation degree of traffic flow and air pollutant concentration is analyzed. By the calculation of the clustering similarity index , it can be found that is a matrix defining that a certain class of the first clustering and each class of the second clustering have the same number of sample points, which is similar to finding the intersection of the results of two times of clustering, so that each class of the first clustering and each class of the second clustering are associated. The value of n is related to the matching matrix, , , is specifically defined as follows:

[0051]

[0052]

[0053]

[0054] like Figure 4 As shown, the relevant pollutant concentration attribute matrix K can be represented as: , , ;in, express The pollutant concentrations at n road segments at given time. Indicates that the i-th road segment is in Pollutant concentration at any given time. The attribute matrix. Integrating into the traffic feature matrix in the form of an augmented matrix In the process, we obtain the augmented matrix P; the augmented matrix part is called the attribute enhancement unit, and we obtain the mapping function. The augmented matrix containing air pollutant concentration data and traffic characteristics at time t is: The prediction result is expressed as .

[0055] Furthermore, in step S3, a neural network model is built using deep learning methods to enhance the mapping function. Modeling is performed; the neural network model comprises an input layer, two layers of GCN model, and one layer of GRU model. The input layer takes into account the traffic feature matrix X, the relevant pollutant concentration attribute matrix K, and the adjacency matrix A. The GCN model captures the spatial features of traffic flow data between road segment nodes, and the GRU model captures the temporal features of traffic flow data and air pollutant concentration between road segment nodes. The constructed neural network model is trained using training set data, and its accuracy is tested using test set data.

[0056] Specifically, such as Figure 5 As shown, each layer of the two-layer GCN model includes T GCN processing units. The GCN processing units in the two layers correspond one-to-one. The enhanced traffic feature matrix at each sampling time corresponds to two GCN processing units. The output of the two-layer GCN processing units is represented as follows: That is, H=h t-T , ..., h t-1 h t The output of the second layer of the GCN model is used as the input of the GRU model. After capturing temporal features, the GRU model outputs a new node, denoted as H', meaning the output of the GRU model is H'. The output of the GRU processing unit is represented as... The overall output of the GRU model is then expressed as: .

[0057] Furthermore, in step S4, the loss function between the predicted value and the actual value is calculated, the parameters of the neural network model are optimized through the backpropagation algorithm, and the Adam optimizer is used to adaptively learn the learning rate to optimize the parameters of the neural network model to make the model more accurate.

[0058] Specifically, when calculating the loss function between the predicted and actual values, the mean squared error is used as the loss function Loss. The formula for calculating the loss function Loss is as follows:

[0059] ;

[0060] in, Represents the true value at time t. This represents the predicted value at time t. Indicates L2 regularization, It is a hyperparameter that controls the regularization rate.

[0061] Furthermore, in step S5, the trained traffic flow prediction model is used to predict the real-time traffic flow data to obtain the prediction results.

[0062] In addition, the root mean square error (RMSE) is used as a performance evaluation metric. The formula for calculating the root mean square error (RMSE) is as follows:

[0063] ;

[0064] in, express The true value of a moment express The predicted value at any given time.

[0065] This invention uses a dataset from http: / / iot.ee.surrey.ac.uk:8080 / datasets.html to conduct experiments and verify the effectiveness of the aforementioned neural network model. The Aarhus, Denmark dataset is used, where traffic flow data records the average vehicle speed every 5 minutes. Data from 100 road segments was selected from 00:05 on August 2, 2014 to 24:00 on August 16, 2014. In the matrix, rows represent road segments, columns represent timestamps, and the traffic feature matrix has a dimension of 100×4320. Air pollution data is recorded every 5 minutes, with the time interval consistent with the traffic flow data. The main monitored air pollutants include ozone, particulate matter, carbon monoxide, sulfur dioxide, and nitrogen dioxide, measured in micrograms per cubic meter, with the same dimension as the traffic feature matrix.

[0066] Figure 6 This is a comparison chart showing the results of street traffic flow predictions every 5 minutes. Figure 6(a) in the figure is a prediction result graph, Figure 6 (b) in the figure is a real result graph. The present prediction reflects traffic flow through traffic speed, and the smaller the traffic speed, the higher the traffic flow. From Figure 6 It can be seen that the error between the prediction result and the real result is very small, indicating that the accuracy of traffic flow prediction by using the neural network model described above in the embodiment of the present application is high.

[0067] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A traffic flow prediction method that fuses air pollution factors, characterized by, The method comprises the following steps: obtaining road network topology information, historical traffic flow time series data sets and air pollutant concentration data sets of a target area; According to the road network topology information, an adjacency matrix A representing road network connectivity is established, according to the historical traffic flow time series data set, a traffic feature matrix X is established, according to the air pollutant concentration data set, a concentration matrix Y of a plurality of target pollutants is established, and the concentration matrix Y of the target pollutant with the highest correlation with the traffic feature matrix X is represented as a related pollutant concentration attribute matrix K; wherein the concentration matrix Y of the target pollutant with the highest correlation with the traffic feature matrix X is represented as the related pollutant concentration attribute matrix K, including: clustering the traffic flow data and the target pollutant concentration data in the traffic feature matrix X and the concentration matrix Y of one target pollutant respectively to obtain two data with clustering labels, and performing matching calculation on the two data with clustering labels to obtain a clustering similarity index ; the concentration matrix Y of the target pollutant with the highest value of the clustering similarity index is represented as the related pollutant concentration attribute matrix K. building a neural network model, training the neural network model using the adjacency matrix A, the traffic feature matrix X and the related pollutant concentration attribute matrix K; wherein an augmented matrix P is established according to the traffic feature matrix X and the related pollutant concentration attribute matrix K to fuse air pollutant concentration data; optimizing parameters of the neural network model according to a loss function through a back propagation algorithm to obtain a traffic flow prediction model; using the traffic flow prediction model to perform traffic flow prediction; The neural network model comprises an input layer, two GCN models and one GRU model; the input layer is used for inputting the traffic feature matrix X , a related pollutant concentration attribute matrix K and an adjacency matrix A The GCN model is used for capturing the spatial features of traffic flow data between road segment nodes, and the GRU model is used for capturing the time features of traffic flow data and air pollutant concentration between road segment nodes.

2. The method of claim 1, wherein, The method for establishing the adjacency matrix A representing the road network connectivity according to the road network topology information comprises: modeling the road network topology information to obtain a road network topology graph G = (V, E) representing the connection relationship between roads, an adjacency matrix A representing the road network connectivity, and a mapping function ; wherein V represents a road segment set, and E represents a set of edges representing the connectivity between two road segments.

3. The method of claim 2, wherein, Establishing a traffic feature matrix X based on the historical traffic flow time-series dataset includes: establishing a traffic feature matrix from historical traffic flow time-series data from time tT to time t. , , ;in, express Traffic flow on n road segments at a given time. Indicates that the i-th road segment is in Traffic flow at any given moment.

4. The method of claim 2, wherein, The concentration matrix Y of the target pollutants is established according to the air pollutant concentration data set, and the concentration matrix Y of the target pollutants is established by the target pollutant concentration data from t-T time to t time , , ; wherein, represents the pollutant concentration of the n road sections at t time, represents the pollutant concentration of the i-th road section at t time. ​ 5. The method of claim 1, wherein, the target pollutant comprises one of carbon monoxide, particulate matter, sulfur dioxide, nitrogen dioxide and ozone.

6. The method of claim 1, wherein, The clustering is K-means clustering.

7. The method of claim 1, wherein, The calculation formula of the loss function is: ； wherein, represents a true value at time t, represents a predicted value at time t, represents L2 regularization, represents a hyperparameter that controls the regularization rate.

8. A traffic flow prediction device characterized by comprising: one or more processors and a memory; the memory is coupled with the one or more processors, the memory is used to store computer program code, the computer program code comprises computer instructions, and the one or more processors invoke the computer instructions to implement the traffic flow prediction method fusing air pollution factors according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, instructions stored in the memory, the instructions are executed by the processor to implement the traffic flow prediction method fusing air pollution factors according to any one of claims 1-7.

Images