Multi-vehicle cooperative control method and system based on dynamic spatiotemporal adaptive swarm game

Abstract

The present application relates to the field of intelligent transportation systems and automatic driving technology, in particular to a kind of dynamic space-time domain adaptive cluster game multi-vehicle cooperative control method and system, by predicting the future trajectory of HDV, ICV in the merging area is divided into dynamic space-time domain adaptive cluster, dynamic space-time domain adaptive cluster game is carried out, the optimal control sequence of each cluster in ICV is generated, the optimal control sequence of each cluster and HDV guide decision mark are obtained, so as to control ICV to execute action and guide instruction according to the optimal control sequence and HDV guide decision mark, coordination between clusters is carried out.The scheme is efficient through dynamic space-time domain adaptive cluster game cooperation, to solve the problem that HDV behavior is uncontrollable in mixed traffic flow, single-vehicle perception is limited and large-scale heterogeneous vehicle group cooperative calculation complexity is high.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation systems and autonomous driving technology, specifically to a multi-vehicle cooperative control method and system based on dynamic spatiotemporal adaptive cluster game theory. Background Technology

[0002] In intelligent connected vehicle environments, unsignalized intersections and merging zones are typical high-complexity, high-stakes scenarios faced by autonomous vehicles. In mixed traffic flows where intelligent connected vehicles (ICVs) and manually driven vehicles (HDVs) coexist, the driving behavior of HDVs is highly random and uncontrollable, posing significant challenges to ICV trajectory planning and cooperative control.

[0003] Traditional autonomous driving decision-making methods generally rely on single-vehicle perception, which suffers from limited field of vision, blind spots, and data redundancy. They fail to acquire crucial information in areas with obstructed vision, easily leading to delayed or erroneous decisions. Meanwhile, most existing multi-vehicle cooperative methods assume all vehicles are 100% integrated vehicle (ICVs), lacking accurate modeling and interactive prediction of HDV behavior. This results in ICVs adopting overly conservative strategies (such as frequent slowing down to yield) in mixed traffic flow scenarios, severely reducing the overall traffic efficiency at intersections and merging zones. Traditional methods struggle to effectively predict HDV responses to ICV behavior in scenarios with strong ICV-HDV interaction, leading to insufficient prediction accuracy and conservative or ineffective decisions. Furthermore, existing cooperative decision-making models often face the "curse of dimensionality" when dealing with large-scale heterogeneous vehicle groups, with computational complexity increasing exponentially, making it difficult to meet millisecond-level real-time requirements. In particular, effectively coordinating fully controllable ICVs and unpredictable, non-directly controllable HDVs within a unified framework, and achieving cross-cluster dynamic coordination and conflict arbitration, are pressing technical challenges in the field of mixed traffic flow cooperative control.

[0004] Therefore, there is an urgent need for a multi-vehicle cooperative control method and system based on dynamic spatiotemporal adaptive cluster game theory, which can achieve efficient cooperation through dynamic spatiotemporal adaptive cluster game theory to solve the problems of uncontrollable HDV behavior, limited perception of single vehicles, and high computational complexity of large-scale heterogeneous vehicle group cooperation in mixed traffic flow. Summary of the Invention

[0005] One of the objectives of this invention is to provide a multi-vehicle cooperative control method based on dynamic spatiotemporal adaptive cluster game theory. This method achieves efficient cooperation through dynamic spatiotemporal adaptive cluster game theory, thereby addressing the problems of uncontrollable HDV behavior, limited perception of individual vehicles, and high computational complexity in large-scale heterogeneous vehicle group cooperation in mixed traffic flows.

[0006] The basic solution provided by this invention is a multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game, comprising the following: Get An input set that contains the status information of all vehicles in the merging zone at all times; Based on the input set, construct a global view containing all vehicle states, predict the future trajectory of the HDV, and obtain a global list of vehicle states and the predicted trajectory of the HDV. Based on the global vehicle status list and the predicted trajectory of HDV, ICVs in the merging zone are divided into dynamic spatiotemporal adaptive clusters to obtain the dynamic cluster structure. Based on the dynamic cluster structure and the predicted trajectory of HDV, dynamic spatiotemporal adaptive cluster game is performed to generate the optimal control sequence for ICV in each cluster and obtain the optimal control sequence and HDV guidance decision flag for each cluster. Based on the optimal control sequence and HDV guidance decision flags, the ICVs are controlled to perform actions and guidance instructions to coordinate between clusters.

[0007] The second objective of this invention is to provide a multi-vehicle cooperative control system based on dynamic spatiotemporal adaptive cluster game theory, which achieves efficient cooperation through dynamic spatiotemporal adaptive cluster game theory to solve the problems of uncontrollable HDV behavior, limited perception of individual vehicles, and high computational complexity of large-scale heterogeneous vehicle group cooperation in mixed traffic flow.

[0008] This invention provides a second basic solution: a multi-vehicle cooperative control system for dynamic spatiotemporal adaptive cluster game, comprising: RCCU, at least one ICV, and HDV as an external traffic participant and sensing object; RCCU, used to obtain The system generates an input set containing the status information of all vehicles within the merging zone at all times. Based on the input set, it constructs a global view containing the status of all vehicles, predicts the future trajectory of the HDVs, and obtains a global vehicle status list and the predicted trajectory of the HDVs. Based on the global vehicle status list and the predicted trajectory of the HDVs, it divides the ICVs within the merging zone into dynamic spatiotemporal adaptive clusters and obtains the dynamic cluster structure. Based on the dynamic cluster structure and the predicted trajectory of the HDVs, it conducts dynamic spatiotemporal adaptive cluster game to generate the optimal control sequence for each ICV in the cluster, and obtains the optimal control sequence and HDV guidance decision flag for each cluster. Based on the optimal control sequence and HDV guidance decision flag, it controls the ICVs to execute actions and guidance instructions, and performs inter-cluster coordination.

[0009] Beneficial effects: This solution yields... The input set contains the status information of all vehicles in the merging zone at all times. It integrates multi-source perception data, eliminates blind spots in single-vehicle perception, and constructs a global view containing the status of all vehicles based on the input set. It predicts the future trajectory of HDVs, obtains a global vehicle status list and the predicted trajectory of HDVs, and thus predicts the behavior of uncontrollable HDVs and incorporates them into collaborative planning as environmental constraints. Based on the global vehicle status list and the predicted trajectory of HDV, ICVs in the merging zone are divided into dynamic spatiotemporal adaptive clusters. The dynamic cluster structure is obtained, and the concept of dynamic spatiotemporal adaptive clusters is introduced. Based on the global vehicle status list and the predicted trajectory of HDV, ICVs are dynamically divided into different cooperative clusters (rather than static clusters), and the cluster boundaries are adaptively adjusted. This effectively reduces the computational dimension of large-scale heterogeneous vehicle group collaboration and significantly improves decision-making efficiency and scalability. Based on the dynamic cluster structure and the predicted trajectory of HDVs, dynamic spatiotemporal adaptive cluster game is conducted to generate the optimal control sequence for each ICV within the cluster. The optimal control sequence and HDV guidance decision flags of each cluster are obtained. Through a multidimensional heterogeneous cluster evolution game framework, the cost is comprehensively optimized. This not only enables deep cooperation among ICVs, but more importantly, it enables interactive prediction of HDV behavior by ICVs (considering the impact of ICV behavior on HDVs) and active guidance (such as creating merging gaps through ICV behavior or VMS). Thus, under the premise of ensuring safety, the overall traffic efficiency of mixed traffic flow is effectively improved.

[0010] Based on the optimal control sequence and HDV guidance decision flags, the ICVs are controlled to execute actions and guidance instructions, and inter-cluster coordination is carried out to ensure the smooth execution of the underlying control.

[0011] In particular, in highly complex and competitive environments such as highway merging zones, this solution enables safe, efficient, and orderly passage of mixed traffic flows of intelligent connected vehicles and manually driven vehicles. It transforms the problem of large-scale heterogeneous vehicle group collaboration into a dynamic cluster game problem that can be solved efficiently, and achieves interactive prediction and proactive guidance of HDV behavior, significantly improving the efficiency and safety of mixed traffic flows.

[0012] In summary, this solution achieves efficient collaboration through dynamic spatiotemporal adaptive cluster game theory, addressing the challenges of uncontrollable HDV behavior, limited single-vehicle perception, and high computational complexity in collaborative large-scale heterogeneous vehicle groups within mixed traffic flows. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an embodiment of the multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game of the present invention; Figure 2This is a schematic diagram of the road conditions in the merging zone of a highway in an embodiment of the multi-vehicle cooperative control method of dynamic spatiotemporal adaptive cluster game of the present invention. Detailed Implementation

[0014] The following detailed description illustrates the specific implementation method: Example 1 This embodiment provides a multi-vehicle cooperative control method based on dynamic spatiotemporal adaptive cluster game theory, as shown in the attached figure. Figure 1 As shown, it includes the following: Get The input set contains the status information of all vehicles within the merging zone at all times. :

[0015] in For the ICV state set, :

[0016] in Intelligent Connected Vehicles (ICVs) The vertical position, in meters; For intelligent connected vehicles Speed, unit: m / s; For intelligent connected vehicles acceleration, unit: m / s² 2 ; For intelligent connected vehicles The driving characteristic parameters are dimensionless continuous variables that satisfy... The higher the value, the more aggressive the driving preference; conservative, standard, and aggressive driving preferences are respectively... 0.5 represents conservative, 0.5 represents standard, and 1 represents aggressive; when using historical driving data for learning, Alternatively, continuous values ​​within the interval can be taken; for ICV, The driver is primarily responsible for pre-setting the parameters in the onboard system; if not pre-set, they can be calculated based on historical vehicle operating data; HDV uses independent driving style estimation parameters. This indicates that it should not be used in conjunction with ICV preset parameters; In this embodiment, when no preset parameters are specified, the parameters can be calculated based on historical vehicle operation data. This includes: when the ICV does not have preset driving characteristic parameters, extracting features such as average headway, speed utilization, root mean square of acceleration, root mean square of jerk, and lane change frequency from historical vehicle operation data, and then calculating the results after normalization and weighted summation. :

[0017] in, For ICV The Classified driving behavior characteristics For the corresponding feature weights, satisfying and ; This means truncating the result to the interval [0,1].

[0018] This represents the number of ICVs in the ICV set.

[0019] For intelligent connected vehicles Lane markings, such as main line or ramp; In this embodiment, the above parameters are obtained by the RCCU (Roadside Cooperative Control Unit) through V2X communication.

[0020] For HDV state set, :

[0021] in Human-driven vehicles (HDVs) The vertical position, in meters; For manually driven vehicles Speed, unit: m / s; For manually driven vehicles The acceleration estimate; in this embodiment, the HDV acceleration estimate inferred by the RCCU based on the historical trajectory; For manually driven vehicles The estimated driving style; in this embodiment, the estimated HDV driving style inferred by the RCCU based on historical trajectories; For manually driven vehicles Lane markings, such as main line or ramp; Since HDVs lack the ability to directly interact with RCCUs via V2X, this embodiment acquires HDV status information through roadside sensing devices (such as high-definition cameras, millimeter-wave radar, and lidar) deployed in the merging control area. Roadside sensing devices include at least high-definition cameras and millimeter-wave radar, and may further include lidar. Data collected by each roadside sensing device is accompanied by a collection timestamp and device number. The RCCU processes the data according to a preset sensing fusion period, which in this embodiment can be 0.1 seconds.

[0022] The RCCU first transforms the camera image coordinates, millimeter-wave radar polar coordinates, and lidar point cloud coordinates to a unified road coordinate system based on pre-calibrated sensor extrinsic parameters. The road coordinate system uses longitudinal arc length coordinates along the lane centerline. And the lateral coordinates perpendicular to the lane centerline The merging conflict point is used as the common longitudinal reference position for all lanes. Camera data is used to obtain vehicle position and category through vehicle target detection, lane line recognition and coordinate mapping; millimeter-wave radar data provides target distance, azimuth angle and radial velocity; lidar data is used to obtain vehicle outline and spatial position through ground point culling and point cloud clustering.

[0023] The RCCU sets association thresholds based on vehicle position, speed, size, and lane location, and uses Hungarian matching or joint probability data association methods to match detection results from different sensors. For the first... Establish state vectors for each HDV:

[0024] The RCCU employs Kalman filtering or extended Kalman filtering based on a constant acceleration model for state prediction and measurement updates. It maintains the target number of the same vehicle through continuous periodic target association and outputs the longitudinal position of the HDV. Longitudinal velocity Longitudinal acceleration estimate and state estimation covariance. Lane markings where the HDV is located. According to the horizontal position The matching results with the lane boundaries of the high-precision map are determined.

[0025] HDV driving style estimation From the recent Online calculation of the effective trajectory length. A timeframe of 3–5 seconds can be used. The RCCU extracts features such as average headway, speed utilization, root mean square of acceleration, root mean square of jerk, and lane change frequency, and normalizes each feature to... The estimated driving style value is:

[0026] in, Determined as conservative. It was determined to be of the standard type. Determined as aggressive; subsequent models directly use continuous values. When there are insufficient valid trajectories or the estimated confidence level is below a threshold, The value is tentatively set to 0.5, and the trajectory prediction uncertainty is increased. If continuous measurement is missing for more than the preset period or the target identity cannot be maintained continuously, the trajectory segment is terminated and not input as a valid HDV state into the subsequent collaborative control model.

[0027] This represents the number of HDVs in the HDV set.

[0028] Application scenarios of this solution, such as Figure 2 As shown, this occurs in highly complex and game-theoretic environments such as merging zones on highways.

[0029] Based on the input set, construct a global view containing all vehicle states, predict the future trajectory of the HDV, and obtain the global vehicle state list and the predicted trajectory of the HDV. Specifically, based on the input set, a global view containing the states of all vehicles is constructed, and the future trajectory of the uncontrollable HDV is predicted as a dynamic environmental constraint for subsequent games; where the predicted trajectory of the HDV... The interactive trajectory prediction for HDV not only provides key dynamic environmental constraints for ICV dynamic spatiotemporal adaptive cluster game, but more importantly, it reflects the potential impact of ICV behavior on HDV, providing a basis for decision-making for subsequent active guidance strategies.

[0030] In this embodiment, the global vehicle status list The ICV states are substituted into the ICV kinematic model for trajectory planning, and the HDV states are substituted into the manually driven vehicle. The acceleration prediction equation (E-IDM model) is used to calculate the predicted acceleration; the solution is obtained iteratively (at each prediction step). Internally, the E-IDM model calculates and predicts acceleration, and then combines this with the vehicle's kinematic equations to update the state at the next moment, recursively deriving the future state of the HDV. Predicted trajectory of the step .

[0031] The specific process is as follows: A global view containing the status of all vehicles is constructed, and the RCCU obtains the current time through fusion of roadside sensing devices and V2X communication. Global vehicle status list ; The global vehicle status list Specifically, it is expressed as follows:

[0032] in, For the ICV state set, For the index set, i.e. the input set ; For HDV state set, For the index set, i.e. the input set ; Global vehicle status list The ICV state set is substituted into the ICV kinematic model for trajectory planning; The ICV kinematic model uses a discretized kinematic model for trajectory planning.

[0033] in, For control input (ICV) At the present moment (acceleration), with constraints as (Default value:) , );in For ICV In the next moment The vertical position, in meters; For ICV At the present moment The vertical position, in meters; For ICV At the present moment Speed, unit: m / s; For ICV In the next moment Speed, unit: m / s; The time step is the discretization time step, which is the time interval between two adjacent moments.

[0034] Global vehicle status list Substituting the HDV state set into the manually driven vehicle The acceleration prediction equation (E-IDM model) is used to calculate the predicted acceleration; Specifically, this embodiment uses an Enhanced Intelligent Driver Model (E-IDM) to predict the longitudinal motion of the HDV; for the first... For each HDV, the preceding constrained vehicle is determined based on its lane and the expected merging path. When the predicted trajectory of an ICV in another lane overlaps with that HDV in the merging conflict zone, the corresponding ICV is included in the model calculation as an interacting vehicle.

[0035] Define the relative speed between the HDV and the vehicle constrained in front as: The effective net spacing is Based on HDV driving style estimates The desired speed, desired headway, maximum acceleration, and comfortable deceleration are continuously adjusted. The expected dynamic spacing of HDV is:

[0036] The basic IDM acceleration is:

[0037] To reflect the impact of ICV cooperative behavior on HDV driving behavior, an interactive correction option is set:

[0038] in, In order to be compatible with HDV The set of ICVs that may have merging conflicts; and These are the estimated remaining times for the ICV and HDV to arrive at the merging conflict point, respectively. The time-normalized scale is in seconds (s). Normalized interaction weights; This is the ICV-HDV interactive correction acceleration scale, used to limit the maximum amplitude of the interactive correction term, with units of m / s². 2 ; is the normalized scale for acceleration, with units of m / s². 2 ; and These are dimensionless calibration coefficients, namely the influence coefficient of arrival time difference and the influence coefficient of ICV acceleration, respectively. For ICV In the prediction step Controlled acceleration; For ICV HDV Normalized interaction weights; due to The output range is (-1, 1), therefore When there is no interactive ICV, let The final predicted acceleration is:

[0039] Based on the predicted acceleration, the HDV state is updated using the general discrete vehicle kinematics equations within each prediction step:

[0040]

[0041] in, , For the prediction step size, the unit is seconds (s). Speed ​​limits are set for roads; non-negative speed constraints and upper and lower acceleration constraints are used to prevent vehicles from reversing or experiencing non-physical acceleration or deceleration.

[0042] The future of artificially driven vehicles Predicted trajectory of the step :

[0043] in, and Based on time Information prediction of human-driven vehicles At any moment The longitudinal position and velocity, To predict the number of time-domain steps.

[0044] Based on the global vehicle status list and the predicted trajectory of the HDV The ICVs within the merging zone are divided into dynamic spatiotemporal adaptive clusters to obtain the dynamic cluster structure. ; To address the heterogeneity and dynamism of mixed traffic flows, ICVs within the merging zone are divided into dynamic spatiotemporal adaptive clusters, significantly reducing computational dimensionality. This scheme innovatively introduces the concept of dynamic spatiotemporal adaptive clusters. By considering vehicle spatiotemporal proximity, driving characteristic similarity, and interaction with HDVs, ICVs are dynamically divided into cooperative clusters. This provides a set of participants for subsequent cluster game optimization and enables adaptive adjustment of cluster boundaries. The specific process is as follows: Define the game participants and the cluster, where the participant set is all ICVs within the merging zone: ; The dynamic cluster structure is as follows: ,satisfy and Unlike traditional static clusters, the formation and disbandment of clusters are dynamically adjusted based on real-time traffic conditions and HDV behavior predictions. The specific adjustment rules are as follows: Cluster Formation and Adaptive Adjustment Rules: The RCCU calculates heterogeneous distances between vehicles based on their spatiotemporal proximity, driving characteristic similarities, and interaction with HDVs. It then constructs a vehicle interaction graph based on a dynamic connectivity threshold, using the connected components of the graph as dynamic clusters. These parameters are updated according to real-time traffic conditions and HDV prediction results, thus achieving adaptive adjustment of cluster boundaries. To avoid direct addition of different physical quantities, the RCCU uses normalized dimensionless heterogeneous distances to define vehicle... The remaining path distance to the merging conflict point is The estimated arrival time is ,vehicle With vehicles The heterogeneous distance is:

[0045] in, The weighting coefficients are for position, speed, driving characteristics, and interaction effects. , and These are the normalized scales for position, velocity, and time, respectively; all weights are dimensionless and the sum of the weights is 1. Indicates vehicle and Indicator variables for whether there are potential interactions between them. For ICV ICV The remaining path distance to the point of convergence conflict; For ICV ICV longitudinal velocity; For ICV ICV Driving characteristic parameters; For ICV ICV The estimated remaining time to reach the merging conflict point. When the safety envelopes of the predicted trajectories of the two vehicles overlap, or the estimated time intervals for the two vehicles to occupy the merging conflict zone overlap, let... Otherwise .use As a distance term, it makes vehicles with potential interactions more likely to enter the same cluster.

[0046] The dynamic threshold is in dimensionless form:

[0047] in, Traffic density is expressed per lane, in vehicles per (km·lane). HDV penetration rate; , and All are dimensionless thresholds, namely the maximum distance threshold, the minimum distance threshold, and the basic distance threshold.

[0048] RCCU uses ICV as a graph node, when At the time in the vehicle With vehicles Establish connections between them, and use the connected components of the graph as the initial clusters, where This refers to the connection threshold for vehicles to join the same cluster; when the number of vehicles exceeds the preset maximum cluster size... When needed, connections are disconnected and split according to heterogeneous distance from largest to smallest. Single ICVs are allowed to form a single-element cluster. To prevent vehicles from frequently joining and leaving the cluster, the following settings are configured: The hysteresis threshold. and Together, they constitute the hysteresis mechanism for joining / leaving the cluster. ,in, This is the hysteresis width. Therefore, for vehicles not yet in the same cluster, only when... Establish a connection only when needed; for vehicles already in the same cluster, only when... The connection is then disconnected. This satisfies the requirement. This design can prevent Slight fluctuations around the threshold cause vehicles to frequently join and leave the cluster, thus avoiding cluster structure jitter.

[0049] When an HDV is located within a critical interaction range and its predicted trajectory conflicts with that of an ICV in both time and space, the RCCU marks the HDV as a virtual obstacle node. Virtual obstacle nodes are not added to the ICV cluster as controllable vehicles; their predicted location, vehicle size, and prediction covariance together constitute the time-varying occupancy envelope. The critical interaction range refers to the area of ​​an HDV located within the RCCU merging control area and expected to enter the merging conflict region within the current prediction time domain. For example: Additional spatial requirements may be necessary. ;in, For HDV Estimated remaining time to reach the point of conflict For MPC prediction of time domain length, This represents the remaining path distance. This is the threshold for the critical interaction space range.

[0050] The safety constraints between ICV and HDV are:

[0051] in, This represents the prediction step index within the MPC prediction time domain. This is the current control moment; Indicates based on the current time Information for future moments The prediction.

[0052] When this constraint renders the optimization problem infeasible, the RCCU prioritizes triggering a safety degradation strategy, controlling the relevant ICVs to decelerate. In the next control cycle, it reacquires the vehicle status and HDV predicted trajectory, performs cluster partitioning and trajectory optimization, that is, re-executes the predicted trajectory based on the global vehicle status list and HDVs. The ICVs within the merging zone are divided into dynamic spatiotemporal adaptive clusters to obtain the dynamic cluster structure. In the next control cycle, the complete closed loop is re-executed based on the newly acquired real-time state: state fusion, HDV trajectory prediction, heterogeneous distance calculation, updating vehicle interaction graph, updating dynamic cluster structure, and resolving MPC.

[0053] Based on dynamic cluster structure And HDV's predicted trajectory Dynamic spatiotemporal adaptive cluster game is performed to generate the optimal control sequence for each cluster's ICV and obtain the optimal control sequence for each cluster. and HDV guiding decision-making mark ; Optimal control sequences are generated for ICVs within each cluster to achieve safe, efficient, and comfortable merging. Global coordination is achieved by solving an MPC optimization problem within each cluster and treating inter-cluster conflicts as external constraints. The specific processing steps are as follows: Construct cluster feature functions (profit model) and quantify clusters. value It is defined as the difference between the total system cost when ICVs within the cluster make independent decisions and the total system cost when they make collaborative decisions. It is used to characterize the collaborative benefits obtained by ICVs within the cluster through information sharing and joint optimization, that is:

[0054] Among them, the total cost of independent decision-making The total cost function for independent decision-making by all ICVs within the cluster; the total cost of collaborative decision-making. This is the total cost function for cluster collaborative decision-making. When the cluster value... Greater than the synergistic benefit threshold (Set to 0 in this embodiment) indicates that collaborative decision-making can reduce the total cost of the cluster system, has positive collaborative benefits, and can trigger MPC collaborative optimization within the cluster; when the cluster value When the synergistic benefit threshold is not exceeded, each ICV maintains independent decision-making.

[0055] Independent decision-making means that the ICVs within the cluster do not perform joint optimization; each ICV only optimizes its own predictive time-domain control sequence. For the cluster... Any ICV ,definition:

[0056] In the independent decision-making process, the predicted trajectories of other ICVs and HDVs are taken as inputs to the given external environment, and not as inputs related to the vehicle. Simultaneously optimized decision variables. Vehicles Solving the model predictive control problem independently:

[0057] Thus, the vehicle was obtained. The optimal control sequence for independent decision-making. After all ICVs within the cluster have completed independent optimization, the total cluster cost under independent decision-making is defined as:

[0058] Therefore, the optimal cost per vehicle in the formula represents the cost of the vehicle. After adopting its independent optimal control sequence in the independent decision-making mode, the corresponding minimum cost in the predicted time domain is not simply a replacement of the cluster joint control sequence in the collaborative decision-making formula with the single-vehicle independent optimal control sequence.

[0059] During collaborative decision-making, cluster The control variables of each ICV are uniformly combined to form a joint control sequence:

[0060] Among them, the The predictive time-domain control sequence for the ICV is as follows:

[0061] The RCCU uses the control sequences of all ICVs within the cluster as joint decision variables to solve the problem.

[0062] This yields the optimal joint control sequence for cluster collaboration, and the total cluster cost under collaborative decision-making is defined as follows:

[0063] For clusters Any ICV The predicted time-domain costs consist of security costs, efficiency costs, and comfort costs. To distinguish between the two decision-making models, they are represented as follows:

[0064] as well as:

[0065] in, , and Weights for safety, efficiency, and comfort. , and Considering safety costs, efficiency costs, and comfort costs; and to ensure the comparability of the benefits of cluster collaboration, independently decide the total cost. and The total cost of collaborative decision-making should be within the same control cycle, the same initial vehicle state, and the same prediction time domain. The calculations are performed under the same reference target, the same dynamic cost weight, and the same basic dynamics and safety constraints. The main difference between the two decision-making modes is that in independent decision-making, each ICV optimizes its own control sequence separately, while in collaborative decision-making, the control sequences of all ICVs in the cluster are optimized uniformly as joint decision variables.

[0066] The security cost is:

[0067] in, For two vehicles and The distance projected from the center along the same or conflicting travel path, in meters; The distance of the safety potential field is expressed in meters. The safe distance is:

[0068] The efficiency cost is:

[0069] Comfort cost measures acceleration and jerk ( Smoothness:

[0070] in, , , and The normalized scales are time, velocity, acceleration, and jerk, respectively. After the above processing, all three types of costs are dimensionless.

[0071] This solution uses dimensionless dynamic weights to combine safety, efficiency, comfort, and cost. Define ICV driving preference mapping quantity Among them, conservative, standard, and radical ICVs correspond to respectively , and .

[0072] HDV risk indicators are defined as follows:

[0073] When the relative speeds of the two vehicles approach each other are greater than 0 Otherwise, ; The variance of HDV predicted location is expressed in meters. 2 The safety weight increases with traffic density, HDV risk, and conservative driving preference; the efficiency weight increases with aggressive driving preference and decreases with increasing safety weight, as detailed below: Define ICV driving preference mapping:

[0074] Conservative type hour Standard type hour radical hour .

[0075] Definition and ICV The biggest risks associated with HDV:

[0076] in, In order to cooperate with ICV A set of HDVs with potentially conflicting relationships.

[0077] The unnormalized weights for security are:

[0078] The efficiency unnormalized weight is:

[0079] The initial value of the comfort weight is calculated using the following formula:

[0080] when , At that time, the initial comfort weights for the conservative, standard, and aggressive ICVs were 0.26, 0.20, and 0.14, respectively.

[0081] in, , and These are the basic weights for the three types of costs; , , , , and These are dimensionless calibration coefficients.

[0082] The dynamic weights are eventually normalized to:

[0083] Since safety, efficiency, and comfort costs have been dimensionless using reference distance, reference time, reference speed, reference acceleration, and reference jerk, respectively, weight normalization is only used to adjust the relative importance of the three types of costs.

[0084]

[0085] The dynamic weights are then normalized as follows:

[0086] Thus satisfying:

[0087] in , and The unnormalized weights for safety, efficiency, and comfort, obtained according to the above dynamic adjustment rules, are summed as the normalization factor.

[0088] In other words, the safety weight is jointly adjusted by traffic density, HDV risk index, and ICV driving preference. Among these factors, the HDV risk index... The collision time-to-trajectory (TTC) risk and HDV trajectory prediction uncertainty are comprehensively characterized; when or hour, Increase the weight to improve safety and suppress efficiency weight accordingly; normalize the three weights after dynamic adjustment by summing them; and unify TTC and prediction variance into the HDV risk indicator.

[0089]

[0090] Where the relative approach speed of the two vehicles is greater than 0, we can take:

[0091] Otherwise .

[0092] Based on the cost function, RCCU targets each cluster. Solve the cooperative model predictive control problem. The control sequence of an ICV in the prediction time domain is defined as follows:

[0093] Cluster The joint control sequence is defined as:

[0094] The RCCU uses the control sequences of all ICVs within the cluster as joint decision variables to solve the problem.

[0095] in, For a collaborative optimization problem, the joint control sequence to be optimized is... This is the optimal joint control sequence obtained after optimization.

[0096] Optimization constraints include:

[0097]

[0098]

[0099]

[0100] in, The last item uses HDV to predict the location variance to set the uncertainty margin.

[0101] The aforementioned optimization constraints are: acceleration constraints, non-negativity of velocity and road speed limit constraints, jerk constraints, and safety constraints on the uncertainty margin of HDV prediction. In addition, the optimization problem should also satisfy the ICV discrete kinematics state transition equation and the HDV interactively enhanced IDM state prediction equation.

[0102] The optimization constraints also include: the discrete kinematic equations of the ICV and the interactively enhanced state prediction equations of the HDV. When the optimization problem has no feasible solution or the solution fails to converge within the specified time, the RCCU stops issuing new cooperative acceleration commands, the relevant ICVs switch to conservative and safe control mode, and the state fusion and optimization are performed again in the next control cycle.

[0103] When conflicts arise between clusters, the RCCU performs asynchronous coordination and conflict arbitration based on candidate control sequences and cost functions obtained from MPC within each cluster. Each cluster involved in the conflict acts as a participant in the coordination game, with cluster priorities... and cost weight adjustment As a policy variable.

[0104] Nash equilibrium candidate solution: Under the condition of fixing the strategies of other conflict clusters, each cluster solves for the optimal response strategy that satisfies vehicle dynamics constraints and safety constraints, and updates iteratively. When the discrete priorities of all clusters no longer change, and the change in continuous weight adjustment is less than a preset threshold, a Nash equilibrium candidate solution is obtained.

[0105] Pareto Selection: A Nash equilibrium candidate solution only indicates that no cluster can further reduce its own cost by unilaterally changing its strategy; it is not directly equivalent to a Pareto optimal solution. RCCU performs Pareto selection on equilibrium candidate solutions that satisfy all hard safety constraints. When multiple Pareto-efficient candidate solutions exist, the candidate solution with the lowest total system cost is selected as the inter-cluster coordination result. If convergence is not achieved within the preset maximum number of iterations or computation time, a first-come-first-served rule based on the expected arrival time at the merging point is used to determine the temporary priority. The total system cost is defined as: the set of conflicting clusters is... Then the total system cost is:

[0106] This can be further expanded to:

[0107] When multiple Pareto efficient candidate solutions satisfying hard safety constraints exist, the total system cost is selected. The smallest candidate solution is taken as the result of inter-cluster coordination.

[0108] For high-priority clusters with conflicts and low-priority clusters Add a merging order constraint to the MPC optimization problem of low-priority clusters:

[0109] in, This is a high-priority conflict cluster; This is a low-priority conflict cluster; The expected merge time for high-priority clusters; The expected merge time for low-priority clusters; For the safe time interval between clusters, this embodiment can be set to 1.5 seconds; This only represents the discrete time step of MPC; in this embodiment, it is taken as 0.1s.

[0110] Regarding the first For each HDV, set up two-dimensional guidance and decision-making signs:

[0111] in For a moment Regarding the first Two-dimensional guidance and decision-making signs for HDV vehicles; VMS (Variable Messaging System) guidance signs for roadside variable message signs; Guiding signs to create safe clearances for controlled acceleration and deceleration of surrounding ICVs; When the HDV is located in a critical merging area and its predicted trajectory conflicts with the ICV, it is necessary to issue suggested speeds, lane indications, or risk warnings via roadside variable message signs. When the uncertainty of HDV trajectory prediction exceeds the corresponding threshold, or when surrounding ICVs need to form a safe gap through controlled deceleration, then... This allows for the simultaneous representation of both VMS bootstrapping and ICV gap creation. Both flags are binary variables (0-1), and the two types of measures can be executed individually or simultaneously.

[0112] Based on the coordination results, the RCCU adjusts the MPC constraints or cost weights of the conflict cluster and re-solves to obtain control instructions that satisfy the safety constraints; if no feasible solution is found after re-solving, the relevant ICV enters the conservative safety control mode.

[0113] Following the Receding Horizon principle, only the first control variable of the optimal control sequence is extracted as the target acceleration command for the current control cycle:

[0114] in, For the first The ICV determines the final optimal control sequence to be used in the current control cycle. > ,but:

[0115] like ≤ ,but:

[0116] in, This indicates the extraction of the sequence belonging to the optimal joint control sequence of the cluster. The optimal control sequence for the ICV; another branch is the first... The optimal control sequence obtained by independent decision-making of each ICV.

[0117] Define ICV The output command is:

[0118] The complete control instruction set for ICVs is as follows:

[0119] in, This is the target acceleration command, in m / s². 2 ; The estimated merging time is expressed in seconds. The estimated merging time is determined based on the predicted steps of the vehicle's first arrival at the merging point in the optimal predicted trajectory; if the vehicle has not yet reached the merging point within the predicted time domain, it is extrapolated based on the remaining distance and end velocity at the end of the predicted time domain. The RCCU sends output commands to the ICV via V2X communication, and the ICV's underlying controller tracks the target acceleration; the estimated merging time is only used for priority coordination and status verification between vehicles.

[0120] Regarding the first HDV vehicles, equipped with two-dimensional guidance decision signs Both components are either 0 or 1. This indicates that lane-level or area-level recommended speeds, lane prompts, or risk warnings are being disseminated via roadside variable message signs; This refers to creating a safety gap through controlled acceleration and deceleration of surrounding ICVs. These two types of measures can be implemented individually or simultaneously.

[0121] ICV behavior guidance is achieved through normal longitudinal control; when the ICV performs controlled deceleration, the vehicle's brake lights illuminate normally according to the vehicle's own control logic, without the RCCU issuing a separate brake light control command to generate a warning signal.

[0122] Compared with existing technologies, this solution has the following advantages: This solution achieves interactive prediction and proactive guidance of HDV behavior, significantly improving the safety and efficiency of mixed traffic flow. It innovatively introduces an enhanced IDM model to realize interactive prediction of HDV behavior, considering the potential impact of ICV behavior on HDVs. Based on this, through the cooperative behavior of ICVs and VMS, it achieves proactive guidance of HDVs, rather than passive avoidance. This effectively overcomes the technical challenge of uncontrollable HDV behavior in mixed traffic flow, significantly reduces collision risk, and improves the efficiency of merging areas. This solution introduces dynamic spatiotemporal adaptive cluster game theory, which significantly reduces computational complexity and ensures real-time performance. The solution dynamically divides ICVs into heterogeneous clusters and adaptively adjusts the cluster boundaries based on real-time traffic conditions and HDV behavior predictions. This dynamic spatiotemporal adaptive cluster game theory method decomposes the global high-dimensional optimization problem into low-dimensional sub-problems and uses Nash equilibrium between clusters for asynchronous coordination and conflict arbitration. This effectively avoids the "curse of dimensionality" and exponential computational complexity caused by traditional global optimization, ensuring that the system meets millisecond-level real-time requirements in complex mixed traffic flow scenarios, improving the ability to meet preset control cycle requirements, and overcoming the technical bottleneck of large-scale collaborative computing. This solution achieves a balance between personalization and comfort for heterogeneous vehicles, enhancing the user experience: This solution innovatively introduces an adaptive weight matrix and incorporates driving characteristic parameters into the cost function. It can dynamically adjust the weights of safety, efficiency, and comfort according to the driving preferences (conservative / aggressive) of different ICVs, realizing personalized collaboration among heterogeneous vehicles. It overcomes the technical difficulty of fixed weights in traditional methods that cannot accommodate different driving styles. Under the premise of ensuring safety, it significantly improves the riding experience and driver acceptance, thereby promoting the widespread application of intelligent connected vehicles. This solution possesses robust closed-loop adaptability and engineering feasibility: It employs a 100ms rolling time-domain closed-loop feedback mechanism to ensure the system can respond in real time to unpredictable behaviors of HDVs such as sudden lane changes and emergency braking, significantly enhancing the system's robustness and adaptability to dynamic environments, and overcoming the technical difficulty of poor system stability in open traffic environments. Simultaneously, the clearly defined RCCU input / output interfaces and formula parameters make this solution highly feasible in engineering, facilitating rapid deployment and application in practical vehicle-road cooperative systems.

[0123] Example 2 This embodiment provides a multi-vehicle cooperative control system based on dynamic spatiotemporal adaptive cluster game theory, used to execute the above-mentioned multi-vehicle cooperative control method based on dynamic spatiotemporal adaptive cluster game theory, including: RCCU (Roadside Cooperative Control Unit), at least one ICV (Intelligent Connected Vehicle), and HDV as an external traffic participant and sensing object. RCCU, used to obtain The system consists of an input set containing the status information of all vehicles within the merging zone at all times; based on the input set, a global view containing the status of all vehicles is constructed to predict the future trajectory of the HDVs, obtaining a global vehicle status list and the predicted trajectory of the HDVs; based on the global vehicle status list and the predicted trajectory of the HDVs, the ICVs within the merging zone are divided into dynamic spatiotemporal adaptive clusters, obtaining the dynamic cluster structure; based on the dynamic cluster structure and the predicted trajectory of the HDVs, dynamic spatiotemporal adaptive cluster game is performed to generate the optimal control sequence for each ICV in the cluster, obtaining the optimal control sequence and HDV guidance decision flag for each cluster; based on the optimal control sequence and HDV guidance decision flag, the ICVs are controlled to execute actions and guidance instructions, and inter-cluster coordination is performed; Specifically, in this embodiment, the system includes: RCCU, roadside sensing equipment, V2X communication equipment, roadside information dissemination equipment, and at least one intelligent connected vehicle (ICV); a manually driven vehicle (HDV) serves as the external traffic participant and sensing object of the system. Roadside sensing devices are used to collect sensing data of ICV and HDV within the merging control zone; V2X communication equipment is used to receive vehicle status reports from ICVs and send control commands to ICVs; The RCCU is used to complete time synchronization, coordinate transformation, target association and state fusion of multi-source sensing data, obtain HDV state and driving style estimates, predict HDV future trajectory, divide dynamic spatiotemporal adaptive clusters, solve MPC control problems within the cluster and coordination game between clusters, and generate ICV optimal control commands and HDV guidance decision flags. ICV is used to track target acceleration commands issued by RCCU through the vehicle controller; Roadside information dissemination equipment is used to disseminate lane-level or area-level recommended speeds, lane prompts, or risk warnings based on HDV guidance decision signs.

[0124] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A multi-vehicle cooperative control method based on dynamic spatiotemporal adaptive cluster game theory, characterized in that, Includes the following: Get An input set that contains the status information of all vehicles in the merging zone at all times; Based on the input set, construct a global view containing all vehicle states, predict the future trajectory of the HDV, and obtain a global list of vehicle states and the predicted trajectory of the HDV. Based on the global vehicle status list and the predicted trajectory of HDV, ICVs in the merging zone are divided into dynamic spatiotemporal adaptive clusters to obtain the dynamic cluster structure. Based on the dynamic cluster structure and the predicted trajectory of HDV, dynamic spatiotemporal adaptive cluster game is performed to generate the optimal control sequence for ICV in each cluster and obtain the optimal control sequence and HDV guidance decision flag for each cluster. Based on the optimal control sequence and HDV guidance decision flags, the ICVs are controlled to perform actions and guidance instructions to coordinate between clusters.

2. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 1, characterized in that, The input set : in For the ICV state set, : in For ICV The longitudinal position; For ICV speed; For ICV The acceleration; For ICV Driving characteristic parameters , 0.5 represents conservative, 0.5 represents standard, and 1 represents radical. For ICV Lane markings The number of ICVs in the ICV set; For HDV state set, : in For HDV The longitudinal position; For HDV speed; For HDV The estimated acceleration value; For HDV Driving style estimates , 0.5 represents conservative, 0.5 represents standard, and 1 represents radical. For HDV Lane markings This represents the number of HDVs in the HDV set.

3. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 2, characterized in that, The process of constructing a global view containing all vehicle states based on the input set, predicting the future trajectory of the HDV, and obtaining the global vehicle state list and the predicted trajectory of the HDV includes: A global view containing the status of all vehicles is constructed, and the RCCU obtains the current time through fusion of roadside sensing devices and V2X communication. Global vehicle status list ; The global vehicle status list Specifically, it is expressed as follows: in, For the ICV state set, For the index set, i.e. the input set ; For HDV state set, For the index set, i.e. the input set ; Global vehicle status list The ICV state set is substituted into the ICV kinematic model for trajectory planning; The ICV kinematic model uses a discretized kinematic model for trajectory planning. in, To control the input, the constraints are as follows: ; For ICV In the next moment The longitudinal position; For ICV At the present moment The longitudinal position; For ICV At the present moment speed; For ICV In the next moment speed; This represents the discretization time step. Global vehicle status list Substituting the HDV state set into the manually driven vehicle The E-IDM model calculates the predicted acceleration; Based on the predicted acceleration, the HDV state is updated using the general discrete vehicle kinematics equations within each prediction step: in, , For the prediction step size, the unit is seconds (s). Speed ​​limits on roads; The future of artificially driven vehicles Predicted trajectory of the step : in, and Based on time Information prediction of human-driven vehicles At any moment The longitudinal position and velocity, To predict the number of time-domain steps.

4. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 3, characterized in that, The step of dividing the ICVs within the merging zone into dynamic spatiotemporal adaptive clusters includes: Define the game participants and the cluster, where the participant set is all ICVs within the merging zone: ; The dynamic cluster structure is as follows: ,satisfy and The formation and disbandment of the clusters are dynamically adjusted based on real-time traffic conditions and HDV behavior predictions, including: Define vehicle The remaining path distance to the point of convergence conflict is The estimated arrival time is ,vehicle With vehicles The heterogeneous distance is: in, The weighting coefficients are for position, speed, driving characteristics, and interaction effects. , and These are the normalized scales for position, velocity, and time, respectively. Indicates vehicle and Indicator variables indicating whether there are potential interactions between them. For ICV ICV The remaining path distance to the point of convergence conflict; For ICV ICV longitudinal velocity; For ICV ICV Driving characteristic parameters; For ICV ICV The estimated remaining time to reach the merging conflict point; when the safety envelopes of the predicted trajectories of the two vehicles overlap, or the estimated time intervals for the two vehicles to occupy the merging conflict zone overlap, let Otherwise ; RCCU uses ICV as a graph node, when At the time in the vehicle With vehicles Establish connections between them, using the connected components of the graph as the initial cluster; when the number of vehicles exceeds the preset maximum cluster size... At that time, the connection is disconnected and the parts are split according to the heterogeneous distance from largest to smallest; When an HDV is located within a critical interaction range and its predicted trajectory conflicts with that of an ICV in a spatiotemporal manner, the RCCU marks the HDV as a virtual obstacle node. Virtual obstacle nodes are not added to the ICV cluster as controllable vehicles, and their predicted position, vehicle size, and prediction covariance together constitute the occupancy envelope that changes over time.

5. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 4, characterized in that, The process involves performing dynamic spatiotemporal adaptive cluster game theory based on the dynamic cluster structure and the predicted trajectory of the HDV, generating optimal control sequences for the ICVs within each cluster, and obtaining the optimal control sequences and HDV guidance decision flags for each cluster, including: Construct cluster feature functions and quantify clusters Value: Among them, the total cost of independent decision-making The total cost function for independent decision-making by all ICVs within the cluster; the total cost of collaborative decision-making. This is the total cost function for cluster collaborative decision-making; For clusters Any ICV ,definition: vehicle Solving the model predictive control problem independently: After all ICVs within the cluster have completed independent optimization, the total cluster cost under independent decision-making is: During collaborative decision-making, cluster The control variables of each ICV are uniformly combined to form a joint control sequence: Among them, the The predictive time-domain control sequence for the ICV is as follows: The RCCU uses the control sequences of all ICVs within the cluster as joint decision variables to solve the problem. The total cost of the cluster under collaborative decision-making is: For clusters Any ICV The predicted time-domain costs consist of security costs, efficiency costs, and comfort costs; to distinguish between the two decision-making models, they are represented as follows: as well as: in, , and Weights for safety, efficiency, and comfort. , and For safety costs, efficiency costs, and comfort costs; Based on the cluster characteristic function, a dynamic spatiotemporal adaptive cluster game optimization problem is solved, providing a basis for cluster optimization. The ICV generates the optimal control sequence and generates the HDV guidance decision flag.

6. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 5, characterized in that, The solution to the dynamic spatiotemporal adaptive cluster game optimization problem includes: like Greater than the synergistic benefit threshold This indicates that collaborative decision-making can obtain sufficient positive collaborative benefits compared to independent decision-making, triggering MPC collaborative optimization within the cluster; like If the return is not greater than the synergistic benefit threshold, each ICV will maintain independent decision-making. When performing MPC collaborative optimization, the priority and cost weight adjustment of conflicting clusters are used as policy variables. The Nash equilibrium candidate solution is obtained through optimal response iteration, and Pareto screening is performed on the equilibrium candidate solution that satisfies hard safety constraints. For low-priority clusters, the merging order constraint is added. If the convergence is not completed within the preset time or there is no feasible solution, the first-come-first-served rule and conservative safety control strategy are adopted.

7. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 5, characterized in that, The security cost is: in, For vehicles and The distance projected from the center along the same or conflicting travel path; For the safe range of the potential field, To maintain a safe distance; The efficiency cost is: The comfort cost is: in, , , and These are the normalized scales for time, velocity, acceleration, and jerk, respectively.

8. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 7, characterized in that, The safety weight, efficiency weight, and comfort weight in the cost function are dynamically adjusted. Safety weight increases with traffic density, HDV risk, and conservative driving preference; efficiency weight increases with aggressive driving preference and decreases with increasing safety weight.

9. The multi-vehicle cooperative control method for dynamic spatiotemporal adaptive cluster game as described in claim 5, characterized in that, The process of controlling the ICV underlying execution actions and guidance instructions based on the optimal control sequence and HDV guidance decision flags to coordinate between clusters includes: Extract the target acceleration command for the current control cycle based on the optimal control sequence: And form ICV The output command is: The complete control instruction set for ICVs is as follows: in, Acceleration command for the target; For the expected merging time; The RCCU sends output commands to the ICV, and the ICV's underlying controller tracks the target acceleration; the merging time is expected to be used only for inter-vehicle priority coordination and status verification. Regarding the first HDV vehicles, equipped with two-dimensional guidance decision signs , where both components take the value of 0 or 1; This indicates that lane-level or area-level recommended speeds, lane prompts, or risk warnings are being disseminated via roadside variable message signs; This indicates that a safety gap is created through controlled acceleration and deceleration of the surrounding ICVs.

10. A multi-vehicle cooperative control system with dynamic spatiotemporal adaptive cluster game theory, characterized in that, include: RCCU, at least one ICV, and HDV as an external traffic participant and sensing object; RCCU, used to obtain The system generates an input set containing the status information of all vehicles within the merging zone at all times. Based on the input set, it constructs a global view containing the status of all vehicles, predicts the future trajectory of the HDVs, and obtains a global vehicle status list and the predicted trajectory of the HDVs. Based on the global vehicle status list and the predicted trajectory of the HDVs, it divides the ICVs within the merging zone into dynamic spatiotemporal adaptive clusters and obtains the dynamic cluster structure. Based on the dynamic cluster structure and the predicted trajectory of the HDVs, it conducts dynamic spatiotemporal adaptive cluster game to generate the optimal control sequence for each ICV in the cluster, and obtains the optimal control sequence and HDV guidance decision flag for each cluster. Based on the optimal control sequence and HDV guidance decision flag, it controls the ICVs to execute actions and guidance instructions, and performs inter-cluster coordination.

Images