阅读说明：本技术 一种高速公路瓶颈区域自动驾驶协同控制方法 (Automatic driving cooperative control method for bottleneck area of expressway ) 是由 曾伟良 廖立 莫菲 黄永慧 孙为军 于 2021-07-26 设计创作，主要内容包括：本发明公开了一种高速公路瓶颈区域自动驾驶协同优化方法,包括：构建考虑车辆通行时间、耗电量的车辆轨迹优化模型；所述模型用于根据车辆的实时速度、加速度来计算成本值；将车辆的通行优先权顺序寻优转换为树搜索以进行树结构的展开,将蒙特卡洛树搜索算法与设定的选择策略相结合,使得搜索不断向总成本更小的方向进行搜索,得到候选的通行优先权顺序；根据所述候选通行优先权顺序对车辆轨迹进行轨迹规划,使得车辆能从初始化状态行驶到候选的通行优先权顺序所对应的状态；通过不断循环迭代,直到所述树结构扩展的节点中包含所有车辆为止,得到对应总成本最小的通行优先权顺序和对应的车辆轨迹,达到最小化通行时间和耗电量的目的。(The invention discloses an automatic driving cooperative optimization method for a bottleneck area of a highway, which comprises the following steps: constructing a vehicle track optimization model considering vehicle passing time and power consumption; the model is used for calculating a cost value according to the real-time speed and the acceleration of the vehicle; optimizing the traffic priority sequence of the vehicle, converting the optimized traffic priority sequence into tree searching to expand a tree structure, and combining a Monte Carlo tree searching algorithm with a set selection strategy to continuously search towards a direction with smaller total cost to obtain a candidate traffic priority sequence; carrying out trajectory planning on the vehicle trajectory according to the candidate passage priority sequence, so that the vehicle can run from an initial state to a state corresponding to the candidate passage priority sequence; and continuously iterating in a circulating manner until all vehicles are contained in the nodes expanded by the tree structure, so that a traffic priority sequence with the minimum corresponding total cost and a corresponding vehicle track are obtained, and the purposes of minimizing traffic time and electricity consumption are achieved.)

1. An automatic driving cooperative optimization method for a bottleneck area of a highway is characterized by comprising the following steps:

setting a control area for a bottleneck area of a highway, and constructing a vehicle track optimization model considering vehicle passing time and power consumption for vehicles in the control area; the model is used for calculating a cost value according to the real-time speed and the acceleration of the vehicle;

optimizing the traffic priority sequence of the vehicle, converting the optimized traffic priority sequence into tree searching to expand a tree structure, and combining a Monte Carlo tree searching algorithm with a set selection strategy to continuously search towards a direction with smaller total cost to obtain a candidate traffic priority sequence;

carrying out trajectory planning on the vehicle trajectory according to the candidate traffic priority sequence, so that the vehicle can drive from an initial state to a state corresponding to the candidate traffic priority sequence, and calculating the total cost of the process according to a vehicle trajectory optimization model;

and continuously carrying out loop iteration until all vehicles are contained in the nodes expanded by the tree structure, obtaining a traffic priority sequence with the minimum corresponding total cost and a corresponding vehicle track, and then controlling the vehicles in a control range by using the vehicle track to achieve the purpose of minimizing traffic time and electricity consumption.

2. The automatic driving collaborative optimization method for the bottleneck area of the expressway according to claim 1, wherein the vehicle trajectory optimization model is expressed as follows:

wherein x (t), u (t) are a state set and an input vector set of the vehicles in the control area at the time t respectively, N is the number of the vehicles in the control area, and omega₁、ω₂Time cost coefficients and power consumption cost coefficients are respectively used for converting the cost into price;andrespectively representing the time when the vehicle i passes through the bottleneck region and the time when the vehicle i enters the bottleneck region,the net power consumption of the vehicle i is represented and calculated by the kinematic data of the vehicle and the conversion efficiency between systems of the vehicle;

the constraints of the vehicle trajectory optimization model include vehicle dynamics constraints, traffic priorities of the vehicles, and vehicle collision avoidance constraints.

3. The automatic driving cooperative optimization method for the bottleneck area of the expressway according to claim 1, wherein the optimizing of the traffic priority order of the vehicles is converted into tree searching for the expansion of a tree structure, and the Monte Carlo tree searching algorithm is combined with the set selection strategy, and the method comprises the following steps:

3.1 performing ID distribution on vehicles in the control area;

3.2, each vehicle is taken as a node, and the data information of the node comprises a character string S consisting of vehicle numbers ID, a father node and child node information, and comprises the number N of traversed times and an award value Q;

3.3, transmitting a starting node P which needs to start searching at present, returning the best node according to a set selection strategy for expansion, and if the node is a leaf node, and all vehicle numbers ID are in a character string S at the moment, directly ending;

3.4 for the returned node, expanding a new node as a child node of the node, wherein the character string S of the new node is formed by adding a new vehicle number ID at the tail of the character string of a parent node of the new node; in this process, the character string S between the child nodes of the same parent node cannot be repeated;

3.5, carrying out reward value evaluation on the new node, planning the passing priority sequence of the node in the process, and calculating the reward value of the new node;

3.6 according to the reward value of the new node, backtracking upwards, and updating the data information of the new node and the upstream father node thereof;

3.7 repeat steps 3.3-3.7 until the string S in the loop start node contains all the vehicle numbers ID, then the string S is the optimal candidate traffic priority order.

4. The automatic driving cooperative optimization method for the bottleneck area of the expressway according to claim 1, wherein the evaluating of the reward value of the new node, the planning of the traffic priority order of the nodes in the process and the calculation of the reward value of the new node comprise:

3.5.1 obtaining the character string S of the node according to the given new node, if S includes all vehicle numbers ID, directly skipping step 3.5.2, otherwise, needing to add the vehicle numbers ID not added to S, and further obtaining the complete traffic priority order Ψ, the specific steps are as follows:

(1) firstly, dividing the vehicle number into three arrays P according to the lanes₁、P₂、P₃Sorting the elements in the array according to the longitudinal distance from the conflict area;

(2) will P₁、P₂、P₃The elements in the string S that have already appeared in the string S are removed, and if the empty array appears, the elements are also removed; randomly selecting an array P from the stored arrays_iAnd taking out the first element P thereof_i(1) Adding the element to the end of S;

(3) repeating the steps (1) and (2) until all the vehicle IDs are contained in S; finally, obtaining a complete traffic priority order psi;

3.5.2, planning the vehicle track according to the complete traffic priority order Ψ given in the step 3.5.1, and finally obtaining the reward value of the new node.

5. The automatic driving cooperative optimization method for the bottleneck area of the expressway according to claim 1, wherein the set selection strategy is as follows:

firstly, finding out the sub-nodes which are not traversed currently, and randomly selecting if a plurality of sub-nodes exist; if all the UCB values are traversed, finding the maximum UCB value, and if the UCB values are equal, randomly selecting; the UCB value for node v' is calculated as follows:

wherein v 'represents the current node, v represents the parent node of v', Q represents the accumulated reward value of the node, N represents the number of access times of the node, and c is a weight parameter.

Further, trajectory planning is performed on the vehicle trajectory according to the candidate passage priority order, so that the vehicle can travel from the initialization state to the state corresponding to the candidate passage priority order, and the total cost of the process is calculated according to a vehicle trajectory optimization model, and the method comprises the following steps:

4.1 constructing a matrix by all vehicle IDs, carrying out differential operation on the matrix and the candidate traffic priority sequence matrix, and processing the result to ensure that the vehicle IDs corresponding to the differential values are consistent with the index values of the vehicle IDs, and finally obtaining a differential matrix so as to determine the target speed of the vehicle;

4.2 planning the track of the single vehicle according to the determined target speed;

4.3 local trajectory planning: step 4.2 is carried out on all vehicles in the control area, and then local track optimization of all vehicles can be obtained;

4.4 Global trajectory planning: according to the local tracks of all the vehicles obtained in the step 4.3, in order to obtain a global track, continuously sampling and updating the position information of the vehicles, and then re-performing the local track planning in the step 4.3 according to the updated position information of the vehicles;

4.5 executing step 4.4 circularly until all vehicles pass through the bottleneck area; finally, obtaining a track planning scheme, total passing time and total power consumption of the whole process of the vehicle, and then calculating the total cost value of the process;

4.6 feeds back the total cost value obtained from step 4.5 back to step 3.5.2 as its prize value q.

6. The automatic driving cooperative optimization method for the bottleneck area of the expressway according to claim 1, wherein the planning of the trajectory of a single vehicle according to the determined target speed comprises:

4.2.1 specifying vehicle position (s, d) in Frenet coordinates based on longitudinal displacement s along a given reference line and lateral offset d from the reference line, while tracing points can be usedIs shown in whichRespectively a first derivative and a second derivative of s,first and second derivatives of d, respectively;

4.2.2, defining a set of vehicle trajectoriesThe trajectories in the set are all composed of s (t) and d (t), and are represented as time polynomials along the Frenet axis:

wherein s (t) and d (t) represent the longitudinal displacement s and the lateral offset d, a at time t_i、c_iIs a polynomial coefficient, tⁱAs a time parameter, t₀As track start time, t_fIs the track end time; therefore, if a complete track set is required, only a needs to be solved_i、c_iThese polynomial coefficients;

4.2.3, set of trajectories under Frenet coordinatesThen, the digital value is converted into a CartesianCoordinates [ x, y, theta, k, v, a ]]Wherein [ x, y]Is the position under a Cartesian coordinate system, k is the curvature, theta is the course angle, v is the speed, and a is the acceleration;

4.2.4, based on the vehicle dynamics constraints and vehicle collision avoidance constraints, it can be refined into hard constraints and soft constraints to exclude trajectories that do not meet the constraints and to select the optimal trajectory

7. The automatic driving cooperative optimization method for bottleneck area of expressway according to claim 1, wherein polynomial coefficient a is solved_i、c_iThe process of obtaining a complete track set is as follows:

(1) in order to maintain the continuity and dynamic feasibility of the previous trajectory, the following constraints need to be made on the trajectory termination:

wherein, t_fIs the track end time; d_fThe final lateral position ignores the lateral velocity of the terminal state, so the derivative of d at the end time and the second derivative are both set to 0; v. of_fThe final longitudinal velocity;

(2) can be initialized by determining the trackAnd track end stateAnd within the time interval T, utilizing quintic and quadric polynomials to make fitting solution to obtain a_i、c_iAnd then completing the trajectory planning between the two states;

(3) to obtain d_f、v_fRespectively setting three different termination statesNamely, the cruising, lane changing and following modes are specifically set as follows:

a) in cruise mode, set d_fD remaining in the initial state₀Same, v_fSetting as a target speed;

b) in lane change mode, set d_fIs adjacent to the center of the lane, v_fSetting as a target speed;

c) in the following mode, d is set_fD remaining in the initial state₀Same, v_fSet as the front vehicle speed.

(4) Through the steps, a track set can be obtained

8. The automatic driving cooperative optimization method for bottleneck areas of expressways according to claim 1, wherein the dynamic constraints and collision avoidance constraints of vehicles can be refined into hard constraints and soft constraints to eliminate tracks which do not meet the constraints and select the optimal trackThe method comprises the following steps:

a) collision avoidance constraints if the trajectories are setIn which there is a collision trajectory with another vehicleThe collision trajectory is directly excluded

b) Hard constraints to exclude traces that do not meet vehicle dynamics performance standards:

in the above formula, k (t), θ (t), a (t) represent curvature k, heading angle θ, and acceleration a at time t; k is a radical of_max、θ_max、v_max、a_maxRepresents the maximum value of the set curvature, heading angle, speed, acceleration, v_minRepresents a set speed minimum;

c) soft constraints, by designing a cost function and then optimizing in a trajectory set using numerical valuesFinding out the track with the minimum cost;

wherein the content of the first and second substances,representing a set of trajectories, ω_d、ω_v、ω_aIs a weight coefficient, d_desRepresenting the center of the target lane, v_des(t) indicates target velocity, a (t) indicates vehicle acceleration, d (t) and v (t) indicate lateral offset and velocity at time t;

d) by screening in steps a), b), c), the track set can be obtainedTo select the optimal track

Technical Field

The invention relates to the field of automatic driving control, in particular to an automatic driving cooperative control method for a bottleneck area of a highway.

Background

The country has also come out related policies to support the development of the automatic driving automobile in succession, thanks to the global new technological revolution and the vigorous development of the industrial revolution, and the electromotion, networking and intellectualization become the development trend and trend of the automobile industry. At present, in the field of research and application of vehicle intellectualization, a highly effective achievement is taken, how to realize cooperative driving on the basis of single-vehicle intellectualization, explore a multi-vehicle cooperative mechanism and research a multi-vehicle cooperative behavior decision method is important for improving traffic safety and relieving traffic jam of an intelligent traffic system.

The bottleneck area is a road section which is easy to cause traffic jam in a road network, is a weak road section in the road network, and is a key road section which influences the level of the whole road network. The bottleneck in the traffic network is mainly caused by the increasing road traffic demand, unreasonable road conditions, traffic flow conditions and emergencies. The bottleneck area of the expressway mainly means that the traffic capacity of the road section is reduced due to the phenomena of flow splitting and converging in the processes of sudden accidents, road construction and vehicle driving, and traffic jam is caused.

One key point of cooperative driving is to determine the right of way (passing order) of vehicles in the control area in the bottleneck area. Strategies for determining traffic sequences in collaborative driving can be broadly divided into two categories, provisional negotiation based and plan based. The temporary negotiation-based method, which mostly coordinates the vehicle trajectory by way of a first-in-first-out-based traffic priority order, can be solved in a very short time, but a first-in-first-out-based traffic priority order is not a general solution to the problem, and in most cases there is a better solution than this order, or in some cases the order cannot be solved. Planning-based strategies, which consider vehicles that are going to arrive within a certain spatial range of the bottleneck area and make a relatively long-term driving plan for the vehicles, solve the model directly by mathematical planning or heuristic algorithms (ant colony algorithm, genetic algorithm, etc.), however, the process takes a lot of time, especially in view of vehicle collision avoidance issues. Most of methods based on temporary negotiation and planning do not consider the lane change situation of vehicles, new research also appears in recent years to consider the lane change situation of vehicles, the traffic management problem based on tracks is expressed as a mixed integer planning problem, a rolling time domain method is developed to improve the calculation efficiency, the optimal tracks are searched for a plurality of automatic driving vehicles, and the conflict problem is solved, however, the lane change maneuver is assumed to be completed within a given time by the method, and the emergency situation in the lane change process is ignored. Most of the above methods only take the transit time minimization as an optimization target, and do not consider the energy consumption problem.

Disclosure of Invention

The invention mainly provides an automatic driving collaborative optimization method for a bottleneck area of a highway, which is mainly oriented to a complex driving scene of a full-automatic driving vehicle, such as a driving scene of a lane number reduction or a driving scene of a bottleneck area of a road.

In order to realize the task, the invention adopts the following technical scheme:

an automatic driving collaborative optimization method for a bottleneck area of a highway comprises the following steps:

setting a control area for a bottleneck area of a highway, and constructing a vehicle track optimization model considering vehicle passing time and power consumption for vehicles in the control area; the model is used for calculating a cost value according to the real-time speed and the acceleration of the vehicle;

optimizing the traffic priority sequence of the vehicle, converting the optimized traffic priority sequence into tree searching to expand a tree structure, and combining a Monte Carlo tree searching algorithm with a set selection strategy to continuously search towards a direction with smaller total cost to obtain a candidate traffic priority sequence;

carrying out trajectory planning on the vehicle trajectory according to the candidate passage priority sequence, so that the vehicle can run from an initial state to a state corresponding to the candidate passage priority sequence;

and continuously carrying out loop iteration until all vehicles are contained in the nodes expanded by the tree structure, obtaining a traffic priority sequence with the minimum corresponding total cost and a corresponding vehicle track, and then controlling the vehicles in a control range by using the vehicle track to achieve the purpose of minimizing traffic time and electricity consumption.

Further, the vehicle trajectory optimization model is represented as follows:

wherein x (t), u (t) are a state set and an input vector set of the vehicles in the control area at the time t respectively, N is the number of the vehicles in the control area, and omega₁、ω₂Time cost coefficients and power consumption cost coefficients are respectively used for converting the cost into price;andrespectively representing the time when the vehicle i passes through the bottleneck region and the time when the vehicle i enters the bottleneck region,the net power consumption of the vehicle i is represented and calculated by the kinematic data of the vehicle and the conversion efficiency between systems of the vehicle;

the constraints of the vehicle trajectory optimization model include vehicle dynamics constraints, traffic priorities of the vehicles, and vehicle collision avoidance constraints.

Further, the optimizing the traffic priority order of the vehicle is converted into tree search to perform tree structure expansion, and the monte carlo tree search algorithm is combined with the set selection strategy, and the method comprises the following steps:

3.1 performing ID distribution on vehicles in the control area;

3.2, each vehicle is taken as a node, and the data information of the node comprises a character string S consisting of vehicle numbers ID, a father node and child node information, and comprises the number N of traversed times and an award value Q;

3.3, transmitting a starting node P which needs to start searching at present, returning the best node according to a set selection strategy for expansion, and if the node is a leaf node, and all vehicle numbers ID are in a character string S at the moment, directly ending;

3.4 for the returned node, expanding a new node as a child node of the node, wherein the character string S of the new node is formed by adding a new vehicle number ID at the tail of the character string of a parent node of the new node; in this process, the character string S between the child nodes of the same parent node cannot be repeated;

3.5, carrying out reward value evaluation on the new node, planning the passing priority sequence of the node in the process, and calculating the reward value of the new node;

3.6 according to the reward value of the new node, backtracking upwards, and updating the data information of the new node and the upstream father node thereof;

3.7 repeat steps 3.3-3.7 until the string S in the loop start node contains all the vehicle numbers ID, then the string S is the optimal candidate traffic priority order.

Further, the evaluating the reward value for the new node, during which the traffic priority order of the nodes is planned, and calculating the reward value of the new node includes:

3.5.1 obtaining the character string S of the node according to the given new node, if S includes all vehicle numbers ID, directly skipping step 3.5.2, otherwise, needing to add the vehicle numbers ID not added to S, and further obtaining the complete traffic priority order Ψ, the specific steps are as follows:

(1) firstly, the vehicles are weavedThe number is divided into three arrays P according to lanes₁、P₂、P₃Sorting the elements in the array according to the longitudinal distance from the conflict area;

(2) will P₁、P₂、P₃The elements in the string S that have already appeared in the string S are removed, and if the empty array appears, the elements are also removed; randomly selecting an array P from the stored arrays_iAnd taking out the first element P thereof_i(1) Adding the element to the end of S;

(3) repeating the steps (1) and (2) until all the vehicle IDs are contained in S; finally, obtaining a complete traffic priority order psi;

3.5.2, planning the vehicle track according to the complete traffic priority order Ψ given in the step 3.5.1, and finally obtaining the reward value of the new node.

Further, the set selection policy is:

firstly, finding out the sub-nodes which are not traversed currently, and randomly selecting if a plurality of sub-nodes exist; if all the UCB values are traversed, finding the maximum UCB value, and if the UCB values are equal, randomly selecting; the UCB value for node v' is calculated as follows:

wherein v 'represents the current node, v represents the parent node of v', Q represents the accumulated reward value of the node, N represents the number of access times of the node, and c is a weight parameter.

Further, trajectory planning is performed on the vehicle trajectory according to the candidate traffic priority order, so that the vehicle can travel from the initialization state to the state corresponding to the candidate traffic priority order, and the trajectory planning method comprises the following steps:

4.1 constructing a matrix by all vehicle IDs, carrying out differential operation on the matrix and the candidate traffic priority sequence matrix, and processing the result to ensure that the vehicle IDs corresponding to the differential values are consistent with the index values of the vehicle IDs, and finally obtaining a differential matrix so as to determine the target speed of the vehicle;

4.2 planning the track of the single vehicle according to the determined target speed;

4.3 local trajectory planning: step 4.2 is carried out on all vehicles in the control area, and then local track optimization of all vehicles can be obtained;

4.4 Global trajectory planning: according to the local tracks of all the vehicles obtained in the step 4.3, in order to obtain a global track, continuously sampling and updating the position information of the vehicles, and then re-performing the local track planning in the step 4.3 according to the updated position information of the vehicles;

4.5 executing step 4.4 circularly until all vehicles pass through the bottleneck area; finally, obtaining a track planning scheme, total passing time and total power consumption of the whole process of the vehicle, and then calculating the total cost value of the process;

4.6 feeds back the total cost value obtained from step 4.5 back to step 3.5.2 as its prize value q.

Further, according to the determined target speed, planning the track of the single vehicle, comprising:

4.2.1 specifying vehicle position (s, d) in Frenet coordinates based on longitudinal displacement s along a given reference line and lateral offset d from the reference line, while tracing points can be usedIs shown in whichRespectively a first derivative and a second derivative of s,first and second derivatives of d, respectively;

4.2.2, defining a set of vehicle trajectoriesThe trajectories in the set are all composed of s (t) and d (t), and are represented as time polynomials along the Frenet axis:

wherein s (t) and d (t) represent the longitudinal displacement s and the lateral offset d, a at time t_i、c_iIs a polynomial coefficient, tⁱAs a time parameter, t₀As track start time, t_fIs the track end time; therefore, if a complete track set is required, only a needs to be solved_i、c_iThese polynomial coefficients;

4.2.3, set of trajectories under Frenet coordinatesThen, the coordinate system is converted into Cartesian coordinates [ x, y, theta, k, v, a ] through numerical transformation]Wherein [ x, y]Is the position under a Cartesian coordinate system, k is the curvature, theta is the course angle, v is the speed, and a is the acceleration;

4.2.4, based on the vehicle dynamics constraints and vehicle collision avoidance constraints, it can be refined into hard constraints and soft constraints to exclude trajectories that do not meet the constraints and to select the optimal trajectory

Further, solving the polynomial coefficient a_i、c_iThe process of obtaining a complete track set is as follows:

(1) in order to maintain the continuity and dynamic feasibility of the previous trajectory, the following constraints need to be made on the trajectory termination:

wherein, t_fIs the track end time; d_fThe final lateral position ignores the lateral velocity of the terminal state, so the derivative of d at the end time and the second derivative are both set to 0; v. of_fThe final longitudinal velocity;

(2) can be initialized by determining the trackAnd track end stateAnd within the time interval T, utilizing quintic and quadric polynomials to make fitting solution to obtain a_i、c_iAnd then completing the trajectory planning between the two states;

(3) to obtain d_f、v_fThe values of (2) are respectively set with three different termination states, namely cruising, lane changing and following modes, and are specifically set as follows:

a) in cruise mode, set d_fD remaining in the initial state₀Same, v_fSetting as a target speed;

b) in lane change mode, set d_fIs adjacent to the center of the lane, v_fSetting as a target speed;

c) in the following mode, d is set_fD remaining in the initial state₀Same, v_fSet as the front vehicle speed.

(4) Through the steps, a track set can be obtained

Further, the dynamic constraint and the collision avoidance constraint of the vehicle can be refined into a hard constraint and a soft constraint to eliminate the tracks which do not meet the constraint condition and select the optimal trackThe method comprises the following steps:

a) collision avoidance constraints if the trajectories are setIn which there is a collision trajectory with another vehicleThe collision trajectory is directly excluded

b) Hard constraints to exclude traces that do not meet vehicle dynamics performance standards:

in the above formula, k (t), θ (t), a (t) represent curvature k, heading angle θ, and acceleration a at time t; k is a radical of_max、θ_max、v_max、a_maxRepresents the maximum value of the set curvature, heading angle, speed, acceleration, v_minRepresents a set speed minimum;

c) soft constraints, by designing a cost function and then optimizing in a trajectory set using numerical valuesFinding out the track with the minimum cost;

wherein the content of the first and second substances,representing a set of trajectories, ω_d、ω_v、ω_aIs a weight coefficient, d_desRepresenting the center of the target lane, c_des(t) indicates target velocity, a (t) indicates vehicle acceleration, d (t) and v (t) indicate lateral offset and velocity at time t;

d) by screening in steps a), b), c), the track set can be obtainedTo select the optimal track

Compared with the prior art, the invention has the following technical characteristics:

1. the optimization function takes into account not only the transit time cost but also the power consumption cost, and converts both into the price cost by the weight coefficient to perform optimization.

2. In the priority sequence optimizing part, the priority sequence is converted into tree search, the Monte Carlo tree search algorithm is used for solving, and the provided selection strategy is combined, so that the search of the optimal priority sequence is accelerated, and the performance of the result is ensured.

3. In the vehicle track planning part, a dynamic track planning method is provided to solve the problem of track planning. And planning the track under the Frenet coordinate, and continuously iterating a local track planning algorithm to further obtain the track plan of the whole change process. The change process of the vehicle lane change maneuver is considered in the process; the multi-vehicle trajectory planning is prevented from falling into a local optimal solution through the inter-vehicle dynamic decision trajectory planning, and meanwhile, the trajectory planning process is accelerated.

4. A double-layer closed-loop solving framework is provided, the traffic optimization problem of a bottleneck area of a highway is divided into two parts (traffic priority optimization and vehicle track planning), the traffic priority optimization is carried out by combining a Monte Carlo tree searching method on the first layer, the vehicle track dynamic planning is carried out on the basis of the first layer on the second layer, the solving complexity is reduced, the two parts are continuously and circularly iterated through the Monte Carlo tree searching method and the vehicle track dynamic planning method, the optimal traffic priority sequence and the vehicle track planning scheme are finally obtained, the solving speed is accelerated, and meanwhile the obtained result is guaranteed to have good performance in the aspects of reducing traffic time and energy consumption.

Drawings

FIG. 1 is a schematic flow chart of a control method according to an embodiment of the present invention;

FIG. 2 is a schematic view of a driving scenario;

FIG. 3 is a schematic diagram of a closed loop solution framework;

FIG. 4 is a diagram of an iterative process of a Monte Carlo tree search algorithm;

FIG. 5 is a schematic diagram of trajectory generation.

Detailed Description

The invention is mainly oriented to the complex driving scenes of the fully-automatic driving vehicle, such as the driving scenes of the reduced number of lanes or the areas with bottlenecks on the roads, as shown in fig. 2. The autonomous driving mode is employed for both vehicles in the driving scenario shown in fig. 2, and with the support of V2X communication technology, their driving states (position, speed, acceleration, etc.) and intentions can be shared with neighboring vehicles and road infrastructure to better coordinate their movements. Therefore, we can express the cooperative driving problem of the bottleneck region of the expressway as a trajectory planning problem, and the constraints comprise vehicle dynamics, safety, physical constraints and traffic priority order.

Referring to the attached drawings, the invention provides an automatic driving cooperative control method for a bottleneck area of a highway, an optimization model established by the method considers the aspects of traffic efficiency and energy consumption, takes the minimized traffic time and the minimized power consumption as the optimization target, and solves the problem that the existing optimization method does not consider the energy consumption at the same time. In order to solve the model, a double-layer closed loop solving framework is adopted, the first layer is used for carrying out traffic priority optimization by combining a Monte Carlo tree searching method, the second layer is used for carrying out dynamic planning on the vehicle track on the basis of the first layer, and finally, the optimal traffic priority sequence and a vehicle track planning scheme are obtained through continuous loop iteration, so that the optimization problem is solved. The method comprises the following steps:

step 1, setting a control area for a bottleneck area of a highway, and constructing a vehicle track optimization model considering vehicle passing time and power consumption for vehicles in the control area; the model is used for calculating cost values according to the real-time speed and the acceleration of the vehicle.

In the step, a control area is set for a bottleneck area of the highway, and a track optimization model is established for all vehicles in the control area; the bottleneck area refers to the condition that partial lanes cannot pass on the road due to traffic accidents, construction, natural disasters and the like. The size of the control area is set according to actual requirements, for example, the control area is a road section 500 meters away from the bottleneck area of the expressway in the embodiment.

1.1 according to the physical and design constraints of the vehicle, establishing vehicle dynamics constraints for N vehicles in the control area range:

wherein x is_i(t) represents the state of the vehicle i at the time t, namely a vector value formed by the current position and the speed of the vehicle; u. of_i(t) represents an input vector of the vehicle i at the time t, namely an input vector consisting of the vehicle speed and the acceleration;is x_iA time derivative of (t); f. of_iThe (·) function represents the vehicle dynamics constraints.

1.2 setting the traffic priority p of the vehicle i_iAnd E, P is a traffic priority matrix of all vehicles in the control area, and when two vehicles have potential collision risks, the vehicle with low priority needs to decelerate to avoid the vehicle with high priority.

1.3 set vehicle collision avoidance restraint, with g_i(x_i(t),p_i) To indicate that the vehicle i is in the state x_i(t) the spatial extent occupied thereby; therefore, the case where the vehicles i and j do not collide at time t can be expressed as:

wherein p is_iAnd p_jIndicating the traffic priorities of vehicles i and j.

1.4 an objective function is constructed by using the time cost and the power consumption cost in the driving process, so that the optimized model can have a good balance between the time cost and the power consumption cost:

wherein, J_i(. h) represents a cost function for vehicle i;andrespectively representing the time when the vehicle i passes through the bottleneck area and the time when the vehicle i enters the bottleneck area; omega₁、ω₂Time cost coefficients and power consumption cost coefficients are respectively used for converting the cost into price; in this embodiment, the values are 0.01 and 0.8, respectively;representing the net power consumption of vehicle i.

For theCalculating an instantaneous power consumption of the vehicle with a vehicle speed and an acceleration per second as inputs; the specific calculation process of the net power consumption is as follows:

1.4.1 calculating the instantaneous Power P at the wheels from vehicle kinematic data (e.g. instantaneous speed, acceleration), road grade, rolling friction, and vehicle characteristics, etc_W(t); the instantaneous required tractive power P is then calculated based on the inter-vehicle system conversion efficiency (i.e., engine, braking system, battery)_T(t) or the generated regenerative power P_R(t); finally, during the running of the vehicle, P is paired_T(t) and P_R(t) performing an integral operation, thereby obtaining energy consumption and energy recovery during the driving process, and obtaining the net power consumption E of the vehicle during the driving process_AC。

1.4.2 instantaneous Power P at the wheel_W(t) is calculated as follows:

wherein m is the vehicle mass; a (t) is the instantaneous acceleration of the vehicle at time t; v (t) is a vehicleThe instantaneous speed of the vehicle at time t; g is gravity acceleration with the value of 9.8066m/s²(ii) a Theta is the road gradient; c_r、c₁And c₂Is a rolling resistance parameter that is related to road surface type, road conditions, and vehicle tires; rho_airThe value of the air mass density is 1.2256kg/m when the sea level temperature is 15 DEG C³；A_fThe area of the front area of the vehicle is the size; c_DIs the aerodynamic drag coefficient of the vehicle.

1.4.3 when the vehicle is in traction mode, energy flows from the electric machine to the wheels. At this time, the power of the motor is higher than that of the wheel, and it is assumed that the power of the wheel is positive; when the vehicle is in a regenerative braking mode, energy flows from the wheels to the electric machine. At this time, the power of the motor is lower than that of the wheel, and the power is assumed to be negative. Therefore P is_T(t) and P_R(t) can be according to P_w(t) is calculated as:

wherein eta is_DLThe conversion efficiency of the transmission system; eta_EMThe conversion efficiency of the motor; eta_BATThe conversion efficiency of the battery system; eta_RB(t) is the regenerative braking system conversion efficiency, which is related to the instantaneous acceleration, defined as follows:

where the value of a depends on the characteristics of the electric vehicle, and is usually between [0,1], and e is a natural logarithmic base.

1.4.4 combining the above formula, the energy consumption E can be directly obtained_CAnd energy recovery E_R：

Wherein, t₀And t_eRespectively representing the start time and the end time of the calculation cycle.

1.4.5 finally, the net consumption of electric energy E of the vehicle during travel_ACThis can be obtained by the following formula:

E_AC＝E_C-E_R

1.5 combining the steps, a complete vehicle track optimization model (cost function) can be obtained, and the vehicle cooperative control problem in the bottleneck region of the expressway can be solved by minimizing the cost value of the model.

Wherein, x (t), u (t) are respectively a state set and an input vector set of the vehicle in the control area at the time t.

And 2, optimizing the traffic priority sequence of the vehicle, converting the traffic priority sequence into tree searching to expand a tree structure, and combining a Monte Carlo tree searching algorithm with a set selection strategy to continuously search in a direction with smaller total cost to obtain a candidate traffic priority sequence.

For the vehicle track optimization model, direct solution is a mixed integer programming problem which is difficult to solve, especially considering collision avoidance between vehicles and lane change. The solution is performed by a two-level closed-loop solution framework, as shown in fig. 3. The first layer is used for carrying out the traffic priority optimization by combining a Monte Carlo tree searching method, so that the problem of the optimization of the traffic priority in the step 1.2 is solved; the second layer dynamically plans the vehicle track on the basis of the first layer, and solves the planning problems in the steps 1.1, 1.3 and 1.4; and (3) continuously and circularly iterating to finally obtain an optimal traffic priority sequence and a vehicle track planning scheme, so that the problem of vehicle cooperative control in a bottleneck region of the expressway is solved, namely the complete vehicle track optimization model provided in the step 1.5 is solved. The specific steps are as step 3 and step 4.

Step 3, combining Monte Carlo tree search method to carry out traffic priority optimization

And (3) aiming at the traffic priority optimizing problem in the step (1.2), converting the traffic priority optimizing problem into a tree searching problem, and solving by utilizing a Monte Carlo tree searching method and a set selection strategy. The character string composed of the vehicle number ID is developed as each node. Wherein, the top layer has the highest priority and the bottom layer has the lowest priority in the expanded tree structure. For example, in the node "2-3-1", the priority of the vehicle number 2 is the highest, it will pass through the bottleneck area first, and when it collides with other vehicles, the other vehicles will avoid. Fig. 4 shows a schematic diagram of tree structure search, which includes the following specific processes:

3.1 the ID assignment is carried out on the vehicles in the control area (the vehicles are numbered according to the time sequence of the vehicles entering the control area, the ID of the vehicle entering the control area at the earliest is 1; if the entering time is the same, the ID assignment is carried out according to the left-to-right sequence of the lanes where the vehicles are located in the advancing direction).

3.2, each vehicle is taken as a node, and the data information of the node comprises a character string S consisting of vehicle numbers ID, a father node and child node information, and comprises the number N of traversed times and an award value Q;

3.3 selection phase: and (3) introducing a starting node P (such as a root node) which needs to start searching currently, returning the best node according to a set selection strategy for expansion (step 3.4), and directly ending if the node is a leaf node (namely all the vehicle numbers ID are in the character string S).

In the scheme, the selection strategy is to find the sub-nodes which are not traversed at present, and if a plurality of sub-nodes exist, the sub-nodes are randomly selected; if all the UCB values are traversed, the UCB value is found to be the largest, and if the UCB values are equal, the UCB values are randomly selected. The UCB value for node v' is calculated as follows:

wherein, v' tableIndicating the current node, v the parent of v', Q the accumulated reward value for that node, N the number of visits the node,is a weight parameter; the first half part of the formula is used for utilization, and the current optimal child node is encouraged to be selected; the second half is used for exploration, encouraging the exploration of more child nodes.

3.4 expansion phase: inputting the node given in 3.3, and expanding a new node on the node as a child node thereof, wherein the character string S of the new node is formed by adding a new vehicle number ID at the tail of the character string of a parent node thereof; and finally, the newly expanded node is transmitted to the step 3.5.

The process needs to ensure that vehicles on the same lane cannot appear in the newly formed character string S, and the priority of the front vehicle is lower than that of the rear vehicle (the number of the sub-nodes expanded by the nodes can be limited by the limiting condition); the character string S between them cannot be repeated for the child nodes of the same parent node.

3.5 simulation phase: and (4) evaluating the reward value according to the new node given in the step (3.4), wherein the specific process is as follows:

3.5.1, acquiring a character string S of the node according to the given new node, if the S contains all vehicle number IDs (namely, a complete traffic priority order Ψ), directly skipping the step 3.5.2, otherwise, adding the vehicle number IDs which are not added to the S, and further obtaining the complete traffic priority order Ψ:

in the step, the character string S is used as input, and a complete pass priority sequence is output, and the specific steps are as follows:

(1) firstly, dividing the vehicle number into three arrays P according to the lanes₁、P₂、P₃The elements in the array are sorted according to the longitudinal distance from the conflict area (from small to large);

(2) will P₁、P₂、P₃The elements in the string S that have already appeared in the string S are removed, and if the empty array appears, the elements are also removed; randomly selecting an array P from the stored arrays_iAnd taking out the first element P thereof_i(1) Adding the element to the end of S;

(3) repeating the steps (1) - (2) until all the vehicle IDs are contained in S; finally, obtaining a complete traffic priority order psi;

3.5.2, the complete traffic priority order Ψ given in the step 3.5.1 is transmitted to the vehicle trajectory planning in the step 4, and finally, a reward value (i.e. the reward value q of the new node) is obtained.

3.6 backtracking stage: according to the reward value Q obtained in step 3.5, an upward backtracking is performed, and the data information (the reward value Q and the traversed times N) of the new node and the upstream parent node thereof is updated, that is, Q is Q + Q, and N is N + 1.

3.7 circularly executing the steps 3.3-3.6 within a limited time, then selecting the optimal child node of the node P as the circular starting node of the next round according to the continuously updated data information and the selection strategy in the step 3.3, and repeating the steps 3.3-3.7 until the character string S in the circular starting node contains all the vehicle numbers ID, so that the character string S is the optimal candidate traffic priority sequence.

And 4, planning the vehicle track according to the candidate passage priority sequence to ensure that the vehicle can drive from the initialization state to the state corresponding to the candidate passage priority sequence.

In the step, the vehicle in the control area is planned by using the candidate traffic priority order Ψ given in the step 3: firstly, the track planning of a single vehicle is solved, the track planning is expanded into a plurality of vehicles to obtain the local tracks of all vehicles, then the track planning of the whole change process is obtained through continuously iterating a local track planning algorithm, and meanwhile, the total cost value (namely the reward value Q) of all vehicles in the whole process is calculated according to the step 1.4. The method comprises the following specific steps:

4.1, constructing a matrix V by the vehicle ID in the step 3.1, carrying out difference operation on the matrix V and the traffic priority sequence matrix psi, and processing the result to make the vehicle ID corresponding to the difference value consistent with the index value thereof, finally obtaining a difference matrix D (if the difference value is greater than 0, acceleration is needed, if the difference value is equal to 0, uniform speed is indicated, and if the difference value is less than 0, deceleration is indicated), and further determining the target speed target _ V of the vehicle.

4.2, planning the vehicle track (single vehicle) according to the vehicle target speed target _ v determined in the step 4.1. The specific process is as follows:

4.2.1, the vehicle track planning method is carried out under a Frenet coordinate system. Vehicle position (s, d) is specified in Frenet coordinates based on longitudinal displacement s along a given reference line (generally defined as the center line of the roadway) arc and lateral offset d from the reference line, while track points are availableIs shown in whichRespectively a first derivative and a second derivative of s,first and second derivatives of d, respectively.

4.2.2, defining a set of vehicle trajectoriesThese trajectories, all consisting of s (t) and d (t), can be expressed as time polynomials along the Frenet axis:

wherein s (t) and d (t) represent the longitudinal displacement s and the lateral offset d, a at time t_i、c_iIs a polynomial coefficient, k is 4, r is 5, tⁱAs a time parameter, t₀As track start time, t_fIs the track end time; therefore, if a complete track set is required, only a needs to be solved_i、c_iThese polynomial coefficients can be obtained by the following specific process:

(1) in order to maintain the continuity and dynamic feasibility of the previous trajectory, the following constraints need to be made on the trajectory termination:

wherein, t_fIs the track end time; d_fThe final lateral position ignores the lateral velocity of the terminal state, so the derivative of d at the end time and the second derivative are both set to 0; v. of_fThe final longitudinal velocity.

(2) Can be initialized by determining the trackAnd track end stateAnd fitting (i.e. solving for a) using quintic and quartic polynomials over time interval T_i、c_i) And then completing the trajectory planning between the two states; in connection with the constraints given in (1), since the initial state and the end time are known, it is critical to determine d in the end state_f、v_f. Therefore, by changing these two parameters, a track set can be obtained

Wherein s is₀、d₀The vertical displacement s and the horizontal displacement d in the initial state.

(3) To obtain d_f、v_fThe value of (2) is that we set up three different termination states respectively, namely cruise, lane change and car following modes, and the specific settings are as follows:

a) in cruise mode, set d_fD remaining in the initial state₀Same, v_fSet to target speed target _ v;

b) in lane change mode, set d_fIs adjacent to the center of the lane, v_fSet to target speed target _ v;

c) in the following mode, d is set_fD remaining in the initial state₀Same, v_fSet as the front vehicle speed.

(4) Through the steps, a track set can be obtainedThe generated trajectory is shown in fig. 5.

4.2.3, set of trajectories under Frenet coordinatesThen, the coordinate system is converted into Cartesian coordinates [ x, y, theta, k, v, a ] through numerical transformation]Wherein [ x, y]For position in a cartesian coordinate system, k is curvature, θ is heading angle, v is velocity, and a is acceleration, so as to eliminate infeasible trajectories and select an optimal trajectory.

4.2.4, the dynamics constraints established in step 1.1 and the collision avoidance constraints of the vehicle in step 1.3 can be refined into hard constraints and soft constraints for eliminating the trajectories which do not meet the constraint conditions and selecting the optimal trajectoryThe method comprises the following specific steps:

a) collision avoidance constraints if the trajectories are setIn which there is a collision trajectory with another vehicleThe collision trajectory is directly excluded

b) Hard constraints to exclude trajectories that do not meet the vehicle dynamics criteria, mainly to make the curvature k, heading angle θ, velocity v and acceleration a within the control range:

in the above formula, k (t), θ (t), a (t) represent curvature k, heading angle θ, and acceleration a at time t; k is a radical of_max、θ_max、v_max、a_maxRepresents the maximum value of the set curvature, heading angle, speed, acceleration, v_minIndicating a set minimum speed value.

c) Soft constraints, mainly considering aspects of safety, reliability, comfort and the like, are obtained by designing a cost function and then optimizing a track set by using numerical valuesFinding out the track with the minimum cost;

wherein the content of the first and second substances,representing a set of trajectories, ω_d、ω_v、ω_aAs a weight coefficient, the first term J₀Indicating minimization of vehicle to target lane center d_desTo ensure safety; second item J_vV representing the vehicle trying to travel at the target speed_des(t) to ensure reliability; third item J_aRepresents the vehicle acceleration a (t) change to ensure comfort. d (t) and v (t) represent the lateral offset and velocity at time t.

d) By screening in steps a), b), c), the track set can be obtainedTo select the optimal track

4.3, local trajectory planning: step 4.2 is carried out on all vehicles in the control area, and then local track optimization of all vehicles can be obtained;

4.4, global trajectory planning: according to the local tracks of all vehicles obtained in the step 4.3, in order to obtain a global track, the vehicle position information needs to be continuously sampled and updated, and then the local track planning in the step 4.3 is executed again according to the updated position information of the vehicles.

4.5, circularly executing the step 4.4 until all vehicles pass through the bottleneck area; finally, a track planning scheme, total passing time and total power consumption of the whole process of the vehicle are obtained, and then the total cost value of the process is calculated according to the step 1.5;

4.6, based on the total cost value obtained in step 4.5, it is fed back to step 3.5.2 as its prize value q.

And 5, continuously carrying out loop iteration until all vehicles are contained in the nodes expanded by the tree structure, obtaining a traffic priority sequence with the minimum corresponding total cost and a corresponding vehicle track, and then controlling the vehicles in a control range by using the vehicle track to achieve the purpose of minimizing traffic time and power consumption.

And (4) by utilizing the mutual loop iteration between the step (3) and the step (4), finally obtaining the optimal traffic priority sequence and the planning scheme of the vehicle track, further obtaining the optimal solution in the step (1.5), and thus completing the solution of the cooperative driving problem in the bottleneck area of the expressway.

The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.