Multi-vehicle cooperative driving double-layer decision optimization method in intelligent network connection mixed traffic environment

Abstract

The invention discloses a multi-vehicle cooperative driving double-layer decision optimization method in an intelligent network connection mixed traffic environment, and the method comprises the steps: adjusting the lane type distribution, and selecting a target lane for a vehicle; the driving intention of the vehicle is quantified, behavior search and target area distribution are carried out, and a driving decision of the vehicle is obtained; and based on the driving decision, enabling each intelligent connected automobile to smoothly lead to the target area from the current position. According to the method, adaptive modeling and real-time optimization of multi-vehicle interaction behaviors can be realized, and the traffic efficiency and the overall traffic stability under complex road conditions are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle traffic control, in particular to a multi-vehicle cooperative driving double-layer decision optimization method in an intelligent and connected hybrid traffic environment. BACKGROUND

[0002] With the deep integration of intelligent and connected technology (V2X) and autonomous driving systems, the traffic system is evolving towards high coordination and real-time optimization. Especially in the urban arterial and highway mainline scenarios with intensive multi-vehicle interaction and complex decision coupling, multi-type vehicles such as intelligent connected vehicles (ICV) and human-driven vehicles (HDV) need to interact and cooperate frequently. How to achieve efficient lane allocation and vehicle scheduling has become a key factor affecting overall traffic efficiency and safety level.

[0003] However, existing traffic regulation relies mainly on centralized optimization or static signal control, which is difficult to respond to sudden traffic conditions and lane resource shortages, and is prone to cause traffic bottlenecks, delays, and even safety risks. In addition, traditional models are generally based on the homogeneity assumption of traffic flow, lacking detailed description of the characteristics and behavior differences of different vehicle types, making it difficult to achieve accurate scheduling and efficient traffic.

[0004] Therefore, in order to solve the above problems, a multi-vehicle cooperative driving double-layer decision optimization method in an intelligent and connected hybrid traffic environment is needed, which can realize adaptive modeling and real-time optimization of multi-vehicle interaction behavior, and improve the traffic efficiency and stability of the overall traffic under complex road conditions. SUMMARY

[0005] Therefore, the purpose of the present application is to overcome the defects in the prior art and provide a multi-vehicle cooperative driving double-layer decision optimization method in an intelligent and connected hybrid traffic environment, which can realize adaptive modeling and real-time optimization of multi-vehicle interaction behavior, and improve the traffic efficiency and stability of the overall traffic under complex road conditions.

[0006] The multi-vehicle cooperative driving double-layer decision optimization method in an intelligent and connected hybrid traffic environment of the present application comprises:

[0007] Adjusting lane type allocation to select a target lane for the vehicle;

[0008] Quantifying the driving intention of the vehicle, performing behavior search and target area allocation, and obtaining the driving decision of the vehicle;

[0009] Based on the driving decision, each intelligent and connected vehicle smoothly passes from the current position to the target area.

[0010] Further, the target lane is selected for the vehicle, specifically comprising:

[0011] Constructing a target function, the target function comprising a lane flow balance function and a lane priority selection function;

[0012] Constraints are set for the objective function, and the objective function is solved under the constraints, and the lane corresponding to the solution of the objective function is taken as the target lane of the vehicle.

[0013] Further, the lane flow balance function is determined according to the following formula:

[0014]

[0015] Wherein, c i,j,t represents the capacity of the lane j of the section i at the time period t.

[0016] Further, the lane priority selection function is determined according to the following formula:

[0017]

[0018] Wherein, x i,j,t,a represents whether the lane j of the section i at the time period t is of the lane type a, x i,j,t,a ∈{0,1} respectively means no and yes; x i,j,t,b represents whether the lane j of the section i at the time period t is of the lane type b, x i,j,t,b ∈{0,1} respectively means no and yes.

[0019] Further, the constraint condition is determined according to the following formula:

[0020]

[0021]

[0022] if the section i contains an exit or an entrance, then:

[0023]

[0024] Wherein, q i,t,n represents the flow of the vehicle type n at the section i at the time period t; M represents a constant; q i,j,t represents the flow of the lane j of the section i at the time period t; x i,j,t,a represents whether the lane j of the section i at the time period t is of the lane type a, x i,j,t,a ∈{0,1} respectively means no and yes; x i,j,t,b represents whether the lane j of the section i at the time period t is of the lane type b, x i,j,t,b ∈{0,1} respectively means no and yes; q i,j,t,n represents the flow of the vehicle type n selecting the lane j to travel at the section i at the time period t.

[0025] Further, the driving intention of the vehicle is quantified, the behavior search and the target area allocation are performed, and the driving decision of the vehicle is obtained, which specifically includes:

[0026] According to the execution action of the vehicle, a plurality of potential target areas are set for the vehicle;

[0027] When a plurality of vehicles intend to enter the same target area, coordination is performed by priority, and a high priority is preferentially allocated to the target area.

[0028] Further, the coordination by priority, and the high priority is preferentially allocated to the target area, specifically includes:

[0029] The priority of the manually driven vehicle is always higher than that of the intelligent connected vehicle, and when the intelligent connected vehicle competes for the area, the straight priority is higher than the lane changing priority, and when the intelligent connected vehicle is for lane changing, the forced lane changing priority is greater than the free lane changing priority;

[0030] When the intelligent connected vehicle is for free lane changing, the intelligent connected vehicle with high priority is determined based on a Monte Carlo tree search algorithm:

[0031] Action space definition: a discrete action set is designed for each vehicle, and the action space is the Cartesian product of the actions of each vehicle, and the number of action combinations is limited within a certain range through pruning strategy;

[0032] A search tree is gradually constructed in an iterative manner, and one iteration process includes four steps of selection, expansion, simulation and back propagation:

[0033] Selection: starting from the root node, the optimal node is selected based on the strategy formula until the selected node is an inextensible node; the strategy formula is:

[0034] wherein, q i is the value score of the child node i; n is the total number of times of exploring the search tree; n i is the number of times of exploring the node i; ω is a weight coefficient;

[0035] Expansion: a new child node is randomly generated for the selected node, and the new node is added to the search tree structure;

[0036] Simulation: starting from the expanded new node, a new child node is continuously generated by executing a random sampling simulation strategy, that is, a vehicle is randomly selected from the vehicles not yet included in the current partial passing order, and is added to the tail of the partial passing order until a complete passing order is generated, and the node value score is updated according to the following formula:

[0037]

[0038] wherein, q i is the value score of the node i; J is a quantitative evaluation value; J i,maxis the maximum value of the score among all the siblings of node i; J i,min is the minimum value of the score among all the siblings of node i;

[0039] Back propagation: the new score q i is propagated up to the root node along the direction of its parent node;

[0040] During the above iteration process, the current optimal passing order is constantly updated, and when the maximum search iteration limit is reached, the iteration process stops and returns the current optimal leaf node, which corresponds to the intelligent connected vehicle with high priority.

[0041] Further, based on the driving decision, each intelligent connected vehicle is smoothly guided from the current position to the target area, specifically including:

[0042] A third-order Bezier curve is used to pre-generate a trajectory smoothly leading from the current position to the target area;

[0043] Based on the trajectory length and the expected time window, an expected speed v is generated:

[0044] where c is an adjustment coefficient in speed planning, D TP is the trajectory length, ΔT is the expected time window, and v0 is the initial speed;

[0045] And a proportional controller is used to generate a longitudinal acceleration a :

[0046] where a cmax is the maximum allowed acceleration, k p is the gain coefficient of the proportional controller, is the speed of the current intelligent connected vehicle i;

[0047] If the closest distance between the trajectory points of the two intelligent connected vehicles is lower than a threshold, it is determined to be a conflict, and the avoidance vehicle is determined by the set priority. A collision ellipse model is used to determine whether there is a trajectory conflict between the intelligent connected vehicle and the manually driven vehicle, and if so, the intelligent connected vehicle reduces the acceleration when it is in low priority to avoid collision;

[0048] A preview point is selected in the trajectory to construct a preview vector :

[0049] where l p represents the distance between the current vehicle head position and the preview point, x target is the horizontal coordinate of the preview point, is the horizontal coordinate of the current vehicle i;

[0050] Based on the bicycle model, the preview vector is converted into the steering angle based on the included angle a between the current direction of the vehicle :

[0051] wherein, l pr is the distance from the rear wheel of the vehicle to the preview point; L wb is the wheelbase of the vehicle;

[0052] The acceleration and the steering angle are sent to the corresponding intelligent connected vehicle, which performs actions according to the corresponding acceleration and steering angle.

[0053] The beneficial effects of the present application are: the disclosed multi-vehicle cooperative driving double-layer group decision optimization method in the intelligent connected hybrid traffic environment, based on the hierarchical architecture of lane level and vehicle level, fuses static structural features and dynamic traffic state information, faces the hybrid traffic control demand in the intelligent connected environment, builds a cooperative optimization mechanism with high computing efficiency, fine scheduling and strong scalability, effectively solves the core problems of traditional single optimization level, weak scheduling ability and lack of hybrid traffic structure modeling, and provides technical support for large-scale deployment of intelligent connected vehicles. BRIEF DESCRIPTION OF DRAWINGS

[0054] The present application will be further described below in conjunction with the drawings and examples:

[0055] Figure 1 is the principle diagram of the multi-vehicle cooperative driving double-layer group decision framework of the present application;

[0056] Figure 2 is the definition diagram of the lane-level decision scenario of the present application;

[0057] Figure 3 is the delineation diagram of the vehicle driving target area of the present application;

[0058] Figure 4 is the screening diagram of the high-risk area of the present application;

[0059] Figure 5 is the priority setting diagram of the road right conflict of the present application;

[0060] Figure 6 is the vehicle optimal decision behavior search diagram based on Monte Carlo tree search of the present application;

[0061] Figure 7 is the flow diagram of the multi-vehicle cooperative driving double-layer group decision framework of the present application;

[0062] Figure 8 The schematic diagram of the principle flow of the double-layer decision implementation of the application is shown in the figure. DETAILED DESCRIPTION

[0063] The application is further described below in combination with the accompanying drawings of the specification, as shown in the figure:

[0064] The embodiment discloses a multi-vehicle cooperative driving double-layer decision optimization method in an intelligent connected hybrid traffic environment, including the following steps:

[0065] S1. Adjusting lane type distribution to select a target lane for a vehicle;

[0066] S2. Quantifying the driving intention of the vehicle, performing behavior search and target area distribution, and obtaining the driving decision of the vehicle;

[0067] S3. Based on the driving decision, each intelligent connected vehicle is smoothly guided from the current position to the target area.

[0068] The application designs a multi-vehicle cooperative driving double-layer group decision framework in an intelligent connected hybrid traffic environment, which contains upper and lower structures in the whole framework. The upper layer is lane management, and the lower layer is vehicle-level multi-vehicle decision, as shown in the figure: Figure 1

[0069] The upper layer (lane-level decision layer): the vehicle management layer mainly filters the selectable lanes according to the types of vehicles (ICV / HDV, small passenger car / small truck / large passenger car / large truck). Then, considering that it is difficult to accurately quantify the lane selection probability formula of various vehicle types, it is assumed that all vehicles prefer to select the lane with the smallest traffic in the allowable driving lane. The specific traffic formula is in the second part. After adjusting the lane type distribution, the allowable lane set and traffic distribution are recalculated, and iteration is performed until convergence.

[0070] The lower layer (vehicle-level decision layer): according to the lane type decision result of the road section, combined with the state and driving intention of the intelligent connected vehicle, the vehicle-level multi-vehicle decision layer is divided into two parts:

[0071] The first part is the decision of the vehicle driving behavior level. The driving intention of the vehicle is quantified by using the grid method, based on which conflicts and obstacle avoidance are detected, and then the target area is allocated based on the behavior search of MCTS. The allocation mode refers to the priority of road right, and the driving decision of each vehicle is output.

[0072] The second part is the specific vehicle trajectory level planning, which is affected by the driving decision set of the first part. A three-order Bezier curve is used to generate an action trajectory for each ICV, and then the specific speed is planned and the conflict is corrected. Finally, the driving action of each vehicle is output based on the bicycle model.

[0073] ​The application combines lane type configuration and vehicle behavior planning by introducing a hierarchical decoupling mechanism, realizes behavior coordination and path optimal allocation of multiple vehicles under shared road resources on the premise of ensuring traffic safety. Based on real-time traffic data and vehicle perception information, lane priority scheme and vehicle behavior instruction are dynamically generated to adapt to various traffic flow structures and environmental changes, significantly improving the adaptive ability and decision efficiency under mixed traffic flow. A unified decision basis and scheduling system is built for multi-vehicle cooperative driving, enabling intelligent networked traffic to have higher cooperative scheduling capability, resource allocation flexibility and multi-dimensional safety protection mechanism, promoting the evolution of intelligent traffic from "single vehicle intelligence" to "group intelligence", promoting the transformation of traffic system from static management to dynamic perception and intelligent response, and finally realizing the multi-objective optimal coordination of traffic efficiency, safety and comfort.

[0074] In this embodiment, step S1 involves lane-level decision making, as shown in FIG. 1. Figure 2 The time period The road segment The lane The vehicle type HDV small car, HDV small truck, HDV bus, HDV truck, ICV small car, ICV small truck, ICV bus, ICV truck, ICV exclusive lane, ICV and HDV mixed lane, HDV exclusive lane, small car lane, car lane, car and truck lane, truck lane.

[0075] The probability of each vehicle type choosing a certain lane to drive is closely related to lane management strategy, proportion of each vehicle type, whether the road segment contains an exit or entrance ramp, driver driving preference, lane speed limit, vehicle performance, etc., that is:

[0076] p i,j,t,n =f(x i,j,t,a ,x i,j,t,b ,r n ,v i,j,t ,…);

[0077] In the formula, p i,j,t,n is the probability of vehicle type n choosing lane j to drive at road segment i at time period t; v i,j,t is the speed limit value of lane j at road segment i at time period t; x i,j,t,a is whether lane j at road segment i at time period t is lane type a, x i,j,t,a ∈{0,1} respectively for no and yes; x i,j,t,b is whether lane j at road segment i at time period t is lane type b, x i,j,t,b ∈{0,1} respectively for no and yes; r nis the proportion of vehicle type n for various vehicle types; q i,j,t,n is the flow of vehicle type n traveling on lane j at link i at time period t; q i,t,n is the flow of vehicle type n at link i at time period t; q i,j,t is the flow of lane j at link i at time period t; c i,j,t is the capacity of lane j at link i at time period t; is the set of allowable lane management strategies;

[0078] The establishment of the vehicle road model needs to be reasonable, and the road type can be divided into two modes:

[0079] 1. According to the driving type of the vehicle, that is, ICV and HDV, it can be divided into ICV special lane, HDV special lane and mixed lane. The lane layout is because ICV can obtain more information, react faster, and the overall speed trend is more gentle, so the lane flow is fast, so it is distributed in the fast lane, that is, the leftmost side. Then the mixed lane can only be set in the middle, and the HDV special lane is the rightmost side.

[0080] 2. According to the original attribute of the vehicle, that is, small passenger car, passenger car, passenger car and truck, the lane layout is the same as above. The small passenger car is the smallest in size and the fastest in speed, so it should be distributed on the right side of the lane. The ordinary passenger car should be on the left side of the small passenger car lane, the passenger car should be on the left side of the ordinary passenger car, and the truck should be on the leftmost side, satisfying an inclusion relationship.

[0081] Although the road type can be divided into two models, only one mode can be determined to divide the road type, and the final value of the maximum lane saturation degree of the two modes is compared to determine the mode. The vehicle saturation degrees of the lane allocation strategies of the two modes are calculated, and the lane allocation strategies of each link are calculated separately. However, if the lane allocation mode of the current link is the driving type allocation of the vehicle, the difference of the same lane type of the adjacent link needs to be controllable (such as avoiding sudden change from ICV special lane to HDV special lane); if the lane allocation mode of the current link is the original attribute allocation of the vehicle, the lane type change of the continuous link needs to be smooth (such as avoiding sudden change from passenger car lane to truck lane). If the link contains an exit / entry ramp, the inner lane vehicle can change lane to the outer lane, allowing dynamic adjustment of lane selection.

[0082] To meet the above requirements, the target lane for the vehicle is selected, which specifically includes:

[0083] A target function is constructed, which includes a lane flow balancing function and a lane priority selection function;

[0084] The constraint conditions are set for the target function, and the target function is solved under the limitation of the constraint conditions. The lane corresponding to the solution of the target function is taken as the target lane of the vehicle.

[0085] where the lane flow balance function involves minimization of the maximum of lane saturation:

[0086] where c i,j,t represents the capacity of link i lane j at period t, taking 2000 pcu / h.

[0087] The lane priority function involves lane priority:

[0088]

[0089] where x i,j,t,a represents whether link i lane j is of lane type a at period t, x i,j,t,a ∈{0,1} for no and yes, respectively; x i,j,t,b represents whether link i lane j is of lane type b at period t, x i,j,t,b ∈{0,1} for no and yes, respectively.

[0090] The constraints are determined according to the following equations:

[0091]

[0092]

[0093] if link i contains an exit or an entrance, then:

[0094]

[0095]

[0096] where q i,t,n represents the flow of vehicle type n at link i at period t, in pcu / h; M represents a constant, which can take the value of 10000; q i,j,t represents the flow of link i lane j at period t, in pcu / h; x i,j,t,a represents whether link i lane j is of lane type a at period t, x i,j,t,a ∈{0,1} for no and yes, respectively; x i,j,t,b represents whether link i lane j is of lane type b at period t, x i,j,t,b ∈{0,1} for no and yes, respectively; q i,j,t,n represents the flow of vehicle type n choosing lane j at link i at period t. β i is a 0-1 variable, β i = 1 if the link i belongs to the upstream 2 km to the downstream 0 m of the diverging area or the upstream 0 m to the downstream 2 km of the converging area, otherwise β i = 0. α i,t is an intermediate variable, in pcu / h.

[0097] It should be noted that the weight coefficients H1 and H2 can also be set for the lane flow equalization function and the lane priority selection function in the target function, respectively, and the values can be 1000 and 1, respectively.

[0098] In this embodiment, step S2 involves vehicle behavior search of vehicle-level behavior decision, quantizing the driving intention of the vehicle, performing behavior search and target region allocation, and obtaining the driving decision of the vehicle, specifically including:

[0099] A Monte Carlo tree search (MCTS) algorithm is used to construct a behavior search tree, and the executable actions of the intelligent connected vehicle (such as left turn, straight, right turn, acceleration, and deceleration) are mapped to numbered regions (such as 0-5 as shown in the figure) in Figure 3 The driving behavior of the vehicle can mainly be divided into two main driving behaviors of following and lane changing, and can be further divided into acceleration and deceleration in following, left turn and right turn in lane changing, and left lane changing and right lane changing, in order to quantitatively and comprehensively evaluate all driving behaviors, six potential target regions are defined as shown in Figure 3 First, high-risk paths with high-risk regions (such as construction, explosion signs) or traffic rule restrictions (such as no-entry signs, traffic lights, forming a feasibility vector) are filtered out, as shown in Figure 4

[0100] As shown in Figure 5 When multiple vehicles intend to enter the same target region, priority is used for coordination, and high priority is given priority to allocate the target region, specifically including:

[0101] The priority of the human-driven vehicle is always higher than that of the intelligent connected vehicle, and when the intelligent connected vehicle competes for the region, the straight priority is higher than the lane changing priority, and when the intelligent connected vehicle is for lane changing, the forced lane changing priority is higher than the free lane changing priority;

[0102] When the intelligent connected vehicle is for free lane changing, the intelligent connected vehicle with high priority is determined based on the Monte Carlo tree search algorithm:

[0103] In the Monte Carlo tree search algorithm, each child node in the search tree has a score representing its own value, and in the application of the right-of-way allocation problem, the evaluation strategy of the child node value score is often related to the corresponding target function. According to the difference of the child node value score, the Monte Carlo tree search strategy will determine which branches of the search tree to preferentially expand.

[0104] Action space definition, design a discrete action set for each vehicle:

[0105] The group action space is the Cartesian product of the actions of each vehicle A G ​= A1x A2x... x A m , the number of action combinations is limited to 5 3 = 125 (when m = 3);

[0106] A search tree is gradually constructed by iteration, and each iteration process includes four steps: selection, expansion, simulation and back propagation, as shown in Figure 6 :

[0107] Selection: Starting from the root node, the optimal node is selected based on the strategy formula until the selected node is an unexpandable node; the strategy formula is:

[0108] Where q i is the value score of child node i; n is the total number of times of exploring the search tree; n i is the number of times of exploring node i; ω is a weight coefficient, which can be 0.5.

[0109] In the iterative loop of Monte Carlo tree search, an expandable node refers to a node that has unexplored child nodes and is not a leaf node. In the "selection" step, the child node with the highest value score will be selected as the next expanded node. The first term in the strategy formula only consists of the node value score, which aims to encourage the selection of the child node with the highest value score in the search process. The second term in the strategy formula considers the number of times of exploring the current node and the total number of explorations, giving a higher value score to nodes that have been explored less, which encourages more exploration of new nodes in the search process.

[0110] The first term of the strategy tends to "conservative" actions, i.e., directly selecting the node with the highest value score, and the second term tends to "adventurous" actions, trying to explore as many new nodes as possible to find the globally optimal solution. The weight ω is used to balance the two opposing search tendencies.

[0111] Of course, three improvements can be introduced based on the above strategy:

[0112] (1) Dynamic depth limit: dynamically adjust the search depth according to the group urgency:

[0113]

[0114] (2) Heuristic reward function:

[0115] The driving benefit is not only related to the state of the vehicle, but also related to the degree of conflict between vehicles, so safety, speed and comfort indicators are used to reflect the reward function of the decision-making subject.

[0116] a. Safety

[0117] Safety is the primary indicator for avoiding collisions during intersection traffic. Assuming ΔT represents the time difference between vehicles reaching the conflict point, the smaller ΔT is, the more intense the conflict is and the smaller the driving benefit is. The safety index is calculated as:

[0118]

[0119] Where: T i 、T j are the times when the i-th target vehicle and the j-th conflicting vehicle arrive at the conflict point at time t.

[0120] b. Speed

[0121] Rapidity refers to the vehicle's expectation to pass through an intersection or merging area in a relatively short time. The rapidity index is reflected by the change in speed, namely:

[0122]

[0123] Where: Δv i , Δv j is the speed change of vehicle i and vehicle j within time Δt.

[0124] c. Comfort

[0125] Comfort mainly considers the change of vehicle longitudinal acceleration. Excessive acceleration change will lead to a decrease in comfort. Therefore, the acceleration change is selected to reflect the comfort index as follows:

[0126]

[0127] Where: Δa i , Δa j is the acceleration change of vehicle i and vehicle j from time t-1 to time t, including all actions in the conflict resolution strategy set.

[0128] Extension: randomly generate a new child node for the selected node and add the new node to the search tree structure;

[0129] Simulation: Starting from the expanded new node, execute the random sampling simulation strategy to continuously generate new child nodes, that is, randomly select a vehicle from the current part of the traffic sequence that is not included, and add it to the end of the traffic sequence until the complete traffic sequence is generated. Figure 6 Taking the CB node in the search process as an example, after executing the random sampling simulation strategy, a new child node (such as CBE) will be randomly expanded and generated, and then the random sampling simulation strategy will continue to be executed on the CBE node, and so on, until a leaf node (such as CBEDA) is generated.

[0130] After obtaining the complete passing order, the node value score is updated according to the following formula:

[0131]

[0132] wherein q i is the value score of node i; J is a quantitative evaluation value, which can take values [-0.33, 66.7]; J i,max is the maximum value score of all sibling nodes of node i; J i,min is the minimum value score of all sibling nodes of node i;

[0133] Back propagation: the new value score q i of the node is continuously transmitted upwards along the direction of its parent node until the root node;

[0134] In the above iteration process, the current optimal passing order is constantly updated, and when the maximum search iteration limit is reached, the iteration process stops and returns the current optimal leaf node, which corresponds to the intelligent connected vehicle with high priority.

[0135] In the embodiment, step S3 involves vehicle trajectory planning of vehicle-level behavior decision, which makes each intelligent connected vehicle smoothly pass from the current position to the target area based on driving decision, specifically including:

[0136] A third-order Bezier curve is used to pre-generate a trajectory smoothly passing from the current position to the target area; wherein the third-order Bezier curve has good smoothness and computational efficiency, and is suitable for path planning in traffic environment. The trajectory is defined by four control points:

[0137]

[0138] Control point setting: the first point is the current vehicle position; the fourth point is the center of the target area allocated by MCTS; the middle control points are offset by a fixed proportion to control the shape of the curve. The trajectory is discretized into 101 points to form a trajectory point set:

[0139] Based on the trajectory length and the expected time window, the expected speed is generated:

[0140] wherein c is an adjustment coefficient in speed planning, which can be set according to actual working conditions, such as taking a value of 1.0; D TP is the trajectory length, ΔT is the expected time window, and v0 is the initial speed;

[0141] And a proportional controller is used to generate the longitudinal acceleration :

[0142] Among them, a cmax is the maximum permissible acceleration, k p is the gain coefficient of the proportional controller. Similarly, the gain coefficient can be set according to the actual working conditions, for example, the value is 0.5; is the current speed of vehicle i;

[0143] Analyze whether the closest distance between the trajectory points of two intelligent connected vehicles is below a threshold. If so, it is determined to be a conflict and the vehicle to be avoided is determined by the set priority;

[0144] A collision ellipse model is used to determine whether there is a trajectory conflict between the intelligent connected vehicle and the manually driven vehicle. If so, the intelligent connected vehicle is at a low priority and reduces acceleration to avoid collision:

[0145] d cmax is the maximum allowable deceleration, v i is the current speed of the intelligent connected vehicle i, v j is the current speed of the manually driven vehicle j. Considering the uncertainty of the manually driven vehicle trajectory, the collision ellipse model is used to judge the potential conflict:

[0146] in,

[0147] Select a forward viewpoint in the trajectory and construct a preview vector :

[0148] Among them, l p Indicates the distance between the current vehicle head position and the front view point, x target is the horizontal coordinate of the foresight point, is the horizontal coordinate of the current vehicle i. The above coordinates belong to the global map coordinate system or the map Cartesian coordinate system, which will not be repeated here.

[0149] Based on the bicycle model, the angle α between the preview vector and the current direction of the vehicle is converted into the steering angle :

[0150] Among them, l pr L is the distance from the rear wheel to the preview point; wb is the vehicle wheelbase;

[0151] The acceleration (longitudinal control) and steering angle (transverse control) in the form of instructions to the vehicle actuators of the corresponding intelligent connected vehicle, which performs actions according to the corresponding acceleration and steering angle.

[0152] The application can be widely applied to multi-vehicle cooperative control scenes in intelligent connected hybrid traffic environments, and is particularly suitable for complex road environments such as high-speed main line lanes, urban expressway intersections, and temporary construction lane-changing areas. By introducing a roadside cooperative control unit for global perception and intelligent guidance of multi-vehicle states, efficient scheduling and dynamic right-of-way allocation of vehicle behavior coordination are realized, effectively improving the traffic efficiency and operational safety level of key road sections.

[0153] The double-layer group decision control method proposed by the application fully utilizes the advantages of lane-level and vehicle-level cooperative control, significantly enhancing the adaptability to traffic flow changes. In the multi-vehicle cooperative scheduling process, real-time adjustment of vehicle travel path, speed and timing is realized through high-frequency V2V / V2I communication and game optimization algorithm, effectively relieving traffic conflicts in the interlaced area and reducing potential traffic risks caused by inconsistent vehicle behavior. At the same time, while improving traffic efficiency, the method also considers driving comfort and operational stability, providing reliable control support for high-density traffic scenarios.

[0154] The application has good scalability and portability, and has wide application potential in future intelligent transportation systems. Its control mechanism can be seamlessly embedded in existing intelligent transportation infrastructure, promoting the upgrading of urban road traffic systems from traditional static management to dynamic, self-adaptive cooperative control mode. Overall, the application not only provides key technical support for large-scale deployment of intelligent connected vehicles, but also lays a solid foundation for realizing a more efficient, safe and green intelligent transportation ecosystem, and has significant engineering application prospects.

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

Claims

1. A two-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment, characterized by: include: Adjust lane type assignment and select target lane for the vehicle; Quantify the vehicle's driving intention, conduct behavior search and target area allocation, and obtain the vehicle's driving decision; Based on driving decisions, each intelligent connected vehicle can smoothly move from its current location to the target area.

2. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 1 is characterized by: Select the target lane for the vehicle, including: Constructing an objective function, wherein the objective function includes a lane flow balancing function and a lane priority selection function; Set constraints for the objective function, solve the objective function under the constraints, and use the lane corresponding to the solution of the objective function as the target lane of the vehicle.

3. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 2 is characterized by: The lane flow balancing function is determined according to the following formula: Among them, c i,j,t It represents the traffic capacity of lane j in section i during time period t.

4. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 2 is characterized by: The lane priority function is determined according to the following formula: Among them, x i,j,t,a Indicates whether lane j on road section i during time period t is lane type a, x i,j,t,a ∈{0,1} are respectively no and yes; x i,j,t,b Indicates whether lane j in section i during time period t is lane type b, x i,j,t,b ∈{0,1} are no and yes respectively.

5. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 2 is characterized by: The constraints are determined according to the following formula: If road segment i contains an exit or an entrance, then: Among them, q i,t,n represents the flow of vehicle type n at road section i in time period t; M represents a constant; q i,j,t represents the flow rate of lane j in section i during time period t; x i,j,t,a Indicates whether lane j on road section i during time period t is lane type a, x i,j,t,a ∈{0,1} are respectively no and yes; x i,j,t,b Indicates whether lane j in section i during time period t is lane type b, x i,j,t,b ∈{0,1} are respectively no and yes; q i,j,t,n It represents the traffic volume of vehicle type n choosing lane j at road section i in time period t.

6. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 1 is characterized by: Quantify the vehicle's driving intention, conduct behavior search and target area allocation, and obtain the vehicle's driving decision, including: Setting multiple potential target areas for the vehicle based on the vehicle's execution actions; When multiple vehicles intend to enter the same target area, they are coordinated based on priority, and the vehicle with higher priority is allocated to the target area first.

7. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 6 is characterized by: Coordinate by priority, and allocate the target area with high priority first, including: Manually driven vehicles always have a higher priority than connected vehicles. When connected vehicles compete for the area, going straight takes precedence over lane changing. When all connected vehicles are changing lanes, forced lane changing takes precedence over free lane changing. When all connected vehicles are free to change lanes, the high-priority connected vehicle is determined based on the Monte Carlo tree search algorithm: Action space definition: a discretized action set is designed for each vehicle. The group action space is the Cartesian product of the actions of each vehicle. The number of action combinations is limited to a certain range through pruning strategies. A search tree is gradually constructed in an iterative manner. One iteration process includes four steps: selection, expansion, simulation, and backpropagation: Selection: Starting from the root node, the optimal node is continuously selected based on the strategy formula until the selected node is a non-scalable node; the strategy formula is: Among them, q i is the value score of child node i; n is the total number of times the search tree is currently explored; n i is the number of times node i has been explored; ω is the weight coefficient; Extension: randomly generate a new child node for the selected node and add the new node to the search tree structure; Simulation: Starting from the newly expanded node, a random sampling simulation strategy is executed to continuously generate new child nodes. That is, a vehicle is randomly selected from the vehicles not yet included in the current partial passage sequence and added to the end of the partial passage sequence until the complete passage sequence is generated. The node value score is then updated according to the following formula: Among them, q i is the value score of node i; J is the quantitative evaluation value; J i,max is the maximum value score among all sibling nodes of node i; J i,min is the minimum value score among all sibling nodes of node i; Back propagation: the new value score q of the node i Continue to pass upward along the direction of its parent node until it reaches the root node; In the above iterative process, the current optimal passage order is constantly updated. When the maximum search iteration limit is reached, the iterative process stops and returns the current optimal leaf node. The optimal leaf node corresponds to the smart connected vehicle with high priority.

8. The dual-layer decision-making optimization method for multi-vehicle cooperative driving in an intelligent connected mixed traffic environment according to claim 1 is characterized by: Based on driving decisions, each intelligent connected vehicle is smoothly guided from its current location to the target area, including: Using a third-order Bezier curve, a trajectory is pre-generated that smoothly leads from the current position to the target area; Generate expected speed based on trajectory length and expected time window Among them, c is the adjustment coefficient in speed planning, D TP is the trajectory length, ΔT is the desired time window, and v0 is the initial velocity; A proportional controller is used to generate longitudinal acceleration Among them, a cmax is the maximum permissible acceleration, k p is the gain coefficient of the proportional controller, is the current speed of the intelligent connected car i; Analyze whether the closest distance between the trajectory points of two intelligent connected vehicles is below a threshold. If so, a conflict is determined, and the vehicle to be avoided is determined based on the set priority. A collision ellipse model is used to determine whether there is a trajectory conflict between the intelligent connected vehicle and the manually driven vehicle. If so, the intelligent connected vehicle, which is at a lower priority, reduces acceleration to avoid a collision. Select a forward viewpoint in the trajectory and construct a preview vector Among them, l p Indicates the distance between the current vehicle head position and the front view point, is the horizontal coordinate of the foresight point, is the horizontal coordinate of the current vehicle i; Based on the bicycle model, the angle α between the preview vector and the current direction of the vehicle is converted into the steering angle Among them, l pr L is the distance from the rear wheel to the preview point; wb is the vehicle wheelbase; The acceleration and steering angle The data is sent to the corresponding intelligent connected vehicle, which then performs actions according to the corresponding acceleration and steering angle.

Images