CAVs car-following queue forming method for variable curvature complex road condition mixed traffic flow

Abstract

The invention discloses a variable curvature complex road condition mixed traffic flow-oriented CAVs car-following queue forming method. The method comprises the steps of obtaining state information of confluence CAVs and a confluence center of an initial CAVs queue; performing confluence scene division according to the state information of the confluence CAVs; for the collaborative lane-changing confluence scene, establishing a collaborative lane-changing game model considering the compromise degree of a human driver based on the confluence CAVs, the confluence center and the position state of HVs in the confluence direction, and generating a collaborative lane-changing confluence decision; for an overtaking lane-merging confluence scene, generating an initial confluence track based on road surface information by using a dynamic programming algorithm; constraining the initial confluence trajectory through a double-layer quadratic programming algorithm to generate an overtaking lane-merging confluence decision; and the confluence CAVs complete confluence through a collaborative lane changing confluence decision or an overtaking lane changing confluence decision, and join the confluence into the initial CAVs queue to form a CAVs car following queue. According to the method, the influence of HVs and road conditions (environmental information) on the formation of the CAVs car-following queue in a mixed traffic scene is considered.

Description

Technical Field

[0001] The present invention relates to the field of intelligent automobile control, and in particular to a method for forming a CAVs following queue for mixed traffic flows under complex road conditions with variable curvature. Background Art

[0002] With the advancement of the national new energy strategy and the rapid development of vehicle communication and control technology, the degree of vehicle intelligence is gradually improving. Intelligent vehicles with certain autonomous driving capabilities and the ability to communicate with other similar vehicles and roadside information units are called intelligent connected vehicles. However, autonomous driving technology is still in its ascending stage, so it is inevitable that CAVs (Connected Autonomous Vehicles, CAVs) and human-driven vehicles (Human-driven Vehicles, HVs) are mixed on today's urban roads.

[0003] By reviewing research papers and patents on CAVs following queues, it is found that CAVs following queues of a certain scale and penetration rate can help improve the efficiency and safety of traffic flow. Therefore, it is meaningful to make CAVs vehicles scattered on the road autonomously form a certain scale of following queues to improve the efficiency and safety of mixed traffic flow. It is necessary to design corresponding CAVs following queue formation schemes according to different traffic scenarios to achieve the formation of CAVs following queues.

[0004] The applicant consulted relevant patents and papers and found that in the current research, the formation of CAVs following queues only focuses on the information of CAVs vehicles themselves and similar CAVs vehicles, and does not fully consider the impact of surrounding heterogeneous HVs and road environment on the formation of CAVs following queues. The patent with publication number CN118298615A designs a CAVs following queue formation control method based on vehicle following distance. The method divides different following queues according to the following distance, and considers the computing device and corresponding storage medium of CAVs, which can effectively divide different CAVs following queues. However, this method does not consider the impact of HVs and road conditions on the formation of CAVs following queues in mixed traffic scenarios. The patent with publication number CN116913072A considers the driving information of similar CAVs vehicles before applying to join the original CAVs queue, queries the driving records of the CAVs applying to join the cloud and gives a credibility report, ensuring that the current CAVs will not affect the safe driving of the original CAVs queue after joining the original queue, and ensures the driving safety of the following queue. However, this method does not consider the impact of heterogeneous vehicle and environmental information. Summary of the invention

[0005] In view of the deficiencies of the above-mentioned prior art, the technical problem to be solved by the present invention is: how to provide a method for forming a CAVs following queue for mixed traffic flows on complex road conditions with variable curvature, establish a collaborative lane changing game model that takes into account the degree of courtesy of human drivers based on the position status of CAVs and HVs in the merging direction, and consider the influence of HVs and road conditions (environmental information) on the formation of CAVs following queues in mixed traffic scenarios, thereby improving the traffic efficiency and safety of CAVs following queues in mixed traffic flows with complex road conditions.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] The CAVs following queue formation method for mixed traffic flow on complex road conditions with variable curvature includes:

[0008] S1: Obtain the status information of the merging CAVs and the merging center of the initial CAVs queue;

[0009] S2: dividing the merging scene according to the status information of the merging CAVs to obtain a coordinated lane-changing merging scene or an overtaking and merging merging scene; executing step S3 for the coordinated lane-changing merging scene, and executing step S4 for the overtaking and merging merging scene;

[0010] S3: Based on the position states of merging CAVs, merging center and HVs in the merging direction, a cooperative lane-changing game model considering the degree of courtesy of human drivers is established to generate a cooperative lane-changing merging decision;

[0011] S4: Generate an initial merging trajectory based on road surface information using a dynamic programming algorithm; constrain the initial merging trajectory using a two-level quadratic programming algorithm to generate an overtaking and merging decision;

[0012] S5: The merging CAVs complete the merging through the coordinated lane-changing merging decision or the overtaking and merging merging decision, and join the initial CAVs queue to form a CAVs following queue.

[0013] Preferably, in step S1, the merging center is the vehicle center of a single CAV selected in the initial CAVs queue or the position center of the queue.

[0014] Preferably, in step S3, the coordinated lane-changing merging decision is generated through the following steps:

[0015] S301: Constructing the profit function of the cooperative lane-changing game model;

[0016] S302: Constructing a strategy combination of a coordinated lane-changing game model, including:

[0017] Strategy combination 1: CAVs change lanes, HVs cooperate;

[0018] Strategy combination 2: CAVs change lanes, HVs do not cooperate;

[0019] Strategy combination three: CAVs do not change lanes, HVs cooperate;

[0020] Strategy combination 4: CAVs do not change lanes, and HVs do not coordinate;

[0021] S303: Setting a corresponding benefit function for each strategy combination based on the benefit function;

[0022] S304: Calculate the benefits of each strategy combination according to the position states of the merging CAVs and the HVs in the merging direction; and use the strategy combination with the greatest benefit as the coordinated lane-changing merging decision of the merging CAVs.

[0023] Preferably, in step S301, the profit function is expressed as:

[0024]

[0025] Where: ω1, ω2, ω3 represent the corresponding income weights, ω1+ω2+ω3=1; accel lon _gain represents the lateral acceleration gain; accel lat _safe represents the lateral safety acceleration benefit; reward represents the reward benefit; collision_punish represents the collision penalty;

[0026] in:

[0027] Dist lon _gain = DG;

[0028]

[0029] Where: DG represents distance gain; Dist_gain represents position gain. Different strategy selections combined with actual road conditions will result in different position gains; width road Indicates the road width; t lc represents the set lane-changing time; reward is the reward item. For CAVs, the reward item is the merging reward queue_reward. For HVs, the reward item is obtained by multiplying the comity reward comity_reward by the comity index λ. The random number λ = rand(0,1) represents the comity psychological state of human drivers; collision_punish is the collision penalty item, which is generated when the CAVs strategy is lane-changing and the HVs strategy is non-cooperative; s HVs 、s CAVs are the longitudinal positions of the center of mass of HVs and CAVs respectively; length car is the vehicle length.

[0030] Preferably, in step S303:

[0031] The profit function of strategy combination 1 is: (U 11 CAVs ,U 11 HVs );

[0032] The profit function of strategy combination 2 is: (U 12 CAVs ,U 12 HVs );

[0033] The profit function of strategy combination three is: (U 21 CAVs ,U 21 HVs );

[0034] The profit function of strategy combination 4 is: (U 22 CAVs ,U 22 HVs );

[0035] in:

[0036] U 11 CAVs =ω1*DG CAVs_1 +ω2*accel lat _safe+ω3*queue_reward;

[0037] U 11 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ;

[0038] U 12 CAVs =ω1*DG CAVs_1 +ω2*accel lat _safe+ω3*queue_reward;

[0039] U 12 HVs =ω1*DG HVs_2 +ω2*0+ω3*0;

[0040] U 21 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0;

[0041] U21 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ;

[0042] U 22 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0.

[0043] Preferably, in step S4, the overtaking and merging decision is generated through the following steps:

[0044] S401: Generate a variable density constraint sampling matrix according to road surface information;

[0045] S402: Generate an initial confluence trajectory based on a variable density constraint sampling matrix through a dynamic programming algorithm;

[0046] S403: performing anti-collision constraints and smoothness constraints on the initial merging trajectory through a first-level quadratic programming algorithm to generate a first-level quadratic programming trajectory;

[0047] S404: applying a direction angle constraint to the first-level quadratic programming trajectory through a second-level quadratic programming algorithm to generate a second-level quadratic programming trajectory as an optimal merging trajectory;

[0048] S405: Using the optimal merging trajectory as a merging decision for overtaking and merging, so that the merging CAVs complete the merging by tracking the optimal merging trajectory and join the initial CAVs queue to form a CAVs following queue.

[0049] Preferably, in step S401, a variable density constraint sampling matrix is ​​constructed based on the lateral dynamics constraint sampling area, the collision constraint sampling area and the parallel sampling area;

[0050] 1) The formula for the lateral dynamics constraint sampling area is expressed as:

[0051]

[0052] Where: R0 represents the minimum turning radius; L represents the vehicle wheelbase; θ max It indicates the maximum turning angle of the outer wheel of the steering wheel; b indicates the front wheel track; M indicates the kingpin center distance.

[0053] 2) The formula for the collision constraint sampling area is expressed as:

[0054]

[0055] Where: l left_boundary ,l rigth_boundaryRespectively represent the lateral positions of the left and right boundaries of the road in the Frenet coordinate system; width car Indicates vehicle width; l car_position ,s car_position is the lateral and longitudinal position of the vehicle in the Frenet coordinate system; l safe Indicates the lateral safety margin; s safe It represents the longitudinal safety margin;

[0056] 3) The formula for the parallel sampling area is expressed as:

[0057]

[0058] Where: Sparse_area_l represents the size of the sparse area; dl represents the horizontal sampling interval; i represents the sparseness of the sampling points in the odd rows; s car_position Indicates the current position of the vehicle; length car represents the length of the vehicle; s represents the symbol of the longitudinal area of ​​the parallel sampling area;

[0059] 4) The formula of the variable density constraint sampling matrix is ​​expressed as:

[0060]

[0061] Where: nan represents invalid sampling points after constraint deletion and sparseness.

[0062] Preferably, in step S402, the dynamic programming algorithm uses the vehicle as the planning starting point, uses the columns of the variable density constraint sampling matrix as the traversal unit, finds the point with the minimum cost in each column as the optimal path point, and after traversing to the last row, uses the vehicle as the planning starting point to connect all the optimal path points in series to obtain the optimal path, i.e., the initial merging trajectory;

[0063] The initial confluence trajectory generated by the DP algorithm satisfies the following conditions:

[0064] min{cost pass a ij}(i=1,2,...,m;j=n);

[0065] cost pass a ij =min{cost pass a i1}+min{cost pass a i2}+...+min{cost pass a in-1};

[0066]

[0067] Where: S planning ,L planning are the total longitudinal and transverse lengths of the plan respectively; cost pass a i1 Represents the cost set of all sampling points in the first column of the variable density constraint sampling matrix, min{cost pass a i1} represents the sampling point with the smallest cost among all the sampling points in the first column, min{cost pass a ij} is the set of all points from the first column to the last column of the variable density constraint sampling matrix, where each sampling point satisfies the condition that it is the point with the minimum cost in the column. The optimal path, i.e. the initial confluence trajectory, is generated by connecting all columns through the sampling point with the minimum cost in units of columns and using the fifth-order polynomial as the connecting line;

[0068] in:

[0069] cost pass =cost similar +cost safe +cost smooth ;

[0070] cost similar =ω similar *(l′*l);

[0071]

[0072] cost smooth =ω smooth_dl *(dl'*dl)+ω smooth_ddl *(ddl'*ddl)+ω smooth_dddl *(dddl'*dddl);

[0073] Square min =min{(l obs_car -l(i)) 2 +(s obs_car -s(i)) 2}(i=1,2,...,10);

[0074] Where: similar ,ω safe ,ω smooth_dl ,ω smooth_ddl ,ω smooth_dddl Respectively represent the reference line similarity cost weight, safety cost weight and smoothness cost set; Square minis the shortest distance from the line segment between the sampling points connected by the quintic polynomial to the vehicle with heterogeneous obstacles; l = [l(1), l(2), ..., l(10)] T It is a set of transverse position values ​​of points on the quintic polynomial line segment, represented by a column vector. dl, ddl, dddl are the first, second, and third derivatives of the transverse coordinate l with respect to the longitudinal coordinate s respectively; l′ represents the first derivative of the transverse coordinate l with respect to the longitudinal coordinate s; l obs_car Indicates the lateral position of the vehicle with heterogeneous obstacles; s obs_car represents the longitudinal position of the vehicle with heterogeneous obstacles; s(i) represents the ordinate value of the quintic polynomial connecting two sampling points.

[0075] Preferably, in step S403, the first-level quadratic programming algorithm generates a first-level quadratic programming trajectory that satisfies the anti-collision inequality constraint and the smoothness equality constraint by solving the following objective function:

[0076] Objective function: f(l i ,l i ′,l i ″)=x T Ux-2x0 T x;

[0077] Where: represents the initial confluence trajectory generated by the dynamic programming algorithm; U represents the unit matrix; l i Represents the i-th horizontal coordinate, and the subscript i represents the vertical coordinate s i Corresponding to; i ′ represents the horizontal coordinate l i For the vertical coordinate s i The first derivative of i ″ represents the horizontal coordinate l i For the vertical coordinate s i The second derivative of

[0078] Anti-collision inequality constraints:

[0079] Smoothness equality constraint: A equation (2i-1:2i,3i-2:3i+3) (2n-2)×3n =A eq_sub (i=1,2,...,n-1);

[0080] b equation =[0 0 0 0 … 0 0] T (2n-2)×1 ;

[0081]

[0082] Δs=(l i+1 -l i );

[0083] Where: max{l rigth_constrained_area},min{l left_constrained_area} are the maximum horizontal coordinates of the right area of ​​the variable density constraint sampling matrix and the minimum horizontal coordinates of the left area of ​​the variable density constraint sampling matrix; width car Indicates vehicle width; A equation is the quadratic programming equality constraint coefficient matrix; b equation is the column vector on the right side of the equality constraint; d1 and d2 represent the distance values ​​from the center of the vehicle to the front and rear respectively; i represents an incrementally assigned variable, which fills the submatrix into the equality constraint coefficient matrix; n represents the number of path points to be optimized.

[0084] Preferably, in step S404, the second-level quadratic programming algorithm obtains the optimized variable value by solving the following objective function: Then y i Find the optimal solution for the next QP algorithm for the initial value of the iteration y i and i+1 Substitute Until |f(y i+1 )-f(y i )|≤10 -3 Stop the iteration when , and get the second-level quadratic planning trajectory

[0085] Objective function:

[0086] in:

[0087] H1=2E,f1=-2x1 T ;

[0088] A second (2i-1:2i,2i-1:2i) 2n×2n =A second_sub (i);

[0089] b second (2i-1:2i,1) 2n×1 =b second_sub (i);

[0090]

[0091] Where: represents the first-level quadratic programming trajectory; E represents the identity matrix;

[0092] Orientation Angle Constraint:

[0093]

[0094] c 1i = -tan(σθ ri )-b 1i ;

[0095] c 2i =tan(σθ ri )-b 1i ;

[0096]

[0097] Where: σ∈(-0.1,0.1) is the heading angle limit coefficient; k ri k r represents the stolen curvature, and the subscript i represents the k corresponding to the i-th optimization variable l. r value; The l in the equation represents the optimized horizontal coordinate value, the superscript 1 represents the optimized value obtained by the first quadratic programming algorithm, and the subscript i represents the number of all optimized variables in this horizontal coordinate; θ ri θ r represents the road direction angle, and the subscript i represents the θ corresponding to the i-th optimization variable l. r value.

[0098] Compared with the prior art, the CAVs following queue formation method for mixed traffic flow on complex road conditions with variable curvature in the present invention has the following beneficial effects:

[0099] For the collaborative lane-changing merging scenario, the present invention establishes a collaborative lane-changing game model that takes into account the degree of courtesy of human drivers based on the position states of merging CAVs (connected autonomous vehicles) and HVs (manually driven vehicles) in the merging direction, and then generates a collaborative lane-changing merging decision. First, the collaborative lane-changing game model can predict and adapt to the driving behavior of HVs, and considers the influence of HVs and road conditions (environmental information) on the formation of CAVs following queues in mixed traffic scenarios, especially considering the degree of courtesy of human drivers, which can reduce conflicts and waiting time during the lane-changing process, thereby improving the traffic efficiency of CAVs following queues under complex road conditions and mixed traffic flows. Secondly, the collaborative lane-changing game model can evaluate the potential risks of CAVs during the lane-changing process and ensure that CAVs change lanes under safe conditions. By considering the driving behavior of HVs, collisions with HVs can be effectively avoided, thereby improving the safety of CAVs following queues under complex road conditions and mixed traffic flows. At the same time, collaborative lane-changing reduces frequent lane changes and sudden braking between vehicles, making the process of CAVs following queue formation smoother and safer.

[0100] For overtaking and merging scenarios, the present invention uses a dynamic programming algorithm to generate an initial merging trajectory based on road surface information; and a double-layer quadratic programming algorithm is used to constrain the initial merging trajectory to generate an overtaking and merging decision. First, the dynamic programming algorithm can generate an initial merging trajectory based on road surface information, and the trajectory can take into account multiple factors such as road curvature, vehicle speed, and vehicle distance. By constraining and optimizing the initial trajectory through a double-layer quadratic programming algorithm, the smoothness and safety of the merging trajectory can be further ensured, thereby improving the safety and stability of the CAVs following queue formed under complex road conditions and mixed traffic flows. Secondly, through the precise calculation and constraint of the merging trajectory, it helps to assist CAVs in overtaking and merging more accurately, thereby improving the success rate of CAVs merging under complex road conditions and mixed traffic flows, and reducing traffic congestion and accident risks caused by merging failures. Finally, by accurately calculating and constraining the merging trajectory, CAVs can maintain a safe distance from other vehicles, reduce the occurrence of traffic conflicts, and contribute to the safety of the CAVs following queue. At the same time, it can also be flexibly adjusted according to different road conditions and traffic conditions, making the autonomous driving system more adaptable to complex traffic environments, thereby improving the practicality and reliability of the autonomous driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0102] Figure 1 The overall flow chart of the CAVs following queue formation method for mixed traffic flow on complex road conditions with variable curvature.

[0103] Figure 2 Select a schematic diagram for the merging center and merging vehicles.

[0104] Figure 3 Select a schematic for Distance Gain.

[0105] Figure 4 Schematic diagram of variable density constraint sampling matrix.

[0106] Figure 5 This is a schematic diagram of the cost calculation part.

[0107] Figure 6 Schematic diagram of the opened drivable area. DETAILED DESCRIPTION

[0108] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but only represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0109] The following is a further detailed description through specific implementation methods:

[0110] Example:

[0111] This embodiment discloses a method for forming a CAVs following queue for mixed traffic flow on complex road conditions with variable curvature.

[0112] like Figure 1 As shown in FIG. 1 , a CAVs following queue formation method for mixed traffic flow on complex road conditions with variable curvature includes:

[0113] S1: Obtain the status information of the merging CAVs and the merging center of the initial CAVs queue;

[0114] S2: dividing the merging scene according to the status information of the merging CAVs to obtain a coordinated lane-changing merging scene or an overtaking and merging merging scene; executing step S3 for the coordinated lane-changing merging scene, and executing step S4 for the overtaking and merging merging scene;

[0115] S3: Based on the position states of merging CAVs, merging center and HVs in the merging direction, a cooperative lane-changing game model considering the degree of courtesy of human drivers is established to generate a cooperative lane-changing merging decision;

[0116] S4: Generate an initial merging trajectory based on road surface information using a dynamic programming algorithm; constrain the initial merging trajectory using a two-level quadratic programming algorithm to generate an overtaking and merging decision;

[0117] S5: The merging CAVs complete the merging through the coordinated lane-changing merging decision or the overtaking and merging merging decision, and join the initial CAVs queue to form a CAVs following queue.

[0118] For the collaborative lane-changing merging scenario, the present invention establishes a collaborative lane-changing game model that takes into account the degree of courtesy of human drivers based on the position states of merging CAVs (connected autonomous vehicles) and HVs (manually driven vehicles) in the merging direction, and then generates a collaborative lane-changing merging decision. First, the collaborative lane-changing game model can predict and adapt to the driving behavior of HVs, and considers the influence of HVs and road conditions (environmental information) on the formation of CAVs following queues in mixed traffic scenarios, especially considering the degree of courtesy of human drivers, which can reduce conflicts and waiting time during the lane-changing process, thereby improving the traffic efficiency of CAVs following queues under complex road conditions and mixed traffic flows. Secondly, the collaborative lane-changing game model can evaluate the potential risks of CAVs during the lane-changing process and ensure that CAVs change lanes under safe conditions. By considering the driving behavior of HVs, collisions with HVs can be effectively avoided, thereby improving the safety of CAVs following queues under complex road conditions and mixed traffic flows. At the same time, collaborative lane-changing reduces frequent lane changes and sudden braking between vehicles, making the process of CAVs following queue formation smoother and safer.

[0119] For overtaking and merging scenarios, the present invention uses a dynamic programming algorithm to generate an initial merging trajectory based on road surface information; and a double-layer quadratic programming algorithm is used to constrain the initial merging trajectory to generate an overtaking and merging decision. First, the dynamic programming algorithm can generate an initial merging trajectory based on road surface information, and the trajectory can take into account multiple factors such as road curvature, vehicle speed, and vehicle distance. By constraining and optimizing the initial trajectory through a double-layer quadratic programming algorithm, the smoothness and safety of the merging trajectory can be further ensured, thereby improving the safety and stability of the CAVs following queue formed under complex road conditions and mixed traffic flows. Secondly, through the precise calculation and constraint of the merging trajectory, it helps to assist CAVs in overtaking and merging more accurately, thereby improving the success rate of CAVs merging under complex road conditions and mixed traffic flows, and reducing traffic congestion and accident risks caused by merging failures. Finally, by accurately calculating and constraining the merging trajectory, CAVs can maintain a safe distance from other vehicles, reduce the occurrence of traffic conflicts, and contribute to the safety of the CAVs following queue. At the same time, it can also be flexibly adjusted according to different road conditions and traffic conditions, making the autonomous driving system more adaptable to complex traffic environments, thereby improving the practicality and reliability of the autonomous driving system.

[0120] In order to better introduce the technical solution of the present invention, this embodiment is described through the following parts.

[0121] 1. Confluence Center

[0122] In this embodiment, the communication distance of a single CAVs approaching other CAVs is the merging radius; the merging center is the vehicle center of a single CAVs selected in the initial CAVs queue or the position center of the queue.

[0123] Inspired by the behavior of wolves in nature, when migrating, wolves always line up in a vertical line towards the migration destination. After encountering an accident, individuals in the wolf pack will use the various biological information left by other members of the same group along the way as a guide to return to the migration queue. Of course, this behavior can only be successful when the individual is within a certain range of the group queue. If it exceeds a certain range, the biological information left by the group will gradually disappear in nature, making it impossible for the individual to return to the migration queue.

[0124] Similarly, the communication distance and perception ability of CAVs in this embodiment are also affected by various factors in nature. The range of a single CAVs vehicle approaching the initial group of other CAVs is limited. This range is called the confluence radius (CR), and the center of the radius is the center point of the selected single CAVs individual or the initial CAVs queue that must be gathered together, which is also called the confluence center. Figure 2 As shown in the figure, the middle position of the two CAVs can be regarded as the merging center. Taking the merging center as the starting point, a merging radius value of a certain distance is set. Homogeneous intelligent connected vehicles within the merging radius can be added to the original population of CAVs through a series of vehicle motion planning to form a CAVs following queue.

[0125] The corresponding confluence center and confluence determination method are shown in Algorithm 1:

[0126]

[0127]

[0128] 2. Scene Division

[0129] In this embodiment, due to the different position states of the merging vehicles relative to the merging center, different merging scenarios need to be divided and different merging solutions are applied, such as Figure 1 As shown in the figure, the merging scenarios are divided according to the lanes where the merging CAVs are located and the position status between the merging CAVs and the initial CAVs queue, and a coordinated lane change merging scenario or an overtaking and merging merging scenario is generated. The scenario where a simple lane change is required to merge is classified as a coordinated lane change merging scenario, otherwise it is classified as an overtaking and merging merging scenario.

[0130] 3. Collaborative Lane-Changing Game Model

[0131] In this embodiment, the coordinated lane-changing merging decision is generated through the following steps:

[0132] S301: Constructing the profit function of the cooperative lane-changing game model;

[0133] S302: Constructing a strategy combination of a coordinated lane-changing game model, including:

[0134] Strategy combination 1: CAVs change lanes, HVs cooperate;

[0135] Strategy combination 2: CAVs change lanes, HVs do not cooperate;

[0136] Strategy combination three: CAVs do not change lanes, HVs cooperate;

[0137] Strategy combination 4: CAVs do not change lanes, and HVs do not coordinate;

[0138] S303: Setting a corresponding benefit function for each strategy combination based on the benefit function;

[0139] S304: Calculate the benefits of each strategy combination according to the position states of the merging CAVs and the HVs in the merging direction; and use the strategy combination with the greatest benefit as the coordinated lane-changing merging decision of the merging CAVs.

[0140] According to the position status of CAVs and HVs, a cooperative lane-changing game model is established. The different benefits corresponding to different strategy choices of the two parties in the game are given in Table 1.

[0141] Table 1. The payoff matrix of CAVs and HVs cooperative lane-changing game

[0142]

[0143] The profit function is expressed as:

[0144]

[0145] Where: ω1, ω2, ω3 represent the corresponding income weights, ω1+ω2+ω3=1; accel lon _gain represents the lateral acceleration gain; accel lat _safe represents the lateral safety acceleration benefit; reward represents the reward benefit; collision_punish represents the collision penalty;

[0146] in:

[0147] Dist lon _gain = DG;

[0148]

[0149] Where: DG represents distance gain; Dist_gain represents position gain. Different strategy selections combined with actual road conditions will result in different position gains; width road Indicates the road width; t lcrepresents the set lane-changing time; reward is the reward item. For CAVs, the reward item is the merging reward queue_reward. For HVs, the reward item is obtained by multiplying the comity reward comity_reward by the comity index λ. The random number λ = rand(0,1) represents the comity psychological state of the human driver; collision_punish is the collision penalty item, which is generated when the CAVs strategy is lane-changing and the HVs strategy is uncooperative; s HVs 、s CAVs are the longitudinal positions of the center of mass of HVs and CAVs respectively; length car is the vehicle length.

[0150] Specific:

[0151] The profit function of strategy combination 1 is: (U 11 CAVs ,U 11 HVs );

[0152] The profit function of strategy combination 2 is: (U 12 CAVs ,U 12 HVs );

[0153] The profit function of strategy combination three is: (U 21 CAVs ,U 21 HVs );

[0154] The profit function of strategy combination 4 is: (U 22 CAVs ,U 22 HVs );

[0155] in:

[0156] U 11 CAVs =ω1*DG CAVs_1 +ω2*accel lat _safe+ω3*queue_reward;

[0157] U 11 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ;

[0158] U 12 CAVs =ω1*DG CAVs_1+ω2*accel lat _safe+ω3*queue_reward;

[0159] U 12 HVs =ω1*DG HVs_2 +ω2*0+ω3*0;

[0160] U 21 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0;

[0161] U 21 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ;

[0162] U 22 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0.

[0163] The specific DG value is as follows: Figure 3 shown.

[0164] IV. Overtaking and merging decisions

[0165] In this embodiment, the overtaking and merging decision is generated through the following steps:

[0166] S401: Generate a variable density constraint sampling matrix according to road surface information;

[0167] S402: Generate an initial confluence trajectory based on a variable density constraint sampling matrix through a dynamic programming algorithm;

[0168] S403: performing anti-collision constraints and smoothness constraints on the initial merging trajectory through a first-level quadratic programming algorithm to generate a first-level quadratic programming trajectory;

[0169] S404: Using a second-level quadratic programming algorithm to constrain the heading angle of the first-level quadratic programming trajectory, and generating a second-level quadratic programming trajectory as an optimal merging trajectory;

[0170] S405: Using the optimal merging trajectory as a merging decision for overtaking and merging, so that the merging CAVs complete the merging by tracking the optimal merging trajectory and join the initial CAVs queue to form a CAVs following queue.

[0171] In the present invention, firstly, a sampling matrix is ​​generated according to real-time road surface information (such as curvature, slope, lane width, etc.), so that the sampling points can more accurately reflect the road characteristics, and the variable density design can increase the sampling density in areas where the road curvature changes greatly or the traffic conditions are complex, and reduce the sampling density in relatively flat or simple areas, thereby enhancing the adaptability of the merging trajectory and reducing the computational burden while ensuring accuracy. Secondly, the dynamic programming algorithm can find the global optimal merging trajectory under given constraints (such as time, distance, speed, etc.), ensuring the efficiency and safety of the merging process, and based on the variable density sampling matrix, the initial merging trajectory can more flexibly adapt to different road conditions and traffic conditions, providing a good foundation for subsequent trajectory optimization. Then, the initial merging trajectory is subjected to anti-collision constraints and smoothness constraints by the first-level quadratic programming algorithm. The anti-collision constraints can ensure that the merging CAVs will not collide with other vehicles during the merging process, improving road safety, and the smoothness constraints make the merging trajectory more continuous and smooth, improving the completeness and stability of the merging. Finally, the heading angle constraint is imposed on the first-level quadratic planning trajectory through the second-level quadratic programming algorithm. The heading angle constraint can ensure that the vehicle maintains a stable driving direction during the merging process, avoiding the vehicle losing control or deviating from the predetermined trajectory due to sudden changes in the heading angle. Through precise control of the heading angle, the merging trajectory can be further optimized to make it closer to actual road conditions and traffic conditions, thereby improving the accuracy and practicality of the trajectory.

[0172] 1. Variable density constraint sampling matrix

[0173] Before opening the initial merging path, a variable density constraint sampling matrix is ​​generated based on the road surface information. Figure 4 As shown in Figure 1, the variable density constraint sampling matrix is ​​constructed based on the lateral dynamics constraint sampling area, the collision constraint sampling area and the parallel sampling area. On the ordinary sampling matrix, considering the lateral dynamics and collision constraints, the sampling points are sparsely spaced in the parallel area to obtain the variable density constraint sampling matrix.

[0174] 1) The formula for the lateral dynamics constraint sampling area is expressed as:

[0175]

[0176] Where: R0 represents the minimum turning radius; L represents the vehicle wheelbase; θ max It indicates the maximum turning angle of the outer wheel of the steering wheel; b indicates the front wheel track; M indicates the kingpin center distance.

[0177] 2) The formula for the collision constraint sampling area is expressed as:

[0178]

[0179] Where: lleft_boundary ,l rigth_boundary Respectively represent the lateral positions of the left and right boundaries of the road in the Frenet coordinate system; width car Indicates vehicle width; l car_position ,s car_position is the lateral and longitudinal position of the vehicle in the Frenet coordinate system; l safe It represents the lateral safety margin, that is, it is necessary to add a certain safety margin for the road boundary and the collision area around the vehicle with heterogeneous obstacles to prevent vehicle collisions caused by vehicle control accuracy or perception system delay; s safe It represents the longitudinal safety margin;

[0180] 3) The formula for the parallel sampling area is expressed as:

[0181]

[0182] Where: Sparse_area_l represents the size of the sparse area; dl represents the horizontal sampling interval; i represents the sparseness of the sampling points in the odd rows; s car_position Indicates the current position of the vehicle; length car Indicates the length of the vehicle; s safe represents the safety margin; s represents the sign of the longitudinal area of ​​the parallel sampling area;

[0183] 4) The formula of the variable density constraint sampling matrix is ​​expressed as:

[0184]

[0185] Where: nan represents invalid sampling points after constraint deletion and sparseness; a m,n-1 Represents the sampling points of m rows and n-1 columns.

[0186] The present invention proposes a new variable density constraint sampling matrix, which can effectively reduce the initial path generation time, improve algorithm efficiency, and cope with complex road conditions with variable curvature.

[0187] 2. Dynamic Programming (DP) Algorithm

[0188] In this embodiment, after the sampling matrix is ​​pruned and sparsely processed, a dynamic programming algorithm is used to obtain a preliminary optimal path. A fifth-order polynomial based on the Frenet coordinate system is used for the connection between discrete path points to calculate the cost of each discrete path point. The dynamic programming algorithm uses the ego vehicle (merging CAVs) as the planning starting point and the column of the variable density constraint sampling matrix as the traversal unit. In each column, the point with the minimum cost is found as the optimal path point. After traversing to the last row, all the optimal path points are connected in series with the ego vehicle as the planning starting point to obtain the optimal path, i.e., the initial merging trajectory;

[0189] The initial confluence trajectory generated by the DP algorithm satisfies the following conditions:

[0190] min{cost pass a ij}(i=1,2,...,m;j=n);

[0191] cost pass a ij =min{cost pass a i1}+min{cost pass a i2}+...+min{cost pass a in-1};

[0192]

[0193] Where: S planning ,L planning are the total longitudinal and transverse lengths of the plan respectively; cost pass a i1 Represents the cost set of all sampling points in the first column of the variable density constraint sampling matrix, min{cost pass a i1} represents the sampling point with the smallest cost among all the sampling points in the first column, and so on. min{cost pass a ij} is the set of all points from the first column to the last column of the variable density constraint sampling matrix, where each sampling point satisfies the condition that it is the point with the minimum cost in the column. The optimal path, i.e. the initial confluence trajectory, is generated by connecting all columns through the sampling point with the minimum cost in units of columns and using the fifth-order polynomial as the connecting line; a ij Represents the sampling point in the i-th row and j-th column of the variable density constraint sampling matrix;

[0194] in:

[0195] cost pass =cost similar +cost safe +cost smooth ;

[0196] cost similar =ω similar *(l′*l);

[0197]

[0198] cost smooth =ω smooth_dl *(dl'*dl)+ωsmooth_ddl *(ddl'*ddl)+ω smooth_dddl *(dddl'*dddl);

[0199] Square min =min{(l obs_car -l(i)) 2 +(s obs_car -s(i)) 2}(i=1,2,...,10);

[0200] Where: similar ,ω safe ,ω smooth_dl ,ω smooth_ddl ,ω smooth_dddl Respectively represent the reference line similarity cost weight, safety cost weight and smoothness cost set; Square min is the shortest distance from the line segment between the sampling points connected by the quintic polynomial to the heterogeneous obstacle vehicles (i.e., HVs in the merging direction of the merging CAVs), as Figure 5 As shown; l = [l(1), l(2), ..., l(10)] T is the set of horizontal position values ​​of points on the quintic polynomial line segment, represented by a column vector. dl, ddl, dddl are the first, second, and third derivatives of the horizontal coordinate l with respect to the vertical coordinate s, respectively, which are used as the consideration standard for the smoothness cost; l′ represents the first derivative of the horizontal coordinate l with respect to the vertical coordinate s; l obs_car represents the lateral position of the heterogeneous obstacle vehicle (i.e., HVs in the merging direction of the merging CAVs); s obs_car represents the longitudinal position of the vehicle with heterogeneous obstacles; s(i) represents the ordinate value of the quintic polynomial connecting two sampling points.

[0201] 3. First-level quadratic programming (QP) algorithm

[0202] In this embodiment, the first-level quadratic programming algorithm generates the first-level quadratic programming trajectory that satisfies the anti-collision inequality constraint and the smoothness equality constraint by solving the following objective function:

[0203] In the actual planning process, only the lateral trajectory value, i.e., the lateral coordinate l, needs to be considered, because the lane change is determined by the lateral coordinate l. The longitudinal trajectory value, i.e., the lateral coordinate s, increases according to the road distance, and it is a fixed value.

[0204] Objective function: f(l i ,l i ′,l i ″)=x T Ux-2x0 T x;

[0205] Where: represents the initial confluence trajectory generated by the dynamic programming algorithm; U represents the unit matrix; l i Represents the i-th horizontal coordinate, and the subscript i represents the vertical coordinate s i Corresponding to; i ′ represents the horizontal coordinate l i For the vertical coordinate s i The first derivative of i ″ represents the horizontal coordinate l i For the vertical coordinate s i The second derivative of

[0206] Anti-collision inequality constraints:

[0207] Smoothness equality constraint: A equation (2i-1:2i,3i-2:3i+3) (2n-2)×3n =A eq_sub (i=1,2,...,n-1);

[0208] b equation =[0000…00] T (2n-2)×1 ;

[0209]

[0210] Δs=(l i+1 -l i );

[0211] Where: max{l rigth_constrained_area},min{l left_constrained_area} are the maximum horizontal coordinates of the right area of ​​the variable density constraint sampling matrix and the minimum horizontal coordinates of the left area of ​​the variable density constraint sampling matrix; width car Indicates vehicle width; A equation is the quadratic programming equality constraint coefficient matrix; b equation is the column vector on the right side of the equality constraint; d1 and d2 represent the distance values ​​from the center of the vehicle to the front and rear respectively; i represents an incrementally assigned variable, which fills the submatrix into the equality constraint coefficient matrix; n represents the number of path points to be optimized.

[0212] The specific derivation process is as follows:

[0213] The first-level quadratic programming algorithm performs corner point collision constraints and further smooth optimization on the initial trajectory of the DP algorithm based on the car corner points in the Frenet coordinate system to eliminate the possibility of collision during the merging process. The linear approximation of the vehicle corner point calculation in the Frenet coordinate system is shown in the following formula:

[0214]

[0215] in: is the lateral position information of the four corner points of the vehicle; d1 and d2 are the distances from the head of the vehicle to the center of mass of the vehicle, and the distances from the tail of the vehicle to the center of mass of the vehicle, respectively; θ h is the vehicle heading angle;

[0216] Further, the inequality constraints of the QP algorithm are constructed according to the anti-collision constraint area in the DP algorithm variable density constraint matrix in step 3, as shown in the following formula:

[0217]

[0218] Where: max{l rigth_constrained_area},min{l left_constrained_area} are the maximum value of the horizontal coordinate of the right area of ​​the variable density constraint matrix and the minimum value of the horizontal coordinate of the left area of ​​the variable density constraint matrix, respectively. Figure 6 shown.

[0219] At the same time, we need to consider the smoothness of the planned trajectory again. The expression of the planned trajectory of the quadratic program is l = f(s). To make the trajectory smooth, it is converted to making the absolute value of each order derivative of f(s) as small as possible. Here, the segmented acceleration optimization method is used to constrain the planned trajectory of the quadratic program. There are two points in the Frenet coordinate system (l i ,s i ),(l i+1 ,s i+1 ), assuming that the connection point l i With l i+1 The third derivative of the quintic polynomial f(s) in is a constant, then the derivatives of f(s) above the fourth order are all zero, which can make the trajectory smoother. i+1 =f(s i+1 ) and l i ′i+1=f′(s i+1 ) are respectively expanded by the third-order and second-order Taylor, as shown in the following formula:

[0220]

[0221] The following matrix form is obtained by combining:

[0222]

[0223] Where: Δs=(l i+1 -l i );

[0224] After obtaining the inequality constraints to prevent collisions and the equality constraints to make the trajectory smooth, we can use them as constraints for quadratic programming to obtain a confluence trajectory of one-layer quadratic programming. The basic form of quadratic programming is shown in the following formula:

[0225]

[0226] Where x = [l1,l1′,l1″,l2,l2″,l2″,...,l n ,l n ′,l n ″] T is the column vector of decision variables, n is the number of variable groups to be optimized; H is the quadratic coefficient matrix, which is a symmetric matrix; f T A is the linear term coefficient column vector; equation is the quadratic programming equality constraint coefficient matrix, b equation is the column vector on the right side of the equality constraint; A is the quadratic programming inequality constraint coefficient matrix, b is the column vector on the right side of the inequality constraint; According to the collision corner inequality constraint and smoothness equality constraint mentioned above, A equation ,b equation The general form of is shown in the following formula:

[0227] A equation (2i-1:2i,3i-2:3i+3) (2n-2)×3n =A eq_sub (i=1,2,...,n-1);

[0228] b equation =[0 0 0 0 … 0 0] T (2n-2)×1 ;

[0229] in:

[0230] The anti-collision inequality constraint is shifted to satisfy the expression of the quadratic programming inequality constraint Ax≤b, and then the decision variables are further separated to obtain the following expression, that is, the inequality constraint A sub x≤b sub in the form of.

[0231]

[0232] Then the expression of A and b is as follows:

[0233] A(8i-7:8i,3i-2:3i) 8n×3n =A sub (i=2,3,...,n);

[0234] b(8i-7:8i,1) 8n×1 =b sub (i=2,3,...,n);

[0235] The initial value of the optimization vector is obtained by the DP algorithm The QP algorithm optimizes the decision vector in accordance with equality and inequality constraints based on the initial value x0 of the optimization vector, that is, it requires that the optimized x value has a certain similarity with x0, which is converted into the expected value of the following formula to be as small as possible. The smaller the value of the following formula, the greater the similarity.

[0236]

[0237] Further separating the variables in the above formula gives the following formula:

[0238]

[0239] In summary, we have f(l i ,l i ′,l i ″)=x T Ux-2x0 T x, U is the unit matrix, and then let H = 2U, f = -2x0 T , and the standard form of the quadratic programming is obtained. By solving the quadratic programming, the first-level quadratic programming trajectory value that satisfies the anti-collision inequality constraint and the smoothness equality constraint is obtained.

[0240] 4. Second-level quadratic programming (QP) algorithm

[0241] In this embodiment, the second-level quadratic programming algorithm obtains the optimized variable value by solving the following objective function: Then y i Find the optimal solution for the next QP algorithm for the initial value of the iteration y i and i+1 Substitute Until |f(y i+1 )-f(y i )|≤10 -3 Stop the iteration when , and get the second-level quadratic planning trajectory

[0242] Objective function:

[0243] in:

[0244] H1=2E,f1=-2x1 T , E=O 2n×2n ;

[0245] A second (2i-1:2i,2i-1:2i) 2n×2n =A second_sub (i);

[0246] b second (2i-1:2i,1) 2n×1 =b second_sub (i);

[0247]

[0248] Where: represents the trajectory value of the first-level quadratic programming; E represents the unit matrix;

[0249] Orientation Angle Constraint:

[0250]

[0251]

[0252] c 1i = -tan(σθ ri )-b 1i ;

[0253] c 2i =tan(σθ ri )-b 1i ;

[0254]

[0255] Where: σ∈(-0.1,0.1) is the heading angle limit coefficient; k ri k r represents the stolen curvature, and the subscript i represents the k corresponding to the i-th optimization variable l. r value; The l in the equation represents the optimized horizontal coordinate value, the superscript 1 represents the optimized value obtained by the first quadratic programming algorithm, and the subscript i represents the number of all optimized variables in this horizontal coordinate; θ ri θ r represents the road direction angle, and the subscript i represents the θ corresponding to the i-th optimization variable l. r value.

[0256] The specific derivation process is as follows:

[0257] On roads with variable curvature, it is hoped that the heading angle of the vehicle can be within a certain range of the reference line direction angle when tracking the planned trajectory.

[0258] The difference between the vehicle head direction and the road heading direction is expected to be within a controllable range to prevent the vehicle from colliding with the road boundary. The heading angle of the planned trajectory represents the heading angle of the vehicle after tracking the trajectory. Therefore, the heading angle of the planned trajectory is limited. The relationship between the actual trajectory heading angle and the reference line heading angle is shown in the following formula:

[0259]

[0260] Among them, θ r ,θ p is the reference line direction angle, i.e. the actual planned trajectory direction angle; l,l′ is the horizontal coordinate value of the actual trajectory at this point and its first-order derivative with respect to s; k r is the curvature value of the reference line at this point.

[0261] If you want to make the actual trajectory azimuth within a certain range of the reference line trajectory direction angle, you only need to consider the value of the variable [l,l′] in the subsequent nonlinear part. Suppose the actual trajectory direction angle limit expression is as shown below:

[0262] (1-σ)θ r ≤θ p ≤(1+σ)θ r ;

[0263] Combining the above two equations, we get the following heading angle constraint inequality:

[0264]

[0265] Where: σ∈(-0.1,0.1) is the heading angle limiting coefficient.

[0266] From the above formula, it can be obtained that when the heading angle limit coefficient is given, the nonlinear factor As long as the variables [l,l′] satisfy the nonlinear inequality constraints given in equation 3.45, a planning trajectory that meets the requirements can be obtained.

[0267] For optimization problems with nonlinear constraints, the second-level quadratic programming algorithm uses the Sequential Quadratic Programming (SQP) algorithm to optimize the variables in accordance with the nonlinear inequality constraints. The SQP algorithm decomposes an optimization problem with nonlinear constraints into multiple QP problems with linear constraints. The trajectory values ​​obtained by the first-level QP algorithm are As the initial iteration point. Then put it at the initial point Performing a finite-term Taylor expansion at the position gives the following formula:

[0268]

[0269] After finding the partial derivatives of each variable and inserting them into the initial iteration point, we get:

[0270]

[0271] Combining the heading angle constraint inequalities, we get:

[0272]

[0273] Separate order Simplifying the above formula, we get:

[0274]

[0275] Let c 1i = -tan(σθ ri )-b 1i , c 2i =tan(σθ ri )-b 1i , Further simplifying the above formula, we get:

[0276]

[0277] The approximate linear inequality constraints are as follows:

[0278]

[0279] For sequential quadratic normalization, the new decision variables y = [l1,l1′,l2,l2′,...,l n ,l n ′] T (i=1,2,...,n), its inequality constraint coefficient matrix A second (2i-1:2i,2i-1:2i) 2n×2n =A second_sub (i), b second (2i-1:2i,1) 2n×1 =b second_sub (i).

[0280] in:

[0281] Similarly, the coefficient matrix of the second-layer QP algorithm quadratic term H1=2E, f1=-2x1 T , where E is the identity matrix and E = O 2n×2n By solving the quadratic programming Get the optimized variable value Then use this as the initial value of the iteration to obtain the optimal solution of the next QP algorithm Substitute them into The iteration stops until the convergence condition shown in the following formula is met, and the optimal trajectory value that satisfies the heading angle constraint is obtained.

[0282] |f(y i+1 )-f(y i )|≤10 -3 .

[0283] 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 the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A CAVs following queue formation method for mixed traffic flow on complex road conditions with variable curvature, characterized in that: include: S1: Obtain the status information of the merging CAVs and the merging center of the initial CAVs queue; S2: dividing the merging scene according to the status information of the merging CAVs to obtain a coordinated lane-changing merging scene or an overtaking and merging merging scene; executing step S3 for the coordinated lane-changing merging scene, and executing step S4 for the overtaking and merging merging scene; S3: Based on the position states of merging CAVs, merging center and HVs in the merging direction, a cooperative lane-changing game model considering the degree of courtesy of human drivers is established to generate a cooperative lane-changing merging decision; S4: Generate an initial merging trajectory based on road surface information using a dynamic programming algorithm; constrain the initial merging trajectory using a two-level quadratic programming algorithm to generate an overtaking and merging decision; S5: The merging CAVs complete the merging through the coordinated lane-changing merging decision or the overtaking and merging merging decision, and join the initial CAVs queue to form a CAVs following queue.

2. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 1, characterized in that: In step S1, the merging center is the vehicle center of a single CAV selected in the initial CAVs queue or the position center of the queue.

3. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 1, characterized in that: In step S3, a coordinated lane-changing merging decision is generated through the following steps: S301: Constructing the profit function of the cooperative lane-changing game model; S302: Constructing a strategy combination of a coordinated lane-changing game model, including: Strategy combination 1: CAVs change lanes, HVs cooperate; Strategy combination 2: CAVs change lanes, HVs do not cooperate; Strategy combination three: CAVs do not change lanes, HVs cooperate; Strategy combination 4: CAVs do not change lanes, and HVs do not coordinate; S303: Setting a corresponding benefit function for each strategy combination based on the benefit function; S304: Calculate the benefits of each strategy combination according to the position states of the merging CAVs and the HVs in the merging direction; and use the strategy combination with the greatest benefit as the coordinated lane-changing merging decision of the merging CAVs.

4. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 3, characterized in that: In step S301, the profit function is expressed as: U=ω1*Dist lon _gain+ω2*accel lat _safe+ω3*(reward / collision_punish); Where: ω1, ω2, ω3 represent the corresponding income weights, ω1+ω2+ω3=1; accel lon _gain represents the lateral acceleration gain; accel lat _safe represents the lateral safety acceleration benefit; reward represents the reward benefit; collision_punish represents the collision penalty; in: Dist lon _gain=DG; Where: DG represents distance gain; Dist_gain represents position gain. Different strategy selections combined with actual road conditions will result in different position gains; width road Indicates the road width; t lc Indicates the set lane change time; reward is the reward item. For CAVs, the reward item is the merging reward queue_reward. For HVs, the reward item is obtained by multiplying the comity reward comity_reward by the comity index λ. The random number λ = rand (0, 1) represents the comity psychological state of the human driver; collision_punish is the collision penalty item, which is generated when the CAVs strategy is lane change and the HVs strategy is non-cooperation; s HVs 、s CAVs are the longitudinal positions of the center of mass of HVs and CAVs respectively; length car is the vehicle length.

5. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 4, characterized in that: In step S303: The profit function of strategy combination 1 is: (U 11 CAVs , U 11 HVs ); The profit function of strategy combination 2 is: (U 12 CAVs , U 12 HVs ); The profit function of strategy combination three is: (U 21 CAVs , U 21 HVs ); The profit function of strategy combination 4 is: (U 22 CAVs , U 22 HVs ); in: U 11 CAVs =ω1*DG CAVs_1 +ω2*accel lat _safe+ω3*queue_reward; U 11 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ; U 12 CAVs =ω1*DG CAVs_1 +ω2*accel lat _safe+ω3*queue_reward; U 12 HVs =ω1*DG HVs_2 +ω2*0+ω3*0; U 21 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0; U 21 HVs =ω1*DG HVs_1 +ω2*accel lat _safe+ω3*comity_reward*λ; U 22 CAVs =ω1*DG CAVs_2 +ω2*0+ω3*0.

6. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 1, characterized in that: In step S4, the overtaking and merging decision is generated through the following steps: S401: Generate a variable density constraint sampling matrix according to road surface information; S402: Generate an initial confluence trajectory based on a variable density constraint sampling matrix through a dynamic programming algorithm; S403: performing anti-collision constraints and smoothness constraints on the initial merging trajectory through a first-level quadratic programming algorithm to generate a first-level quadratic programming trajectory; S404: applying a direction angle constraint to the first-level quadratic programming trajectory through a second-level quadratic programming algorithm to generate a second-level quadratic programming trajectory as an optimal merging trajectory; S405: Using the optimal merging trajectory as a merging decision for overtaking and merging, so that the merging CAVs complete the merging by tracking the optimal merging trajectory and join the initial CAVs queue to form a CAVs following queue.

7. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 6, characterized in that: In step S401, a variable density constraint sampling matrix is ​​constructed based on the lateral dynamics constraint sampling area, the collision constraint sampling area and the parallel sampling area; 1) The formula for the lateral dynamics constraint sampling area is expressed as: Where: R0 represents the minimum turning radius; L represents the vehicle wheelbase; θ max It indicates the maximum turning angle of the outer wheel of the steering wheel; b indicates the front wheel track; M indicates the kingpin center distance. 2) The formula for the collision constraint sampling area is expressed as: Where: l left_boundary , l rigth_boundary Respectively represent the lateral positions of the left and right boundaries of the road in the Frenet coordinate system; width car Indicates vehicle width; l car_position ,s car_position is the lateral and longitudinal position of the vehicle in the Frenet coordinate system; l safe Indicates the lateral safety margin; s safe It represents the longitudinal safety margin; 3) The formula for the parallel sampling area is expressed as: Where: Sparse_area_l represents the size of the sparse area; dl represents the horizontal sampling interval; i represents the sparseness of the sampling points in the odd rows; s car_position Indicates the current position of the vehicle; length car represents the length of the vehicle; s represents the symbol of the longitudinal area of ​​the parallel sampling area; 4) The formula of the variable density constraint sampling matrix is ​​expressed as: Where: nan represents invalid sampling points after constraint deletion and sparseness.

8. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 7, characterized in that: In step S402, the dynamic programming algorithm uses the vehicle as the planning starting point and the columns of the variable density constraint sampling matrix as the traversal unit, finds the point with the minimum cost in each column as the optimal path point, and after traversing to the last row, connects all the optimal path points in series with the vehicle as the planning starting point to obtain the optimal path, i.e., the initial merging trajectory; The initial confluence trajectory generated by the DP algorithm satisfies the following conditions: min{cost pass a ij }(i=1,2,...,m;j=n); cost pass a ij =min{cost pass a i1 }+min{cost pas a i2 }+...+min{cost pass a in-1 }; Where: S planning , L planning are the total longitudinal and transverse lengths of the plan respectively; cost pass a i1 Represents the cost set of all sampling points in the first column of the variable density constraint sampling matrix, min{cost pass a i1 } represents the sampling point with the smallest cost among all the sampling points in the first column, min{cost pass a ij } is the set of all points from the first column to the last column of the variable density constraint sampling matrix, where each sampling point satisfies the condition that it is the point with the minimum cost in the column. The optimal path, i.e. the initial confluence trajectory, is generated by connecting all columns through the sampling point with the minimum cost in units of columns and using the fifth-order polynomial as the connecting line; in: cost pass =cost similar +cost safe +cost smooth ; cost similar =ω similarr *(l′*l); cost smooth =ω smooth_dl *(dl′*dl)+ω smooth_ddl *(ddl′*ddl)+ω smooth_dddl *(dddl′*dddl); Square min =min{(l obs_car -l(i)) 2 +(s obs_car -s(i)) 2 }(i=1,2,...,10); Where: similar ,ω safe ,ω smooth_dl ,ω smooth_ddl ,ω smooth_dddl Respectively represent the reference line similarity cost weight, safety cost weight and smoothness cost set; Square min is the shortest distance from the line segment between the sampling points connected by the quintic polynomial to the heterogeneous obstacle vehicle; l = [l(1), l(2), ..., l(10)] T It is a set of transverse position values ​​of points on the quintic polynomial line segment, represented by a column vector. dl, ddl, dddl are the first, second, and third derivatives of the transverse coordinates with respect to the longitudinal coordinate s respectively; l′ represents the first derivative of the transverse coordinate with respect to the longitudinal coordinate s; l obs_car Indicates the lateral position of the vehicle with heterogeneous obstacles; s obs_car represents the longitudinal position of the vehicle with heterogeneous obstacles; s(i) represents the ordinate value of the quintic polynomial connecting two sampling points.

9. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 6, characterized in that: In step S403, the first-level quadratic programming algorithm generates a first-level quadratic programming trajectory that satisfies the anti-collision inequality constraint and the smoothness equality constraint by solving the following objective function: Objective function: f(l i , l i ′,l i ″)=x T Ux-2x0 T x; Where: represents the initial confluence trajectory generated by the dynamic programming algorithm; U represents the unit matrix; l i Represents the i-th horizontal coordinate, and the subscript i represents the vertical coordinate s i Corresponding to; i ′ represents the horizontal coordinate l i For the vertical coordinate s i The first derivative of i ″ represents the horizontal coordinate l i For the vertical coordinate s i The second derivative of Anti-collision inequality constraints: Smoothness equality constraint: A equatio n(2i-1∶2i,3i-2:3i+3) (2n-2)×3n =A eq_sub (i=1, 2, ..., n-1); Δs=(l i+1 -l i ); Where: max{l rigth_constrained_area },min{l left_constrained_area } are the maximum horizontal coordinates of the right area of ​​the variable density constraint sampling matrix and the minimum horizontal coordinates of the left area of ​​the variable density constraint sampling matrix; width car Indicates vehicle width; A equation is the quadratic programming equality constraint coefficient matrix; b equation is the column vector on the right side of the equality constraint; d1 and d2 represent the distance values ​​from the center of the vehicle to the front and rear respectively; i represents an incrementally assigned variable, which fills the submatrix into the equality constraint coefficient matrix; n represents the number of path points to be optimized.

10. The method for forming a CAVs following queue for mixed traffic flow on a road with complex variable curvature as claimed in claim 6, characterized in that: In step S404, the second-level quadratic programming algorithm obtains the optimized variable value by solving the following objective function: Then y i Find the optimal solution for the next QP algorithm for the initial value of the iteration y i and i+1 Substitute Until |f(y i+1 )-f(y i )|≤10 -3 Stop the iteration when , and get the second-level quadratic planning trajectory Objective function: in: <h2 style=";text-align:left;direction:ltr">H1 = 2E = f1 = 2x1<h2 style=";text-align:left;direction:ltr"> T <h2 style=";text-align:left;direction:ltr"> ; And second (2i-1:2i,2i-1:2i) 2n×2n =A second_sub (i); b second (2i-1:2i,1) 2n×1 =b secind_sub (and); Where: represents the first-level quadratic programming trajectory; E represents the identity matrix; Orientation constraint: <h2 style=";text-align:left;direction:ltr">c<h2 style=";text-align:left;direction:ltr"> 1i <h2 style=";text-align:left;direction:ltr"> =-tan(σθ<h2 style=";text-align:left;direction:ltr"> ri <h2 style=";text-align:left;direction:ltr"> )-b<h2 style=";text-align:left;direction:ltr"> 1i <h2 style=";text-align:left;direction:ltr"> ; <h2 style=";text-align:left;direction:ltr">c<h2 style=";text-align:left;direction:ltr"> 2i <h2 style=";text-align:left;direction:ltr"> =tan(σθ<h2 style=";text-align:left;direction:ltr"> ri <h2 style=";text-align:left;direction:ltr"> )-b<h2 style=";text-align:left;direction:ltr"> 1i <h2 style=";text-align:left;direction:ltr"> ; Where: σ∈(-0.1, 0.1) is the heading angle limit coefficient; k ri k r represents the stolen curvature, and the subscript i represents the k corresponding to the i-th optimization variable l. r value; In the above, the lateral coordinates after optimization are represented. The superscript 1 represents the optimization value obtained by the first quadratic programming algorithm. The subscript i represents the number of all optimization variables in this lateral coordinate. ri θ r represents the road direction angle, the subscript i represents the θ corresponding to the i-th optimization variable r value.

Images