Highway traffic state estimation method and system fusing traffic flow model and graph neural network

Abstract

The application discloses a highway traffic state estimation method and system fusing a traffic flow model and a graph neural network, and the method comprises the following steps: discretizing a highway into cells with a fixed length, and representing the traffic state of each cell by three parameters of flow, density and speed; determining the active influence area of the cell in a time period and the area to be affected by using the traffic wave speed between any two cells, and establishing a cell traffic state space-time evolution model; converting the discretized cell traffic state and adjacent information into graph structure data, and designing a graph neural network to update the traffic state of all cells; and based on the updated cell traffic state, using the gradient descent method to optimize the cell traffic state space-time evolution model, and obtaining the highway traffic state estimation result. The application can effectively improve the explainability, accuracy and generalization ability of the highway traffic flow space-time evolution modeling, and accurately grasp the highway traffic flow space-time evolution mechanism.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of traffic control, and relates to highway traffic state estimation, in particular to a highway traffic state estimation method and system fusing a traffic flow model and a graph neural network. BACKGROUND

[0002] A precise and reliable macroscopic traffic flow spatiotemporal evolution model is crucial for modeling and explaining traffic flow phenomena, estimating and predicting traffic states, deducing congestion propagation, optimizing ramp control, and managing highway emergencies. An effective traffic flow model should have three key characteristics: (1) strong interpretability in capturing traffic flow spatiotemporal evolution; (2) high accuracy in estimating and predicting key traffic state variables such as flow, speed, and density; (3) universal applicability under various traffic conditions.

[0003] The key to building a reliable highway macroscopic traffic flow model lies in a modeling method with a solid theoretical foundation. Existing macroscopic traffic flow modeling methods can be broadly divided into three categories: (1) models based on physical principles; (2) models based on machine learning; (3) hybrid models. Models based on physical principles excel in describing traffic state evolution, capturing a wide range of traffic flow phenomena, and estimating traffic states when historical data is scarce. Historically, researchers have drawn analogies between highway traffic flow and fluid dynamics, thereby deriving macroscopic models based on fluid dynamics principles. These models conceptualize highway traffic as a continuous flow with specific properties and use aggregate variables such as flow, average speed, and density to model traffic operation at a macroscopic level, which facilitates the derivation of mathematical formulas for traffic dynamics. Models based on machine learning, on the other hand, start from the data itself and use the data mining capabilities of machine learning to extract spatiotemporal correlations between traffic flow parameters for traffic state estimation. Models based on machine learning perform excellently when there is sufficient training data, but they have high requirements for the quality and quantity of training data.

[0004] Although there are currently limited hybrid models that explore how to combine traffic flow models and machine learning-based models, the main focus is on exploring the loss function of machine learning-based models. In this case, the results obtained by the model conform to certain traffic flow physical models, but the interpretability of the model itself is poor. SUMMARY

[0005] Purpose of the invention: To overcome the shortcomings of existing technologies, this invention provides a method and system for estimating highway traffic state by integrating traffic flow models and graph neural networks. This method can effectively improve the interpretability, accuracy, and generalization ability of highway traffic flow spatiotemporal evolution modeling, accurately grasp the mechanism of highway traffic flow spatiotemporal evolution, and help traffic design and management departments optimize highway design and renovation plans, formulate reasonable and effective real-time control and guidance strategies, thus having practical application value for alleviating highway traffic congestion.

[0006] Technical Solution: To achieve the above objectives, this invention provides a method for highway traffic state estimation that integrates traffic flow models and graph neural networks, comprising the following steps:

[0007] S1: Discretize the highway into cells of fixed length, and represent the traffic state of each cell through three parameters: flow rate, density, and speed;

[0008] S2: Use the traffic wave velocity between any two cells to determine the active influence area and the area that will be affected by a cell within a time period;

[0009] S3: Transform the discrete cellular traffic state and adjacency information into graph structure data, design a graph neural network to update the traffic state of all cells, and establish a spatiotemporal evolution model of cellular traffic state.

[0010] S4: Based on the updated cellular traffic state, the gradient descent method is used to optimize the spatiotemporal evolution model of the cellular traffic state to obtain the highway traffic state estimation results.

[0011] Furthermore, in step S1, the main highway is cut into sections of equal length according to the cellular automata model. The cell, whose number is ; by time interval The vehicle movement trajectories on the road segment are divided into continuous time series, and the traffic state of each cell in different time periods is calculated based on the vehicle trajectory information in each cell. including density ,speed and traffic .

[0012] Further, step S2 specifically includes:

[0013] For any cell Utilizing the density of adjacent cells and traffic Calculate the direction of traffic wave propagation between the cell and the upstream and downstream cells. , and speed , Based on this, the upstream and downstream spatial influence range of each cell within a time period is determined. and Thus, the corresponding number of upstream and downstream influencing cells can be obtained. and The corresponding upstream influence cell region is The downstream affected area is ;

[0014] For any cell traversing cells If cells Downstream affected areas Contains cells Then the cell was found. The farthest upstream cell Thus, the influence of cells was obtained. Number of upstream cells The traversal stops; for any cell traversing cells If cells upstream influence area Contains cells Then the cell was found. The downstream farthest impact cell Thus, the influence of cells was obtained. downstream cell number The traversal stops;

[0015] For any cell Number of upstream cells and the number of downstream cells Calculate the maximum number of cells that generate influence corresponding to the cell. Iterate through the number of upstream and downstream influencing cells of all cells, and determine the number of the farthest influencing cells among all cells. .

[0016] Furthermore, in step S3, for any given cell, the traffic state of that cell is updated using the traffic states of all cells that affect that cell within a certain time period. This specifically includes:

[0017] Combining methods for updating density, velocity, and flow using cellular automata, a hybrid model with a cyclic traffic state update structure is proposed. ; through the longest number of impacts Determine the number of times to update in a loop;

[0018] In an update In the first update, the traffic status of all cells observed at the previous time step will be updated. As input, subsequent updates utilize the previous update. Output As for this update The input and output are Among them, the updated output Depend on , and Composition; during the update process, for each cell Utilizing the cell's own traffic state Adjacent cell traffic status and Traffic wave speed and Perform cell Calculation of upstream and downstream traffic state changes: and Then utilize the cell's own traffic state. cell Changes in upstream and downstream traffic conditions and Perform cell state update ;

[0019] After completing the update, iterate through all cells, and if the current count is... Reached cell The longest generation affects the number of cells Then the cell is completed. Traffic status update .

[0020] Furthermore, the specific method by which the graph neural network updates the traffic state of all cells in step S3 is as follows:

[0021] We propose a graph neural network that conforms to the upstream and downstream adjacency relationships of cells, transforming the traffic states and adjacency relationships of cells on a road segment into a graph structure. Furthermore, a message passing and update network consistent with the spatiotemporal transition of traffic states is constructed within a graph neural network. Based on the logic of traffic state transition, the hidden spatiotemporal correlations within the data are extracted, thereby completing the modeling of traffic flow evolution within a time period and the estimation of cell states for the next time period. ;

[0022] Each update involves updating the traffic status of all cells observed at the previous time step. As input, subsequent updates utilize the previous update. Output As for this update Input; in updating In the method, a cell connection relationship matrix is initialized first wherein represents a directed edge of a cell connection relationship, wherein represents the number of directed connection edges of all cells; and respectively represent the starting cell and the arrival cell of the directed edge . represents the traffic wave speed of the directed edge . represents the traffic state transition amount of the directed edge . Then, the traffic wave speed and the traffic state transition amount of each cell connection directed edge in the cell connection relationship matrix are calculated, and the expression of the traffic wave speed of the directed edge is as follows:

[0023]

[0024] The expression of the traffic state transition amount of each cell connection directed edge is as follows:

[0025]

[0026] wherein , , , , and are parameter matrices of a three-layer fully connected neural network;

[0027] Finally, the connection relationship matrix is converted into a matrix with a dimension of , and the expression of the state update calculation process of all cells is as follows:

[0028]

[0029] wherein represents matrix point multiplication; and are parameter matrices; represents a matrix full of 1s;

[0030] After the update is completed, all cells are traversed, and if the current number reaches the longest number of cells that have an impact on the cell , the traffic state update of the cell is completed.

[0031] Furthermore, step S4 optimizes the spatiotemporal evolution model of cellular traffic states using gradient descent, specifically including:

[0032] First, utilize the observed state values ​​of all cells. Calculate the average values ​​of flow rate, density, and velocity. , , and standard deviation , , Then for The estimated and observed values ​​for the time period are normalized, and the normalization calculation is as follows:

[0033]

[0034]

[0035]

[0036]

[0037]

[0038]

[0039] in, express Traffic flow estimates for a given time period; express Density estimates for the time period; express Speed ​​estimates for a given time period; express Traffic flow observations over a given time period; express Density observations over a time period; express Velocity observations over a time period;

[0040] The mean absolute error is calculated using the normalized traffic state values. Calculations are performed, and based on the obtained error, the gradient descent method is used to model the problem. The parameters are updated until the error is reduced to a stable level, at which point the optimal spatiotemporal evolution model of traffic conditions is obtained.

[0041] This invention also provides a highway traffic state estimation system that integrates traffic flow models and graph neural networks, comprising:

[0042] The discretization module is used to discretize the highway into cells of fixed length, and the traffic state of each cell is represented by three parameters: flow rate, density, and speed.

[0043] a model establishing module, configured to determine an active influence area of a cell and an area to be influenced within a time period by using traffic wave speed between any two cells, and establish a cell traffic state space evolution model;

[0044] a state updating module, configured to convert the discretized cell traffic state and adjacency information into graph structure data, and design a graph neural network to update all cell traffic states;

[0045] a model optimization module, configured to optimize the cell traffic state space evolution model by using a gradient descent method, and obtain a highway traffic state estimation result.

[0046] The application further provides a computer storage medium storing a program of a highway traffic state estimation method fusing a traffic flow model and a graph neural network, the program of the highway traffic state estimation method fusing the traffic flow model and the graph neural network being executed by at least one processor to implement steps of the highway traffic state estimation method fusing the traffic flow model and the graph neural network.

[0047] The application constructs a theoretical model based on a cellular automaton model, and improves model interpretability. The purpose of interpretability is to make the model more understandable, have explicit parameter relationships and change processes, and have a trend description of traffic flow evolution process, so as to avoid the influence of a small amount of training set data, cause model overfitting, and guide the model to learn in accordance with the trend.

[0048] Meanwhile, the discretized cell traffic state and adjacency information are converted into graph structure data, a message passing and updating mechanism of the graph neural network is designed to complete calculation and design of the theoretical model, and errors between a traffic state estimated by the model at the next moment and an observed traffic state are used to adjust and optimize model parameters.

[0049] The application introduces a nonlinear modeling of the graph neural network, better mines all cell hidden space-time relationships in accordance with traffic flow, and improves model accuracy and generalization ability. The flow, density and speed of training data are used to determine mean values and standard deviations of corresponding traffic flow parameters, normalize the traffic state estimated by the model at the next moment and the observed traffic state, calculate mean absolute errors of the normalized estimated value and the observed value, and adjust and optimize model parameters, so as to eliminate imbalance caused by numerical values of traffic flow parameters.

[0050] Beneficial effects: Compared with existing technologies, this invention can effectively improve the interpretability, accuracy and generalization ability of highway traffic flow spatiotemporal evolution modeling, accurately grasp the mechanism of highway traffic flow spatiotemporal evolution, help traffic design and management departments optimize highway design and renovation plans, formulate reasonable and effective real-time control and guidance strategies, and has practical application value for alleviating highway traffic congestion. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the framework of the present invention;

[0052] Figure 2 This is a flowchart for quantifying the cell influence area within a unit time period. Detailed Implementation

[0053] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0054] Example 1:

[0055] like Figure 1 As shown in the figure, this embodiment provides a method for highway traffic state estimation that integrates traffic flow models and graph neural networks, including the following steps:

[0056] S1: Discretize the highway into cells of fixed length, and represent the traffic state of each cell through three parameters: flow rate, density, and speed;

[0057] In this embodiment, the highway main road is cut into equal lengths based on the cellular automata model. The cell, whose number is ; by time interval The vehicle movement trajectories on the road segment are divided into continuous time series, and the traffic state of each cell in different time periods is calculated based on the vehicle trajectory information in each cell. including density ,speed and traffic A dataset is constructed, with 80% consisting of the training dataset and 20% consisting of the test dataset.

[0058] S2: Use the traffic wave velocity between any two cells to determine the active influence area and the area that will be affected by a cell within a time period;

[0059] Reference Figure 2 For any cell Utilizing the density of adjacent cells and traffic Calculate the direction of traffic wave propagation between the cell and the upstream and downstream cells. , and speed , Based on this, the upstream and downstream spatial influence range of each cell within a time period is determined. and Thus, the corresponding number of upstream and downstream influencing cells can be obtained. and The corresponding upstream influence cell region is The downstream affected area is ;

[0060] For any cell traversing cells If cells Downstream affected areas Contains cells Then the cell was found. The farthest upstream cell Thus, the influence of cells was obtained. Number of upstream cells The traversal stops; for any cell traversing cells If cells upstream influence area Contains cells Then the cell was found. The downstream farthest impact cell Thus, the influence of cells was obtained. downstream cell number The traversal stops;

[0061] For any cell Number of upstream cells and the number of downstream cells Calculate the maximum number of cells that generate influence corresponding to the cell. Iterate through the number of upstream and downstream influencing cells of all cells, and determine the number of the farthest influencing cells among all cells. .

[0062] S3: Based on the cell region determination method in step S2, the discrete cell traffic state and adjacency information are transformed into graph structure data. A graph neural network is designed to update the traffic state of all cells and establish a spatiotemporal evolution model of cell traffic state.

[0063] For any given cell, update its traffic state by utilizing the traffic states of all cells that affect it within a given time period. This includes:

[0064] A hybrid model with cyclic traffic state update structure is proposed based on the method of updating density, speed and flow in cellular automaton ; by the maximum number of influences Determine the number of cyclic updates;

[0065] In one update , if it is the first update, the observed traffic state of all cells at the last time is taken as input, and for subsequent updates, the output of the last update is taken as input , and the output is ; wherein the output of the update is composed of , , and ; during the update process, for each cell , the traffic state of the cell itself , the traffic state of the adjacent cell and , the traffic wave speed and are used to calculate the upstream and downstream traffic state change of the cell : and ; then the traffic state of the cell itself , the upstream and downstream traffic state change of the cell and are used to update the cell state ;

[0066] After the update is completed, all cells are traversed, and if the current number reaches the maximum number of influencing cells of the cell , the traffic state update of the cell is completed .

[0067] The specific way of updating the traffic state of all cells by graph neural network is as follows:

[0068] A graph neural network consistent with the upstream and downstream adjacency relationship of cells is proposed, which converts the traffic state of cells on the road segment and the adjacency relationship into a graph structure , and constructs a message passing and updating network consistent with the spatiotemporal transfer of traffic state in the graph neural network; on the basis of conforming to the traffic state transfer logic, the hidden spatiotemporal correlation in the data is extracted, thereby completing the modeling of traffic flow evolution in a time period and the estimation of cell state in the next time period​ ;

[0069] Each time update, the traffic state of all cells observed at the last time is used as input, for the subsequent update, the output of the last update is used as input for this update; in the update , first initialize a cell connection relationship matrix , where represents a directed edge of a cell connection relationship, where represents the number of directed connection edges of all cells; and respectively represent the departure cell and the arrival cell of the directed edge ; v represents the traffic wave speed of the directed edge ; s represents the traffic state transition amount of the directed edge ; then calculate the traffic wave speed and the traffic state transition amount of each cell connection directed edge in the cell connection relationship matrix, the expression of the traffic wave speed of the directed edge is:

[0070]

[0071] The expression of the traffic state transition amount of each cell connection directed edge is:

[0072]

[0073] where , , , , and are the parameter matrices of the three-layer fully connected neural network;

[0074] Finally, the connection relationship matrix is converted into a matrix with a dimension of , and the expression of the state update calculation process of all cells is as follows:

[0075]

[0076] where represents matrix multiplication; and are parameter matrices; represents a matrix full of 1;

[0077] After completing the update​​ Then, iterate through all cells, if the current count is... Reached cell The longest generation affects the number of cells Then the cell is completed. Traffic status update .

[0078] S4: Based on the updated cellular traffic state, the gradient descent method is used to optimize the spatiotemporal evolution model of the cellular traffic state to obtain the highway traffic state estimation results.

[0079] First, utilize the observed state values ​​of all cells. Calculate the average values ​​of flow rate, density, and velocity. , , and standard deviation , , Then for The estimated and observed values ​​for the time period are normalized, and the normalization calculation is as follows:

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] in, express Traffic flow estimates for a given time period; express Density estimates for the time period; express Speed ​​estimates for a given time period; express Traffic flow observations over a given time period; express Density observations over a time period; express Velocity observations over a time period;

[0087] The mean absolute error is calculated using the normalized traffic state values. Calculations are performed, and based on the obtained error, the gradient descent method is used to model the problem. The parameter update is performed until the error is reduced to stable, and the optimal traffic state space evolution model is obtained.

[0088] Embodiment 2:

[0089] Based on the method of embodiment 1, the embodiment provides a highway traffic state estimation system fusing a traffic flow model and a graph neural network, comprising:

[0090] A discretization module is configured to discretize the highway into cells with a fixed length, and represent the traffic state of each cell by three parameters of flow, density and speed;

[0091] A model establishment module is configured to determine the active influence area of the cell in a time period and the area to be influenced by using the traffic wave speed between any two cells, and establish a cell traffic state space evolution model;

[0092] A state update module is configured to convert the discretized cell traffic state and adjacency information into graph structure data, and design a graph neural network to update the traffic state of all cells;

[0093] A model optimization module is configured to optimize the cell traffic state space evolution model by using a gradient descent method, and obtain the highway traffic state estimation result.

[0094] The embodiment also provides a computer storage medium storing a computer program, and the computer program can implement the above-described method when a processor executes the computer program. The computer readable medium can be considered as tangible and non-transitory. Non-limiting examples of the non-transitory tangible computer readable medium include non-volatile memory circuits (such as flash memory circuits, erasable programmable read-only memory circuits or mask read-only memory circuits), volatile memory circuits (such as static random access memory circuits or dynamic random access memory circuits), magnetic storage media (such as analog or digital magnetic tapes or hard disk drives), and optical storage media (such as CDs, DVDs or Blu-ray discs), etc. The computer program includes processor executable instructions stored on at least one non-transitory tangible computer readable medium. The computer program can also include or depend on stored data. The computer program can include a basic input / output system (BIOS) interacting with hardware of a special computer, a device driver interacting with a specific device of the special computer, one or more operating systems, user application programs, background services, background application programs, etc.

[0095] Those skilled in the art will appreciate that embodiments of the application can be readily used as a method, a system or a computer program product. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code thereon.

[0096] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. Figure 1 The flowchart and / or block diagram in the figure(s) illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to the present application. In this regard, each block in the flowchart and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable Figure 1 The flowchart and / or block diagram in the figure(s) illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to the present application. In this regard, each block in the flowchart and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable

[0097] Embodiment 3

[0098] In order to prove the effectiveness of the present application, this embodiment uses trajectory data collected by millimeter wave radar on Chengdu Expressway for case analysis, as follows:

[0099] Trajectory data collection: using roadside millimeter wave radar as data collection equipment, using two sections of Chengdu Expressway as target areas, two millimeter wave radars covering a detection area of 525 meters each, and using 5 days of collected trajectory data as experimental data source;

[0100] The length of the two target sections is 525 meters, and the spatial interval is 25 meters Discretize the road section to form 21 cells. Take 5 seconds as the time interval Divide the trajectory data. The flow, density and speed of each cell in each time interval are calculated by the following model:

[0101]

[0102]

[0103]

[0104] After obtaining the flow, density and speed of each cell in each time period, two experiments are designed to verify the accuracy of the method. In experiment 1, the traffic flow parameter data of the first 4 days of road section 1 is used as the training data set, and the data of the 5th day of road section 1 is used as the test data set. In experiment 2, the traffic flow parameter data of the first 4 days of road section 1 is used as the training data set, and the data of the 5th day of road section 2 is used as the test data set.

[0105] Table 1 shows the comparison results of the traditional cell transmission model (CTM), gated recurrent unit (GRU) model, graph convolution network (GCN) model and the proposed hybrid model in free flow state and congestion state. In free flow state and congestion state, the hybrid model performs well in all evaluation indicators.

[0106] In free flow state, the hybrid model has the lowest mean absolute error (MAE): 4.29 km / h for speed, 2.11 vehicles / km / lane for density, and 127.16 vehicles / hour / lane for flow. It also has the lowest mean absolute percentage error (MAPE) in speed, density and flow, which are 5.87%, 17.67% and 14.82% respectively. In terms of speed estimation, compared with the traditional CT model, GRU model and GCN model, the MAE of the hybrid model is reduced by 34.8%, 42.8% and 17.7% respectively, and the MAPE value is reduced by 1.99%, 3.24% and 0.72% respectively. In terms of density estimation, the MAE of the hybrid model is reduced by 23.6%, 31.7% and 8.3% respectively, and the MAPE value is reduced by 6.81%, 8.27% and 2.12% respectively. In terms of flow estimation, the MAE of the hybrid model is reduced by 26.2%, 33.3% and 4.4% respectively, and the MAPE value is reduced by 7.06%, 7.55% and 1.39% respectively.

[0107] In the congested state, the hybrid model also performs well in all evaluation indicators. The MAE of the hybrid model is the lowest: 1.66 km / h for speed, 3.48 vehicles / km / lane for density, and 84.43 vehicles / hour / lane for flow. It also has the smallest MAPE in speed, density, and flow, which are 14.66%, 6.13%, and 9.97%, respectively. In terms of speed estimation, compared with the traditional CTM model, GRU model, and GCN model, the MAE of the hybrid model is reduced by 23.1%, 25.9%, and 10.3%, respectively, and the MAPE value is reduced by 8.03%, 12.68%, and 1.58%, respectively. In terms of density estimation, compared with the three baseline models, the MAE of the hybrid model is reduced by 41.3%, 38.3%, and 7.2%, respectively, and the MAPE value is reduced by 1.98%, 1.71%, and 0.5%, respectively. In terms of flow estimation, the MAE of the hybrid model is reduced by 24.8%, 30.2%, and 9.4%, respectively, and the MAPE value is reduced by 7.98%, 12.17%, and 3.15%, respectively.

[0108] Table 1. Comparison of the hybrid model and the base model in Experiment 1

[0109]

[0110] Table 2 shows the comparative results of the traditional traffic flow model (CTM), the gated recurrent unit (GRU) model, the graph convolutional network (GCN) model, and the proposed hybrid model in free-flow and congested states, respectively. In free-flow and congested states, the hybrid model performs well in all evaluation indicators. In the free-flow state, the hybrid model has the lowest mean absolute error (MAE): 4.29 km / h for speed, 2.11 vehicles / km / lane for density, and 127.16 vehicles / hour / lane for flow. It also achieves the lowest mean absolute percentage error (MAPE) in speed, density, and flow, which are 5.87%, 17.67%, and 14.82%, respectively. In terms of speed estimation, compared with the traditional CTM, GRU model, and GCN model, the hybrid model reduces the MAE by 34.8%, 42.8%, and 17.7%, respectively, and the MAPE by 1.99%, 3.24%, and 0.72%, respectively. In terms of density estimation, the hybrid model reduces the MAE by 23.6%, 31.7%, and 8.3%, respectively, and the MAPE by 6.81%, 8.27%, and 2.12%, respectively. In terms of flow estimation, the hybrid model reduces the MAE by 26.2%, 33.3%, and 4.4%, respectively, and the MAPE by 7.06%, 7.55%, and 1.39%, respectively.

[0111] In the congestion state, the hybrid model exhibits excellent performance in all evaluation indicators. The hybrid model has the lowest MAE: 1.66 km / h for speed, 3.48 vehicles / km / lane for density, and 84.43 vehicles / hour / lane for flow. It also has the lowest MAPE for speed, density, and flow, which are 14.66%, 6.13%, and 9.97%, respectively. In terms of speed estimation, compared with the traditional CTM, GRU model, and GCN model, the hybrid model reduces the MAE by 23.1%, 25.9%, and 10.3%, respectively, and reduces the MAPE by 8.03%, 12.68%, and 1.58%, respectively. In terms of density estimation, compared with the three baseline models, the hybrid model reduces the MAE by 41.3%, 38.3%, and 7.2%, respectively, and reduces the MAPE by 1.98%, 1.71%, and 0.5%, respectively. In terms of flow estimation, the hybrid model reduces the MAE by 24.8%, 30.2%, and 9.4%, respectively, and reduces the MAPE by 7.98%, 12.17%, and 3.15%, respectively. The results show that integrating machine learning methods into the spatiotemporal discretized traffic flow model can improve the estimation accuracy under different traffic conditions.

[0112] Table 2. Comparison of the hybrid model and the base model in Experiment 2

[0113]

Claims

1. A method for highway traffic state estimation that integrates traffic flow models and graph neural networks, characterized in that, Includes the following steps: S1: Discretize the highway into cells of fixed length, and represent the traffic state of each cell through three parameters: flow rate, density, and speed; S2: Use the traffic wave velocity between any two cells to determine the active influence area and the area that will be affected by a cell within a time period; S3: Transform the discrete cellular traffic state and adjacency information into graph structure data, design a graph neural network to update the traffic state of all cells, and establish a spatiotemporal evolution model of cellular traffic state. S4: Based on the updated cellular traffic state, the gradient descent method is used to optimize the spatiotemporal evolution model of the cellular traffic state to obtain the highway traffic state estimation results. In step S1, the main highway is cut into equal lengths according to the cellular automata model. The cell, whose number is ; by time interval The vehicle movement trajectories on the road segment are divided into continuous time series, and the traffic state of each cell in different time periods is calculated based on the vehicle trajectory information in each cell. including density ,speed and traffic ; Step S2 specifically includes: For any cell Utilizing the density of adjacent cells and traffic Calculate the direction of traffic wave propagation between the cell and the upstream and downstream cells. , and speed , Based on this, the upstream and downstream spatial influence range of each cell within a time period is determined. and Thus, the corresponding number of upstream and downstream influencing cells can be obtained. and The corresponding upstream influence cell region is The downstream affected area is ; For any cell traversing cells If cells Downstream affected areas Contains cells Then the cell was found. The farthest upstream cell Thus, the influence of cells was obtained. Number of upstream cells The traversal stops; for any cell traversing cells If cells upstream influence area Contains cells Then the cell was found. The downstream farthest impact cell Thus, the influence of cells was obtained. downstream cell number The traversal stops; For any cell Number of upstream cells and the number of downstream cells Calculate the maximum number of cells that generate influence corresponding to the cell. Iterate through the number of upstream and downstream influencing cells of all cells, and determine the number of the farthest influencing cells among all cells. .

2. The highway traffic state estimation method integrating traffic flow model and graph neural network according to claim 1, characterized in that, In step S3, for any given cell, the traffic state of that cell is updated using the traffic states of all cells that affect that cell within a given time period. Specifically, this includes: Combining methods for updating density, velocity, and flow using cellular automata, a hybrid model with a cyclic traffic state update structure is proposed. ; through the longest number of impacts Determine the number of times to update in a loop; In an update In the first update, the traffic status of all cells observed at the previous time step will be updated. As input, subsequent updates utilize the previous update. Output As for this update The input and output are Among them, the updated output Depend on , and Composition; during the update process, for each cell Utilizing the cell's own traffic state Adjacent cell traffic status and Traffic wave speed and Perform cell Calculation of upstream and downstream traffic state changes: and Then utilize the cell's own traffic state. cell Changes in upstream and downstream traffic conditions and Perform cell state update ; After completing the update, iterate through all cells, and if the current count is... Reached cell The longest generation affects the number of cells Then the cell is completed. Traffic status update .

3. The method for highway traffic state estimation by fusing traffic flow model and graph neural network according to claim 2, characterized in that, The specific method by which the graph neural network updates the traffic state of all cells in step S3 is as follows: We propose a graph neural network that conforms to the upstream and downstream adjacency relationships of cells, transforming the traffic states and adjacency relationships of cells on a road segment into a graph structure. Furthermore, a message passing and update network consistent with the spatiotemporal transition of traffic states is constructed within a graph neural network. Based on the logic of traffic state transition, the hidden spatiotemporal correlations within the data are extracted, thereby completing the modeling of traffic flow evolution within a time period and the estimation of cell states for the next time period. ; Each update involves updating the traffic status of all cells observed at the previous time step. As input, subsequent updates utilize the previous update. Output As for this update Input; in updating First, a cell connection matrix is ​​initialized. ,in A directed edge represents a cell connection. ,in This represents the number of directed edges connecting all cells; and They represent directed edges respectively. The originating cell and the arriving cell; Represents a directed edge Traffic wave speed; Represents a directed edge The traffic state transition amount is calculated; then, the traffic wave velocity and traffic state transition amount of the directed edges connected to each cell in the cell connection matrix are calculated. The expression for the traffic wave velocity of the directed edges is: ； The expression for the traffic state transition of each cell connected to a directed edge is: ； in , , , , and This is the parameter matrix of a three-layer fully connected neural network; Finally, the connection matrix will be... Transform into a dimension matrix The expression for the state update calculation process of all cells is as follows: ； in Represents matrix dot product; and For parameter matrices; Represents a matrix consisting entirely of 1s; After the update is completed Then, iterate through all cells, if the current count is... Reached cell The longest generation affects the number of cells Then the cell is completed. Traffic status update .

4. The highway traffic state estimation method integrating traffic flow model and graph neural network according to claim 3, characterized in that, Step S4 optimizes the spatiotemporal evolution model of cellular traffic states using gradient descent, specifically including: First, utilize the observed state values ​​of all cells. Calculate the average values ​​of flow rate, density, and velocity. , , and standard deviation , , Then for The estimated and observed values ​​for the time period are normalized, and the normalization calculation is as follows: ； ； ； ； ； ； in, express Traffic flow estimates for a given time period; express Density estimates for the time period; express Speed ​​estimates for a given time period; express Traffic flow observations over a given time period; express Density observations over a time period; express Velocity observations over a time period; The mean absolute error is calculated using the normalized traffic state values. Calculations are performed, and based on the obtained error, the gradient descent method is used to model the problem. The parameters are updated until the error is reduced to a stable level, at which point the optimal spatiotemporal evolution model of traffic conditions is obtained.

5. A highway traffic state estimation system integrating traffic flow model and graph neural network, characterized in that, For implementing the method of claim 1, the system comprises: The discretization module is used to discretize the highway into cells of fixed length, and the traffic state of each cell is represented by three parameters: flow rate, density, and speed. The model building module is used to determine the active influence area and the area that will be affected by the cell within a time period by using the traffic wave velocity between any two cells, and to build a spatiotemporal evolution model of the cell traffic state. The state update module is used to transform the traffic state and adjacency information of discrete cells into graph structure data, and a graph neural network is designed to update the traffic state of all cells. The model optimization module is used to optimize the spatiotemporal evolution model of cellular traffic state using the gradient descent method to obtain highway traffic state estimation results.

6. A computer storage medium, characterized in that: The computer storage medium stores a program for a highway traffic state estimation method that integrates a traffic flow model and a graph neural network. When the program for a highway traffic state estimation method that integrates a traffic flow model and a graph neural network is executed by at least one processor, it implements the steps of the highway traffic state estimation method that integrates a traffic flow model and a graph neural network as described in any one of claims 1 to 4.

Images