A CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN

Abstract

The present invention discloses a CAV guidance control method for a mixed vehicle group in a near-signal area based on CP-CNN, the method comprising obtaining single-vehicle data of a mixed heterogeneous vehicle group in a three-lane near-signal area, mapping the physical feature information of the single vehicle to a multi-dimensional feature graph, and constructing a model data set; constructing a macro traffic flow prediction model for a mixed vehicle group based on a convolutional neural network, namely, CP-CNN, and training the model; constructing a dynamic optimization target for a mixed vehicle group according to the signal lights in the near-signal area and the real-time traffic status; designing a CAV guidance control method for a mixed vehicle group in a near-signal area based on CP-CNN under mixed traffic near-signal areas; and designing a variable speed limit optimization strategy based on a particle swarm optimization algorithm. The present invention adaptively controls the speed of CAV vehicles in response to changes in traffic flow and signal lights in the near-signal area, thereby achieving effective control of the bottleneck of traffic at the intersection.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent traffic information technology, and specifically relates to a CAV guidance control method for a near-signal area vehicle group under a new mixed traffic condition, which can be applied to the near-signal area vehicle group optimization and bottleneck negative effect suppression in an intelligent network environment. Background Art

[0002] Signalized intersections are bottleneck areas of urban road congestion and high-incidence areas of accidents, with concentrated conflicts between vehicles and roads, complex control decisions, and low traffic efficiency. In order to improve the control effect and increase the means of control, domestic and foreign scholars have studied intersections and their adjacent extended areas as near-signal control areas. With the development of intelligent vehicle technology, heterogeneous traffic flows consisting of connected automatic vehicles and connected human-driven vehicles have emerged, providing a new way to solve the congestion problem at signalized intersections. Therefore, it is worth exploring whether to use connected automatic vehicles in new mixed traffic flows as an entry point for the negative impact of the bottleneck near the signal area.

[0003] Existing research on signal intersection congestion mostly focuses on traffic flow prediction to control the timing of traffic lights. For example, patent publication CN109697866A discloses an edge computing traffic light control system and control method, which dynamically adjusts and learns to obtain the first traffic light timing strategy based on the collected traffic flow, and globally optimizes the first traffic light timing strategy of all intersections according to the preset algorithm to obtain the optimal traffic light timing strategy for all intersections. This method focuses on adjusting the timing of traffic lights, and as traffic flow and traffic density increase, the regulatory role played by adjusting the timing of traffic lights becomes smaller and smaller. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a novel CAV guidance control method for a vehicle group near a signal area under mixed traffic conditions, the method comprising the following steps:

[0005] Step 1: Obtain the single-vehicle data of a mixed heterogeneous vehicle group in the three-lane near-signal area, map the physical feature information of the single vehicle to a multi-dimensional feature map, and construct a model dataset;

[0006] Step 2: Build a mixed vehicle group macro traffic flow prediction model CP-CNN based on convolutional neural network and train the model;

[0007] Step 3: Construct a dynamic optimization target for the mixed vehicle group based on the traffic lights near the signal area and the real-time traffic status;

[0008] Step 4: Design a CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN in mixed traffic near signal areas;

[0009] Step 5: Design a variable speed limit optimization strategy based on particle swarm optimization algorithm.

[0010] Furthermore, the step 1 comprises:

[0011] Step 11: Select a road section s with a length of L near the signal area, and obtain the vehicles n1, n2, ..., n in the road section s at time t. k The physical feature information includes vehicle information, vehicle position information and motion state information, wherein the vehicle information includes vehicle ID and vehicle type p, which is used to distinguish between human-driven HV and connected automatic vehicle CAV; the vehicle position information includes lane c and lane longitudinal position x (vehicle head position); the motion state information includes speed v, acceleration a, yaw angle h, and target turning lane td;

[0012] Step 12: Based on the single-vehicle data in road section s at time t, divide road section s into c×m units, where c is the number of lanes and m is the number of units divided in each lane. Determine the length l of a single unit division in each lane. min , expressed as:

[0013]

[0014] Where, L represents the length of the road section s;

[0015] Step 13: Determine the unit position for each vehicle to be inserted in each lane. The specific operation is to first construct a feature matrix of c×m×f, where f is the number of feature layers, and then insert the physical feature information of the vehicle into the corresponding unit according to the relative position of the vehicle in the road section. For each lane, the calculation formula for the unit position for each vehicle to be inserted is:

[0016]

[0017]

[0018] In the formula, m i and m i ′ represents the range of the horizontal position inserted into the feature matrix, that is, m i ' indicates the starting position of insertion, m i Indicates the end position of insertion, x j is the longitudinal position of the jth vehicle, s min is the initial position of the road segment s, l j is the length of the j-th vehicle, and “[·]” indicates the rounding function;

[0019] Step 14: Select τ as the time interval, and use the feature matrix constructed at all moments within the selected control time T as the input of the convolutional neural network, expressed as:

[0020] X=[M1,M2,M3,…,M n ]

[0021] Where M t is the vehicle group feature matrix at time t, and n is the amount of data obtained within time T, that is, the size of the training set data input to the CNN model.

[0022] Furthermore, the step 2 comprises:

[0023] Step 21: Select the average traffic volume in the approach road near the signal area at time t+τ as the training label and build a CNN deep network;

[0024] Step 22: The constructed multi-dimensional feature map of the road section s at time t is normalized and input into the CNN model for training. The formula for data normalization is:

[0025]

[0026] In the formula, f i * is the standardized data, f i is the feature value that needs to be standardized in sample i, f min is the minimum value of the eigenvalue that needs to be standardized, f max is the maximum value of the eigenvalue that needs to be standardized;

[0027] Step 23: After multiple convolutions and pooling, the model is trained through the fully connected layer output. The output of the CP-CNN model is the predicted average traffic flow q in the approach road near the signal area at time t+τ.

[0028] Furthermore, the step 3 comprises:

[0029] Step 31: Calculate the maximum traffic capacity of the approach area under different signal light conditions in the near signal area.

[0030] When the traffic light is red,

[0031] When the traffic light is green, the maximum traffic capacity is calculated as follows:

[0032]

[0033] In the formula, is the maximum traffic capacity of the approach road, p c 、p h is the penetration rate of CAV and HV, and p h =1-p c , Δx cc is the headway time between CAV and CAV, Δxch is the headway time between the CAV and the HV, Δx hc is the headway time of the HV following the CAV, Δx hh The headway time of the HV following the HV;

[0034] Step 32: The maximum approach capacity under different traffic lights is used as the dynamic optimization target of the mixed vehicle group and compared with the approach traffic volume predicted by the CP-CNN model. If the approach traffic volume at time t+τ is greater than the maximum approach capacity at time t+τ, variable speed limit control is performed.

[0035] Furthermore, the step 4 comprises:

[0036] Step 41: Calculate the maximum traffic capacity at time t+τ based on the approach area Compared with the approach traffic volume predicted by the CP-CNN model, if No control is performed;

[0037] Step 42: If The CAV vehicle in the near-signal area is controlled, and the variable speed limit value is calculated using an integral controller:

[0038]

[0039] Where V SL (t) is the speed control value proposed for the CAV vehicle using the CAV guidance control algorithm, K I is the undetermined control parameter, and the optimal control speed is obtained by optimization based on the feedback. The optimization goal is:

[0040]

[0041] In the formula, q * (t+τ) is the time when CAV guidance control is implemented at time t. SL (t) replaces the CAV vehicle speed v in the multidimensional feature graph with the value, and the approach traffic flow at time t+τ predicted by the CP-CNN model; q(t+τ) is the approach traffic flow at time t+τ predicted by the CP-CNN model without variable speed limit control; is the maximum capacity of the ramp at time t+τ calculated using the saturated flow rate method; |·| is the absolute value symbol.

[0042] Furthermore, the step 5 comprises:

[0043] According to the optimization goal, based on the feedback idea and particle swarm optimization algorithm, the K I Optimization includes the following steps:

[0044] Step 51: Set the relevant parameters in the particle swarm optimization algorithm, such as the number of particles n used in the algorithm and the archive threshold of the Archive set. As well as the parameter M for the number of grid divisions, the setting of constraints, and the number of iterations of the algorithm;

[0045] Step 52: Under the constraint conditions, randomly generate uniform distribution and initialize the speed and position of each particle;

[0046] Step 53: According to the fitness function, the fitness of each particle is calculated and used as its own local optimal solution Pbest;

[0047] Step 54: According to the Pareto dominance rule, the non-inferior solutions in the initialization particles are added to the Archive set;

[0048] Step 55: Based on the grid method, obtain the density estimation value of each particle in the Archive set, and select the global optimal solution gBest of the particles according to the Pareto optimal solution search algorithm based on the adaptive grid;

[0049] Step 56: Update the speed and position of each particle to generate the next generation of particle groups;

[0050] Step 57: Calculate the fitness of each particle according to the fitness function, and update the local optimal solution Pbest of each particle;

[0051] Step 58: According to the Pareto dominance rule, add the current particle's non-inferior solution to the Archive set, update the Archive set, and determine whether the number of particles in the Archive set reaches the archive threshold. If so, the Archive set needs to be truncated;

[0052] Step 59: Return to step 55: Update the global optimal solution of the particle;

[0053] Step 510: Determine the current number of iterations. If it has not reached the set maximum value, return to step 56. If it has reached the maximum value, end the optimization process.

[0054] The beneficial effects of the present invention are:

[0055] The present invention proposes a CAV guidance and control method for mixed vehicle groups near signal areas under new mixed traffic conditions. This method aims to solve the problem that few scholars have explored how to alleviate the congestion at signal intersections by taking connected automatic vehicles as the entry point from the perspective of vehicle control. The present invention utilizes the single-vehicle information obtained in a networked environment, fully considers the spatial physical information of single vehicles in a new type of mixed vehicle group, maps the vehicle's motion state and spatial distribution characteristics to a multi-dimensional feature map, and uses a convolutional neural network for learning to achieve an accurate characterization of the evolution law of mixed vehicle groups. On this basis, according to the signal lights in the near-signal area and the actual traffic conditions, the saturated flow rate method is combined to calculate the maximum traffic capacity of the approach area under different signal light states in the near-signal area, and the results are combined with the CAV speed guidance and control algorithm, and the particle swarm optimization algorithm is used for optimization, and a CAV guidance and control algorithm scheme for mixed vehicle groups near signal areas based on CP-CNN is designed. The present invention proposes a CAV guidance control method for mixed vehicle groups in near-signal areas based on CP-CNN, which can adaptively control the speed of CAV vehicles according to changes in traffic flow and signal lights in near-signal areas, thereby achieving effective control of intersection bottlenecks. The method has the advantages of wide applicability and good control effect.

[0056] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0058] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0059] Figure 2 Schematic diagram of vehicle group feature extraction and multi-dimensional feature vector construction in step 1;

[0060] Figure 3 This is a schematic diagram of the deep learning network in step 2;

[0061] Figure 4 Flowchart of particle swarm optimization algorithm. DETAILED DESCRIPTION

[0062] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.

[0063] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0064] It should be noted that the illustrations provided in this embodiment are only used to illustrate the basic concept of the present invention in a schematic manner. Therefore, the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0065] like Figure 1 As shown, a CAV guidance control method for a mixed vehicle group near a signal area based on CP-CNN, the method comprises the following steps:

[0066] Step 1) obtains the single-vehicle data of a mixed heterogeneous vehicle group in a three-lane near-signal area, maps the physical feature information of the single vehicle to a multi-dimensional feature map, and constructs a model data set. In some embodiments, this step specifically includes the following sub-steps:

[0067] Step 11: Select a road section s with a length of L near the signal area, and obtain the number of vehicles, n1, n2, ..., n in the road section s at time t. k The physical feature information includes vehicle information, vehicle position information and motion state information, wherein the vehicle information includes vehicle id, vehicle type (type) p (used to distinguish between human-driven HV and connected automatic vehicle CAV); the vehicle position information includes the lane c, the longitudinal position x (front position); the motion state information includes speed v, acceleration a, yaw angle (yaw angle) h, target turning lane td, for details, please refer to Figure 2 The schematic diagram of vehicle group feature extraction and multi-dimensional feature vector construction is shown in FIG. Figure 2 In the figure, the overall rectangular area 505*3 represents the near-signal area section s with a length of L and its three longitudinal lanes. The small rectangles of different shades in the rectangular area represent the current vehicle location and its physical information stored in five channels (5 channels), each of which is speed, acceleration, yaw angle, target turning lane and vehicle type. Since the 5 information physical characteristic values ​​of each vehicle are different, when they are superimposed and converted into RGB values ​​and displayed in the example image, the colors presented will be different. After being converted into a grayscale image, the depth of the color block presented will be different. It should be noted that Figure 2 The data shown in the figures are only exemplary and are not intended to limit the present invention.

[0068] Step 12: Based on the single-vehicle data in road section s at time t, divide road section s into c×m units, where c is the number of lanes and m is the number of units divided in each lane. Determine the length l of a single unit division in each lane. min , which can be expressed by the following formula:

[0069]

[0070] Where, L represents the length of the road section s;

[0071] Step 13: Determine the cell position for each vehicle in each lane. The specific operation is to first construct a c×m×f feature matrix, where f is the number of feature layers, and then insert the physical feature information of the vehicle into the corresponding cell according to the relative position of the vehicle in the road section. Since there are cases where a vehicle spans multiple cells, the insertion position needs to be determined.

[0072] In some embodiments, for each lane, the calculation formula for the unit position of each vehicle inserted can be expressed as:

[0073]

[0074]

[0075] In the formula, m i and m i ′ represents the range of the horizontal position inserted into the feature matrix, that is, m i ' indicates the starting position of insertion, m i Indicates the end position of insertion, x j is the longitudinal position of the jth vehicle, s min is the initial position of the road segment s, l j is the length of the j-th vehicle, and “[·]” indicates the rounding function;

[0076] Step 14: Select τ as the time interval, and use the feature matrix constructed at all moments within the selected control time T as the input of the convolutional neural network, expressed as:

[0077] X=[M1,M2,M3,…,M n ]

[0078] Where M t is the vehicle group feature matrix at time t, and n is the amount of data obtained within time T, that is, the size of the training set data input to the CNN model.

[0079] Step 2) constructs a mixed vehicle group macro traffic flow prediction model based on a convolutional neural network, namely CP-CNN, and trains the model, which specifically includes the following steps:

[0080] Step 21: Select the average traffic flow in the approach road near the signal area at time t+τ as the training label and build a CNN deep network, such as Figure 3 The network structure shown;

[0081] Step 22: Normalize the constructed multi-dimensional feature map of the road section s at time t and input it into CNN for training.

[0082] The formula for data normalization is:

[0083]

[0084] In the formula, f i * is the standardized data, f i is the feature value that needs to be standardized in sample i, f min is the minimum value of the eigenvalue that needs to be standardized, f max is the maximum value of the eigenvalue that needs to be standardized;

[0085] Step 23: After multiple convolutions and poolings, the data is output through the fully connected layer (FClayer) to complete the model training. In some embodiments, the data can also be flattened by the flattening layer before the fully connected layer processing. The output of the CP-CNN model is the predicted average vehicle flow q in the approach road near the signal area at time t+τ.

[0086] Step 3) constructs a dynamic optimization target for a mixed vehicle group based on the traffic lights near the signal area and the real-time traffic status, which specifically includes the following steps:

[0087] Step 31: Based on the saturation flow rate method proposed in the U.S. Highway Capacity Manual (HCM), calculate the maximum traffic capacity of the approach area under different signal light conditions in the near-signal area.

[0088] When the traffic light is red,

[0089] When the traffic light is green, the maximum traffic capacity is calculated as follows:

[0090]

[0091] In the formula, is the maximum traffic capacity of the approach road, p c 、p h is the penetration rate of CAV and HV, and p h =1-p c . Δx cc is the headway time between the CAV and the CAV. Similarly, Δx chis the headway time between the CAV and the HV, Δx hc is the headway time of the HV following the CAV, Δx hh The headway time of the HV following the HV;

[0092] Step 32: The maximum approach capacity under different traffic lights is used as the dynamic optimization target of the mixed vehicle group and compared with the approach traffic volume predicted by the CP-CNN model. If the approach traffic volume at time t+τ is greater than the maximum approach capacity at time t+τ, variable speed limit control is performed.

[0093] Step 4), combined with the neural network model constructed in step 3), designs a CAV guidance control algorithm for vehicle groups in mixed traffic near signal areas based on CP-CNN, which specifically includes the following steps:

[0094] Step 41: Calculate the maximum traffic capacity at time t+τ based on the approach area Compared with the approach traffic volume predicted by the CP-CNN model, if No control is performed;

[0095] Step 42: If The CAV vehicle in the near-signal area is controlled, and the variable speed limit value is calculated using an integral controller:

[0096]

[0097] Where V SL (t) is the speed limit value proposed for CAV vehicles using the CAV speed control algorithm; K I is the control parameter to be determined. According to the feedback, the optimization is performed to obtain the optimal control speed, and the optimization target is:

[0098]

[0099] In the formula, q * (t+τ) is the value of V after CAV speed control is implemented at time t. SL (t) replaces the CAV vehicle speed v in the multidimensional feature graph with the value, and the approach traffic flow at time t+τ predicted by the CP-CNN model; q(t+τ) is the approach traffic flow at time t+τ predicted by the CP-CNN model without variable speed limit control; is the maximum capacity of the ramp at time t+τ calculated using the saturated flow rate method; |·| is the absolute value symbol.

[0100] Step 5), according to the optimization goal designed in step 4), based on the feedback idea and particle swarm optimization algorithm, K is used to calculate the control parameters. I Optimization, reference Figure 4 , specifically including the following steps:

[0101] Step 51: Set the relevant parameters in the particle swarm optimization algorithm, such as the number of particles n used in the algorithm and the archive threshold of the Archive set. As well as the parameter M for the number of grid divisions, the setting of constraints, and the number of iterations of the algorithm;

[0102] Step 52: Under the constraint conditions, randomly generate uniform distribution and initialize the speed and position of each particle;

[0103] Step 53: According to the fitness function, the fitness of each particle is calculated and used as its own local optimal solution Pbest;

[0104] Step 54: According to the Pareto dominance rule, the non-inferior solutions in the initialization particles are added to the Archive set;

[0105] Step 55: Based on the grid method, obtain the density estimation value of each particle in the Archive set, and select the global optimal solution gBest of the particles according to the Pareto optimal solution search algorithm based on the adaptive grid;

[0106] Step 56: Update the speed and position of each particle to generate the next generation of particle groups;

[0107] Step 57: Calculate the fitness of each particle according to the fitness function, and update the local optimal solution Pbest of each particle;

[0108] Step 58: According to the Pareto dominance rule, add the current particle's non-inferior solution to the Archive set, update the Archive set, and determine whether the number of particles in the Archive set reaches the archive threshold. If so, the Archive set needs to be truncated;

[0109] Step 59: Return to step 55 and update the global optimal solution of the particle;

[0110] Step 510: Determine the current number of iterations. If it has not reached the set maximum value, return to step 56. If it has reached the maximum value, end the optimization process.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. 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 solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN, characterized in that: include: Step 1: Obtain the single-vehicle data of a mixed heterogeneous vehicle group in the three-lane near-signal area, map the physical feature information of the single vehicle to a multi-dimensional feature map, and construct a model dataset; Step 2: Build a mixed vehicle group macro traffic flow prediction model CP-CNN based on convolutional neural network and train the model; Step 3: Construct a dynamic optimization target for the mixed vehicle group based on the traffic lights near the signal area and the real-time traffic status; The step 3 comprises: Step 31: Calculate the maximum traffic capacity of the approach area under different signal light conditions in the near signal area. When the traffic light is red, When the traffic light is green, the maximum traffic capacity is calculated as follows: In the formula, is the maximum traffic capacity of the approach road, p c 、p h is the penetration rate of CAV and HV, and p h =1-p c , Δx cc is the headway time between CAV and CAV, Δx ch is the headway time between the CAV and the HV, Δx hc is the headway time of the HV following the CAV, Δx hh The headway time of the HV following the HV; Step 32: The maximum capacity of the approach road under different traffic lights is used as the dynamic optimization target of the mixed vehicle group, and is compared with the approach road traffic volume predicted by the CP-CNN model. If the approach road traffic volume at time t+τ is greater than the maximum capacity of the approach road area at time t+τ, variable speed limit control is performed; Step 4: Design a CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN in mixed traffic near signal areas; The step 4 comprises: Step 41: Calculate the maximum traffic capacity at time t+τ based on the approach area Compared with the approach traffic volume predicted by the CP-CNN model, if No control is performed; Step 42: If The CAV vehicle in the near-signal area is controlled, and the variable speed limit value is calculated using an integral controller: Where V SL (t) is the speed control value proposed for the CAV vehicle using the CAV guidance control algorithm, K I is the undetermined control parameter, and the optimal control speed is obtained by optimization based on the feedback. The optimization goal is: In the formula, q * (t+τ) is the time when CAV guidance control is implemented at time t. SL (t) replaces the CAV vehicle speed v in the multidimensional feature graph with the value, and the approach traffic flow at time t+τ predicted by the CP-CNN model; q(t+τ) is the approach traffic flow at time t+τ predicted by the CP-CNN model without variable speed limit control; is the maximum capacity of the approach at time t+τ calculated according to the saturation flow rate method; |·| is the absolute value symbol; Step 5: Design a variable speed limit optimization strategy based on particle swarm optimization algorithm; The step 5 comprises: According to the optimization goal, based on the feedback idea and particle swarm optimization algorithm, the K I Optimization includes the following steps: Step 51: Setting relevant parameters in the particle swarm optimization algorithm, including the number of particles n used in the algorithm, the archive threshold N of the Archive set, the parameter M of the number of grid divisions, the setting of constraint conditions, and the number of iterations of the algorithm; Step 52: Under the constraint conditions, randomly generate uniform distribution and initialize the speed and position of each particle; Step 53: According to the fitness function, the fitness of each particle is calculated and used as its own local optimal solution Pbest; Step 54: According to the Pareto dominance rule, the non-inferior solutions in the initialization particles are added to the Archive set; Step 55: Based on the grid method, obtain the density estimation value of each particle in the Archive set, and select the global optimal solution gBest of the particles according to the Pareto optimal solution search algorithm based on the adaptive grid; Step 56: Update the speed and position of each particle to generate the next generation of particle groups; Step 57: Calculate the fitness of each particle according to the fitness function, and update the local optimal solution Pbest of each particle; Step 58: According to the Pareto dominance rule, the non-inferior solution of the current particle is added to the Archive set, the Archive set is updated, and it is determined whether the number of particles in the Archive set reaches the archive threshold N. If so, the Archive set needs to be truncated; Step 59: Return to step 55: Update the global optimal solution of the particle; Step 510: Determine the current number of iterations. If it has not reached the set maximum value, return to step 56. If it has reached the maximum value, end the optimization process.

2. The CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN according to claim 1 is characterized in that: The step 1 comprises: Step 11: Select a road section s with a length of L near the signal area, and obtain the vehicles n1, n2, ..., n in the road section s at time t. k The physical feature information includes vehicle information, vehicle position information and motion state information, wherein the vehicle information includes vehicle ID and vehicle type p, which is used to distinguish between human-driven HV and connected automatic vehicle CAV; the vehicle position information includes lane c and lane longitudinal position x; the motion state information includes speed v, acceleration a, yaw angle h, and target turning lane td; Step 12: Based on the single-vehicle data in road section s at time t, divide road section s into c×m units, where c is the number of lanes and m is the number of units divided in each lane. Determine the length l of a single unit division in each lane. min , expressed as: Where, L represents the length of the road section s; Step 13: Determine the unit position for each vehicle to be inserted in each lane. The specific operation is to first construct a feature matrix of c×m×f, where f is the number of feature layers, and then insert the physical feature information of the vehicle into the corresponding unit according to the relative position of the vehicle in the road section. For each lane, the calculation formula for the unit position for each vehicle to be inserted is: In the formula, m i and m i ′ represents the range of the horizontal position inserted into the feature matrix, that is, m i ' indicates the starting position of insertion, m i Indicates the end position of insertion, x j is the longitudinal position of the jth vehicle, s min is the initial position of the road segment s, l j is the length of the j-th vehicle, "[·]" represents the rounding function; Step 14: Select τ as the time interval, and use the feature matrix constructed at all moments within the selected control time T as the input of the convolutional neural network, expressed as: X=[M1,M2,M3,···,M n ] Where M t is the vehicle group feature matrix at time t, and n is the amount of data obtained within time T, that is, the training set data size input to the CNN model.

3. The CAV guidance control method for mixed vehicle groups near signal areas based on CP-CNN according to claim 1 is characterized in that: The step 2 comprises: Step 21: Select the average traffic volume in the approach road near the signal area at time t+τ as the training label and build a CNN deep network; Step 22: The constructed multi-dimensional feature map of the road section s at time t is normalized and input into the CNN model for training. The formula for data normalization is: In the formula, f i * is the standardized data, f i is the feature value that needs to be standardized in sample i, f min is the minimum value of the eigenvalue that needs to be standardized, f max is the maximum value of the eigenvalue that needs to be standardized; Step 23: After multiple convolutions and pooling, the model is trained through the fully connected layer output. The CP-CNN model output is the predicted average traffic flow q in the approach road near the signal area at time t+τ.

Images