System and method for net-connected autonomous vehicle platoon to perform lane changing and aggregation in bottleneck area

Abstract

This invention discloses a system and method for lane-changing and clustering of connected autonomous driving fleets in bottleneck areas. It generates corresponding lane-changing and clustering schemes based on the type of vehicle ahead, calculates the lateral offset of the vehicle ahead, and utilizes a cooperative search intelligent optimization algorithm to improve the accuracy of the lateral offset. The impact values ​​of collision factors and the safety thresholds of collision impact factors are input into a collision risk assessment model to determine whether there is a collision risk, thereby controlling the connected autonomous driving fleet. Finally, based on an information matrix, the types of lagging vehicles in the target lane are discussed to complete the lane-changing and clustering process. This invention can improve traffic efficiency in bottleneck areas while ensuring driving safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent highway vehicle aggregation and traffic control technology, specifically relating to a system and method for connected autonomous driving fleets to aggregate and change lanes in bottleneck areas. Background Technology

[0002] With the continuous development of autonomous driving and vehicle-to-everything (V2X) technologies, connected and autonomous vehicles (CAVs) are increasingly appearing on the roads. For a considerable period in the future, a new situation will emerge where CAVs and traditional human-driven vehicles (HVs) coexist. Highway ramps are nodes connecting highways to other roads, and are also frequent sites of traffic congestion and accidents. Vehicles are prone to collisions near ramps, creating bottleneck areas. Currently, the methods for merging vehicles at highway ramps are only suitable for HVs. With the increasing number of CAVs, new problems that urgently need to be addressed will inevitably arise, affecting traffic efficiency and safety. The main manifestations are: (1) When CAVs face complex traffic conditions, such as when vehicles merge into the main road at highway ramps and collide with or form bottleneck areas, it is difficult for them to make timely and reasonable driving decisions like human drivers; (2) The traffic conditions at ramps will inevitably be affected by the difference between human and non-human control in CAVs and HVs, leading to increased conflicts and decreased efficiency. For example, there are significant differences between CAVs and HVs in terms of driving behavior. HV drivers may be affected by a variety of factors such as emotions, judgment, and experience, which may lead to some irregular driving behaviors and even induce accidents; however, CAVs do not have similar situations.

[0003] While significant progress has been made in the field of connected autonomous driving, numerous key challenges remain. The field has evolved from initially studying single connected autonomous vehicles to exploring multi-vehicle cooperative operation, achieving some success in areas such as platooning and autonomous lane changing. However, in practical applications, particularly in ensuring safe lane changing and merging on complex and dynamic highways, many problems remain to be solved. Efficient cooperative operation of connected autonomous driving platoons is crucial for ensuring safe distances between vehicles and smooth lane changing operations. In this area, in-depth research is needed on how to achieve platoon coordination to ensure sufficient safe distances between vehicles and to develop effective lane-changing strategies. Furthermore, achieving effective platooning, especially in highway bottleneck areas, is a pressing issue. This requires considering the dynamic changes in traffic flow to achieve coordinated vehicle operation, thereby improving overall traffic efficiency. However, achieving this goal requires overcoming many technical and safety challenges, further in-depth research, and the search for innovative solutions.

[0004] Therefore, designing a reasonable safe lane-changing and clustering method for connected autonomous vehicle (CAV) fleets in bottleneck areas of highways to improve road traffic efficiency and safety is an urgent problem to be solved. Currently, on the one hand, there is a lack of guiding theories and technical methods for CAV mixed-traffic control; on the other hand, there is a lack of relevant guiding theories and technical methods for combining micro-level vehicle collision avoidance with macro-level vehicle clustering. This naturally leads to many new technological demands. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a system and method for connected autonomous driving fleets to perform lane-changing and aggregation in bottleneck areas, thereby improving traffic efficiency in bottleneck areas and ensuring driving safety.

[0006] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0007] A method for connected autonomous driving fleets to cluster together in bottleneck areas:

[0008] (1) The vehicle in front of the connected autonomous driving convoy is a traditional manually driven vehicle.

[0009] ① The connected autonomous driving fleet forms a fleet-level following action with the vehicle in front based on CACC following action, and ensures the safety of the connected autonomous driving fleet following the vehicle in front based on the actual minimum safe distance between them;

[0010] ② Determine whether a collision will occur when the vehicle ahead changes lanes, and control the connected autonomous driving fleet.

[0011] Calculate the lateral offset d of the vehicle ahead and use a collaborative search intelligent optimization algorithm to improve the accuracy of d;

[0012] The impact values ​​of the collision factor and the safety threshold of the collision impact factor are input into the collision risk assessment model to determine whether there is a collision risk. If there is a collision risk, the speed, acceleration, and lateral and longitudinal distances of each vehicle in the convoy are controlled to avoid the collision risk. If there is no collision risk, the safe driving state is maintained based on ACC.

[0013] The influence value of the collision factor is:

[0014] When η∈[0,1 / 2), then:

[0015] When η∈[1 / 2,1], then:

[0016] Where: η is the lane change incomplete rate. This represents the numerical values ​​of the four collision impact factors. Let g(v,x), h(v,x), m(v,x), and n(v,x) be the constraint functions for the bidirectional collision factors, respectively. Let g(v,x), h(v,x), m(v,x), and n(v,x) be the quantization functions of the collision influence factors with respect to v and x. Let G(·) be the function obtained by fitting g(v,x) to η, H(·) be the function obtained by fitting h(v,x) to η, M(·) be the function obtained by fitting m(v,x) to η, and N(·) be the function obtained by fitting n(v,x) to η.

[0017] The safety threshold for the collision impact factor is:

[0018] When η∈[0,1 / 2), the safety threshold for the collision impact factor is: g=0, h=0. n∈[0,1-m];

[0019] When η∈[1 / 2,1], the safety threshold for the collision impact factor is: g∈(0,g1(2η-1)], h∈(0,h1(2η-1)]. n∈[0,1-mgh];

[0020] Where h1, g1, m1, and n1 are the initial values ​​of the collision influence factors h, g, m, and n, respectively, and h1 + g1 + m1 + n1 = 1;

[0021] ③ Based on the information matrix, discuss the types of vehicles lagging behind in the target lane and complete lane-changing aggregation;

[0022] (2) The vehicle in front of the connected autonomous driving convoy is a connected autonomous driving vehicle.

[0023] ① The connected autonomous driving fleet, based on CACC (Computer Acceleration and Adaptive Cruise Control), forms a fleet-level following driving system with the vehicle in front.

[0024] ② Determine whether a collision will occur when the vehicle ahead changes lanes, and control the connected autonomous driving fleet.

[0025] Calculate the lateral offset d of the vehicle ahead and use a collaborative search intelligent optimization algorithm to improve the accuracy of d;

[0026] The collision factor impact value and the safety threshold of the collision impact factor are input into the collision risk assessment model to determine whether there is a collision risk. If there is a collision risk, the speed, acceleration, and lateral and longitudinal distances of each vehicle in the convoy are controlled to avoid the collision risk. If there is no collision risk, the safe driving state is maintained based on ACC.

[0027] The impact value of the collision factor is:

[0028] When η∈[0,1 / 2), we have:

[0029] When η∈[1 / 2,1], we have:

[0030] in: The values ​​represent the impact values ​​of the four collision impact factors. a(v,x), b(v,x), k(v,x), and e(v,x) are the quantification functions of the collision impact factors with respect to v and x. A(·) is the function obtained by fitting a(v,x) to η, B(·) is the function obtained by fitting b(v,x) to η, E(·) is the function obtained by fitting e(v,x) to η, and K(·) is the function obtained by fitting k(v,x) to η.

[0031] The safety threshold for the collision impact factor is:

[0032] When η∈[0,1 / 2), the safety threshold for the collision impact factor is: b = 0, e∈[0,1-abk];

[0033] When η∈[1 / 2,1], the safety threshold for the collision impact factor is: b∈[0,b1(2η-1)], e∈[0,1-abk];

[0034] Where: a1, b1, k1, e1 are the initial values ​​of collision influence factors a, b, k, e respectively, and a1+b1+k1+e1=1;

[0035] ③ Based on the information matrix, discuss the types of vehicles lagging behind in the target lane and complete lane-changing aggregation.

[0036] Furthermore, based on the actual minimum safe distance between them, the safety of the connected autonomous driving fleet in following the vehicle in front is ensured, specifically as follows:

[0037] like The convoy can then continue following the vehicle in front until it changes lanes;

[0038] like The convoy will then coordinate to slow down and send advance warning signals to vehicles behind to maintain a safe distance in order to ensure driving safety.

[0039] Wherein, min(X) HV -X C1 () indicates the actual minimum safe distance. This represents the ideal minimum safe distance.

[0040] Furthermore, the accuracy of d is improved by utilizing a collaborative search intelligent optimization algorithm, specifically as follows:

[0041] 1) A collaborative team consisting of a cloud control platform, multiple roadside units, and multiple vehicle-mounted units serves as the initial population. All members participate through a specific method. Randomly generate M individuals from the initial population, each ∈ [1, I]; where i ∈ [1, I], j ∈ [1, J], k = 1, and i is the number of solutions in the current population. Let be the j-th position of the i-th individual in the k-th iteration, ε(·) be a function of uniformly distributed random numbers, I represent the total number of individuals, and J represent the total number of positions. x j It represents the lower limit of a certain dimension. Indicates the upper limit of a certain dimension;

[0042] 2) Let the cloud control platform, roadside unit group, and vehicle-mounted unit group be the "Chairman," "Board of Directors," and "Supervisory Board," respectively, within the team. Team communication includes: "Chairman's Knowledge A," "Board of Directors' Collective Knowledge B," and "Supervisory Board's Collective Knowledge C." K represents the number of iteration steps; all members of the roadside unit group and the vehicle-mounted unit group are assigned the same position when calculating B and C, and thus:

[0043]

[0044]

[0045]

[0046] in, It is the j-th value of the ith individual in the (k+1)-th iteration; It represents the j-th value of the optimal solution for the i-th individual after the k-th iteration; This refers to knowledge acquired from a cloud control platform randomly selected from a pool of external elites. These represent the average knowledge gained from discovering M global optima and i individual optima to date; α and β are the adjustment... The learning coefficient that influences the degree of influence; This represents the j-th value in the global optimal solution obtained by the m-th individual in the k-th iteration. This represents the j-th value in the globally optimal solution obtained through the cloud control platform in the k-th iteration;

[0047] 3) The onboard unit gains new knowledge by summarizing its own experience:

[0048]

[0049]

[0050]

[0051]

[0052] in: c represents the value of the new knowledge acquired by the on-board unit in the k-th iteration. j This represents the threshold value for determining when the onboard unit acquires new knowledge. This represents the value of new knowledge summarized by the on-board unit based on its own experience. This represents the value of new knowledge obtained by the vehicle unit from the cloud control platform;

[0053] 4) Extract the value from the team that is closest to the measured value of the lateral offset d during movement, as follows:

[0054]

[0055] Where: F(·) is the fitness value of the solution, derived from... have x j It is the j-th value in the solution x. It is the penalty coefficient for the e-th inequality constraint. It is the penalty coefficient for the f-th inequality constraint, E represents the number of inequality constraints, F represents the number of equality constraints, and g is the penalty coefficient for the f-th inequality constraint. e (x) represents a function related to the e-th inequality constraint, h f (x) represents the function associated with the f-th equality constraint;

[0056] 5) Obtain the optimal lateral offset of the motion.

[0057] Furthermore, based on the information matrix, the types of vehicles lagging behind in the target lane are discussed to complete lane-change aggregation, specifically as follows:

[0058] If the lagging vehicles in the target lane for lane-changing aggregation are manually driven vehicles, and the connected autonomous vehicles in the original lane meet the CAV lane-changing aggregation conditions, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal; the lane-changing aggregation conditions are:

[0059]

[0060] in: This indicates the position of the following vehicle in the adjacent lane at time t. This represents the speed of the following vehicle in the adjacent lane at time t+1. This indicates the position of the vehicle behind in the same lane at time t+1. This indicates the position of the vehicle in front in the adjacent lane of the adjacent lane at time t+1;

[0061] If the lagging vehicles in the target lane for lane-changing aggregation are connected autonomous vehicles, and the connected autonomous vehicles in the original lane meet the conditions for CAV lane-changing aggregation, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal; the conditions for lane-changing aggregation are:

[0062]

[0063] in, This represents the position of the vehicle in the current lane at time t. This represents the speed of the vehicle in the current lane at time t+1. This indicates the position of the vehicle in the adjacent lane at time t+1. G represents the position of the vehicle in the current lane at time t+1. safe This is the minimum safety parameter.

[0064] Furthermore, the lane change incomplete rate η is: the ratio of the difference between the total lateral deviation D of the vehicle ahead and its lateral offset d to the total lateral deviation D.

[0065] Furthermore, when the vehicle ahead is a traditional manually driven vehicle, the information matrix includes a main scene information matrix based on the center line and lane lines of the two lanes. And the sub-scene information matrix formed by dividing the main scene into sub-scenes. It also updates the sub-scene information matrix in real time, providing vehicle position changes and motion status for lane-changing and aggregation of connected autonomous driving fleets.

[0066] Furthermore, when the vehicle ahead is a connected autonomous vehicle, assuming that every factor in the scene is within the perception range of the CAV-Agent, the CAV-Agent acquires vehicle and road information in the scene. Using the lane centerline and lane lines as references, a main scene information matrix C is constructed, dividing the scene into sub-scene regions. Combining the information acquired by the CAV-Agent, a sub-scene information matrix is ​​constructed. Take the single vehicle information set of the original lane connected autonomous driving fleet from the sub-scene information matrix, input it into the sub-scene information data processing set, filter the vehicle information that meets the sub-scene aggregation conditions, input it into the main information data processing set, and finally output the connected autonomous driving vehicles that meet the lane-changing aggregation conditions.

[0067] Furthermore, the collision impact factors include: the two-way collision impact factors between the vehicle in front and the connected autonomous driving fleet behind it, the one-way collision impact factors between the vehicle in front and the lane change in the bottleneck area ahead, the two-way collision impact factors of the connected autonomous driving fleet in the target lane, and the two-way collision impact factors of the vehicle in front in the target lane.

[0068] A system for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area includes:

[0069] The information acquisition module acquires vehicle type, vehicle motion status, vehicle location information, and road safety information.

[0070] The data processing module filters, analyzes, and processes the acquired information to determine the collision impact factors and their real-time values. It then compares the real-time values ​​of the collision impact factors with the safety threshold to generate a pre-formed collision avoidance decision.

[0071] The instruction generation module generates commands for following, collision avoidance, and convergence.

[0072] The execution module is used to issue production instructions.

[0073] The beneficial effects of this invention are as follows:

[0074] (1) This invention addresses the impact of the type of preceding vehicle (connected autonomous vehicle or manually driven vehicle) on the following and lane changing of connected autonomous vehicle fleets, and can generate corresponding solutions. During the following process, the connected autonomous vehicle fleet will choose different following models due to the different types of preceding vehicles. During the lane changing process, the connected autonomous vehicle fleet will select the best connected autonomous vehicles for lane changing from within the fleet according to the different types of preceding vehicles.

[0075] (2) The present invention optimizes the error in measuring the lateral offset d of the vehicle's movement by using a collaborative search intelligent optimization algorithm. When the cloud control platform, roadside unit, and vehicle unit form an intelligent team to work together, the team repeatedly competes to select the value of d that is closest to the actual value, which can ensure the reliability of the invention.

[0076] (3) This invention can reduce the collision risk of connected autonomous driving fleets during lane-changing and clustering; by the target vehicle's perception of the surrounding environment, including other vehicles and bottleneck areas, a unidirectional (bidirectional) collision impact factor quantification function f(x,v) and a lane-changing incomplete rate η are introduced. As the lane-changing process progresses, the real-time collision impact weight is obtained, and a collision risk assessment model is used. and safety threshold limits The system outputs whether the vehicle is at risk of collision. At the same time, the particle swarm optimization algorithm is used to address the impact of the interaction between the dynamic driving potential field and proposes a constraint function E(x,v), which can make the real-time collision impact value output more accurate, so as to achieve the expected collision avoidance goal.

[0077] (4) By constructing a main scene information matrix, the present invention can acquire vehicle information (motion state, vehicle type); by dividing the sub-scene information matrix, and the sub-scene information matrix is ​​updated in real time as the target element moves, the safety of lane changing and gathering of connected autonomous driving fleets can be guaranteed. Attached Figure Description

[0078] Figure 1 This is a system framework diagram of the connected autonomous driving fleet described in this invention for lane-changing aggregation in bottleneck areas;

[0079] Figure 2 The CAV aggregation flowchart described in the invention;

[0080] Figure 3 The invention provides a mixed traffic flow diagram for a two-lane highway with a bottleneck area.

[0081] Figure 4 This is a schematic diagram of the ACC model structure described in the invention;

[0082] Figure 5 The invention provides a diagram illustrating the HV driving time division.

[0083] Figure 6 A diagram showing the lateral displacement during lane changing in the HV / CAV configuration described in the invention;

[0084] Figure 7 A schematic diagram illustrating the collision impact factors generated by the HV lane change described in the invention;

[0085] Figure 8 A schematic diagram illustrating the collision impact factors generated by lane-changing aggregation of the CAV(C0) as described in the invention;

[0086] Figure 9 The invention provides a screening chart for CAVs that meet the lane-changing conditions. Detailed Implementation

[0087] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0088] This invention provides a system and method for connected autonomous driving fleets to perform lane-changing aggregation in bottleneck areas, adaptable to applications such as... Figure 3 The scenario shown is a two-lane highway with a bottleneck area. This scenario is frequently encountered on highways and is typical. The traffic flow depicted is a mixed flow of connected autonomous vehicles and traditionally driven vehicles. The scenario is further divided into an upstream area and a lane-changing convergence area to facilitate the implementation of this invention.

[0089] like Figure 1As shown, a system for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area comprises an information acquisition module, a data processing module, an instruction generation module, and an execution module. Information is passed step-by-step from the bottom-level information acquisition module to the top-level execution module. First, the information acquisition module, based on the vehicle-road cooperative environment, uses GPS for precise vehicle positioning; it also uses sensors to perceive road conditions and acquire environmental information; the onboard unit transmits vehicle-to-vehicle information through a communication unit; the roadside unit monitors the vehicle's driving status and interacts with the onboard unit; the cloud control platform interacts with the roadside unit and the onboard unit, and can also provide reliable information for the onboard unit's decision-making to a certain extent. Secondly, the data processing module is divided into three sub-modules: an information collection sub-module, a data processing sub-module, and a pre-decision-making sub-module. The information collection sub-module filters the acquired information (vehicle type, vehicle motion status, vehicle location information, and road safety information). The data processing sub-module analyzes and processes the filtered information to determine the collision impact factors and their real-time values, enabling collision risk assessment and proactive risk reduction. The pre-decision-making sub-module compares the real-time values ​​of the collision impact factors with safety thresholds to pre-generate collision avoidance decisions. Next, the information processed by the data processing module is input to the instruction generation module, which generates corresponding follow-along, collision avoidance, and convergence instructions. Finally, the execution module executes the instructions issued by the instruction generation module.

[0090] Figure 2 The following is a flowchart of a method for lane-changing and clustering of a connected autonomous vehicle fleet in a bottleneck area according to the present invention. The method is explained in detail below under two scenarios: (1) the vehicle in front of the connected autonomous vehicle fleet is a traditional manually driven vehicle (HV); (2) the vehicle in front of the connected autonomous vehicle fleet is a connected autonomous vehicle (CAV). For scenarios (1) and (2): all vehicles are instructed to travel along the center line of the lane; the connected autonomous vehicle fleet is located at... Figure 3 For ease of explanation, the lanes where the bottleneck area is located are labeled C1 and C2 for the front and rear vehicles in the convoy along the direction of travel, respectively. In actual application, the number of vehicles in the convoy is not necessarily 2.

[0091] 1. The vehicle ahead is manually driven.

[0092] Step (1), the process of the convoy following the HV in front.

[0093] Step (1.1): Set C1 and C2 to form the longitudinal safety distance when the convoy is driving.

[0094] Let the model difference between CAVs be parameter β( (As the minimum safe distance), the longitudinal safe distances of C1 and C2 can be expressed as D. safe =D default +T θ ·v+β, where D default T represents the default safe distance. θ This represents the minimum safe headway between C1 and C2, where v is the cooperative speed of C1 and C2 (v = v). C1 =v C2 ), where β represents the difference in body length between C1 and C2.

[0095] Step (1.2): The (C1, C2) convoy follows the vehicle ahead using CACC (Cooperative Adaptive Cruise Control) to form a (C2-C1)-HV convoy-level following system.

[0096] In a connected autonomous driving fleet, C1 and C2 vehicles follow the driver's path based on CACC (Car Accelerator and Cruise Control). The CACC model expression is as follows:

[0097] e = d′-s0-lt c v

[0098]

[0099] Where: v prev Let be the vehicle's speed at the previous moment, and e be the error between the actual and expected vehicle spacing. The differential of the vehicle spacing error; l is the vehicle length, t c For the desired workshop time interval, k p k d d′ represents the distance between the front of the vehicle and s0 represents the minimum safe distance.

[0100] The distance between the merged vehicle and the vehicle in front (d) C1-C2 ) and its own speed (v) C2 ), acceleration (a C2 Through the CACC model, the following speed v of C2 relative to C1 is output, achieving a state of cooperative driving between C2 and C1, and ultimately forming a (C2-C1)-HV (hybrid vehicle) following state. Simultaneously, the current vehicle changes from CAV (Cooperative Availability) to HV (Hybrid Availability), and the following vehicle degrades from CACC-based cooperative driving to ACC (Adaptive Cruise Control)-based driving. Figure 4 As shown, based on the ACC model, the acceleration and velocity of C1 relative to the target HV are output, so that (C2-C1) can achieve a stable following state with the target HV.

[0101] Step (1.3): To ensure the safety of (C2-C1) following the target HV, calculate the distance between the convoy and the target HV.

[0102] The distance traveled by the convoy is based on the distance traveled by C1, let the distance traveled by C1 be X. C1 The target HV's travel distance is X. HV (Obtained from roadside unit), and X HV >X C1 Meanwhile, the minimum safe following distance is related to driving speed. At high speeds (v ≥ 100 km / h), the minimum safe following distance is 100 meters; at medium speeds (50 km / h ≤ v ≤ 100 km / h), the minimum safe following distance is 50 meters; at low speeds (20 km / h ≤ v ≤ 50 km / h), the minimum safe following distance is 20 meters; and at very low speeds (0 km / h ≤ v ≤ 200 km / h), the minimum safe following distance is 10 meters. Let the minimum safe following distance be... Considering the complexity of factors affecting vehicle movement, and assuming the error influence distance is... Therefore, the ideal minimum safe distance is The safe distance between the connected autonomous driving fleet and the HV during the following process should be greater than or equal to 1. Its expression is

[0103] Meanwhile, let the optimal speed for coordinated travel be Vs; therefore, the travel distance of C1 is expressed as:

[0104]

[0105] In the formula: V0 represents the speed at which C1 just enters the upstream region, T C1 T1 is the travel time of the lead vehicle C1, T2 is the moment when the target vehicle HV enters the lane-changing convergence area, and T2 is the moment when the HV begins lane changing. The coordinated acceleration a of the convoy is given. CAV =a C1 =a C2 The time division diagram for T1 and T2 is as follows: Figure 5 As shown.

[0106] The information acquisition module obtains the target HV and C1 position information, and the data processing module calculates the actual minimum safe distance min(X) between them. HV -X C1 ), and compare it with the set ideal minimum safe distance. Comparison:

[0107] a. If The convoy can then continue following the target HV until it changes lanes;

[0108] b. If Therefore, it is necessary to make corresponding decisions (such as coordinated deceleration of the convoy and issuing early warning signals to vehicles behind to maintain a safe distance) to ensure driving safety.

[0109] Step (2) Detect and avoid collisions that may occur to surrounding vehicles when the HV ahead of the convoy changes lanes.

[0110] Step (2.1) involves estimating the lateral offset d of the target HV's motion and improving measurement accuracy.

[0111] Assuming the vehicle travels along the center line of the lane, the total lateral offset of the target HV during the lane change process is the distance D between the center lines of the two lanes. When the target HV changes from the center line of the left lane to the center line of the right lane, the lateral offset of the vehicle's movement is d (d≤D).

[0112] (1) Measurement method for d: Using millimeter-wave radar (a sensor in the information acquisition module) to measure the lateral distance between the vehicle and the target position.

[0113] 1) Set the center line of the target lane as the endpoint;

[0114] 2) Millimeter-wave radar transmits Chrip signals toward the target endpoint;

[0115] 3) After time Tc (delay time), the radar receives the signal reflected back from the target, with a frequency difference of S. Tc ;

[0116] 4) Obtain the frequency f0 of the intermediate frequency signal, f0 = S Tc ;

[0117] 5) The lateral distance (R) between the target and the radar can be obtained from the frequency of the intermediate frequency signal, according to the formula: f0=S·Tc=S·2R / c, R=f0c / 2S (c is the speed of light, S is the slope of the Chrip signal);

[0118] 6) Calculate the lateral offset d of the vehicle's movement, d = DR.

[0119] (2) Improve the accuracy of d measurement: Collaborative search intelligent optimization algorithm

[0120] 1) A collaborative team consisting of a cloud control platform, multiple roadside units, and multiple vehicle-mounted units serves as the initial population. All members participate through a specific method. Randomly generated individuals are selected from the initial population, M ∈ [1, I] individuals, corresponding to the cloud control platform, roadside unit, and vehicle-mounted unit in this invention; where i is the number of solutions in the current population. Let be the j-th position of the i-th individual in the k-th iteration, ε(·) be a function of uniformly distributed random numbers, I represent the total number of individuals, and J represent the total number of positions. x j It represents the lower limit of a certain dimension. This indicates the upper limit of a certain dimension.

[0121] 2) Let the cloud control platform, roadside unit group, and vehicle-mounted unit group be the "Chairman," "Board of Directors," and "Supervisory Board," respectively, within the team. Team communication includes three parts: "Chairman's Knowledge A," "Board of Directors' Collective Knowledge B," and "Supervisory Board's Collective Knowledge C." Therefore, there is... K represents the iteration step number; the cloud control platform first selects the roadside unit currently observable for vehicle movement from the roadside unit group. When a vehicle leaves the monitoring range of the current roadside unit, that roadside unit automatically takes over from the roadside unit ahead; when a subsequent roadside unit takes over from a previous roadside unit, the latter acquires the data held by the former; all members of the roadside unit group and the vehicle-mounted unit group are assigned the same position when calculating B and C, namely:

[0122]

[0123]

[0124]

[0125] in, It is the j-th value of the ith individual in the (k+1)-th iteration; It represents the j-th value of the optimal solution for the i-th individual after the k-th iteration; This refers to knowledge acquired from a cloud control platform randomly selected from a pool of external elites. These represent the average knowledge gained from discovering M global optima and i individual optima to date; α and β are the adjustment... The learning coefficient that influences the degree of influence; This represents the j-th value in the global optimal solution obtained by the m-th individual in the k-th iteration. This represents the j-th value in the globally optimal solution obtained through the cloud control platform during the k-th iteration.

[0126] 3) In addition to acquiring knowledge from the cloud control platform, the vehicle-mounted unit can also acquire new knowledge by summarizing its own experience, as specifically expressed below:

[0127]

[0128]

[0129]

[0130]

[0131] in: c represents the value of the new knowledge acquired by the on-board unit in the k-th iteration. j This represents the threshold value for determining when the onboard unit acquires new knowledge. This represents the value of new knowledge summarized by the on-board unit based on its own experience. This represents the value of new knowledge obtained by the vehicle unit from the cloud control platform.

[0132] 4) Extract the value of d from the team that is closest to the actual value, as shown below: F(·) is the fitness value of solution x, derived from... have Where x j It is the j-th value in the solution x. It is the penalty coefficient for the e-th inequality constraint. It is the penalty coefficient for the f-th inequality constraint, E represents the number of inequality constraints, F represents the number of equality constraints, and g is the penalty coefficient for the f-th inequality constraint. e (x) represents a function related to the e-th inequality constraint, h f (x) represents the function associated with the f-th equality constraint.

[0133] 5) Finally, output the optimal lateral offset of the motion.

[0134] Step (2.2) constructs the quantification function f(x,v) for the lane change incomplete rate η and the collision impact factor.

[0135] The lane change incomplete rate η is defined as: the ratio of the difference between the total lateral displacement D of the target HV and its lateral offset d to the total lateral displacement D. η changes as d changes;

[0136] The initial quantization function corresponding to the collision impact factor is f i (x i ,v i )=e1·x i +e2·v i (e1 and e2 are weighting coefficients, set to e1 = 0.4 and e2 = 0.6), and after normalization, the collision impact factor quantification function is obtained.

[0137] Using f(x,v) and η, the collision influence factor values ​​are obtained;

[0138] Based on f(x,v) and η, construct a collision risk assessment model F and its safety threshold.

[0139] Figure 6 The diagram shows the lateral displacement during a lane change in a high-velocity (HV) or low-velocity (CAV) configuration. The HV lane change is divided into three time points: the start of the lane change, the crossing of the centerline, and the end of the lane change. A unidirectional and bidirectional collision influence factor is introduced, and the establishment and formal expression of the collision influence factors existing during the lane change process will be further explained.

[0140] Step (2.3) establishes unidirectional (bidirectional) impact factors (quantification functions), numerical calculation models of collision impact, collision risk assessment models, and safety thresholds.

[0141] Based on the impact of the target HV's speed and position changes during lane changes on the surrounding environment, we assume that there are bidirectional collision impact factors between the target HV and the connected autonomous driving vehicle fleet behind it (i.e., they influence each other), a unidirectional collision impact factor for the target HV's lane change in the bottleneck area ahead, a bidirectional collision impact factor for the connected autonomous driving vehicle fleet in the target lane, and a bidirectional collision impact factor for the HV ahead in the target lane. We denote the corresponding collision impact factors as h, g, m, and n, respectively. Figure 7 The diagram shows the impact factors of a collision caused by a lane change at target HV, and the corresponding quantization functions are h(v,x), g(v,x), m(v,x), and n(v,x).

[0142] ① Collision impact factor preprocessing:

[0143] When the target HV begins a lane change, d = 0, η = 1 (i.e., no lane change), and the initial values ​​of h, g, m, and n are h1, g1, m1, and n1, respectively, with h1 + g1 + m1 + n1 = 1. The entire lane change process of the target HV is divided into two stages: the lane change to the center line of the two lanes stage, and the lane change from the center line of the two lanes to the end of the lane change stage (if the values ​​of the following influencing factors are 0, it means there is no influence and they will not be used as real-time parameter inputs or outputs).

[0144] (1) When changing lanes to the center line of the two lanes, η changes from 1 to 1 / 2; the one-way influence factor g changes from the initial value g1 to 0, and the two-way influence factors h, m, and n change from h1 to 0, m from m1 to m2, and n from n1 to n2 respectively. Therefore, when η changes from 1 to 1 / 2, that is, when the center line of the two lanes is reached, g and h do not have an impact on the HV lane change (m2 and n2 represent the influence values ​​of important nodes, which can be solved according to the corresponding equation relationship).

[0145] (2) From the center line of the two lanes to the end of the lane change, η changes from 1 / 2 to 0, and the one-way influence factor g no longer has any influence; at the same time, HV leaves the original lane, and h also has no influence; however, as HV fully enters the target lane, it has a strong influence on the changes of the two-way influence factors m and n, which change from m2 and n2 to m3 and n3 respectively (m3 and n3 represent the influence values ​​of important nodes, which can be solved according to the corresponding equation relationship).

[0146] ②Establish safety thresholds for collision impact factors at different stages:

[0147] I. First, construct the impact values ​​of the collision influence factors for the three important nodes.

[0148] When η = 0, the safe values ​​corresponding to the collision impact factor are: g = 0, h = 0. n = 1 - m;

[0149] When η = 1 / 2, the safe values ​​corresponding to the collision impact factor are: g = 0, h = 0. n = 1 - m;

[0150] When η = 1, the safe values ​​corresponding to the collision impact factor are: g = g1(2η-1), h = h1(2η-1), n = 1 - mgh.

[0151] II. Then, construct a phased safety threshold by analyzing the impact values ​​of collision factors at three key nodes.

[0152] When η∈[0,1 / 2), the safety threshold corresponding to the collision impact factor is g=0, h=0. n∈[0,1-m];

[0153] When η∈[1 / 2, 1], the safety threshold corresponding to the collision impact factor is g∈(0, g1(2η-1)], h∈(0, h1(2η-1)]. n∈[0,1-mgh].

[0154] ③ Establish a numerical calculation model for the impact of collisions.

[0155] (1) When η∈[0,1 / 2), then:

[0156] (2) When η∈[1 / 2,1], then:

[0157] The value of η is unique and constant at any given moment; g(v,x), h(v,x), m(v,x), and n(v,x) are the quantization functions of the collision influence factor with respect to v and x; G(·) is the function obtained by fitting g(v,x) to η, H(·) is the function obtained by fitting h(v,x) to η, M(·) is the function obtained by fitting m(v,x) to η, and N(·) is the function obtained by fitting n(v,x) to η. For real-time values ​​within the range of η.

[0158] ④ Utilize particle swarm optimization algorithm to plan the optimal path and improve collision avoidance accuracy.

[0159] Step 1: C1, the connected autonomous driving fleet in the target lane, the HV ahead of the target lane, and the target HV are respectively defined by ω1, ω2, ω3, and ω4 as the optimal solutions for their respective speed and position relationships during the lane change process of the target HV;

[0160] Step 2: In each iteration, C1, the connected autonomous driving fleet in the target lane, the HV ahead of the target lane, and the target HV find a common optimal solution ω;

[0161] Step 3: Update the velocity and position of each by tracking the two optimal solutions {ω1, ω2, ω3, ω4} and ω respectively; after finding these two optimal values, use the following formula to calculate the velocity and position of each, taking the target HV as an example;

[0162] v HV ′=v HV +c1×rand()×(pbest HV -x HV )+c2×rand()×(gbest HV -x HV )

[0163] x HV ′=x HV +v HV ′

[0164] Among them, v HV It is the current velocity of the target HV, v HV ' is the velocity after the target HV is updated, rand() is a random number between (0, 1), x HV x is the current position of the target HV. HV ′ represents the updated position of the target HV, c1 and c2 are learning factors, and pbest HV gbest represents the individual's optimal location for the target HV. HV This indicates the globally optimal location of the target HV;

[0165] C1. The connected autonomous driving fleet in the target lane and the HV in front of the target lane update their speed and position in the above manner to improve the accuracy of the real-time value of the two-way collision factor.

[0166] Step 4: Let the constraint function of the particle swarm optimization algorithm on the bidirectional collision factor be... Combining the collision impact factor quantization function f(x,v) and η, we have:

[0167] When η∈[0, 1 / 2), then:

[0168] When η∈[1 / 2,1], then:

[0169] in, Constraint symbols; This is the constraint function for the bidirectional collision factor.

[0170] ⑤ Input the impact value and safety threshold, and output the results through the collision risk assessment model.

[0171] The result The input is fed into the collision risk assessment model F, and the output is used to determine whether there is a collision risk. The algorithm flow of F is as follows:

[0172]

[0173] In this algorithm: F(f) i (x,v),η) is Functional form, The threshold limits are generated in real time in the safety threshold assessment model F.

[0174] If there is a risk of collision, the vehicle's speed and acceleration are controlled, and the lateral and longitudinal distances to surrounding vehicles are controlled through vehicle-to-vehicle communication and vehicle-to-infrastructure information interaction to avoid collision risks; if there is no risk of collision, the vehicle maintains a safe driving state based on ACC.

[0175] Step (3), control of safe lane changing and aggregation of connected autonomous driving fleet

[0176] For a single CAV-Agent that meets the platooning conditions for lane changing and clustering, it can form a new platoon with neighboring CAV-Agents by changing lanes. The CAV-Agents in the original lane platoon can not only obtain vehicle information that meets the clustering conditions through vehicle-to-vehicle communication, but also identify the type of driving vehicles (CAV or HV) in the neighboring area, and convey coordinated lane-changing instructions between vehicles, thereby achieving a better clustering effect.

[0177] ① Establish an information matrix within the scene

[0178] Through information exchange between various vehicle-agents, a main scene information matrix C is formed, based on the center line and lane lines of the two lanes. (The elements in the matrix represent vehicles-Agents); then the main scene is divided into sub-scenes, with... Through real-time updates of the sub-scene information matrix: It can provide effective information such as vehicle position changes and motion status for lane-changing aggregation of connected autonomous driving fleets; in the formula, L represents the left lane and R represents the right lane. This indicates bottleneck region information (including the spatial location occupied by the bottleneck region and the reasons for its formation). This indicates important information such as C2's location and speed. This indicates important information such as the location and speed of vehicle C1. This indicates important information such as the target HV's location and speed. This indicates important information such as the position and speed of vehicles following the convoy in the target lane. These represent important information such as the position and speed of the CAVs within the target lane convoy. This represents important information such as the position and speed of the vehicle ahead in the target lane. Sub-information matrix C 子 Each element contains important information such as the vehicle's type, location, and speed.

[0179] ② Based on the information matrix, discuss the lagging vehicles in the target lane to complete lane-change aggregation.

[0180] The STCA-I lane-changing model can reduce the impact of vehicles in adjacent lanes on vehicles changing lanes. While maintaining a small braking safety distance, it uses the speed of vehicles in adjacent lanes as the safe lane-changing condition for vehicles changing lanes; and improves the lane-changing flexibility of vehicles while ensuring safety.

[0181] I. If the lagging vehicle type in the target lane for CAV lane-changing aggregation is HV, and the original lane's CAVs meet the CAV lane-changing aggregation conditions, the CAV-Agent communicates with other vehicle agents in the area to release the lane-changing aggregation signal. Based on the STCA-I lane-changing model, the lane-changing conditions must be met. It can be represented as:

[0182]

[0183] in: This indicates the position of the vehicle following in the adjacent lane at time t. This represents the speed of the following vehicle in the adjacent lane at time t+1. This indicates the position of the vehicle following in the same lane at time t+1. This indicates the position of the vehicle in front in the adjacent lane at time t+1.

[0184] II. If the lagging vehicle type in the target lane for CAV lane-changing aggregation is CAV, and the original lane's CAVs meet the conditions for CAV lane-changing aggregation, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal. Based on the STCA-I lane-changing model, the lane-changing conditions must be met. It can be represented as:

[0185]

[0186] in: This represents the position of the vehicle in the current lane at time t. This represents the speed of the vehicle in the current lane at time t+1. This indicates the position of the vehicle in front in the adjacent lane at time t+1. G represents the position of the vehicle in the current lane at time t+1. safe This is the minimum safety parameter.

[0187] The positions and velocities mentioned above are each element in the information matrix.

[0188] II. The vehicle ahead is a connected autonomous driving vehicle.

[0189] Step (1), the process of the convoy following the CAV in front.

[0190] Will Figure 3 In the bottleneck area, the vehicle ahead of the connected autonomous vehicle convoy in the lane is designated as C0 (CAV). In this case, the following behavior between the connected autonomous vehicle convoy and the preceding connected autonomous vehicle differs from the following behavior between the connected autonomous vehicle convoy and the preceding manually driven vehicle in the first part. It is entirely based on CACC (Continuous Acceleration and Adaptive Cruise Control), forming a C2-C1-C0 following pattern to achieve a stable driving state.

[0191] Step (2) Detect and avoid collisions that may occur to surrounding vehicles when the CAV in front of the convoy changes lanes.

[0192] C0 lane-change aggregation is divided into three stages: before aggregation, during aggregation, and after aggregation. The influencing factors of CAV(C0) lane-change aggregation are as follows: Figure 8 As shown.

[0193] Step (2.1) establishes unidirectional (bidirectional) impact factors (quantification functions), collision risk assessment models, and safety thresholds (the estimation and error optimization of d are consistent with Part 1).

[0194] Repeat the steps regarding collision influence factors described above; during the accumulation process, C0 is affected by a unidirectional collision influence factor b from the bottleneck region, and bidirectional collision influence factors a, k, and e, as well as their safety thresholds, exist between C1, HV, and CAV. See Part 1.

[0195] ① Establish safety threshold limits for collision impact factors at different stages.

[0196] I. First, construct the impact values ​​of the collision influence factors for the three important nodes.

[0197] The initial values ​​of a, b, k, and e are a1, b1, k1, and e1, respectively, and a1 + b1 + k1 + e1 = 1.

[0198] When η = 0, the safe values ​​corresponding to the collision impact factor are: a = 0, b = 0. e = 1 - abk;

[0199] When η = 1 / 2, the safe value corresponding to the collision impact factor is: b = b1(2η-1), n = 1 - abk;

[0200] When η = 1, the safety value corresponding to the collision impact factor is: b = b1(2η-1), n = 1 - abk.

[0201] II. Then, construct a phased safety threshold by analyzing the impact values ​​of collision factors at three key nodes.

[0202] When η∈[0,1 / 2), the safety threshold limit corresponding to the collision impact factor is: b = 0, e∈[0,1-abk];

[0203] When η∈[1 / 2,1], the safety threshold limit corresponding to the collision influence factor is: b∈[0,b1(2η-1)], e∈[0,1-abk].

[0204] ②Establish a numerical calculation model for the impact of collisions.

[0205] (1) When η∈[0, 1 / 2), then:

[0206] (2) When η∈[1 / 2,1], then:

[0207] The value of η is unique and constant at any given moment; a(v,x), b(v,x), k(v,x), and e(v,x) are the quantization functions of the collision influence factor with respect to v and x; A(·) is the function obtained by fitting a(v,x) to η, B(·) is the function obtained by fitting b(v,x) to η, E(·) is the function obtained by fitting e(v,x) to η, and K(·) is the function obtained by fitting k(v,x) to η. For real-time values ​​within the range of η.

[0208] ③ Utilize particle swarm optimization algorithm to plan the optimal path and improve collision avoidance accuracy.

[0209] The preceding vehicle represents the handling of collision impact factors in the manually driven vehicle section. Let the constraint function of the particle swarm optimization algorithm on the bidirectional collision factors be: The function that combines the collision impact factor quantification function with the function fitted to η is:

[0210] When η∈[0,1 / 2), we have:

[0211] When η∈[1 / 2,1], we have:

[0212] ④ Input the impact values ​​and safety thresholds, and output the results through the collision risk assessment model.

[0213] The result The input is fed into the collision risk assessment model F, and the output is used to determine whether there is a collision risk. The algorithm flow of F is as follows:

[0214]

[0215]

[0216] In this algorithm: F(f) j (x,v),η) is The function form, The threshold limits are generated in real time in the safety threshold assessment model F.

[0217] If there is a risk of collision, the vehicle's speed and acceleration are controlled, and the lateral and longitudinal distances to surrounding vehicles are controlled through vehicle-to-vehicle communication and vehicle-to-infrastructure information interaction to avoid collision risks; if there is no risk of collision, the vehicle maintains a safe driving state based on ACC.

[0218] Step (3) Control of safe lane changing and aggregation of connected autonomous driving fleet

[0219] ① During the lane-changing and clustering process of a connected autonomous driving fleet, the introduced collision impact factors can be used for collision assessment, enabling active collision avoidance; the lane-changing and clustering of a connected autonomous driving fleet consists of the following steps:

[0220] Step 1: (Information Perception Module: Sensors perceive surrounding vehicles and generate corresponding information; then the communication unit realizes vehicle-to-vehicle information interaction; in addition, it also integrates information interaction between roadside units and vehicle-mounted units) Assuming that any factor in the scene is within the perception range of CAV-Agent, information such as vehicles and roads in the scene is obtained through CAV-Agents. Based on this, and using the lane centerline and lane lines as references, a main scene information matrix is ​​constructed:

[0221]

[0222] Step 2: (Data Processing Module) Divide the scene into sub-scene areas;

[0223] Step 3: Based on the sub-scene division and the information obtained by CAV-Agents, construct a sub-scene information matrix:

[0224]

[0225] Step 4: Take the single vehicle information set of the original lane connected autonomous driving fleet from the sub-scene information matrix, input it into the sub-scene information data processing set, filter the vehicle information that meets the sub-scene aggregation conditions, input it into the main information data processing set, and finally output the CAV that meets the lane change aggregation conditions. Figure 9 Screening chart for CAVs that meet lane-changing conditions;

[0226] Step 5: (Instruction Generation Module + Execution Module) By inputting the vehicle information of the lane-changing and clustering CAVs from Step 4 and the information of surrounding vehicles, a lane-changing and clustering strategy is generated, and the generated clustering strategy is implemented through the execution module.

[0227] ② Discuss vehicles lagging behind in the target lane and complete lane-changing aggregation:

[0228] I. If the lagging vehicle type in the target lane for CAV lane-changing aggregation is HV, and the original lane's CAV meets the lane-changing aggregation conditions, the CAV-Agent communicates with other vehicle agents in the area to release the lane-changing aggregation signal. Based on the STCA-I lane-changing model, the lane-changing conditions must be met. It can be represented as:

[0229]

[0230] in: This indicates the position of the following vehicle in the adjacent lane at time t. This represents the speed of the following vehicle in the adjacent lane at time t+1. This indicates the position of the vehicle behind in the same lane at time t+1. This indicates the position of the vehicle in front in the adjacent lane at time t+1.

[0231] II. If the lagging vehicle type in the target lane for CAV lane-changing aggregation is CAV, and the original lane's CAVs meet the lane-changing aggregation conditions, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal. Based on the STCA-I lane-changing rules, lane-changing conditions must be met. It can be represented as:

[0232]

[0233] in, This represents the position of the vehicle in the current lane at time t. This represents the speed of the vehicle in the current lane at time t+1. This indicates the position of the vehicle in the adjacent lane at time t+1. G represents the position of the vehicle in the current lane at time t+1. safe This is the minimum safety parameter.

[0234] Thus, the proposed connected autonomous driving vehicle lane-changing aggregation system and bottleneck area control method, which addresses the impact of the preceding vehicle type (CAV or HV) on lane-changing aggregation of connected autonomous driving fleets and on road traffic safety and efficiency, has been implemented in highway bottleneck areas.

[0235] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for a connected autonomous driving fleet to cluster in a bottleneck area, characterized in that: (1) The vehicles in front of the connected autonomous driving fleet are traditional manually driven vehicles; Within the connected autonomous driving fleet, the vehicles in front and behind along the direction of travel are designated as C1 and C2, respectively. The connected autonomous driving fleet forms a fleet-level following action with the vehicle in front based on CACC following action. The safety of the connected autonomous driving fleet following the vehicle in front is ensured based on the actual minimum safe distance between the connected autonomous driving fleet and the vehicle in front. Determine whether a collision will occur when the vehicle ahead changes lanes, and control the connected autonomous driving fleet. Calculate the lateral offset d of the vehicle ahead and use a collaborative search intelligent optimization algorithm to improve the accuracy of d; The impact values ​​of the collision impact factors and the safety threshold of the collision impact factors are input into the collision risk assessment model to determine whether there is a collision risk. If there is a collision risk, the speed, acceleration, and lateral and longitudinal distances of each vehicle in the convoy are controlled to avoid the collision risk. If there is no collision risk, the safe driving state is maintained based on ACC. The impact value of the collision impact factor is: when When ∈[0,1 / 2), then: when When ∈[1 / 2,1], then: in: To account for the incomplete lane change rate, , , , This represents the numerical values ​​of the four collision impact factors. Let be the constraint function for the bidirectional collision impact factor. , , , It is about four collision influencing factors , quantization function, yes and The fitted function, H yes and M yes and N yes and , This represents the coordinated driving speed of the vehicles in front and behind within the connected autonomous driving fleet along the driving direction; the four collision impact factors refer to the two-way collision impact factor between the traditional manual driving vehicle and the connected autonomous driving fleet behind it, the one-way collision impact factor for the lane change of the traditional manual driving vehicle in the bottleneck area ahead, the two-way collision impact factor of the connected autonomous driving fleet in the target lane, and the two-way collision impact factor of the manual driving vehicle in front in the target lane, respectively denoted as h, g, m, and n; The safety threshold for the collision impact factor is: when At that time, the safe threshold for the collision impact factor is: , , , ; when At that time, the safe threshold for the collision impact factor is: , , , ; in: , , , Let h, g, m, and n be the initial values ​​of the collision influence factors, respectively. + =1; Based on the information matrix, the types of vehicles lagging behind in the target lane are discussed to complete lane-change aggregation; (2) The vehicle in front of the connected autonomous driving fleet is a connected autonomous driving vehicle; The connected autonomous driving fleet, based on CACC (Compass Accelerator and Adaptor) driving, forms a fleet-level driving system with the vehicle in front. Determine whether a collision will occur when the vehicle ahead changes lanes, and control the connected autonomous driving fleet. Calculate the lateral offset d of the vehicle ahead and use a collaborative search intelligent optimization algorithm to improve the accuracy of d; The collision impact factor values ​​and safety thresholds are input into the collision risk assessment model to determine whether there is a collision risk. If there is a collision risk, the speed, acceleration, and lateral and longitudinal distances of each vehicle in the convoy are controlled to avoid the collision risk. If there is no collision risk, the ACC (Adaptive Cruise Control) is used to maintain a safe driving state. The impact value of the collision impact factor is: when When ∈[0,1 / 2), we have: when When ∈[1 / 2,1], we have: in: , , , This represents the numerical values ​​of the four collision impact factors. , , , It is the collision impact factor about , quantization function, yes and The fitted function, yes and The fitted function, E yes and The fitted function, yes and The fitted function; the four collision impact factors refer to the one-way collision impact factor b from the bottleneck area that C0 is affected by during the aggregation process, and the two-way collision impact factors a, k, and e that exist with C1, manually driven vehicles, and connected autonomous vehicles. C0 refers to the connected autonomous vehicle in front of the connected autonomous vehicle fleet in the lane where the bottleneck area is located. The safety threshold for the collision impact factor is: when At that time, the safe threshold for the collision impact factor is: , , , ; when At that time, the safe threshold for the collision impact factor is: , , , ; in: , , , These are the initial values ​​of the collision influence factors a, b, k, and e, respectively. + + + =1; Based on the information matrix, the types of vehicles lagging behind in the target lane are discussed to complete lane-changing aggregation.

2. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 1, characterized in that, Based on the actual minimum safe distance between the connected autonomous driving fleet and the vehicle in front, the safety of the connected autonomous driving fleet following the vehicle in front is ensured, specifically as follows: like If the vehicle changes lanes, the convoy can continue following the vehicle in front until it changes lanes. like If the convoy slows down in a coordinated manner and sends a warning signal to the vehicles behind to maintain a safe distance, then the convoy will slow down in a coordinated manner and send a warning signal to the vehicles behind to maintain a safe distance in order to ensure driving safety. in, This indicates the actual minimum safe distance. This represents the ideal minimum safe distance. The driving distance of C1 This refers to the distance traveled by a manually driven vehicle. For minimum safe distance, The distance is affected by error.

3. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 1, characterized in that, The accuracy of d is improved by using a collaborative search intelligent optimization algorithm, specifically as follows: 1) A collaborative team consisting of a cloud control platform, multiple roadside units, and multiple vehicle-mounted units serves as the initial population. All members participate through a specific method. Randomly generated, selected from the initial population. Individuals; among them , , , where i is the number of solutions in the current population. It is the j-th position of the i-th individual in the k-th iteration. It is a function of uniformly distributed random numbers, where I represents the total number of individuals and J represents the total number of locations. It represents the lower limit of a certain dimension. Indicates the upper limit of a certain dimension; 2) Let the cloud control platform, roadside unit group, and vehicle-mounted unit group be the "Chairman," "Board of Directors," and "Supervisory Board," respectively, within the team. Team communication includes: "Chairman's Knowledge A," "Board of Directors' Collective Knowledge B," and "Supervisory Board's Collective Knowledge C." , , , K represents the number of iteration steps; all members of the roadside unit group and the vehicle-mounted unit group are assigned the same position when calculating B and C, and thus: in, It is the j-th value of the ith individual in the (k+1)-th iteration; It represents the j-th value of the optimal solution for the i-th individual after the k-th iteration; This refers to knowledge acquired from a cloud control platform randomly selected from a pool of external elites. These are the average knowledge obtained from discovering M global optimal solutions and i individual optimal solutions so far; , It is an adjustment The learning coefficient that influences the degree of influence; This represents the j-th value in the global optimal solution obtained by the m-th individual in the k-th iteration. This represents the j-th value in the globally optimal solution obtained through the cloud control platform in the k-th iteration; 3) The vehicle-mounted unit gains new knowledge by summarizing its own experience: in: This represents the value of the new knowledge acquired by the on-board unit in the k-th iteration. This represents the threshold for determining when the onboard unit acquires new knowledge. This represents the value of new knowledge summarized by the on-board unit based on its own experience. This represents the value of new knowledge obtained by the vehicle unit from the cloud control platform; 4) Extract the value from the team that is closest to the measured value of the lateral offset d during movement, as follows: ， ， ， Where: F( ) is the fitness value of the solution, derived from ,have , It is the j-th value in the solution x. It is the penalty coefficient for the e-th inequality constraint. is the penalty coefficient for the f-th inequality constraint, E represents the number of inequality constraints, and F represents the number of equality constraints. This represents a function related to the e-th inequality constraint. Represents the function associated with the f-th equality constraint; 5) Obtain the optimal lateral offset of the motion.

4. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 1, characterized in that, Based on the information matrix, the types of vehicles lagging in the target lane are discussed to complete lane-change aggregation, specifically: If the lagging vehicles in the target lane for lane-changing aggregation are manually driven vehicles, and the connected autonomous vehicles in the original lane meet the CAV lane-changing aggregation conditions, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal; the lane-changing aggregation conditions are: in: express The position of the car behind in the adjacent lane at any time. express The speed of the car behind in the adjacent lane at all times. express The position of the car behind in the same lane at any time. express The position of the vehicle in front in the adjacent lane at any given time; If the lagging vehicles in the target lane for lane-changing aggregation are connected autonomous vehicles, and the connected autonomous vehicles in the original lane meet the conditions for CAV lane-changing aggregation, the CAV-Agent communicates with other vehicle agents in the area to release a lane-changing aggregation signal; the conditions for lane-changing aggregation are: in, express The current position of the vehicle in the current lane. express The speed of the vehicle in the current lane at any given time. express The position of the car in front in the adjacent lane at any time. express The current position of the vehicle in the current lane. This is the minimum safety parameter.

5. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 1, characterized in that, The lane change incomplete rate η is: the ratio of the difference between the total lateral deviation D and the lateral offset d of the vehicle ahead to the total lateral deviation D. .

6. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 1, characterized in that, When the vehicle ahead is a traditional manually driven vehicle, the information matrix includes a main scene information matrix based on the center line and lane lines of the two lanes. And the sub-scene information matrix formed by dividing the main scene into sub-scenes. The system updates the sub-scene information matrix in real time, providing vehicle position changes and motion status for lane-changing aggregation of connected autonomous driving fleets; where L represents the left lane and R represents the right lane. This indicates the position and speed of vehicle C2. This indicates the position and speed of vehicle C1. Indicates the position and speed of manually driven vehicles. Indicates bottleneck area information. This indicates the position and speed of the following vehicles in the target lane convoy. These represent the position and speed of the connected autonomous vehicles within the target lane convoy, respectively. This indicates the position and speed of manually driven vehicles in the target lane.

7. The method for lane-changing and clustering of a connected autonomous driving fleet in a bottleneck area according to claim 6, characterized in that, When the vehicle ahead is a connected autonomous vehicle, assuming that every factor in the scene is within the perception range of the CAV-Agent, the CAV-Agent acquires vehicle and road information in the scene, and constructs a main scene information matrix using the lane center line and lane lines as references. The scene is divided into sub-scene regions, and a sub-scene information matrix is ​​constructed by combining the information obtained by CAV-Agents. ; Take the single vehicle information set of the original lane connected autonomous driving fleet from the sub-scene information matrix, input it into the sub-scene information data processing set, filter the vehicle information that meets the sub-scene aggregation conditions, input it into the main information data processing set, and finally output the connected autonomous driving vehicles that meet the lane-changing aggregation conditions.

8. A system for implementing the method of lane-changing and clustering of a connected autonomous driving fleet as described in any one of claims 1-7, characterized in that, include: The information acquisition module acquires vehicle type, vehicle motion status, vehicle location information, and road safety information. The data processing module filters, analyzes, and processes the acquired information to determine the collision impact factors and their real-time values. It then compares the real-time values ​​of the collision impact factors with the safety threshold to generate a pre-formed collision avoidance decision. The instruction generation module generates commands for following, collision avoidance, and convergence.

9. The system according to claim 8, characterized in that, It also includes an execution module for issuing the generated instructions.

Images