Vehicle platoon cooperative lane-changing method and system based on trajectory prediction and game optimization

Abstract

This invention discloses a collaborative lane-changing method and system for vehicle platoons based on trajectory prediction and game theory optimization. First, the invention acquires the operational status information of intelligent connected vehicle platoons and surrounding human-driven vehicles in a mixed traffic environment, constructs a vehicle interaction dataset, and generates scene feature vectors. Then, it uses a conditional diffusion model to generate multiple future candidate trajectories, calculates prediction uncertainty based on trajectory distribution differences, and dynamically adjusts the safety distance. Next, it establishes a comprehensive lane-changing cost function that considers driving safety, driving efficiency, acceleration changes, and platoon integrity, and obtains the optimal lane-changing strategy through game theory optimization. Finally, it employs a model predictive control method that integrates control obstacle functions to constrain the vehicle motion safety boundary. This invention effectively addresses the uncertainty of human driving behavior, improving lane-changing efficiency while ensuring safety, and is suitable for collaborative lane-changing scenarios of intelligent connected vehicle platoons in complex traffic environments such as highways.

Description

Technical Field

[0001] This invention belongs to the field of intelligent traffic control, and in particular relates to an intelligent connected vehicle platooning cooperative lane-changing method and system based on conditional diffusion trajectory prediction and uncertainty perception game optimization. Background Technology

[0002] With the development of Vehicle-to-Everything (V2X) communication technology, autonomous driving perception technology, and vehicle cooperative control technology, connected and automated vehicles (CAVs) are gradually becoming an important component of future intelligent transportation systems. Multiple CAVs can form a platoon through vehicle-to-vehicle communication and cooperative control technologies, enabling information sharing and collaborative driving among vehicles. Compared to traditional single-vehicle autonomous driving, platooning can shorten following distances, improve road capacity, and reduce vehicle energy consumption, thus showing broad application prospects in highway transportation, intelligent logistics, and autonomous commercial vehicles. However, for a considerable period in the future, the road traffic environment will remain in a mixed traffic phase where connected and human-driven vehicles (HDVs) coexist. In this mixed traffic environment, the driving behavior of human drivers is influenced by driving habits, road conditions, and traffic status, exhibiting significant uncertainty and randomness. For example, during lane changes within a platoon, vehicles in the target lane may exhibit various driving behaviors such as yielding, maintaining speed, or suddenly accelerating to block, leading to continuous changes in the available lane-changing space within the platoon. If intelligent connected vehicles cannot accurately predict the future movement trends of surrounding vehicles, it can easily lead to overly aggressive lane-changing decisions, causing collision risks, or overly conservative decisions, reducing road traffic efficiency.

[0003] Currently, research on lane-changing problems in intelligent connected vehicles mainly includes rule-based methods, learning-based methods, and optimization control-based methods. Rule-based methods typically determine whether to perform a lane change based on vehicle distance, speed difference, and a fixed safety threshold. For example, a minimum safe distance is set to determine if the target lane meets the lane-changing conditions; if the vehicle distance exceeds the threshold, the vehicle is allowed to perform a lane change. This type of method is simple in structure and computationally efficient, but fixed rules are difficult to adapt to complex mixed traffic environments. When facing human drivers with different driving styles, a fixed safety distance may lead to overly conservative vehicle behavior or an inability to handle sudden risks in time. Deep learning-based methods can learn vehicle motion patterns using historical traffic data and predict the future trajectories of surrounding vehicles through methods such as recurrent neural networks and graph neural networks. Although this type of method improves prediction accuracy, most methods only output a single deterministic trajectory and cannot describe the various future behaviors that human drivers may produce. At the same time, the prediction model lacks an evaluation of its own prediction reliability, making it difficult for the autonomous driving decision-making system to dynamically adjust the driving strategy based on the predicted risks. In recent years, diffusion models, as a generative model, have shown strong capabilities in modeling complex distributions, and can generate multiple possible results through progressive denoising. However, existing diffusion models for vehicle trajectory prediction primarily focus on improving prediction accuracy, lacking integration with the collaborative decision-making process of vehicle platoons, and thus failing to directly address safety constraints in lane changing within vehicle platoons. Furthermore, existing lane-changing methods typically prioritize individual vehicle interests or overall platoon planning, neglecting consideration of human-vehicle interactions in the surrounding environment. When the behavior of vehicles in the target lane is unknown, platoons are prone to lane-changing failures, platoon breakage, or insufficient safe distances.

[0004] In terms of vehicle control, traditional model predictive control (MPC) can achieve vehicle trajectory tracking, but it mainly relies on preset constraints. When there is a large error in predicting vehicle behavior, it may not be able to guarantee that the vehicle is always in a safe state.

[0005] Therefore, how to establish an integrated method that can simultaneously achieve vehicle future behavior prediction, prediction risk assessment, vehicle platoon collaborative decision-making, and safe trajectory control is an important technical problem that needs to be solved for the safe operation of intelligent connected vehicle platoons in mixed traffic environments. Summary of the Invention

[0006] To address the challenge of balancing safety and efficiency in lane changing within a platoon due to the uncertainty of human driving behavior in mixed traffic environments, this invention provides a collaborative lane changing method based on conditional diffusion trajectory prediction and uncertainty-aware game optimization. The method generates future multimodal trajectories of vehicles using a conditional diffusion model to describe the uncertainty of human driving behavior. It dynamically adjusts safety constraints through uncertainty evaluation to balance safety and efficiency, achieves collaborative decision-making within the vehicle platoon through game optimization to obtain the optimal lane changing strategy, and enhances vehicle motion safety by incorporating a control obstacle function. Ultimately, this enables safe, efficient, and stable lane changing for intelligent connected vehicle platoons in complex traffic environments.

[0007] According to one aspect of the present invention, a vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization is provided, comprising: constructing a vehicle historical motion sequence based on the operating status information of intelligent connected vehicle platoons and surrounding human-driven vehicles in a target road environment within a continuous time range; determining vehicle interaction relationships based on the spatial distance and relative motion states between vehicles; combining the vehicle historical motion sequence, vehicle interaction relationships, and road information into a vehicle interaction dataset; inputting the vehicle historical motion sequence into a scene encoding network to extract motion features of the vehicle's own motion state; calculating the attention weights between the target vehicle and surrounding vehicles using an attention mechanism to obtain vehicle interaction features; fusing the motion features, interaction features, and road information into a scene feature vector for describing the current traffic state; and using the scene feature vector as a condition, performing forward diffusion... The process involves progressively adding random noise to the real vehicle trajectories to transform them into a random distribution. A reverse denoising network is then used to gradually recover the future vehicle trajectories, generating multiple candidate trajectories through multiple sampling. The uncertainty of future vehicle behavior is calculated based on the distribution differences among these predicted trajectories, and the lane-changing safety distance is dynamically adjusted according to the magnitude of this uncertainty. Based on this safety distance, the intelligent connected vehicles in the vehicle queue are treated as game participants. A comprehensive lane-changing optimization function is established, taking into account driving safety, driving efficiency, acceleration changes, and queue integrity. Through game optimization, the speed adjustment strategies and lane-changing execution times for each vehicle are obtained, generating the target lane-changing trajectory. Based on the obtained target lane-changing trajectory, model predictive control methods are used to calculate the future control inputs for the vehicles. Simultaneously, a control obstacle function is introduced to restrict the vehicle's motion safety boundary, completing the coordinated lane-changing of the intelligent connected vehicle queue.

[0008] As a further technical solution, the formation of the vehicle interaction dataset also includes: acquiring the operating status information of intelligent connected vehicle queues and surrounding human-driven vehicles in the target road environment within a continuous time window through on-board sensing devices, vehicle-to-everything (V2X) communication devices, or vehicle trajectory datasets. The operating status information includes the vehicle's longitudinal position, lateral position, speed, acceleration, and heading angle; acquiring the continuous motion state of vehicles according to a fixed sampling period to construct a vehicle historical motion sequence containing the continuous motion state of each vehicle within a preset time period; calculating the distance between vehicles and determining that there is an interaction between vehicles when the distance is less than the sensing range; and combining the vehicle historical motion sequence, the vehicle interaction relationship matrix, and road information into a vehicle interaction dataset.

[0009] As a further technical solution, the generation of the scene feature vector also includes: inputting the vehicle's historical motion sequence into a scene coding network, extracting vehicle speed changes, acceleration changes, and driving trend information as vehicle motion features; constructing a query vector based on the target vehicle's motion state, constructing a key vector based on the motion states of surrounding vehicles, calculating the attention weights between the target vehicle and each surrounding vehicle through an attention mechanism, and obtaining vehicle interaction features by weighted summation; and fusing the motion features, interaction features, and road information to generate a scene feature vector describing the current traffic state.

[0010] As a further technical solution, the generation of future motion trajectories also includes: during the model training phase, gradually adding random Gaussian noise to the real vehicle trajectory to convert it into a random noise distribution, so as to learn the evolution law of the vehicle trajectory from the real motion state to the random distribution state; during the trajectory prediction phase, using the random noise trajectory as the initial input and the scene feature vector as the conditional constraint, gradually removing noise through a parameterized denoising network to recover the future motion trajectory of the vehicle; and generating multiple future candidate trajectories by sampling the conditional diffusion model multiple times, with each trajectory corresponding to a different driving behavior mode to describe the uncertainty of human driving behavior.

[0011] As a further technical solution, the uncertainty of future vehicle behavior is calculated based on the distribution differences among the multiple predicted trajectories, and the lane-changing safety distance is dynamically adjusted according to the magnitude of the uncertainty, including: calculating the average value of the predicted trajectories based on the multiple predicted trajectories output by the conditional diffusion model. ,in, This represents the mean of the predicted trajectory. This represents the total number of candidate trajectories generated; Represent the k-th candidate trajectory; calculate the uncertainty in predicting the vehicle's future behavior based on the degree of deviation between different predicted trajectories: ,in, This represents the uncertainty in predicting the future trajectory of vehicles; a dynamic safety distance model is established based on this uncertainty. ,in, Indicates dynamic safety distance. Indicates the basic safe distance between vehicles. This represents the risk adjustment factor.

[0012] As a further technical solution, generating the target lane-changing trajectory also includes: treating the intelligent connected vehicle platoon performing the lane-changing task and the vehicles affecting the target lane as participants in a traffic game; each vehicle selecting a control strategy including longitudinal acceleration and steering angle based on the predicted trajectories of surrounding vehicles and its own driving objective; and establishing a safety cost based on a dynamic safety distance. Establish efficiency costs based on the difference between the vehicle's actual speed and the expected speed. Establish comfort costs based on the rate of change of acceleration Establish queue maintenance cost based on the deviation between vehicle spacing and desired queue spacing. Taking into account driving safety, driving efficiency, acceleration changes, and queue integrity, a comprehensive lane-changing optimization function is established: ,in, These represent the weights of different evaluation objectives; the optimal cooperative driving strategy for the vehicle platoon is obtained by minimizing the comprehensive lane-changing optimization function, and the target lane-changing trajectory is generated.

[0013] As a further technical solution, the speed adjustment strategy and lane-changing execution time of each vehicle are obtained through game theory optimization, including: generating initial driving strategies for participating vehicles based on the current traffic environment and vehicle queue operation requirements; fixing the driving strategies of other vehicles except the target vehicle in each iteration, and optimizing the control strategy of the target vehicle according to the comprehensive lane-changing optimization function; after completing the single optimization of all vehicles, updating the vehicle queue strategy set, and re-inputting the updated strategy into the game model for the next round of vehicle interaction calculation; calculating the change between two adjacent iterations, and determining that the vehicles have reached a stable interaction state when the change is less than a set convergence threshold, and outputting the vehicle queue collaborative lane-changing strategy.

[0014] As a further technical solution, after generating the target lane-changing trajectory, the process also includes: calculating the available lane-changing space between vehicles in front and behind the target lane based on the predicted future trajectories of vehicles in the target lane, and determining whether the current traffic environment meets the overall lane-changing requirements of the vehicle platoon by combining the dynamic safety distance; when the available lane-changing space is greater than the sum of the total length of the vehicle platoon and the dynamic safety distance, the vehicle platoon adopts a direct lane-changing method, synchronously adjusting its speed according to the game optimization results while maintaining the platoon structure to complete the overall lane-changing; when the available lane-changing space is less than the sum of the total length of the vehicle platoon and the dynamic safety distance, the vehicle platoon enters a collaborative adjustment phase, actively creating lane-changing space that meets safety constraints by changing the longitudinal speed of the vehicles in the platoon and the positional relationship between them and the vehicles in the target lane; after the vehicles enter the target lane, they readjust their vehicle states according to the desired platoon spacing to restore a stable platoon driving state.

[0015] As a further technical solution, completing the cooperative lane-changing of intelligent connected vehicle platoons also includes: establishing a vehicle state model containing the vehicle's longitudinal position, lateral position, speed, and heading angle, as well as a vehicle control input model containing the vehicle's acceleration and front wheel steering angle, to predict the future state of the vehicle; establishing a model to predict control optimization objectives based on the deviation between the planned trajectory and the actual vehicle state, and obtaining the future control input of the vehicle through rolling optimization; introducing a control obstacle function to define a safety function. Where d represents the actual distance of the vehicle. This indicates the minimum permissible safe distance, which must be met during vehicle operation. 0; Under the conditions of satisfying vehicle dynamics constraints and safety constraints, the optimal control quantity is obtained, and the control quantity is sent to the vehicle actuator to complete the coordinated lane changing of the intelligent connected vehicle platoon.

[0016] According to one aspect of this invention, a vehicle platooning cooperative lane-changing system based on trajectory prediction and game theory optimization is provided, comprising: a vehicle information acquisition module, configured to construct a vehicle historical motion sequence based on the operating status information of intelligent connected vehicle platoons and surrounding human-driven vehicles in a target road environment within a continuous time range, and determine vehicle interaction relationships based on the spatial distance and relative motion states between vehicles, and combine the vehicle historical motion sequence, vehicle interaction relationships, and road information into a vehicle interaction dataset; and a trajectory prediction module, connected to the vehicle information acquisition module, configured to input the vehicle historical motion sequence into a scene encoding network, extract motion features of the vehicle's own motion state, calculate the attention weights between the target vehicle and surrounding vehicles using an attention mechanism to obtain vehicle interaction features, fuse the motion features, interaction features, and road information into a scene feature vector, and use the scene feature vector as a condition to perform positive noise addition and... The reverse denoising process generates multiple candidate trajectories through multiple samplings. An uncertainty perception module, connected to the trajectory prediction module, calculates the uncertainty of future vehicle behavior based on the distribution differences between the multiple predicted trajectories and dynamically adjusts the lane-changing safety distance according to the magnitude of the uncertainty. A collaborative lane-changing decision module, also connected to the uncertainty perception module, uses the safety distance and intelligent connected vehicles in the vehicle queue as game participants. It establishes a comprehensive lane-changing optimization function, considering driving safety, driving efficiency, acceleration changes, and queue integrity. Through game optimization, it obtains the speed adjustment strategy and lane-changing execution time for each vehicle, generating the target lane-changing trajectory. A vehicle safety control module, connected to the collaborative lane-changing decision module, calculates the future control input of the vehicle using model predictive control methods based on the obtained target lane-changing trajectory. Simultaneously, it introduces a control obstacle function to restrict the vehicle's motion safety boundary, completing the collaborative lane-changing of the intelligent connected vehicle queue.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Predicting future vehicle trajectories in mixed traffic environments using a conditional diffusion model. Compared to traditional single-trajectory prediction methods, this invention generates multiple candidate future trajectories that conform to the current traffic environment conditions through a diffusion model's random sampling process, thereby expressing the different driving intentions that vehicles may have. This method can effectively describe the uncertain behavior of human drivers and improve the ability of intelligent connected vehicles to understand complex traffic environments.

[0018] 2. A method for uncertainty perception based on predicted trajectory distribution is proposed. By analyzing the dispersion among multiple predicted trajectories, the risk of future vehicle behavior is evaluated, and the vehicle's safety distance is dynamically adjusted based on the prediction uncertainty. When the behavior of surrounding vehicles is difficult to judge, the system automatically increases the safety margin; when vehicle behavior is relatively stable, unnecessary safety distances are reduced, improving lane-changing efficiency. Therefore, it can avoid the problems of over-conservatism or insufficient safety in traditional fixed safety distance methods.

[0019] 3. Establish a collaborative game-theoretic decision-making model for intelligent connected vehicle platoons. During lane changing, vehicle safety, driving efficiency, passenger comfort, and platoon integrity are considered simultaneously. Through inter-vehicle strategy optimization, the vehicle platoon can proactively adjust its speed and lane-changing timing based on the traffic conditions of the target lane, thereby improving the success rate of lane changing in mixed traffic environments.

[0020] 4. Model predictive control method using fused control barrier functions By establishing a vehicle motion safety boundary through a control obstacle function, hazard control inputs are limited in real time during trajectory tracking. Compared to the traditional MPC method, this invention can further reduce the collision risk caused by prediction errors and improve the safety of vehicle lane changing.

[0021] 5. Construct a complete closed-loop system encompassing "environmental perception—trajectory prediction—risk assessment—collaborative decision-making—vehicle control". This system is applicable to intelligent truck platoons on highways, cooperative driving of autonomous vehicles, and vehicle-road cooperative traffic scenarios, and has good engineering application value. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating the vehicle queuing cooperative lane-changing method based on trajectory prediction and game optimization provided in this embodiment of the invention; Figure 2 This is a schematic diagram illustrating the construction of a vehicle interaction scenario in a mixed traffic environment, as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the conditional diffusion vehicle trajectory prediction model structure provided in an embodiment of the present invention; Figure 4 This is a flowchart of the vehicle trajectory prediction uncertainty perception and dynamic safety constraint generation provided in an embodiment of the present invention; Figure 5This is a schematic diagram of the intelligent connected vehicle platoon collaborative lane-changing game decision-making process provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a two-stage coordinated lane-changing process for intelligent connected vehicle platoons provided in an embodiment of the present invention. Figure 7 This is a schematic diagram of the model predictive control structure for the fusion control barrier function provided in an embodiment of the present invention; Figure 8 This is a comparison chart of vehicle trajectory prediction results in a mixed traffic environment simulation experiment provided by an embodiment of the present invention; Figure 9 This is a schematic diagram of the three-dimensional trajectory changes during the coordinated lane-changing process of vehicle platoons, provided in an embodiment of the present invention. Figure 10 The figure shows the experimental comparison results of different lane-changing control methods provided in the embodiments of the present invention in terms of safety and efficiency indicators. Detailed Implementation

[0024] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0026] This invention provides a vehicle platoon cooperative lane-changing method based on trajectory prediction and game theory optimization, such as... Figure 1 As shown, the steps include the following.

[0027] S1: Establish a vehicle interaction dataset for a mixed traffic environment.

[0028] This process acquires operational status information of intelligent connected vehicle platoons and surrounding human-driven vehicles within a continuous timeframe in the target road environment. The collected vehicle position, speed, acceleration, and lane information are arranged chronologically to form a historical vehicle motion sequence. Based on the spatial distance and relative motion states between vehicles, the interaction relationships between them are determined, and a vehicle interaction dataset describing the dynamic changes of vehicles in a mixed traffic environment is established. Vehicle status information includes longitudinal position, lateral position, longitudinal velocity, lateral velocity, longitudinal acceleration, lateral acceleration, heading angle, vehicle length, vehicle width, and the lane number of the vehicle.

[0029] The process of establishing the vehicle interaction dataset in step S1 is as follows: S101, Vehicle Status Acquisition.

[0030] Within a continuous time window, the vehicle's operating status is acquired through onboard sensing devices, vehicle-to-everything (V2X) communication devices, or vehicle trajectory datasets. The status of the i-th vehicle within the continuous time window is represented as follows: ,in, Indicates the longitudinal position of the vehicle; Indicates the lateral position of the vehicle; Indicates vehicle speed; Indicates vehicle acceleration; Indicates the vehicle's heading angle.

[0031] S102, Construct the historical movement sequence of the vehicle.

[0032] The continuous motion status of the vehicle is obtained according to a fixed sampling period: ,in, Indicates the vehicle's historical movement sequence; This indicates the length of the historical observation. This sequence is used to describe the trend of vehicle movement over a past period of time.

[0033] S103, establish vehicle interaction relationships.

[0034] Calculate the distance between vehicles: ,when It is assumed that there are interactive influences between vehicles. Here, D represents the vehicle's perception range. The closer the vehicle is, the more significant its influence on the future movement of the target vehicle.

[0035] S104 generates vehicle interaction data.

[0036] Combine vehicle historical movement sequences, vehicle interaction relationships, and road information: ,in, Indicates the vehicle's motion status; Indicates vehicle interaction relationships; This represents road information. This data serves as input to the conditional diffusion trajectory prediction model.

[0037] S2, Establish a vehicle interaction scenario coding model.

[0038] The vehicle historical motion sequence obtained in step S1 is input into the scene encoding network to extract features of the vehicle's own motion state and the interaction relationship with surrounding vehicles. The degree of influence between different vehicles is calculated using an attention mechanism. The vehicle historical motion features, the interaction features of neighboring vehicles, and the road environment information are fused to generate a scene feature vector that describes the current traffic state.

[0039] The scene encoding method in step S2 is as follows: S201, Vehicle Motion Feature Encoding.

[0040] Input the vehicle's historical motion sequence into the scene encoding network: ,in, Indicates the characteristics of vehicle motion; This represents the encoding function. This process extracts information about vehicle speed changes, acceleration changes, and driving trends.

[0041] S202, Vehicle Interaction Feature Encoding.

[0042] The attention mechanism is used to calculate the influence between vehicles: ,in, Represents the query feature vector of the target vehicle; This represents the key feature vector corresponding to the j-th vehicle surrounding it.

[0043] Further calculations , thereby obtaining vehicle interaction characteristics. Let represent the feature vector corresponding to the j-th vehicle in the vicinity.

[0044] S203, Scene Feature Fusion.

[0045] Vehicle motion characteristics Vehicle interaction features Road information The result of fusion: ,in, This represents a latent variable in the current traffic environment. This variable serves as a conditional input to the diffusion prediction model.

[0046] S3. Establish a vehicle future trajectory prediction model based on the conditional diffusion model.

[0047] The latent variables of the traffic scene obtained in step S2 are used as conditional information input into the diffusion model. First, random noise is gradually added to the real vehicle trajectories through a forward diffusion process, so that the vehicle trajectories are transformed into a random distribution state;

[0048] Then, an inverse denoising network is used to gradually reconstruct the vehicle's future trajectory based on traffic environment information. By sampling multiple times, multiple candidate future trajectories that meet the vehicle's motion constraints are generated, enabling the prediction of various potential behaviors of human drivers.

[0049] In step S3, the method for establishing the vehicle trajectory prediction model based on the conditional diffusion model is as follows: S301, Establish the forward diffusion process of the vehicle's future trajectory. First, obtain the vehicle's actual motion trajectory within the predicted future time range, and represent the vehicle's future trajectory as: ,in, Indicates the vehicle's true future trajectory; This indicates the length of the future prediction time step.

[0050] In order to learn the data distribution of future vehicle trajectories in a mixed traffic environment, random Gaussian noise is gradually added to the real vehicle trajectories during the model training process, so that the vehicle trajectories with clear motion patterns are gradually transformed into random noise distributions.

[0051] The diffusion process at step t is represented as: ,in, This represents the trajectory state after the t-th diffusion step; This represents random noise that follows a standard normal distribution. This parameter represents the proportion of noise controlled during the diffusion process.

[0052] Through the aforementioned positive diffusion process, the model is able to learn the changing patterns of vehicle trajectories from their actual motion state to their random distribution state, providing a foundation for subsequent trajectory generation.

[0053] S302, Establish the conditional reverse denoising trajectory generation process.

[0054] In the trajectory prediction stage, random noise trajectories are used as the initial input, and the latent variables of the traffic scene obtained in step S2 are also introduced: As conditional constraint information.

[0055] Establish a noise reduction network This is used to estimate the noise information contained in the current trajectory state. The vehicle trajectory is then gradually recovered based on the predicted noise. ,in, This represents a parameterized denoising network; This indicates the current diffusion time step.

[0056] Because the vehicle's historical motion state and the interaction with surrounding vehicles are incorporated into the denoising process, the generated trajectory can simultaneously satisfy both the vehicle's own motion trend and traffic environment constraints.

[0057] S303, establish an attention-enhanced denoising network.

[0058] To improve the ability to extract vehicle interaction information in complex traffic scenarios, an attention-enhanced network structure is introduced during the backdiffusion process. First, the input trajectory features are encoded to obtain vehicle motion information at different time steps. Then, an attention mechanism is used to calculate the correlation between historical trajectories and future motion states, enabling the model to focus on important time segments and key interacting vehicles that influence future vehicle behavior. Finally, the output is a future trajectory that conforms to the vehicle's motion patterns.

[0059] S304 generates the future trajectory of a multimodal vehicle.

[0060] Considering the randomness of human driving behavior, different driving behaviors may occur under the same traffic conditions, such as maintaining speed, accelerating, decelerating, or changing lanes. Therefore, the conditional diffusion model is sampled multiple times: ,in, This represents the total number of candidate trajectories generated; This represents the k-th candidate trajectory; Indicates the first The sequence number of each candidate trajectory.

[0061] By generating multiple candidate trajectories, the uncertain behavior of human-driven vehicles can be described.

[0062] S4. Establish a vehicle trajectory prediction uncertainty perception model.

[0063] Based on the multiple predicted trajectories generated in step S3, the distribution differences between different prediction results are calculated, and the uncertainty evaluation results of the future behavior of vehicles are obtained according to the degree of trajectory dispersion. The safety distance of the vehicle platoon during lane changing is adjusted according to the magnitude of uncertainty, so that the vehicles maintain a greater safety margin when the predicted risk is high.

[0064] The uncertainty perception method for vehicle trajectory prediction in step S4 is as follows:

[0065] S401, calculate the statistical characteristics of the predicted trajectory.

[0066] Based on the multiple predicted trajectories output by the conditional diffusion model, statistical analysis is performed on the future motion state of the vehicle.

[0067] Calculate the average value of the predicted trajectory: , in, This represents the mean of the predicted trajectory; This represents the total number of candidate trajectories generated. The mean of the predicted trajectories is used to represent the most likely future movement trend of the vehicle.

[0068] S402, calculate the uncertainty of vehicle behavior prediction.

[0069] The uncertainty of the vehicle's future behavior is calculated based on the degree of deviation between different predicted trajectories: , in, This indicates the uncertainty in predicting the future trajectory of a vehicle.

[0070] When the differences between different predicted trajectories are small, it indicates that the future movement trend of the vehicle is relatively clear; when the differences between different predicted trajectories are large, it indicates that the vehicle may have multiple driving intentions, and the risk of vehicle behavior is increased.

[0071] S403, Establish uncertainty-aware security constraints.

[0072] Traditional lane-changing methods typically employ a fixed safety distance, which cannot be dynamically adjusted based on changes in the behavior of surrounding vehicles. Therefore, this invention establishes a dynamic safety distance model based on prediction uncertainty: ,in, Indicates dynamic safety distance; Indicates the basic safe distance between vehicles; This represents the risk adjustment factor.

[0073] When the uncertainty of surrounding vehicle predictions increases, the safety distance requirement is raised, enabling intelligent connected vehicles to adopt a more conservative lane-changing strategy; when the prediction results are stable, the safety redundancy is reduced, and the lane-changing efficiency of the vehicle platoon is improved.

[0074] S5, establish a game model for collaborative lane-changing in intelligent connected vehicle platoons.

[0075] Based on the dynamic safety constraints obtained in step S4, the intelligent connected vehicles in the vehicle queue are taken as the participants in the game. Taking into account the vehicle safety distance, driving efficiency, acceleration change and vehicle queue integrity, a vehicle lane-changing benefit evaluation function is established. Through game optimization, the speed adjustment strategy and lane-changing execution time of each vehicle in the vehicle queue are obtained.

[0076] In step S5, the method for establishing the intelligent connected vehicle platoon cooperative lane-changing game model is as follows: S501, Establish a vehicle game object model.

[0077] The intelligent connected vehicle platoon performing the lane-changing task, along with the vehicles affecting the target lane, are all considered as participants in the traffic game. The vehicle set is represented as: ,in, This represents the i-th vehicle participating in the game; This indicates the number of vehicles. Each vehicle selects a control strategy based on the predicted trajectories of surrounding vehicles and its own driving objectives.

[0078] The vehicle control strategy is defined as follows: ,in, Indicates the longitudinal acceleration of the vehicle; Indicates the vehicle's steering angle.

[0079] S502, Establish the safety evaluation function.

[0080] Based on the uncertainty-perceived safe distance obtained in step S4, calculate the safety risks between vehicles.

[0081] The distance between vehicles is expressed as Establishing security comes at a cost: , in, Indicates security risk; Indicates the dynamic safety distance.

[0082] When a vehicle approaches the safety boundary, the safety cost increases rapidly, thus preventing dangerous lane-changing behavior.

[0083] S503, establish efficiency and comfort evaluation functions.

[0084] Vehicle efficiency evaluation is used to describe the difference between a vehicle's actual speed and its expected speed: ,in, Indicates the vehicle's current speed; Indicates the desired speed.

[0085] Vehicle comfort evaluation is established based on the rate of change of acceleration: , By limiting sudden acceleration changes, the vehicle's rapid acceleration and deceleration are reduced.

[0086] S504, establish the vehicle queuing maintenance evaluation function.

[0087] During lane changing in a vehicle platoon, a stable distance must be maintained between vehicles.

[0088] definition ,in, Indicates the current vehicle spacing; This represents the desired queue spacing. This function is used to reduce the risk of vehicle queue breakage.

[0089] S505, establish a comprehensive lane-changing optimization function.

[0090] Taking into account safety, efficiency, comfort, and queue integrity, the following system is established: , in, These represent the weights of different evaluation objectives. The optimal cooperative driving strategy for the vehicle platoon is obtained by minimizing the comprehensive cost function.

[0091] In step S5, the solution method for the vehicle platoon cooperative lane-changing game model is as follows:

[0092] S506, Initialize vehicle lane-changing strategy.

[0093] Based on the current traffic conditions and vehicle platooning requirements, initial driving strategies are first generated for the vehicles participating in the game. The initial strategy for the i-th vehicle is expressed as: ,in, Indicates the initial longitudinal control quantity of the vehicle; This indicates the initial lateral control quantity of the vehicle.

[0094] The set of strategies for all participating vehicles is represented as: This initial strategy serves as input for subsequent iterative optimization processes.

[0095] S507 calculates the optimal response strategy for a single vehicle.

[0096] In each iteration, the driving strategies of all vehicles except the target vehicle are fixed. Based on the comprehensive lane-changing optimization function established in step S505, the control strategy of the target vehicle is optimized. The target vehicle optimization process is represented as follows: ,in, This represents the current optimal control strategy for the i-th vehicle; Let represent the set of strategies for all vehicles except the i-th vehicle.

[0097] In this way, each vehicle can continuously adjust its driving strategy based on changes in the behavior of surrounding vehicles.

[0098] S508, Update the overall vehicle queue strategy.

[0099] After completing a single optimization for all vehicles, update the vehicle queuing strategy set: , in, This indicates the current iteration number. The updated strategy is re-inputted into the game model for the next round of vehicle interaction calculations.

[0100] S509, determine whether the game optimization has converged.

[0101] Calculate the change between two consecutive iterations: , When satisfied At that point, a stable interaction state between the vehicles is considered to have been reached. This indicates that a convergence threshold is set.

[0102] At this point, the vehicle queue coordinated lane-changing strategy is output, including: vehicle lane-changing start time, vehicle longitudinal speed adjustment scheme, and target lane-changing trajectory. If the convergence condition is not met, return to step S507 to continue optimization.

[0103] Vehicle platooning cooperative lane-changing strategies include direct game lane-changing mode and cooperative adjustment lane-changing mode, and their implementation process is as follows: S508, Target Lane Lane Changing Space Judgment.

[0104] Based on the predicted future trajectories of vehicles in the target lane obtained in step S3, calculate the available lane-changing space between vehicles in front of and behind the target lane: , in, Indicates the available clearance in the target lane; Indicates the predicted position of the vehicle ahead in the target lane; This indicates the predicted position of the vehicle behind in the target lane.

[0105] Simultaneously, based on the dynamic safety distance obtained in step S4 It determines whether the current traffic conditions meet the requirements for the overall lane-changing of the vehicle platoon.

[0106] S509, execute the direct game lane-changing strategy.

[0107] When the target lane meets the following conditions: At that time, the vehicle queue uses a direct lane-changing method. Among them, This indicates the total length of the vehicle queue.

[0108] In this mode, the vehicle queue adjusts its speed synchronously based on the game optimization results, and maintains the queue structure to complete the overall lane change.

[0109] S510 executes a coordinated lane-changing strategy.

[0110] When the target lane meets the following conditions: When this occurs, it indicates insufficient space in the target lane, and directly changing lanes poses a safety risk. At this point, the vehicle platoon enters a coordinated adjustment phase, adjusting the longitudinal speed of the vehicles in the platoon relative to their position in the target lane. The vehicle speed adjustment is expressed as: , in, Indicates the current speed; Indicates vehicle acceleration; Indicates the control cycle.

[0111] By proactively adjusting vehicle spacing, lane-changing space is created that meets safety constraints.

[0112] S511, complete the queue structure restoration.

[0113] After a vehicle enters the target lane, its status is readjusted according to the desired queue spacing.

[0114] The goal is ,in, Indicates the actual distance between vehicles; This indicates the desired vehicle spacing. The system eventually returns to a stable queuing state.

[0115] S6, Establish a vehicle trajectory tracking and control model.

[0116] Based on the target lane-changing trajectory obtained in step S5, the future control input of the vehicle is calculated using the model predictive control method. At the same time, a control obstacle function is introduced to limit the vehicle's motion safety boundary, so that the vehicle meets the collision avoidance constraint during the lane-changing process, thereby realizing safe and coordinated lane changing of intelligent connected vehicle platoons.

[0117] The model predictive control method for fusing the control barrier function in step S6 is as follows: S601, Establish a vehicle dynamics prediction model.

[0118] To achieve vehicle lane-changing trajectory tracking, a vehicle state model is established: ,in, y represents the vehicle's longitudinal position; v represents the vehicle's lateral position; θ represents the vehicle's speed; and θ represents the vehicle's heading angle.

[0119] Vehicle control inputs are represented as follows: , where a represents the vehicle acceleration; δ represents the front wheel steering angle.

[0120] The prediction of the future state of a vehicle is expressed as follows: .

[0121] S602, Establish model predictive control optimization objectives.

[0122] An optimization objective is established based on the deviation between the planned trajectory and the actual vehicle status: , in, This represents the predicted state of the vehicle at time k. Indicates the target reference trajectory; This represents the vehicle control input; N represents the prediction time range. Future vehicle control inputs are obtained through rolling optimization.

[0123] S603 introduces safety constraints for control barrier functions.

[0124] To ensure that safety requirements are always met during vehicle movement, a safety function is defined: Where d represents the actual distance of the vehicle; This indicates the minimum permissible safe distance.

[0125] The following must be met during vehicle operation: This ensures that vehicles do not enter areas with a risk of collision.

[0126] S604 outputs vehicle control commands.

[0127] Under the conditions of satisfying vehicle dynamics constraints and safety constraints, the optimal control quantity is obtained: The control signal is sent to the vehicle actuator to enable safe lane changing for intelligent connected vehicles.

[0128] To verify the effectiveness of the proposed intelligent connected vehicle platooning cooperative lane-changing method based on conditional diffusion trajectory prediction and uncertainty perception game optimization, a mixed traffic environment lane-changing experimental scenario was constructed. The experiment combined real vehicle trajectory data with a simulation environment. First, human-driven vehicle motion data was obtained using publicly available vehicle trajectory datasets to train the conditional diffusion trajectory prediction model, enabling the model to learn vehicle motion patterns under real driving conditions. Then, an intelligent connected vehicle platooning lane-changing scenario was established in an autonomous driving simulation platform to verify the prediction, decision-making, and control modules.

[0129] I. Establishment of the experimental scenario.

[0130] like Figure 2 As shown in Table 1, a mixed traffic environment for a two-way expressway is established. The road parameters are shown in Table 1.

[0131] Table 1 Mixed Traffic Environment Parameters .

[0132] CAV vehicles are controlled using the method of this invention; HDV vehicles are controlled using real driving data or random driving models.

[0133] II. Training of the Conditional Diffusion Trajectory Prediction Model

[0134] Enter the vehicle's past movement status (3 seconds).

[0135] Predicting the future: 5-second vehicle trajectories.

[0136] Input status includes: Vehicle location: speed: Acceleration: And the relationship with surrounding vehicles.

[0137] The diffusion model structure is as follows Figure 3 As shown in Table 2, the model training parameters are as follows.

[0138] Table 2 Diffusion Model Parameters .

[0139] After training, the model is used to predict the future behavior of vehicles in the target lane.

[0140] III. Verification of trajectory prediction performance.

[0141] The prediction performance was evaluated using the average displacement error (ADE) and the final displacement error (FDE).

[0142]

[0143]

[0144] Experimental results are as follows Figure 8 As shown in Table 3.

[0145] Table 3 Comparison of trajectory prediction results .

[0146] The results show that the conditional diffusion generation method of the present invention can more accurately describe the future motion trend of vehicles.

[0147] IV. Uncertainty-aware dynamic security constraint verification.

[0148] To verify the effect of the uncertainty perception method proposed in this invention on improving the safety of lane changing in vehicle platoons, human-driven vehicles with different driving behaviors were set up in a mixed traffic environment, and the performance of the fixed safety distance method and the dynamic safety distance method of this invention were compared.

[0149] During the experiment, vehicles in the target lane exhibited three typical driving behaviors:

[0150] The first type is stable driving vehicles, which maintain their current speed and have minimal future trajectory changes;

[0151] The second type is randomly driven vehicles, which exhibit random acceleration and deceleration behavior.

[0152] The third type is interactive driving vehicles, which actively adjust their speed based on the lane-changing behavior of intelligent connected vehicles in a platoon.

[0153] For the different vehicle behaviors mentioned above, multiple future candidate trajectories are generated through a conditional diffusion trajectory prediction model, and the degree of dispersion between the predicted trajectories is calculated.

[0154] The uncertainty of vehicle behavior is calculated as follows: , in, This indicates uncertainty in vehicle behavior; This represents the k-th predicted trajectory; This represents the mean of the predicted trajectory. The vehicle safety distance is adjusted based on the calculated uncertainty. .

[0155] The experimental results, comparing different methods for generating safe distances, are shown in Table 4. The uncertainty perception process is as follows: Figure 4 As shown.

[0156] Table 4 Experimental results of different safety distance adjustment methods .

[0157] As shown in Table 4, traditional fixed safety distance methods cannot adjust according to changes in the driving behavior of surrounding vehicles, and are prone to insufficient safety distance when facing randomly driven vehicles. This invention, by performing uncertainty analysis on the predicted trajectory, proactively increases the safety distance when the vehicle behavior risk is high, and reduces safety redundancy when the vehicle behavior is stable. Therefore, it can improve lane-changing efficiency of vehicle platoons while reducing collision risk.

[0158] V. Experimental verification of coordinated lane changing in intelligent connected vehicle platoons.

[0159] like Figure 5 As shown, to further verify the game-theoretic optimization lane-changing strategy proposed in this invention, a lane-changing scenario of a platoon of intelligent connected vehicles was constructed in a simulation environment. The experimental vehicle platoon consisted of 3 intelligent connected vehicles, and the target lane contained multiple human-driven vehicles. The initial state of the vehicle platoon is shown in Table 5.

[0160] Table 5 Initial parameters of intelligent connected vehicle platoon .

[0161] The experiment compared the traditional rule-based lane-changing method, the single-vehicle optimized lane-changing method, and the collaborative game-theoretic lane-changing method of this invention.

[0162] The evaluation indicators for lane changing include: the time required for the vehicle platoon to complete the lane change; the minimum distance between vehicles; changes in vehicle acceleration; and the maintenance of the platoon structure. The experimental results are shown in Table 6.

[0163] Table 6 Performance Comparison of Different Lane Changing Decision Methods .

[0164] The experimental results show that: Rule-based lane-changing methods only consider the current vehicle state and do not predict the future behavior of vehicles in the target lane, thus easily leading to conservative decisions. While single-vehicle optimization methods can improve the lane-changing performance of individual vehicles, they do not consider the overall coordination of the vehicle platoon. This invention establishes a game theory relationship between vehicles, enabling platoon vehicles to proactively adjust their speed and lane-changing order according to the traffic environment, thereby shortening lane-changing time while maintaining platoon stability. The two-stage lane-changing process is as follows: Figure 6 As shown. The three-dimensional trajectory changes are as follows. Figure 9 As shown.

[0165] VI. Experimental verification of CBF-MPC vehicle safety control.

[0166] like Figure 7 As shown, in order to verify the safety of the vehicle control layer, under the same lane-changing trajectory conditions, the traditional PID control, ordinary MPC control and the CBF-MPC control method adopted in this invention are compared.

[0167] The lateral error of a vehicle is defined as: ,in, Indicates the actual lateral position of the vehicle; Indicates the horizontal position of the reference trajectory.

[0168] Vehicle comfort is evaluated using the rate of change of acceleration. The experimental results are shown in Table 7.

[0169] Table 7 Comparison of Trajectory Tracking Performance of Different Control Methods .

[0170] As can be seen from Table 7, traditional PID control cannot fully consider future changes in the vehicle's state, resulting in a large trajectory error. MPC control reduces tracking error through predictive optimization, but still poses safety risks when faced with sudden vehicle behavior. This invention uses a control obstacle function to constrain the vehicle's safety boundary in real time, ensuring that the vehicle always meets safety conditions during trajectory tracking, while also improving the vehicle's operational stability.

[0171] VII. Comprehensive Experimental Analysis.

[0172] like Figure 10 As shown in the above experimental results, it can be seen that the intelligent connected vehicle platooning cooperative lane-changing method based on conditional diffusion trajectory prediction and uncertainty perception game optimization proposed in this invention can effectively solve the problems of unpredictable human driving behavior, insufficient vehicle coordination, and insufficient safety constraints in the process of intelligent connected vehicle platooning lane changing in mixed traffic environments.

[0173] By generating multiple possible future vehicle trajectories through a conditional diffusion model, the ability to predict vehicle behavior in complex traffic environments is improved; by establishing a dynamic safety distance through an uncertainty perception mechanism, the safety of vehicle lane-changing processes is improved; by coordinating vehicle platoon behavior through a game-theoretic optimization method, the efficiency of vehicle platoon lane-changing is improved; and by using a model predictive control method that integrates control obstacle functions, the safety and stability of vehicle execution processes are improved.

[0174] Experimental results demonstrate that this invention is applicable to intelligent connected vehicle platooning on highways, cooperative control of autonomous vehicles, and vehicle-road cooperative transportation systems.

[0175] Based on the same inventive concept as the above-described method embodiments, this embodiment provides an intelligent connected vehicle platoon cooperative lane-changing system based on conditional diffusion trajectory prediction and uncertainty perception game optimization, including:

[0176] The vehicle information acquisition module is used to acquire the operating status information of intelligent connected vehicle queues and surrounding human-driven vehicles in the target road environment within a continuous time range, construct the vehicle historical motion sequence, determine the vehicle interaction relationship based on the spatial distance and relative motion state between vehicles, and combine the vehicle historical motion sequence, vehicle interaction relationship and road information into a vehicle interaction dataset.

[0177] The trajectory prediction module, connected to the vehicle information acquisition module, is used to input the vehicle's historical motion sequence into the scene coding network, extract the motion features of the vehicle's own motion state, calculate the attention weight between the target vehicle and surrounding vehicles using an attention mechanism to obtain vehicle interaction features, fuse the motion features, interaction features and road information into a scene feature vector, and use the scene feature vector as a condition to generate multiple future candidate trajectories through multiple sampling processes of forward noise addition and reverse noise reduction of the diffusion model.

[0178] An uncertainty perception module, connected to the trajectory prediction module, is used to calculate the uncertainty of the vehicle's future behavior based on the distribution differences between the multiple predicted trajectories, and dynamically adjust the lane-changing safety distance according to the magnitude of the uncertainty.

[0179] The collaborative lane-changing decision module, connected to the uncertainty perception module, is used to take the intelligent connected vehicles in the vehicle queue as game participants based on the safe distance, and establish a comprehensive lane-changing optimization function by comprehensively considering driving safety, driving efficiency, acceleration changes and queue integrity. Through game optimization, the speed adjustment strategy and lane-changing execution time of each vehicle are obtained, and the target lane-changing trajectory is generated.

[0180] The vehicle safety control module, connected to the cooperative lane-changing decision module, is used to calculate the future control input of the vehicle based on the obtained target lane-changing trajectory using model predictive control methods, and at the same time introduces a control obstacle function to limit the vehicle's motion safety boundary, thereby completing the cooperative lane-changing of the intelligent connected vehicle platoon.

[0181] The execution methods of each module of the system described in this embodiment correspond one-to-one with the steps of the aforementioned method embodiments. For details, please refer to the descriptions in the aforementioned method embodiments, which will not be repeated here.

[0182] In summary, this invention discloses a method and system for cooperative lane changing in intelligent connected vehicle platoons based on conditional diffusion trajectory prediction and uncertainty perception game optimization, belonging to the field of intelligent connected vehicle cooperative driving technology. First, continuous operational status information of the intelligent connected vehicle platoon and surrounding traffic-participating vehicles in a mixed traffic environment is acquired, and a vehicle interaction scenario model is constructed based on vehicle historical trajectories, motion states, and spatial positional relationships. Then, a conditional vehicle motion diffusion model is used to predict the future motion behavior of traffic-participating vehicles. The vehicle trajectory distribution is learned through the forward noise addition and reverse denoising process of the diffusion model, thereby generating future candidate trajectories with multiple driving intentions. Further, the vehicle prediction uncertainty is calculated based on the distribution differences of the generated trajectories in the potential space, and the safety constraints during the lane changing process of the vehicle platoon are dynamically adjusted according to the magnitude of the uncertainty. Subsequently, an intelligent connected vehicle platoon cooperative game decision-making model considering driving safety, traffic efficiency, ride comfort, and platoon stability is established to optimize speed adjustment, lane changing timing, and vehicle cooperative behavior during the lane changing process. Finally, a model predictive control method incorporating control obstacle functions is used to constrain and optimize the lateral and longitudinal motion of vehicles, achieving safe, efficient, and stable lane changing for intelligent connected vehicle platoons.

[0183] This invention addresses the challenges of highly random human driving behavior in mixed-driving environments, difficulty in determining safe lane-changing gaps in vehicle platoons, and the difficulty of balancing safety and efficiency using traditional methods, thereby improving the adaptability of intelligent connected vehicle platoons in complex traffic environments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle platoon cooperative lane-changing method based on trajectory prediction and game theory optimization, characterized in that, include: Based on the intelligent connected vehicle queue and the surrounding human-driven vehicles' operating status information within a continuous time range in the target road environment, a vehicle historical motion sequence is constructed. The vehicle interaction relationship is determined based on the spatial distance and relative motion state between vehicles. The vehicle historical motion sequence, vehicle interaction relationship and road information are combined into a vehicle interaction dataset. The vehicle's historical motion sequence is input into a scene coding network to extract the motion features of the vehicle's own motion state. The attention weights between the target vehicle and surrounding vehicles are calculated using an attention mechanism to obtain vehicle interaction features; the motion features, interaction features, and road information are then fused into a scene feature vector to describe the current traffic state. The generation of the scene feature vector further includes: inputting the vehicle's historical motion sequence into a scene coding network, extracting vehicle speed changes, acceleration changes, and driving trend information as vehicle motion features; constructing a query vector based on the target vehicle's motion state, constructing a key vector based on the motion states of surrounding vehicles, calculating the attention weights between the target vehicle and each surrounding vehicle through an attention mechanism, and obtaining vehicle interaction features by weighted summation; and fusing the motion features, interaction features, and road information to generate a scene feature vector describing the current traffic state. Using the scene feature vector as a condition, random noise is gradually added to the real vehicle trajectory through a forward diffusion process to convert the vehicle trajectory into a random distribution. The future motion trajectory of the vehicle is gradually recovered using a reverse denoising network, and multiple future predicted trajectories are generated through multiple samplings. The uncertainty of future vehicle behavior is calculated based on the distribution differences among multiple predicted trajectories, and the lane-changing safety distance is dynamically adjusted according to the magnitude of the uncertainty. Based on the aforementioned safe distance, the intelligent connected vehicles in the vehicle queue are taken as the participants in the game. A comprehensive lane-changing optimization function is established by comprehensively considering driving safety, driving efficiency, acceleration changes and queue integrity. The speed adjustment strategy and lane-changing execution time of each vehicle are obtained through game optimization, and the target lane-changing trajectory is generated. Based on the obtained target lane-changing trajectory, the model predictive control method is used to calculate the future control input of the vehicle. At the same time, a control obstacle function is introduced to limit the vehicle's motion safety boundary, thus completing the intelligent connected vehicle platoon cooperative lane-changing.

2. The vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, The formation of the vehicle interaction dataset also includes: The system acquires the operational status information of intelligent connected vehicle platoons and surrounding human-driven vehicles in a continuous time window within a target road environment through vehicle-mounted sensing devices, vehicle-to-everything (V2X) communication devices, or vehicle trajectory datasets. The operational status information includes the vehicle's longitudinal position, lateral position, speed, acceleration, and heading angle. The continuous motion state of the vehicle is obtained according to a fixed sampling period, and a historical motion sequence of the vehicle containing the continuous motion state of each vehicle within a preset time period is constructed. Calculate the distance between vehicles, and determine that there is an interaction between vehicles when the distance is less than the perception range; The vehicle historical motion sequence, vehicle interaction relationship matrix and road information are combined into a vehicle interaction dataset.

3. The vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, The generation of future motion trajectories also includes: During the model training phase, random Gaussian noise is gradually added to the real vehicle trajectory to convert it into a random noise distribution, so as to learn the evolution law of the vehicle trajectory from the real motion state to the random distribution state. In the trajectory prediction stage, random noise trajectory is used as the initial input, and the scene feature vector is used as the condition constraint. Noise is gradually removed through a parameterized denoising network to recover the future trajectory of the vehicle. By sampling the conditional diffusion model multiple times, multiple future prediction trajectories are generated, each corresponding to a different driving behavior pattern, in order to describe the uncertainty of human driving behavior.

4. The vehicle queuing cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, The uncertainty of future vehicle behavior is calculated based on the distribution differences among multiple predicted trajectories, and the lane-changing safety distance is dynamically adjusted according to the magnitude of the uncertainty, including: Calculate the average value of the predicted trajectories based on the multiple predicted trajectories output by the conditional diffusion model: ; in, This represents the mean of the predicted trajectory. Indicates the total number of predicted trajectories generated; Indicates the first Predicted trajectory; The uncertainty in predicting future vehicle behavior is calculated based on the degree of deviation between different predicted trajectories: ; in, This indicates the uncertainty in predicting the future trajectory of a vehicle; Establish a dynamic safety distance model based on prediction uncertainties: ; in, Indicates dynamic safety distance. Indicates the basic safe distance between vehicles. This represents the risk adjustment factor.

5. The vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, Generating the target lane change trajectory also includes: The intelligent connected vehicle platoon performing the lane-changing task and the vehicles affecting the target lane are all treated as participants in the traffic game. Each vehicle selects a control strategy that includes longitudinal acceleration and steering angle based on the predicted trajectory of surrounding vehicles and its own driving goal. Establish security costs based on dynamic security distances Establish efficiency costs based on the difference between the vehicle's actual speed and the expected speed. Establish comfort costs based on the rate of change of acceleration Establish queue maintenance cost based on the deviation between vehicle spacing and desired queue spacing. ; Taking into account driving safety, driving efficiency, acceleration changes, and queue integrity, a comprehensive lane-changing optimization function is established: ; in, These represent the weights of different evaluation objectives; The optimal cooperative driving strategy for the vehicle platoon is obtained by minimizing the comprehensive lane-changing optimization function, and the target lane-changing trajectory is generated.

6. The vehicle queuing cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 5, characterized in that, The speed adjustment strategies and lane-changing execution times for each vehicle are obtained through game theory optimization, including: Based on the current traffic environment and vehicle queue operation requirements, an initial driving strategy is generated for the vehicles participating in the game. In each iteration, the driving strategies of all vehicles except the target vehicle are fixed, and the control strategy of the target vehicle is optimized according to the comprehensive lane-changing optimization function. After completing a single optimization for all vehicles, the vehicle queue strategy set is updated, and the updated strategy is re-input into the game model for the next round of vehicle interaction calculation. Calculate the change between two adjacent iterations. When the change is less than the set convergence threshold, determine that the vehicles have reached a stable interaction state and output the vehicle queue cooperative lane-changing strategy.

7. The vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, After generating the target lane change trajectory, the following is also included: Based on the predicted future trajectories of vehicles in the target lane, the available lane-changing space between vehicles in front and behind the target lane is calculated, and the dynamic safety distance is combined to determine whether the current traffic environment meets the overall lane-changing requirements of the vehicle platoon. When the available lane-changing space is greater than the sum of the total length of the vehicle queue and the dynamic safety distance, the vehicle queue adopts a direct lane-changing method, and adjusts its speed synchronously according to the game optimization results while maintaining the queue structure to complete the overall lane-changing. When the available lane-changing space is less than the sum of the total length of the vehicle queue and the dynamic safety distance, the vehicle queue enters the coordinated adjustment phase. By changing the longitudinal speed adjustment of the vehicles in the queue and the positional relationship between the vehicles in the target lane, lane-changing space that meets safety constraints is actively created. After the vehicles enter the target lane, their status is readjusted according to the desired queue spacing to restore a stable queue driving state.

8. The vehicle platooning cooperative lane-changing method based on trajectory prediction and game theory optimization according to claim 1, characterized in that, Completing coordinated lane changing for intelligent connected vehicles also includes: Establish a vehicle state model that includes the vehicle's longitudinal position, lateral position, speed, and heading angle, as well as a vehicle control input model that includes the vehicle's acceleration and front wheel steering angle, to predict the vehicle's future state. A model is established to predict control optimization objectives based on the deviation between the planned trajectory and the actual state of the vehicle, and the future control input of the vehicle is obtained through rolling optimization. Introduce a control barrier function to define a safety function: ； in, Indicates the actual distance of the vehicle. This indicates the minimum permissible safe distance, which must be met during vehicle operation. ; The optimal control quantity is obtained under the conditions of satisfying vehicle dynamics constraints and safety constraints, and the control quantity is sent to the vehicle actuator to complete the coordinated lane changing of the intelligent connected vehicle platoon.

9. A vehicle platooning cooperative lane-changing system based on trajectory prediction and game theory optimization, characterized in that, include: The vehicle information collection module is used to construct a vehicle historical motion sequence based on the operating status information of intelligent connected vehicle queues and surrounding human-driven vehicles in the target road environment within a continuous time range, and to determine the vehicle interaction relationship based on the spatial distance and relative motion state between vehicles. The vehicle historical motion sequence, vehicle interaction relationship and road information are combined into a vehicle interaction dataset. The trajectory prediction module, connected to the vehicle information acquisition module, is used to input the vehicle's historical motion sequence into the scene coding network, extract the motion features of the vehicle's own motion state, calculate the attention weight between the target vehicle and surrounding vehicles using an attention mechanism to obtain vehicle interaction features, fuse the motion features, interaction features and road information into a scene feature vector, and use the scene feature vector as a condition to generate multiple future predicted trajectories through multiple sampling processes of forward noise addition and reverse noise reduction of the diffusion model. The generation of the scene feature vector further includes: inputting the vehicle's historical motion sequence into a scene coding network, extracting vehicle speed changes, acceleration changes, and driving trend information as vehicle motion features; constructing a query vector based on the target vehicle's motion state, constructing a key vector based on the motion states of surrounding vehicles, calculating the attention weights between the target vehicle and each surrounding vehicle through an attention mechanism, and obtaining vehicle interaction features by weighted summation; and fusing the motion features, interaction features, and road information to generate a scene feature vector describing the current traffic state. An uncertainty perception module, connected to the trajectory prediction module, is used to calculate the uncertainty of the vehicle's future behavior based on the distribution differences between multiple predicted trajectories, and dynamically adjust the lane-changing safety distance according to the magnitude of the uncertainty. The collaborative lane-changing decision module, connected to the uncertainty perception module, is used to take the intelligent connected vehicles in the vehicle queue as game participants based on the safe distance, and establish a comprehensive lane-changing optimization function by comprehensively considering driving safety, driving efficiency, acceleration changes and queue integrity. Through game optimization, the speed adjustment strategy and lane-changing execution time of each vehicle are obtained, and the target lane-changing trajectory is generated. The vehicle safety control module, connected to the cooperative lane-changing decision module, is used to calculate the future control input of the vehicle based on the obtained target lane-changing trajectory using model predictive control methods, and at the same time introduces a control obstacle function to limit the vehicle's motion safety boundary, thereby completing the cooperative lane-changing of the intelligent connected vehicle platoon.

Images