Collaborative decision-making and control method for intelligent connected vehicles in mixed traffic in weaving areas of highways

Abstract

The present invention relates to a collaborative decision-making and control method for intelligent connected vehicles in mixed traffic for highway weaving areas, and belongs to the field of intelligent connected vehicles. The method first forms a unified macro and micro mixed traffic driving rule based on high-precision map data information and road right-of-way information in a fixed geometric reference system of the weaving area. On this basis, a master-slave game model is established, and a particle swarm optimization method is combined to generate single-vehicle / multi-vehicle decision results that comply with traffic rules. The vehicle side establishes a single-vehicle input state / multi-vehicle input state and corresponding constraints respectively, and according to the decision result, the vehicle controller based on reinforcement learning outputs the corresponding steering wheel angle and vehicle longitudinal acceleration to the vehicle bottom layer for execution, thereby realizing the decision-making and control of intelligent connected vehicles in highway weaving areas. The present invention ensures the safety, stability and efficiency of highway weaving areas in mixed traffic.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent connected vehicles and relates to a collaborative decision-making and control method for intelligent connected vehicles in mixed traffic in an interweaving area of ​​a highway. Background Art

[0002] With the increase in transportation volume, congestion and safety issues on highways have become increasingly prominent, especially in highway weaving areas, which serve as transition sections for vehicles entering and exiting highways. Vehicles here need to frequently change speed and lanes to complete the merging / exiting process. Therefore, the operation of vehicles in weaving areas is more complicated than other sections of the road, and it is a high-incidence area for traffic congestion and traffic accidents. Intelligent connected vehicles use the Internet of Vehicles to achieve vehicle-to-vehicle collaboration and vehicle-road collaboration, which can achieve traffic optimization with the goals of "safety, efficiency, comfort, and energy saving", and improve the operating efficiency of highways and vehicle driving safety. Before full intelligence and networking, mixed traffic environments where intelligent connected vehicles and traditional manually driven vehicles share the road will become the norm for a long time in the future. In mixed traffic, vehicles interfere with each other, and the traffic flow operation pattern will undergo structural changes. The collaborative decision-making and control of intelligent connected vehicles that consider the relationship between vehicles and the uncertainty of manually driven vehicles can ensure driving safety and road traffic efficiency in weaving areas.

[0003] Currently, existing collaborative decision-making and control technologies for intelligent connected vehicles in highway weaving zones, as well as publicly available patent literature, focus on vehicle control while ignoring the impact of road traffic regulations. For example, patent application CN115424456A, titled "Collaborative Adaptive Cruise Control Optimization Control Method for Highway Weaving Zones," proposes a collaborative adaptive cruise control optimization control method for highway weaving zones that incorporates a rolling horizon-based model predictive control algorithm to identify the traffic behavior of vehicles in adjacent lanes. However, this method ignores the constraints of the vehicle's own state and the impact of road traffic regulations, which can negatively impact road safety. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a collaborative decision-making and control method for intelligent connected vehicles in mixed traffic in highway weaving areas to ensure the safety, stability and efficiency of the entire collaborative decision-making and control process.

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

[0006] A collaborative decision-making and control method for intelligent connected vehicles in mixed traffic in a weaving area of ​​a highway, the method comprising:

[0007] 1) Based on the road coordinate system of the HD map, road right information is integrated with the HD map data to form road right rules and assign priorities to vehicles on the road;

[0008] 2) Within the HD map's road coordinate system, the driving state of each vehicle is parametrically expressed, establishing safety boundary rules and behavioral target rules for each vehicle's surroundings.

[0009] 3) A master-slave game model is established in the cloud, and combined with a particle swarm optimization algorithm, a vehicle decision is generated that complies with road rights, safety boundaries, and behavioral objectives. The cloud then sends the decision to the vehicle via the V2X device.

[0010] 4) The vehicle side models the vehicle driving process as a Markov decision process. The vehicle controller outputs the corresponding steering wheel angle and vehicle longitudinal acceleration to the vehicle bottom layer for execution based on the decision results, thereby realizing the decision-making and control of intelligent connected vehicles in the weaving area of ​​the highway.

[0011] Furthermore, the right-of-way rules include: during the merging phase, lowering the priority of ramps and increasing the priority of main roads; during the merging phase, increasing the priority of ramps and lowering the priority of main roads.

[0012] Furthermore, the vehicle's safety boundary rule is expressed as:

[0013] R s (s t )=w c [S c (s t )+v i (s t )v j (s t )]·ξ(S c (s t ))+w d [S s (s t )+v i (s t )v i (s t )]·ξ(S s (s t ))

[0014] Where ξ(·) indicates whether there is overlap, which is 1 if there is an overlap, and 0 otherwise; w c and w d is the evaluation weight of the corresponding collision area overlap and safety area overlap; S c (s t ) represents the overlapping area of ​​the vehicle collision zone at time t, S s (s t ) represents the overlapping area of ​​the vehicle safety zone at time t, v i (s t ) and vj (s t ) represents the speed of adjacent vehicles.

[0015] The vehicle's behavior target rule is expressed as:

[0016] R v (s t )=w v (v(s t )) 2

[0017]

[0018] R j (s t )=w jx (j x (s t )) 2 +w jy (j y (s t )) 2

[0019] Where R ν (s t ) represents the task completion rule, which represents the expectation that the vehicle completes the task as soon as possible, R lk (s t ) represents the lane keeping rule, R j (s t ) Comfort rules; w v 、w l 、 w jx and w jy Both represent weight factors; v(s t ) represents the vehicle speed, Δl t (s t ) represents the distance deviation between the vehicle and the lane centerline in the Frenet coordinate system at time t, Indicates the angular deviation between the vehicle and the current lane direction; j x (s t ) and j y (s t ) represent the lateral and longitudinal jerk respectively.

[0020] Furthermore, establishing the master-slave game model includes:

[0021] ① Divide the vehicle group control granularity into a single intelligent connected vehicle and a fleet consisting of multiple intelligent connected vehicles;

[0022] Among them, when the vehicle group control granularity is a single intelligent connected vehicle, the strategy set includes single-vehicle acceleration / deceleration / holding, lane changing, and merging / exiting; when the vehicle group control granularity is a fleet composed of multiple intelligent connected vehicles, the strategy set includes platooning, platoon lane changing, and platoon merging / exiting;

[0023] ② Define leaders and followers, where the leader has a relatively high right of way and the follower has a relatively low right of way;

[0024] ③Let A l and A f Denote the decision sets corresponding to the leader and follower respectively, let R l (a l ,a f ) and R f (a l ,a f ) represent the payoffs of the leader and follower respectively, then the equilibrium solutions of the master-slave game are θ l and γ f ; Then the leader’s decision is expressed as:

[0025]

[0026] Among them, a l and a f denote the decisions of the leader and follower respectively, a′ f represents the rational decision of the follower, A′ f represents the rational decision set of the follower, Q′(a l ) represents the minimum reward;

[0027] The follower's decision is expressed as:

[0028]

[0029] Where Q f (a f ) represents the minimum reward.

[0030] Based on the established game model, several candidate decisions are obtained, and the particle swarm optimization algorithm is used to calculate the benefit of each candidate decision. The candidate decision corresponding to the maximum benefit is selected as the final decision result.

[0031] Furthermore, when the control granularity of the vehicle group is a single intelligent connected vehicle, in the Markov decision model, the vehicle state is designed as:

[0032]

[0033] Where, v hrepresents the vehicle speed, x represents the longitudinal position of the controlled vehicle, y represents the lateral position of the controlled vehicle, represents the yaw angle, that is, the angle between the controlled vehicle heading and the road tangent direction, Δx i and Δy i They represent the longitudinal and lateral relative positions of the controlled vehicle relative to the i-th manually driven vehicle, respectively, and n represents the number of manually driven vehicles within the sensing range of the controlled vehicle;

[0034] The action design is: a(t)={a h ,θ h}, a h represents the vehicle acceleration, θ h Indicates the steering wheel angle;

[0035] The constraint condition design includes collision avoidance constraints, vehicle state constraints and road rule constraints; the collision avoidance constraints are expressed as;

[0036] Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0037] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0038] Where p represents the risk level, which represents the degree of consideration of the randomness of manually driven vehicles; d y,min represents the set of lateral safety distances between the controlled vehicle and the vehicles within the perception range, d x,min represents the set of longitudinal safety distances between the controlled vehicle and the vehicles within the perception range, g x,k 、g y,k is the system parameter matrix, ξ k Represents the motion state vector of the surrounding HDV vehicles;

[0039] The road rule constraint is the vehicle speed constraint: v min ≤v h ≤v max , v min and v max Respectively represent the minimum speed and maximum speed allowed on the road section;

[0040] The vehicle state constraint is expressed as:

[0041] θ min ≤θ h ≤θ max

[0042] a min ≤ah ≤a max

[0043] Where θ min and θ max Represent the minimum turning angle and maximum turning angle of the vehicle, respectively, a min and a max Respectively represent the maximum acceleration and minimum acceleration of the vehicle;

[0044] The reward function is designed as:

[0045] J=min(w1J1+w2J2)

[0046] st:Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0047] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0048] v min ≤v h ≤v max

[0049] a min ≤a h ≤a max

[0050] θ min ≤θ h ≤θ max

[0051] Where w1 and w2 represent proportional coefficients, J1 represents the fluctuation of lateral and longitudinal acceleration, and J2 represents the consumption of wheel angle change;

[0052] The Markov decision model established by TD3 algorithm is solved to obtain the optimal output action a(t) = {a h ,θ h} and output it to the vehicle bottom layer for execution.

[0053] Furthermore, when the vehicle group control granularity is a convoy consisting of multiple intelligent connected vehicles, in the Markov decision model, the state of the vehicles in the convoy is designed as follows:

[0054]

[0055] Where, and They represent the relative speed error, longitudinal spacing error, and lateral spacing error between the vehicle with V2X communication connection and the own vehicle within the convoy, respectively. L Indicates vehicle driving intention;

[0056] The vehicle motion is designed as follows: a(t) = {a h ,θ h};

[0057] The constraint design includes collision avoidance constraints, vehicle state constraints, road rule constraints and intra-platoon collision avoidance constraints; the collision avoidance constraints are expressed as;

[0058] Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0059] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0060] The road rule constraint is the vehicle speed constraint: v min ≤v h ≤v max ;

[0061] The vehicle state constraint is expressed as:

[0062] θ min ≤θ h ≤θ max

[0063] a min ≤a h ≤a max

[0064] The intra-platoon collision avoidance constraint is expressed as:

[0065]

[0066] Where g min Indicates the safe horizontal distance, l wid Indicates vehicle width, l len Indicates the vehicle length, d min Indicates the safe longitudinal distance, h min Indicates the minimum following distance on the road section, v tra Indicates the average road operating speed;

[0067] The reward function is designed as:

[0068] J=min(w1J1+w2J2)

[0069] st:Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0070] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0071]

[0072] v min ≤v h ≤v max

[0073] a min ≤a h ≤a max

[0074] θ min ≤θ h ≤θ max

[0075] The Markov decision process established by TD3 algorithm is solved to obtain the optimal output action a(t) = {a h ,θ h} and output it synchronously to all vehicles in the fleet for execution.

[0076] The beneficial effects of the present invention are that the present invention can ensure the safety, stability and efficiency of the weaving area of ​​the expressway in mixed traffic.

[0077] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following description or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0079] Figure 1 This is a schematic diagram of mixed traffic scene in the weaving area of ​​a highway;

[0080] Figure 2 Schematic diagram of the process of the present invention;

[0081] Figure 3Schematic diagram of collaborative decision-making and control method for intelligent connected vehicles;

[0082] Figure 4 Describe the diagram for the rules of the road;

[0083] Figure 5 Design schematic diagrams for static road priorities;

[0084] Figure 6 Define a schematic diagram for the vehicle's safety boundary;

[0085] Figure 7 This is the flow chart of the particle swarm optimization algorithm;

[0086] Figure 8 This is the structural block diagram of the TD3 algorithm. DETAILED DESCRIPTION

[0087] The following describes the embodiments of the present invention by means of 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. It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0088] In one embodiment of the present invention, considering Figure 1 The mixed traffic scenario shown in the highway weaving section shows two types of vehicles: ICVs, which use onboard sensors to sense their surroundings and the Internet of Vehicles to obtain status information (position, speed, and acceleration) from other connected vehicles, as well as roadside information, enabling autonomous driving. The second type consists of manually driven vehicles, whose drivers steer based on their own perceptions. In the merging areas of entrance and exit ramps, mixed traffic flows composed of these vehicles can occur. Complex interactions and influences between ICVs and manually driven vehicles can also occur. Furthermore, the uncertainty of manually driven vehicles poses significant challenges to traffic efficiency and safety in weaving sections.

[0089] To solve the problems faced by the above scenarios, an embodiment of the present invention proposes a collaborative decision-making and control method for intelligent connected vehicles in mixed traffic in highway weaving areas, ensuring the safety, stability and efficiency of highway weaving areas in mixed traffic. Figure 2As shown, the method mainly includes: first, based on high-precision map data information and road right-of-way information, a unified macro and micro mixed traffic driving rule is formed in a fixed geometric reference system in the weaving area; on this basis, a master-slave game model is established, and the particle swarm optimization method is combined to generate single-vehicle / multi-vehicle decision results that comply with traffic rules, where single-vehicle decision-making includes single-vehicle acceleration / deceleration / maintenance, lane changing, merging / exiting, and multi-vehicle decision-making includes platoon driving, platoon lane changing, and platoon merging / exiting; the cloud sends the decision results to the vehicle side through the V2X device, and the vehicle side establishes the single-vehicle input state / multi-vehicle input state and the corresponding constraints respectively. According to the decision results, the reinforcement learning-based vehicle controller outputs the corresponding steering wheel angle and vehicle longitudinal acceleration to the vehicle bottom layer for execution, thereby realizing the decision-making and control of intelligent connected vehicles in the weaving area of ​​the highway.

[0090] The specific contents of this method are as follows:

[0091] 1. Based on high-precision map data and road right-of-way information, a unified macro and micro mixed traffic driving rule is formed in a fixed geometric reference system in the weaving area.

[0092] 1. Describe road rules based on the data structure of the HD map, such as Figure 4 As shown in the figure, the road right information is further integrated and the priority description of the road is added to allocate the priority of the vehicles on the road in advance, as shown in the figure. Figure 5 Specifically, during the merging phase, the priority of branch roads is lowered and that of fast lanes is increased, forcing vehicles to yield to lanes adjacent to the merging entrance, thereby facilitating the merging of branch vehicles onto the main road as quickly as possible. During the merging phase, the opposite is true: branch roads are prioritized and fast lanes are lowered, pushing departing vehicles toward the slow lanes and facilitating their departure from the main road as quickly as possible. Right-of-way rules are implemented within the tree structure of the HD map. These macro-level rules are combined with navigation data through an A* search method to determine the initial range of vehicle decisions under normal driving conditions.

[0093] 2. Further parameterize the driving status of each vehicle and establish safety boundary rules and behavioral target rules for each vehicle around itself.

[0094] 1) Among them, the vehicle safety boundary rules are as follows Figure 6 As shown in the figure, it is defined as follows: the vehicle collision area is the thick solid line area, and the safety area is the thick dotted line area. t is the state of the vehicle at time t, (x t ,y t ) is the center coordinate of the rear axle of the vehicle at time t, I ef with I er I is the distance between the front and rear edges of the vehicle and the center of the rear axle for collision judgment. sf and Isr The distance between the front and rear edges of the vehicle and the center of the rear axle is reserved for driving safety, w e The width of the vehicle for collision determination, w s The vehicle width is reserved for vehicle safety. sf ,I sr ,w s ) is a function of vehicle speed v(t), which is calculated by the safety-sensitive safety model. According to the model, the overlapping area of ​​the collision zone and the overlapping area of ​​the safety zone between vehicles can be defined. Let S c (s t ) is the overlapping area of ​​the vehicle collision zone at time t, S s (s t ) is the overlapping area of ​​the vehicle safety zone at time t. The speed of the adjacent vehicle is defined as v i (s t ) and v j (s t ), then the definition of the security boundary rule is as follows:

[0095] R s (s t )=w c [S c (s t )+v i (s t )v j (s t )]·ξ(S c (s t ))+w d [S s (s t )+v i (s t )v i (s t )]·ξ(S s (s t ))

[0096] Where ξ(·) indicates whether there is overlap. If there is an overlapping area, the result is 1, and if there is no overlap, the result is 0. c and w d is the evaluation weight of the corresponding collision area overlap and safety area overlap, w c is negative infinity to avoid collisions, and w d Make adjustments based on actual conditions and try to avoid overlapping safety zones between vehicles.

[0097] 2) Vehicle behavior target rules mainly include two contents: task completion and comfort. The completion index describes the vehicle's expectation of reaching the desired location by driving at the fastest speed, and the comfort index describes the lateral and longitudinal smoothness of the vehicle during driving.

[0098] ①Task completion takes the completion time as the target. The driving rules expect the vehicle to complete the task as soon as possible, so its completion degree R v (s t ) is calculated by the vehicle's speed:

[0099] R v (s t )=w v (v(s t )) 2

[0100] Among them, w v is the corresponding weight.

[0101] In addition to the completion time, the task completion also includes the results of lane keeping during the vehicle's driving process. The vehicle should try to follow the center line of the lane. t is the lateral coordinate of the vehicle in the Frenet coordinate system at time t, Δl t is the distance deviation between the vehicle and the lane centerline in the Frenet coordinate system at time t, is the angle of the vehicle at time t, is the angular deviation between the vehicle and the current lane direction, then the lane keeping part R lk (s t ) is defined as follows:

[0102]

[0103] Among them, w l and are the weights of radial distance deviation and angular deviation respectively.

[0104] ② Comfort is expressed by jerkiness, where the lateral and longitudinal jerkiness are expressed as j x (s t ) and j y (s t ), the corresponding weight factor is w jx and w jy , then the comfort rule is expressed as:

[0105] R j (s t )=w jx (j x (s t ))2 +w jy (j y (s t )) 2

[0106] Both macro and micro scenarios unfold in the road coordinate system corresponding to the high-precision map. Therefore, the corresponding vehicle's own results and the road and lane information provided by the map can be converted into each other, forming direct data interaction between static and dynamic scenarios, and constructing driving rules under a unified reference system.

[0107] 2. Establish a master-slave game model and combine it with the particle swarm optimization method to generate single-vehicle / multi-vehicle decision-making results that comply with traffic rules.

[0108] The cloud uses a master-slave game model to determine single-vehicle and multi-vehicle decision-making results that meet vehicle driving requirements. Single-vehicle decisions include acceleration / deceleration / holding, lane changing, and merging / exiting. Multi-vehicle decisions include platooning, platoon lane changing, and platoon merging / exiting. The cloud sends these decision results to the vehicle via the V2X device.

[0109] First, the control granularity of vehicle groups is divided into different levels. When the controlled object within the vehicle group is a single intelligent connected vehicle, the decision-making layer's goal is to plan the acceleration / deceleration / holding, lane changing, and merging / exiting of the single vehicle. When the controlled object within the vehicle group is a fleet of vehicles, the decision-making goals are platoon driving, platoon lane changing, and platoon merging / exiting.

[0110] Define the decision making of the leader and follower vehicles that generate the interaction l and a f , the leading vehicle is the vehicle with higher road right, and the following vehicle is the vehicle with lower road right. Let A l and A f They represent the decision sets corresponding to the leading vehicle and the following vehicle, namely

[0111]

[0112] In the game, the vehicle achieves maximum benefit by choosing the appropriate strategy. The benefits of the leading vehicle and the following vehicle can be expressed as R according to the driving rules. l (a l ,a f ) and R f (a l ,a f ), then the equilibrium solutions of the master-slave game are defined as γ l and γ f, the decisions of the leading vehicle and the following vehicles affect each other, and due to the master-slave game characteristics caused by road rights, the leading vehicle has the advantage of priority decision-making, so the decision of the leading vehicle directly affects the rational decision set A′ of the following vehicles f (a l ), as shown below:

[0113]

[0114] Among them, a′ f represents the rational decision of the following vehicle, Q′(a l ) represents the minimum reward, argmin represents the minimum Q′(a l ) is the equilibrium solution.

[0115] At the same time, considering the spatiotemporal uncertainty of the following vehicle's observation of the leading vehicle, the following vehicle adopts the decision of maximizing its own minimum benefit through the maximization hypothesis:

[0116]

[0117] The particle swarm algorithm is further used to solve the problem, such as Figure 7 As shown in the figure, when the game vehicle has M decision results, the time window length is H, the particle dimension is set to be the same as the prediction time length H, and the decision set size is A, the spatial position of each particle is the potential sequence of the decision space. After the space formed by the maximum acceleration and the maximum deceleration is divided equally, a high-dimensional cubic structure is formed in the constraint space. The computational complexity changes from exponential form O(M·A H ) can be reduced to a linear form O(A·H·MaxStep). Using the established game model and driving rules, the decision-making payoffs for a period of time in the future are deduced from the current moment and the decision results are generated. These include single-vehicle decisions: accelerating / decelerating / holding, lane changing, merging / exiting; and multi-vehicle decisions: platooning, platoon lane changing, and platoon merging / exiting.

[0118] Third, the vehicle side establishes the single-vehicle input state / multi-vehicle input state and corresponding constraints. Based on the decision results, the reinforcement learning-based vehicle controller outputs the corresponding steering wheel angle and vehicle longitudinal acceleration to the vehicle bottom layer for execution, thereby realizing the decision-making and control of intelligent connected vehicles in the weaving area of ​​the highway. The specific steps are as follows:

[0119] The vehicle side models the driving process as a Markov decision process: divided into state design, action design, and reward design.

[0120] The state design fully represents the decision-making state of the intelligent connected vehicle. The vehicle state design under the single vehicle decision is:

[0121]

[0122] Where, v h represents the speed of the own vehicle, x represents the longitudinal position of the controlled vehicle, y represents the lateral position of the controlled vehicle, is the yaw angle, that is, the angle between the controlled vehicle heading and the road tangent direction, Δx i =x i -x and Δy i =y i -y represent the longitudinal and lateral relative positions of the controlled vehicle relative to the i-th HDV, respectively, and x i Indicates the vertical position of the i-th HDV, y i represents the lateral position of the i-th HDV, and n represents the number of HDVs within the sensing range of the controlled vehicle.

[0123] Further determine the constraints, mainly considering collision avoidance constraints, vehicle status and road rules constraints.

[0124] Collision avoidance constraints: Collision avoidance constraints take into account the state randomness of the manually driven vehicle to ensure that the controlled vehicle can still maintain driving safety under the speed and lateral position fluctuations of the manually driven vehicle.

[0125] Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0126] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0127] Where p is the risk level, which represents the degree of consideration of the randomness of manually driven vehicles; d y,min represents the set of lateral safety distances between the controlled vehicle and the perceived vehicle, d x,min The set of longitudinal safety distances between the controlled vehicle and the perceived vehicle. g x,k 、g y,k is the system parameter matrix, ξ k Represents the motion state vector of the surrounding HDV vehicles.

[0128] Consider the speed limit on a highway, where the speed of a vehicle is restricted to a certain range. The speed of a vehicle can be restricted to:

[0129] v min ≤v h ≤v max

[0130] where v min and v maxIndicates the minimum and maximum speeds allowed on this road segment.

[0131] Considering the vehicle state constraints, the constraints are set as follows:

[0132] θ min ≤θ h ≤θ max

[0133] a min ≤a h ≤a max

[0134] where θ min and θ max Indicates the minimum and maximum turning angles of the vehicle, a min and a max Indicates the maximum and minimum acceleration of the vehicle.

[0135] The Actor network output action is set as:

[0136] a(t)={a h ,θ h}

[0137] where a h represents the vehicle acceleration, θ h Indicates the steering wheel angle.

[0138] Furthermore, considering the uncertainty of manually driven vehicles, the vehicle's own state, and safety constraints, with the goal of minimizing the lateral and longitudinal acceleration fluctuations J1 and the wheel angle change consumption J2, the reward function involving multiple control objectives is designed as follows:

[0139] J=min(w1J1+w2J2)

[0140] st:Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0141] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0142] v min ≤v h ≤v max

[0143] a min ≤a h ≤a max

[0144] θmin ≤θ h ≤θ max

[0145] Where w1 and w2 represent the proportional coefficients.

[0146] The vehicle state design under multi-vehicle decision-making is:

[0147]

[0148] in, and They represent the relative speed error, longitudinal spacing error, and lateral spacing error between the vehicle with V2X communication connection and the own vehicle within the convoy, respectively. L Indicates the vehicle's driving intention.

[0149] On the basis of the single-vehicle decision constraints, the internal collision avoidance constraints of the team are added:

[0150]

[0151] where g min Indicates the safe horizontal distance, l wid Indicates vehicle width, l len Indicates the vehicle length, d min Indicates the safe longitudinal distance, h min Indicates the minimum following distance on the road section, v tra Indicates the average road operating speed.

[0152] Therefore, the reward function of the multi-vehicle control objective is designed as:

[0153] J=min(w1J1+w2J2)

[0154] st:Pr(-(g x,k ) T ξ k ≤-d x,min )≥p

[0155] Pr(-(g y,k ) T ξ k ≤-d y,min )≥p

[0156]

[0157] v min ≤v h ≤v max

[0158] a min ≤a h ≤a max

[0159] θ min ≤θ h ≤θ max

[0160] Where w1 and w2 represent the proportional coefficients.

[0161] Finally, the optimal output action a(t)={a h ,θ h} to the vehicle bottom layer for execution, such as Figure 8 As shown, the decision-making and control of intelligent connected vehicles in the weaving area of ​​the highway are finally realized.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions, which should all be included in the scope of the claims of the present invention.

Claims

1. A collaborative decision-making and control method for intelligent connected vehicles in mixed traffic on highway weaving areas, characterized by: Based on the road coordinate system of the HD map, the road right information is integrated with the HD map data to form road right rules and assign priorities to vehicles on the road; In the road coordinate system of the high-precision map, the driving status of each vehicle is parameterized and the safety boundary rules and behavioral target rules of each vehicle around it are established; A master-slave game model is established in the cloud, and combined with a particle swarm optimization algorithm, vehicle decision results are generated that comply with road rights, safety boundary rules, and behavioral target rules. The cloud sends the decision results to the vehicle through the V2X device. The vehicle side models the vehicle driving process as a Markov decision process. The vehicle controller outputs the corresponding steering wheel angle and vehicle longitudinal acceleration based on the decision results to the vehicle bottom layer for execution, thereby realizing the decision-making and control of intelligent connected vehicles in the weaving area of ​​the highway. Wherein, establishing the master-slave game model includes: 1) Divide the vehicle group control granularity into a single intelligent connected vehicle and a fleet consisting of multiple intelligent connected vehicles; 2) defining a leader and a follower, wherein the leader has a relatively high right of way and the follower has a relatively low right of way; 3) Let A l and A f Denote the decision sets corresponding to the leader and follower respectively, let R l (a l ,a f ) and R f (a l ,a f ) represent the benefits of the leader and follower respectively, let γ l and γ f Denote the decisions of the leader and follower respectively; the decision of the leader is expressed as: Where a l and a f denote the decisions of the leader and follower respectively, a′ f represents the rational decision of the follower, A′ f represents the rational decision set of the follower, Q′(a l ) represents the minimum reward; The follower's decision is expressed as: Where Q f (a f ) represents the minimum reward; The vehicle side models the vehicle driving process as a Markov decision process including: When the control granularity of the vehicle group is a single intelligent connected vehicle, in the Markov decision model, the vehicle state is designed as: Where, v h represents the vehicle speed, x represents the longitudinal position of the controlled vehicle, y represents the lateral position of the controlled vehicle, represents the yaw angle, that is, the angle between the controlled vehicle heading and the road tangent direction, Δx i and Δy i They represent the longitudinal and lateral relative positions of the controlled vehicle relative to the i-th manually driven vehicle, respectively, and n represents the number of manually driven vehicles within the sensing range of the controlled vehicle; The action design is: a(t)={a h ,θ h }, a h represents the vehicle acceleration, θ h Indicates the steering wheel angle; The constraint condition design includes collision avoidance constraints, vehicle state constraints and road rule constraints; the collision avoidance constraints are expressed as; Pr(-(g x,k , T ξ k ≤-d x,min )≥p Pr(-(g y,k , T ξ k ≤-d y,min )≥p Where p represents the risk level, which represents the degree of consideration of the randomness of manually driven vehicles; d y,min represents the set of lateral safety distances between the controlled vehicle and the vehicles within the perception range, d x,min represents the set of longitudinal safety distances between the controlled vehicle and the vehicles within the perception range, g x,k 、g y,k is the system parameter matrix, ξ k Represents the motion state vector of the surrounding human-driven vehicles; The road rule constraint is the vehicle speed constraint: v min ≤v h ≤v max , v min and v max Respectively represent the minimum speed and maximum speed allowed on the road section; The vehicle state constraint is expressed as: i min ≤θ h ≤θ max a min ≤a h ≤a max Where θ min and θ max Represent the minimum turning angle and maximum turning angle of the vehicle, respectively, a min and a max Respectively represent the maximum acceleration and minimum acceleration of the vehicle; The reward function is designed as: J=min(w1J1+w2J2) st:Pr(-(g x,k , T ξ k ≤-d x,min )≥p Pr(-(g y,k , T ξ k ≤-d y,min )≥p in min ≤in h ≤in max a min ≤a h ≤a max i min ≤θ h ≤θ max Where w1 and w2 represent proportional coefficients, J1 represents the fluctuation of lateral and longitudinal acceleration, and J2 represents the consumption of wheel angle change; The Markov decision model established by TD3 algorithm is solved to obtain the optimal output action a(t) = {a h ,θ h }, and output to the vehicle bottom layer for execution; When the vehicle group control granularity is a convoy consisting of multiple intelligent connected vehicles, in the Markov decision model, the state of the vehicles in the convoy is designed as follows: Where, and They represent the relative speed error, longitudinal spacing error, and lateral spacing error between the vehicle with V2X communication connection and the own vehicle within the convoy, respectively. L Indicates vehicle driving intention; The vehicle motion is designed as follows: a(t) = {a h ,θ h }; The constraint design includes collision avoidance constraints, vehicle state constraints, road rule constraints and intra-platoon collision avoidance constraints; the collision avoidance constraints are expressed as; Pr(-(g x,k , T ξ k ≤-d x,min )≥p Pr(-(g y,k , T ξ k ≤-d y,min )≥p The road rule constraint is the vehicle speed constraint: v min ≤v h ≤v max ; The vehicle state constraint is expressed as: i min ≤θ h ≤θ max a min ≤a h ≤a max The intra-platoon collision avoidance constraint is expressed as: Where g min Indicates the safe horizontal distance, l wid Indicates vehicle width, l len Indicates the vehicle length, d min Indicates the safe longitudinal distance, h min Indicates the minimum following distance on the road section, v tra Indicates the average road operating speed; The reward function is designed as: J=min(w1J1+w2J2) st:Pr(-(g x,k , T ξ k ≤-d x,min )≥p Pr(-(g y,k , T ξ k ≤-d y,min )≥p in min ≤in h ≤in max a min ≤a h ≤a max i min ≤θ h ≤θ max The Markov decision process established by TD3 algorithm is solved to obtain the optimal output action a(t) = {a h ,θ h } and output it synchronously to all vehicles in the fleet for execution.

2. The intelligent connected vehicle collaborative decision-making and control method according to claim 1, characterized in that: The right-of-way rules include: during the merging phase, lowering the priority of ramps and raising the priority of main roads; during the merging phase, raising the priority of ramps and lowering the priority of main roads.

3. The intelligent connected vehicle collaborative decision-making and control method according to claim 1, characterized in that: When the granularity of vehicle group control is a single intelligent connected vehicle, the strategy set includes single-vehicle acceleration / deceleration / holding, lane changing, and merging / exiting; When the vehicle group control granularity is a fleet consisting of multiple intelligent connected vehicles, the strategy set includes platoon driving, platoon lane changing, and platoon merging / exiting.

4. The intelligent connected vehicle collaborative decision-making and control method according to claim 1, characterized in that: Based on the established game model, several candidate decisions are obtained, and the particle swarm optimization algorithm is used to calculate the benefit of each candidate decision. The candidate decision corresponding to the maximum benefit is selected as the final decision result.

Images