Method for constructing macroscopic evolution model of mixed traffic flow based on vehicle microscopic information

Abstract

The present invention belongs to the field of intelligent transportation and relates to a method for constructing a macro-evolution law model for mixed traffic flow based on vehicle micro-information. The method comprises: S1. dividing mixed traffic flow in bottleneck sections within highway tunnels into vehicle groups and selecting modeling parameters; S2. layering and mapping vehicle data collected in a networked environment to form a multidimensional feature matrix; and S3. extracting the vehicle state feature matrix using a convolutional neural network to establish a macro-evolution law model for traffic flow based on vehicle micro-information. The present invention addresses the problem that single micro- or macro-modeling cannot accurately and comprehensively depict traffic flow evolution. By leveraging the advantages of neural network models in processing spatial matrix data, the method proposes and establishes a law model that effectively reduces prediction errors for vehicle volume in bottleneck sections.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent transportation, and in particular relates to a method for constructing a macroscopic evolution law model of mixed traffic flow based on vehicle microscopic information. Background Art

[0002] In recent years, the rapid evolution of perception and communication technologies has led to significant progress in intelligent connected vehicle technology. "Made in China 2025" documents the positive impact of intelligent driving technology on road traffic. However, the full upgrade from human-driven vehicles to connected automated vehicles requires a long period of technological accumulation and practical optimization. Road traffic patterns will gradually shift and remain in a mixed traffic flow pattern consisting of connected automated vehicles (CAVs) and connected human vehicles (CHVs). Road bottlenecks within highway tunnels will also exist in this traffic system. While the interaction of heterogeneous vehicles in this mixed traffic pattern increases the complexity of the traffic system, it also provides new solutions to traffic problems.

[0003] Macroscopic traffic models analyze the macroscopic behavior of traffic flow from the perspective of vehicle groups. Compared to microscopic traffic models, they offer advantages such as lower computational complexity, more comprehensive characterization, and stronger systematicity. Macroscopic traffic flow models are primarily used to characterize the macroscopic evolution of traffic flow within a road section, such as volume, average speed, and spatial distribution. Under new mixed traffic patterns, macroscopic traffic models consider more intelligent factors and the coupling relationship between vehicle groups. Partial differential equations are often used to characterize the collective motion of traffic. Within the research field of traffic dynamics, a variety of macroscopic models exist, including the Lighthill-Whitham-Richards model (LWR) based on fluid dynamics theory, lattice models based on discrete space, cellular automata models that simulate individual behavior, and the Cell Transmission Model (CTM) for simulating vehicle transmission processes. Markos Papageorgiou considered the influence of CV penetration in a second-order macroscopic METANET model, achieving predictions of the total density and flow of traffic in mixed traffic systems and providing stable operating conditions for the model. Shubin Li et al., in their study of the design of variable speed limit schemes for highway traffic efficiency, simplified system complexity and reduced control delay by loading predictive information variables and introducing a macro-level model. Tuerpraser et al. proposed a new multi-class cellular transmission model, expanding the application scope of traditional models and enabling better simulation and prediction of traffic flow in complex road networks with different types of vehicles. Jin et al. constructed a macro-model of lane change for LWRs that integrates multiple vehicle types, considering different lane-changing tendencies and difficulties. They also created a basic parameter graph that comprehensively considered the influence of vehicle following behavior, lane-changing characteristics, road layout, and traffic composition. Qin Yanyan proposed a basic graph model that describes mixed traffic flows consisting of vehicles equipped with cooperative adaptive cruise control (CACC), vehicles with only adaptive cruise control (ACC), and human-driven vehicles. They analyzed the impact of connected autonomous driving technology on the capacity of mixed traffic flows through experimental simulations. Zhanbo Sun et al. addressed the coordination problem of ramp merging in mixed traffic flows. Based on the evolution model of the traffic flow macro-model, they developed a two-level optimization method based on state-constrained optimal control to efficiently implement dynamic planning of vehicle trajectories. Macro traffic modeling research has broad application prospects in vehicle group route planning, system traffic state prediction, etc. In new mixed traffic modes, macro traffic models are conducive to formulating more systematic and efficient traffic control strategies. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for constructing a macro-evolution law model of mixed traffic flow based on vehicle micro-information.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for constructing a macroscopic evolution law model of mixed traffic flow based on vehicle microscopic information includes the following steps:

[0007] S1. Segment mixed traffic flows in a bottleneck section within a highway tunnel and select modeling parameters.

[0008] S2. Perform hierarchical mapping on the vehicle data collected in the connected environment to form a multi-dimensional feature matrix;

[0009] S3. Use convolutional neural networks to extract vehicle state feature matrices and establish a macroscopic traffic flow evolution model based on vehicle microscopic information.

[0010] Furthermore, in step S1, the mixed traffic flow is divided into a lane-changing conflict vehicle group and a following conflict vehicle group;

[0011] Following conflicting vehicles: A no-entry sign is set up 200 meters upstream of the bottleneck. The vehicles in the left lane downstream of the no-entry sign with only one passable lane are called following conflicting vehicles.

[0012] Lane-changing conflict vehicle group: The vehicle group from the upstream of the no-entry sign to the tunnel entrance lane is called a lane-changing conflict vehicle group.

[0013] Furthermore, the modeling parameters include vehicle speed v, acceleration a, road position x, vehicle category p, upstream vehicle departure volume, CAV penetration rate, number of emergency brakes, average traffic speed, and traffic volume;

[0014] Among them, the upstream departure volume is the number of vehicles entering the bottleneck area of ​​the road in the target scenario within one hour; the traffic flow is the number of vehicles passing through the bottleneck within one hour.

[0015] Furthermore, in step S2, the vehicle data is processed by an image digitization method to perform layered mapping, and the continuous analog image is converted into a discrete digital image form through sampling and quantization steps.

[0016] Furthermore, step S2 includes the following sub-steps:

[0017] S2.1 Image sampling;

[0018] The microscopic information of vehicles on the target road section is collected through intelligent roadside equipment or vehicle communication systems. The lane area where the mixed vehicle group is located is regarded as an image and spatially divided into a c×m grid array.

[0019] Among them, a single grid represents a pixel point of the image, and m×n defines the spatial resolution of the image;

[0020] S2.2 Image quantification;

[0021] The road area where the mixed vehicle group is located is spatially divided into grids. A mapping layer is established for each vehicle status information. Vehicle information is quantitatively filled into the corresponding mapping layer according to the vehicle's relative spatial position, completing the spatial representation of the vehicle microscopic information in the mixed traffic flow.

[0022] S2.3 construct a multidimensional feature matrix;

[0023] A road space S of length L is selected upstream of the bottleneck section inside the highway tunnel. This road space S is treated as an image and gridded into a C×M grid. C is the number of lanes, and M is the number of cells divided into a single lane in the direction of vehicle travel.

[0024] Get vehicles n1,n2,…,n in the road space S k For each type of vehicle state information, a corresponding C×M state mapping layer is constructed. Then, each vehicle micro-state information is mapped to the corresponding cell according to its relative spatial position in the state mapping layer to form F state feature layers; where F is the number of vehicle micro-information states. The state feature layers are superimposed to construct a C×M×F multi-dimensional feature matrix.

[0025] S2.4 Hierarchical mapping of vehicle microscopic information;

[0026] I. Map each vehicle to the corresponding grid of each state feature layer according to its relative position in the road space S. There are vehicles n1, n2, ..., n in the road space S. k The starting and ending positions of the vehicles in each lane are expressed as follows:

[0027]

[0028]

[0029] Where m ia 、m ib They represent the starting unit and the ending unit of the vehicle insertion feature graph respectively; x i is the position of the vehicle head; l veh is the vehicle length; s0 is the starting position of the road space S; [·] indicates rounding up;

[0030] II. Complete the quantitative filling of vehicle status information, where the physical characteristic quantization value includes vehicle speed acceleration Lane and vehicle type The quantitative relationship is expressed as:

[0031]

[0032]

[0033]

[0034]

[0035] Where, v i Indicates vehicle n i The actual speed; v max Indicates the maximum speed limit of the road; a i For vehicle n i The actual acceleration of a max 、b max are the maximum acceleration and maximum deceleration of the vehicles in the group, respectively.

[0036] Furthermore, in step S2.2, the traffic flow information includes vehicle speed, acceleration, road position and vehicle category.

[0037] Furthermore, in step S2.3, the vehicle microscopic information includes the vehicle's own parameters: body length l and vehicle category p, where vehicle category p is used to distinguish between CHVs and CAVs; vehicle motion data: speed v and acceleration a; and vehicle position data: lane c and vehicle head position x.

[0038] Furthermore, in step S3, the traffic flow macro-evolution law model based on vehicle micro-information includes the following parts:

[0039] I. Model Input

[0040] The model inputs the vehicle microscopic information feature matrix. The expression of the model input X is as follows:

[0041] X=[M1,M2,M3,…M n ]

[0042] Where M i represents the C×M×F multi-dimensional feature matrix obtained after hierarchical mapping of the i-th mixed vehicle group data; n represents the number of samples used for CNN model input training;

[0043] II. Vehicle microscopic information and spatial feature extraction;

[0044] Through convolution and pooling operations, the microscopic information and spatial features of the vehicle are extracted. The output of the first convolution and pooling layer in the CNN model and the model input satisfy the following relationship:

[0045]

[0046] Where σ is the activation function; pool(·) represents the pooling operation; l is the depth of the convolutional neural network in the convolution operation; Represents the output and input of each layer of the network respectively; c j Filters for each layer of the network; is the parameter matrix of each layer of the network;

[0047] The output and input relationship of the first convolution and pooling layers in the model are expressed as follows:

[0048]

[0049] In the formula, k represents the layer number of the neural network;

[0050] III. Feature Flattening

[0051] The vehicle state and spatial features in the model input are screened and extracted to obtain a feature matrix that meets the dimensionality requirements, which is in the following form:

[0052]

[0053] Where L represents the depth of the convolutional neural network in the flattening operation; flatten(·) is the flattening operation;

[0054] IV. Model Output

[0055] After completing the feature extraction input and feature flattening, the final model output result is output through the fully connected layer in the form of:

[0056]

[0057] Where w f and b f is the weight coefficient of the fully connected layer; is the model output matrix, which represents the motion state information of the hybrid vehicle group in the next control cycle.

[0058] Furthermore, in step S3, the specific structure of the convolutional neural network is:

[0059] (1) Convolutional layer

[0060] Different convolution kernels extract different vehicle features. The input and output of each convolution layer are expressed as follows:

[0061]

[0062] Where m and n correspond to the convolution kernel dimensions; d ab It is the value corresponding to the position (a, b) of the input matrix X; Represents the weight at the (a, b) position; yconv is the output result after the convolution operation;

[0063] (2) Pooling layer

[0064] By adopting the maximum pooling mechanism to reduce the scale of the CNN model, the recognition and screening of input features are completed. The expression is as follows:

[0065] y pool =max(d ab )

[0066] Where y pool is the output result of the pooling layer;

[0067] (3) Activation function

[0068] Relu(·) is selected as the model activation function, and the expression is as follows:

[0069]

[0070] (4) Design Optimizer

[0071] The optimizer objective is to minimize the MSE value. The calculation expression of MSE is as follows:

[0072]

[0073] Set the network parameters to According to the standard back-propagation algorithm, the expression of Θ can be obtained as follows:

[0074]

[0075] Where K represents the layer number of the neural network;

[0076] (5) Design model network parameters

[0077] The initial layer is composed of convolution and pooling operations, followed by a convolution layer to further integrate and enhance the features extracted by the initial layer. The model output is then connected to the fully connected layer through a flattening layer.

[0078] Beneficial effects:

[0079] 1. The present invention addresses the characteristics of high mixed traffic volume and high speed inside highway tunnels, combines vehicle micro-state information with the macro-performance of traffic flow, and considers the impact of vehicle micro-information on macro-traffic flow, thus solving the problem that single-perspective modeling is difficult to comprehensively and accurately depict the evolution of traffic flow.

[0080] 2. In addition, after collecting microscopic information such as the spatial position and status of vehicles, the information was hierarchically mapped and a vehicle status matrix was constructed. By utilizing the CNN model's processing advantages in spatial matrix data, the feature matrix was extracted and processed, and a modeling method for the macroscopic evolution law of mixed traffic flow based on vehicle microscopic information was proposed and established. This method more accurately depicts the evolution law of mixed traffic flow and effectively reduces the prediction error of the vehicle volume in bottleneck sections.

[0081] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of the method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information according to the present invention;

[0083] Figure 2 This is a schematic diagram of vehicle group division based on traffic flow characteristics;

[0084] Figure 3 Schematic diagram of hierarchical mapping of heterogeneous vehicle microscopic information;

[0085] Figure 4 Schematic diagram of the structure of the CNN model for the macro-evolution law of mixed traffic flow. DETAILED DESCRIPTION

[0086] To make the technical solutions, advantages, and purposes of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0087] like Figure 1 As shown, the present invention provides a method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information, comprising the following steps:

[0088] S1. Segment mixed traffic flows in a bottleneck section within a highway tunnel and select modeling parameters.

[0089] In step S1, the traffic flow characteristics of the bottleneck section of the highway tunnel are analyzed, and the vehicle group is divided into a lane-changing conflict vehicle group and a following conflict vehicle group. Figure 2As shown in the figure, a following conflict group occurs when a no-entry sign is posted 200 meters upstream of a bottleneck. A following conflict group occurs when a group of vehicles is in the only passable lane to the left of the no-entry sign. A lane-changing conflict group occurs when a group of vehicles is in the lane from upstream of the no-entry sign to the tunnel entrance. Micro-information such as vehicle speed v, acceleration a, road position x, and vehicle category p, as well as upstream vehicle departures, CAV penetration rate, number of emergency brakes, average traffic speed, and traffic volume (traffic throughput) are selected as relevant modeling parameters. Upstream vehicle departures are defined as the number of vehicles entering the bottleneck area in this scenario within an hour, and traffic volume is defined as the number of vehicles passing through the bottleneck within an hour.

[0090] S2. Perform hierarchical mapping on the vehicle data collected in the connected environment to form a multi-dimensional feature matrix;

[0091] In step S2, the vehicle data is processed using an image digitization method, and the continuous analog image is converted into a discrete digital image form through sampling and quantization steps. The specific steps are:

[0092] S2.1 Image sampling;

[0093] The microscopic information of vehicles on the target road section is collected through intelligent roadside equipment or vehicle communication systems. The lane area where the mixed vehicle group is located is regarded as an image and spatially divided into a c×m grid array.

[0094] Among them, a single grid represents a pixel point of the image, and m×n defines the spatial resolution of the image;

[0095] S2.2 Image quantification;

[0096] The road area where the mixed vehicle group is located is spatially divided into grids. A mapping layer is created for each vehicle state information. Vehicle information (vehicle speed, acceleration, road position, vehicle category, etc.) is quantified and filled into the corresponding mapping layer based on the vehicle's relative spatial position, completing the spatial representation of the vehicle micro-information in the mixed traffic flow.

[0097] S2.3 construct a multidimensional feature matrix;

[0098] A road space S of length L is selected upstream of the bottleneck section inside the highway tunnel. This road space S is treated as an image and gridded, converting it into a C×M grid. C is the number of lanes, and M is the number of cells in a single lane along the vehicle's travel direction. A single grid cell is treated as a pixel, and the grid size C×M is considered the resolution.

[0099] Get vehicles n1,n2,…,n in the road space S kFor each type of vehicle state information, a corresponding C×M state mapping layer is constructed. Then, each vehicle micro-state information is mapped to the corresponding cell in its state mapping layer according to its relative spatial position to form F state feature layers; where F is the number of vehicle micro-information states. The state feature layers are superimposed to construct a C×M×F multi-dimensional feature matrix.

[0100] Vehicle microscopic information includes vehicle parameters: body length l and vehicle category p, which is used to distinguish between CHVs and CAVs; vehicle motion data: velocity v and acceleration a; and vehicle position data: lane c and vehicle head position x.

[0101] S2.4 Hierarchical mapping of vehicle microscopic information;

[0102] I. Map each vehicle to the corresponding grid of each state feature layer according to its relative position in the road space S, as follows: Figure 3 As shown, there are vehicles n1, n2, ..., n in the road space S. k The starting and ending positions of the vehicles in each lane are expressed as follows:

[0103]

[0104]

[0105] Where m ia 、m ib They represent the starting unit and the ending unit of the vehicle insertion feature graph respectively; x i is the position of the vehicle head; l veh is the vehicle length; s0 is the starting position of the road space S; [·] indicates rounding up;

[0106] II. Complete the quantitative filling of vehicle status information, where the physical characteristic quantization value includes vehicle speed acceleration Lane and vehicle type The quantitative relationship is expressed as:

[0107]

[0108]

[0109]

[0110] Where, v i Indicates vehicle n i The actual speed; v max Indicates the maximum speed limit of the road; a i For vehicle ni The actual acceleration of a max 、b max are the maximum acceleration and maximum deceleration of the vehicles in the group respectively;

[0111] In addition, a road feature layer is added, and its quantitative expression is as follows:

[0112]

[0113] Where x i For vehicle n i The position in the road space S.

[0114] S3. Use a convolutional neural network to extract the vehicle state feature matrix and establish a macroscopic traffic flow evolution model based on vehicle microscopic information.

[0115] On the basis of S2, the macroscopic evolution law model of traffic flow based on vehicle microscopic information in step S3 includes the following four parts: Figure 4 As shown, specifically

[0116] I. Model Input

[0117] The model inputs the vehicle microscopic information feature matrix. The expression of the model input X is as follows:

[0118] X=[M1,M2,M3,...M n ]

[0119] Where M i represents the C×M×F multi-dimensional feature matrix obtained after hierarchical mapping of the i-th mixed vehicle group data; n represents the number of samples used for CNN model input training;

[0120] II. Vehicle microscopic information and spatial feature extraction;

[0121] Through convolution and pooling operations, the microscopic information and spatial features of the vehicle are extracted. The output of the first convolution and pooling layer in the CNN model and the model input satisfy the following relationship:

[0122]

[0123] Where σ is the activation function; pool(·) represents the pooling operation; l is the depth of the convolutional neural network in the convolution operation; Represents the output and input of each layer of the network respectively; c j Filters for each layer of the network; is the parameter matrix of each layer of the network;

[0124] The output and input relationship of the first convolution and pooling layers in the model are expressed as follows:

[0125]

[0126] In the formula, k represents the layer number of the neural network;

[0127] III. Feature Flattening

[0128] The vehicle state and spatial features in the model input are screened and extracted to obtain a feature matrix that meets the dimensionality requirements, which is in the following form:

[0129]

[0130] Where L represents the depth of the convolutional neural network in the flattening operation; flatten(·) is the flattening operation; this step needs to be performed between the convolutional layer and the fully connected layer;

[0131] IV. Model Output

[0132] After completing the feature extraction input and feature flattening, the final model output result is output through the fully connected layer in the form of:

[0133]

[0134] Where w f and b f is the weight coefficient of the fully connected layer; is the model output matrix, which represents the motion state information of the hybrid vehicle group in the next control cycle.

[0135] The specific structure of the convolutional neural network is:

[0136] (1) Convolutional layer

[0137] Different convolution kernels extract different vehicle features. The input and output of each convolution layer are expressed as follows:

[0138]

[0139] Where m and n correspond to the convolution kernel dimensions; d ab It is the value corresponding to the position (a, b) of the input matrix X; Represents the weight at the (a, b) position; y conv is the output result after the convolution operation;

[0140] (2) Pooling layer

[0141] By adopting the maximum pooling mechanism to reduce the scale of the CNN model, the recognition and screening of input features are completed. The expression is as follows:

[0142] y pool =max(d ab )

[0143] Where y pool is the output result of the pooling layer;

[0144] (3) Activation function

[0145] Relu(·) is selected as the model activation function, and the expression is as follows:

[0146]

[0147] (4) Design Optimizer

[0148] The optimizer objective is to minimize the MSE value. The calculation expression of MSE is as follows:

[0149]

[0150] Set the network parameters to According to the standard back-propagation algorithm, the expression of Θ can be obtained as follows:

[0151]

[0152] Where K represents the layer number of the neural network;

[0153] (5) Design model network parameters

[0154] The initial layer is composed of convolution and pooling operations, followed by a convolution layer to further integrate and enhance the features extracted by the initial layer. The model output is then connected to the fully connected layer through a flattening layer.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of protection of the present invention.

Claims

1. A method for constructing a macroscopic evolution law model of mixed traffic flow based on vehicle microscopic information, characterized in that: The following steps are involved: S1. Segment mixed traffic flows in a bottleneck section within a highway tunnel and select modeling parameters. S2. Perform hierarchical mapping on the vehicle data collected in the connected environment to form a multi-dimensional feature matrix; In step S2, the vehicle data is processed using an image digitization method to perform layered mapping, and the continuous analog image is converted into a discrete digital image form through sampling and quantization steps; S2.1 Image sampling; The microscopic information of vehicles on the target road section is collected through intelligent roadside equipment or vehicle communication systems. The lane area where the mixed vehicle group is located is regarded as an image and spatially divided into a c×m grid array. Among them, a single grid represents a pixel point of the image, and m×n defines the spatial resolution of the image; S2.2 Image quantification; The road area where the mixed vehicle group is located is spatially divided into grids. A mapping layer is established for each vehicle status information. Vehicle information is quantitatively filled into the corresponding mapping layer according to the vehicle's relative spatial position, completing the spatial representation of the vehicle microscopic information in the mixed traffic flow. S2.3 construct a multidimensional feature matrix; A road space S of length L is selected upstream of the bottleneck section inside the highway tunnel. This road space S is treated as an image and gridded into a C×M grid. C is the number of lanes, and M is the number of cells divided into a single lane in the direction of vehicle travel. Get vehicles n1,n2,…,n in the road space S k For each type of vehicle state information, a corresponding C×M state mapping layer is constructed. Then, each vehicle micro-state information is mapped to the corresponding cell according to its relative spatial position in the state mapping layer to form F state feature layers; where F is the number of vehicle micro-information states. The state feature layers are superimposed to construct a C×M×F multi-dimensional feature matrix. S2.4 Hierarchical mapping of vehicle microscopic information; I. Map each vehicle to the corresponding grid of each state feature layer according to its relative position in the road space S. There are vehicles n1, n2, ..., n in the road space S. k The starting and ending positions of the vehicles in each lane are expressed as follows: Where m ia 、m ib They represent the starting unit and the ending unit of the vehicle insertion feature graph respectively; x i is the position of the vehicle head; l veh is the vehicle length; s0 is the starting position of the road space S; [·] indicates rounding up; II. Complete the quantitative filling of vehicle status information, where the physical characteristic quantization value includes vehicle speed acceleration Lane and vehicle type The quantitative relationship is expressed as: Where, v i Indicates vehicle n i The actual speed; v max Indicates the maximum speed limit of the road; a i For vehicle n i The actual acceleration of a max 、b max are the maximum acceleration and maximum deceleration of the vehicles in the group respectively; S3. Use a convolutional neural network to extract the vehicle state feature matrix and establish a macroscopic traffic flow evolution model based on vehicle microscopic information. In step S3, the traffic flow macro-evolution law model based on vehicle micro-information includes the following parts: I. Model input; The model inputs the vehicle microscopic information feature matrix. The expression of the model input X is as follows: X=[M1,M2,M3,...M n ] Where M i represents the C×M×F multi-dimensional feature matrix obtained after hierarchical mapping of the i-th mixed vehicle group data; n represents the number of samples used for CNN model input training; II. Vehicle microscopic information and spatial feature extraction; Through convolution and pooling operations, the microscopic information and spatial features of the vehicle are extracted. The output of the first convolution and pooling layer in the CNN model and the model input satisfy the following relationship: Where σ is the activation function; pool(·) represents the pooling operation; l is the depth of the convolutional neural network in the convolution operation; Represent the output and input of each layer of the network respectively; is the parameter matrix of each layer of the network; The output and input relationship of the first convolution and pooling layers in the model are expressed as follows: In the formula, k represents the layer number of the neural network; III. Feature Flattening The vehicle state and spatial features in the model input are screened and extracted to obtain a feature matrix that meets the dimensionality requirements, which is in the following form: Where flatten(·) is the flattening operation; L represents the depth of the convolutional neural network in the flattening operation; IV. Model Output After completing the feature extraction input and feature flattening, the final model output result is output through the fully connected layer in the form of: Where w f and b f is the weight coefficient of the fully connected layer; is the model output matrix, which represents the motion state information of the hybrid vehicle group in the next control cycle.

2. The method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information according to claim 1, characterized in that: In step S1, the mixed traffic flow is divided into a lane-changing conflict vehicle group and a following conflict vehicle group; Following conflicting vehicles: A no-entry sign is set up 200 meters upstream of the bottleneck. The vehicles in the left lane downstream of the no-entry sign with only one passable lane are called following conflicting vehicles. Lane-changing conflict vehicle group: The vehicle group from the upstream of the no-entry sign to the tunnel entrance lane is called a lane-changing conflict vehicle group.

3. The method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information according to claim 1 is characterized by: The modeling parameters include vehicle speed v, acceleration a, road position x, vehicle category p, upstream vehicle departure volume, CAV penetration rate, number of emergency brakes, average traffic speed, and traffic volume; Among them, the upstream departure volume is the number of vehicles entering the bottleneck area of ​​the road in the target scenario within one hour; the traffic flow is the number of vehicles passing through the bottleneck within one hour.

4. The method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information according to claim 1 is characterized by: In step S2.2, the traffic flow information includes vehicle speed, acceleration, road position and vehicle category.

5. The method for constructing a mixed traffic flow macro-evolution law model based on vehicle micro-information according to claim 1 is characterized by: In step S2.3, the vehicle microscopic information includes the vehicle's own parameters: body length l and vehicle category p, where vehicle category p is used to distinguish between CHVs and CAVs; vehicle motion data: speed v and acceleration a; and vehicle position data: lane c and vehicle head position x.

Images