CN-121999636-A - Vehicle team dynamic formation method and system based on extensible alliance game

CN121999636ACN 121999636 ACN121999636 ACN 121999636ACN-121999636-A

Abstract

The invention provides a vehicle team dynamic formation method and system based on an extensible alliance game, the method comprises the steps of constructing a vehicle single-vehicle game cost function, configuring corresponding weight coefficients, weighting and summing, constructing an extensible alliance game model based on the single-vehicle game cost function, determining a core alliance and a representative vehicle, dividing non-core alliance vehicles in the vehicle team into a front vehicle group and a rear vehicle group according to positions, uniformly representing vehicle costs in the groups by the corresponding representative vehicles, solving the extensible alliance game model to obtain a vehicle team longitudinal optimal closing time and an optimal longitudinal control sequence, constructing an external vehicle lane changing transverse reference track based on a longitudinal optimal decision result, tracking the transverse reference track by adopting a curvature self-adaptive transverse composite control method, and outputting a transverse optimal control steering angle sequence according to a driving scene switching control strategy. The invention can realize the cooperative control of dynamic formation of the motorcade, and is suitable for external vehicle merging scenes of motorcades with different scales.

Inventors

LV PEIYUAN
LIANG YONGYU
WANG JIANXING
WU PENGHUI
LI XIANGYANG
LIU SHIHAN
YANG XUEKUN

Assignees

北京航天发射技术研究所

Dates

Publication Date: 20260508
Application Date: 20260311

Claims (10)

1. The dynamic vehicle team formation method based on the extensible alliance game is characterized by comprising the following steps of: Constructing a vehicle single-vehicle game cost function, wherein the vehicle single-vehicle game cost function fuses driving safety cost, speed maintenance cost, control input cost and lane risk cost, and each cost item is configured with a corresponding weight coefficient and weighted and summed; An expandable alliance game model is built based on a single-vehicle game cost function, a core alliance and a representative vehicle are determined, non-core alliance vehicles in a vehicle team are divided into a front vehicle group and a rear vehicle group according to positions, and vehicle cost in the group is uniformly represented by the corresponding representative vehicle; solving the extensible alliance game model, obtaining a longitudinal optimal closing time and an optimal longitudinal control sequence of a fleet through a double-layer optimization method, and controlling the fleet to coordinate a longitudinal gap for closing of external vehicles; Based on a longitudinal optimal decision result, constructing an external vehicle lane change transverse reference track by adopting a B spline function; Tracking the transverse reference track by adopting a curvature self-adaptive transverse compound control method, fusing an LTV-MPC algorithm with automatically updated parameters and a fuzzy PI feedback control algorithm, outputting a transverse optimal control steering angle sequence according to a driving scene switching control strategy, realizing safe merging of external vehicles, and completing dynamic formation of a vehicle team.
2. The method according to claim 1, wherein the step of calculating the driving safety cost in constructing the vehicle single game cost function is: Acquiring actual following distance, expected following distance, front vehicle length and lane state parameters of a vehicle i in a kth time step, wherein the lane state parameters are 1 for the internal vehicles and 0 for the external vehicles of a vehicle team at the beginning, and the lane state parameters are only increased and not reduced; selecting corresponding safety coefficients according to the type of the front vehicle and the lane state parameters, and setting each safety coefficient to 0 if the actual vehicle following distance is larger than the expected vehicle following distance; Calculating the distance deviation, constructing a square term weighting model by combining the lane state parameters and the safety coefficient, and solving to obtain the running safety cost of the vehicle i in the kth time step, wherein the running safety cost of the vehicle of the fleet leader is constant at 0.
3. The method of claim 1, wherein the step of calculating the speed maintenance cost in constructing the vehicular single-car game cost function is: Acquiring the actual speed of the vehicle i in the kth time step, the global planning basic expected speed, the actual following distance, the expected following distance and the preset distance adjustment time; Calculating a distance adjustment speed item according to the deviation of the actual distance between the vehicles and the expected distance between the vehicles and the preset distance adjustment time; constructing a weight function related to the actual following distance and the expected following distance, and solving to obtain a dynamic weight coefficient; combining the dynamic weight coefficient, and carrying out weighted fusion on the distance adjustment speed item and the global planning basic expected speed to obtain the dynamic expected speed of the vehicle; The square of the deviation of the actual speed from the dynamic desired speed is calculated as the speed maintenance cost of the vehicle i at the kth time step.
4. The method according to claim 1, wherein the step of calculating the control input cost in constructing the vehicular single-car game cost function is: configuring a control input weight matrix, wherein the weight matrix is used for quantifying the influence degree of longitudinal control input in cost; Acquiring longitudinal control input of the vehicle i at a kth time step, wherein the longitudinal control input comprises at least one of acceleration, accelerator opening or brake opening; And (3) performing product operation on the longitudinal control input and the weight matrix, and squaring an operation result to obtain the control input cost of the vehicle i in the kth time step.
5. The method according to claim 1, wherein the step of calculating the lane risk cost in constructing the vehicle single-car game cost function is: Configuring a position of an expected merging point, a negative slope coefficient and a positive intercept coefficient, and acquiring a position parameter and a lane state parameter of a vehicle i at a kth time step; If the vehicle is in the internal lane of the motorcade, the lane state parameter is 1, and the lane risk cost is directly judged to be 0; If the vehicle does not finish the merging, the lane state parameter is 0, a linear cost model is built based on the expected merging point position, the vehicle position parameter, the negative slope coefficient and the positive intercept coefficient, and the lane risk cost of the vehicle i in the kth time step is obtained through solving.
6. The method according to claim 1, wherein the step of calculating the coalition game cost for constructing the extensible coalition game model based on the single-car game cost function comprises the steps of: Determining a member set of the alliance, configuring weight coefficients of each cost item for each vehicle i in the alliance, and coupling weight coefficients of other vehicles j in the alliance, wherein the sum of the weight coefficients of each cost item and the coupling weight coefficients is 1; Solving four single vehicle costs of driving safety, speed maintenance, control input and lane risk of each vehicle in the alliance; Summing the coupling weighted values of the vehicle cost and the other vehicle cost for each price item, and summarizing the results of all the price items to obtain the individual coalition game cost of each vehicle i in the coalition; and summing the individual alliance game costs of all vehicles in the alliance to obtain the alliance total game cost.
7. The method of claim 1, wherein the step of solving the extensible coalition gaming model is: Constructing a core alliance, wherein the core alliance comprises external vehicles, front representative vehicles and rear representative vehicles, the front representative vehicles represent a front vehicle group of a vehicle team, and the rear representative vehicles represent a rear vehicle group of the vehicle team; Solving the speed maintenance cost and the control input cost of each vehicle in the front vehicle group and the rear vehicle group, and respectively summarizing to obtain total cost in the two groups; Summing the cost of the representing vehicle and the total cost in the corresponding group to obtain the comprehensive cost of the front representing vehicle and the rear representing vehicle; solving the total cost of the single vehicle of the external vehicle, and combining the comprehensive cost of the two representative vehicles to obtain the total cost of the core alliance; and solving to obtain an optimal control strategy of the core alliance by taking the minimum total cost of the core alliance as an optimization target, wherein the optimal control strategy comprises control input and lane state parameters of an external vehicle, a front representative vehicle and a rear representative vehicle.
8. The method according to claim 1, wherein the step of obtaining the fleet longitudinal optimal entry timing and the optimal longitudinal control sequence by the double-layer optimization method comprises the steps of: Initializing double-layer optimization parameters, including slow thread update period, fast thread control period, MPC prediction time domain and mixed integer programming solver parameters, and initializing decision buffer areas; Starting a slow thread and a fast thread, and initializing a decision buffer area, wherein the slow thread is an upper optimizing thread, and the execution frequency is lower than that of the fast thread serving as a lower optimizing thread; The method comprises the steps of circularly acquiring real-time state information of a fleet and external vehicles by a slow thread, constructing a mixed integer programming problem with a closing time as a discrete decision variable and a core alliance total cost as a target, calling a solver to solve to obtain an optimal closing time, and writing the optimal closing time into a decision buffer area; The fast thread reads the optimal closing time from the decision buffer according to higher frequency circulation, acquires real-time state information of the vehicle, predicts the future state based on the vehicle dynamics model, constructs an MPC optimization problem by taking the optimal closing time as constraint, solves to obtain an optimal longitudinal control sequence, and issues the first control quantity of the sequence to the vehicle for execution; after the external vehicle is completely engaged, the double-layer optimization cycle is terminated, and the conventional cruise control of the fleet is resumed.
9. The method of claim 1, wherein tracking the lateral reference trajectory using a curvature-adaptive lateral compound control method comprises: acquiring real-time state feedback information and a transverse reference track of a vehicle, wherein the state feedback information comprises a yaw angle, a transverse position deviation, a course angle deviation and a vehicle speed; Inputting state feedback information into a fuzzy regulator by adopting a fuzzy PI feedback control algorithm, dynamically adjusting the proportional gain and the integral gain of the PI controller, and solving to obtain a first rotation angle control signal based on tracking error; An LTV-MPC control algorithm is adopted to obtain the nominal state quantity of the current prediction step, a time-varying system matrix and a control matrix are updated, the nominal state quantity of the next prediction step is predicted, a finite time domain optimization problem is constructed, and a second corner control signal is obtained; according to a driving scene switching control strategy, only a second corner control signal is adopted when a lane is changed, and a first corner control signal and a second corner control signal are fused when the lane is kept, so that a transverse optimal control steering angle sequence is output.
10. A fleet dynamic formation system based on scalable coalition gaming, comprising: The single-car cost construction module is used for constructing a vehicle single-car game cost function, wherein the single-car game cost function fuses driving safety cost, speed keeping cost, control input cost and lane risk cost, and each cost item is configured with a corresponding weight coefficient and weighted and summed; The coalition game modeling module is used for constructing an expandable coalition game model based on a single-vehicle game cost function, determining core coalitions and representative vehicles, dividing non-core coalition vehicles in a motorcade into a front vehicle group and a rear vehicle group according to positions, and uniformly representing the vehicle cost in the group by the corresponding representative vehicles; The longitudinal decision solving module is used for solving the extensible alliance game model, obtaining the longitudinal optimal closing time and the optimal longitudinal control sequence of the motorcade through a double-layer optimization method, and controlling the motorcade to coordinate a longitudinal gap for closing of external vehicles; The transverse track planning module is used for constructing an external vehicle lane change transverse reference track by adopting a B spline function based on a longitudinal optimal decision result; and the transverse compound control module is used for tracking the transverse reference track by adopting a curvature self-adaptive transverse compound control method, fusing an LTV-MPC algorithm with automatically updated parameters and a fuzzy PI feedback control algorithm, outputting a transverse optimal control steering angle sequence according to a driving scene switching control strategy, realizing safe merging of external vehicles, and completing dynamic formation of a vehicle team.

Description

Vehicle team dynamic formation method and system based on extensible alliance game Technical Field The invention relates to the technical field of fleet control, in particular to an intelligent network alliance fleet dynamic formation method and system based on extensible alliance gaming. Background The rapid development of intelligent traffic systems promotes the progress of intelligent network linkage vehicle team cooperative control technology, which is a key way for improving road traffic efficiency, relieving traffic jam, reducing energy consumption and improving traffic safety. The key point of the core coordination behavior of dynamic formation of the motorcade comprises queue merging and separating, which is common in the scenes that external vehicles apply for joining the motorcade, long motorcade is converged into a main line lane through an intersection, a ramp vehicle and the like, is that the position and the time of merging/separating are accurately determined, and feasible control input is solved to execute corresponding actions. If the performance of the decision control module is insufficient, the coordination time is prolonged, so that the energy consumption is increased, the road utilization rate is reduced, and the probability of traffic accidents is possibly increased. The existing intelligent network linkage vehicle team decision control method is mainly divided into two types, namely a centralized method relies on a central controller to collect global information and solve a multi-vehicle cooperation problem, the cooperation efficiency can be improved under the condition of high intelligent network linkage vehicle permeability, the method has high requirements on communication equipment performance, the calculation load is increased rapidly along with the increase of the vehicle team scale, the system expandability and the robustness are poor, and the central node failure easily causes global failure, and the distributed method autonomously decides by individual vehicles based on local perception information, reduces the communication and calculation requirements, improves the system expandability and the robustness, but is easy to bias to local optimum due to lack of a global optimization mechanism, and reduces the overall coordination performance and the traffic efficiency of the vehicle team. In order to take the advantages of the centralized method and the distributed method into consideration, the prior art tries to combine the two methods, but the problems of insufficient utilization of road information, single scene adaptability and the like still exist, and particularly when the scale of a motorcade is enlarged, the algorithm is difficult to ensure calculation instantaneity and control optimality simultaneously. In the whole, the existing intelligent network linkage vehicle team dynamic formation decision control technology faces four major core challenges, namely a communication and calculation bottleneck, a centralized method has high requirements on communication bandwidth and calculation resources and is difficult to adapt to large-scale dynamic formation, robustness and expandability are insufficient, a centralized architecture has poor fault tolerance, a distributed method has weak optimization capability, the scene adaptability is limited, the existing method is designed aiming at a specific scene, universality is lacking, optimizing efficiency and instantaneity are contradicted, and an algorithm under the large-scale vehicle team is difficult to achieve both solving precision and calculation instantaneity. Therefore, a dynamic formation method is needed that integrates the advantages of the centralized and distributed methods, combines the algorithm real-time performance and the optimization performance, and can effectively and cooperatively control the fleet of different scales. Disclosure of Invention Aiming at the problems of communication calculation bottleneck, poor expandability, limited scene adaptability, contradiction between optimization and real-time performance and the like in the existing intelligent network linkage vehicle team dynamic formation technology, the invention provides a vehicle team dynamic formation method, system, equipment and storage medium based on an extensible alliance game, which realize high-efficiency and safe cooperative control of vehicle team dynamic formation and simultaneously give consideration to expandability and real-time performance of an algorithm, and are suitable for vehicle team external vehicle integration scenes of different scales. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the present invention provides a dynamic formation method for a fleet of vehicles based on extensible alliance gaming, including the following steps: Constructing a vehicle single-vehicle game cost function, wherein the vehicle single-vehicle game cost function fuses driving