Search

CN-121989939-A - Automatic lane change control method and system for vehicle and vehicle

CN121989939ACN 121989939 ACN121989939 ACN 121989939ACN-121989939-A

Abstract

The invention belongs to the technical field of vehicle lane change control, and provides a vehicle automatic lane change control method, a system and a vehicle, wherein a deep reinforcement learning model is adopted by the vehicle, the current state of a target vehicle, lane line information, surrounding vehicle states and environment information are taken as inputs, the current lane change intention and a target lane of the target vehicle are output, the decision robustness is improved in a complex scene, a lane change track is re-planned according to real-time surrounding environment changes in the lane change process, and the feasibility of the track under different vehicle speeds or different road conditions is ensured.

Inventors

  • YE YUE
  • GUO FEN

Assignees

  • 奇瑞新能源汽车股份有限公司

Dates

Publication Date
20260508
Application Date
20260122

Claims (10)

  1. 1. An automatic lane change control method for a vehicle, comprising: Acquiring the current state, lane line information, surrounding vehicle states and environment information of a target vehicle; Based on a deep reinforcement learning model, taking the fused vehicle state, lane line information, surrounding vehicle state and environment information as input, and optimizing a decision strategy through a reward function, and learning and outputting the current lane changing intention and a target lane of a target vehicle, wherein the reward function consists of safe rewards, efficiency rewards, comfortable rewards and successful rewards; And generating a reference lane change track according to the lane change intention, controlling the vehicle to execute lane change based on the reference lane change track, re-planning the lane change track according to real-time surrounding environment change in the lane change process of the vehicle, and controlling the lane change track after the vehicle executes re-planning.
  2. 2. The automatic lane change control method for a vehicle according to claim 1, wherein the safety rewards give forward rewards, zero rewards or strong penalties according to the distance between a target vehicle and surrounding vehicles, the efficiency rewards are that lane change is completed within a set first time under the premise of rewarding the safety of the vehicle, the punishment speed is lower than the low-efficiency lane change of the set speed, or the long-term low-speed following of the lane change is performed at a speed lower than the set speed and the lane change time is longer than the set time, the comfort rewards are determined based on transverse movement smoothness, and the success rewards give forward rewards or punishment according to the result of lane change success or lane change interruption or lane change failure.
  3. 3. The automatic lane-changing control method of claim 1, wherein the deep reinforcement learning model is trained by a near-end strategy optimization algorithm, and a clipping objective function of the near-end strategy optimization algorithm is as follows: Wherein, the A security risk value for the current state; Is a security risk threshold; is a safety punishment coefficient; The desire for time step t; Policy ratio; Estimating the dominance of the time step t; is a clipping function of the PPO; Parameters are tailored for PPO.
  4. 4. The automatic lane-change control method according to claim 1, wherein the reference lane-change trajectory is generated according to the lane-change intention, specifically: Generating an initial reference lane change track through a polynomial curve according to the lane change intention, wherein the lane change start time and the target lane travel time are taken as key nodes, and boundary conditions for generating the initial reference lane change track are set; and enabling the generated initial reference lane change track to meet vehicle dynamics constraint, safety constraint and traffic rule constraint, and obtaining the optimized reference lane change track.
  5. 5. The automatic lane-changing control method according to claim 1, wherein the lane-changing control method is characterized by controlling the vehicle to execute lane-changing based on the reference lane-changing track, and re-planning the lane-changing track according to real-time ambient environment changes during lane-changing of the vehicle, and controlling the lane-changing track after the vehicle executes re-planning, specifically comprising: And the self-adaptive model predictive control is used for tracking a reference lane change track to obtain a vehicle optimal control sequence, wherein the self-adaptive model predictive control takes a minimized track tracking error as an objective function, and takes vehicle dynamics constraint, control quantity constraint and safe distance constraint as constraint conditions to solve, so that the vehicle optimal control sequence is obtained.
  6. 6. The method of claim 5, wherein the adaptive model predictive control parameters are adjusted by real-time road surface adhesion coefficients, and the weight matrix for balancing track tracking errors and control quantity smoothness is adjusted according to the vehicle speed.
  7. 7. The automatic lane-change control method as claimed in any one of claims 1 to 6, further comprising constructing a probability generation model of driving behavior by a hierarchical probability model based on the driver history lane-change behavior data, modifying the driver style probability based on the probability generation model of driving behavior by a bayesian filtering or a variational reasoning technique, and converting the modified driver style probability into an adjustment of a weight matrix in an adaptive model predictive control objective function or an a priori distribution of a deep reinforcement learning model decision strategy.
  8. 8. A vehicle, characterized in that lane change is performed by using a vehicle automatic lane change control method according to any one of claims 1 to 7.
  9. 9. An automatic lane-changing control system for a vehicle, comprising: The acquisition module is configured to acquire the current state, lane line information, surrounding vehicle states and environment information of the target vehicle; The prediction module is configured to learn and output the current lane changing intention of the target vehicle and the target lane by taking the fused vehicle state, lane line information, surrounding vehicle state and environment information as inputs and optimizing a decision strategy through a reward function based on a deep reinforcement learning model, wherein the reward function consists of safe rewards, efficiency rewards, comfortable rewards and successful rewards; The lane change control module is configured to generate a reference lane change track according to the lane change intention, control the vehicle to execute lane change based on the reference lane change track, re-plan the lane change track according to real-time ambient environment change in the lane change process of the vehicle, and control the lane change track after the vehicle executes re-planning.
  10. 10. An electronic device comprising a memory and a processor and computer instructions stored on the memory and running on the processor, which when executed by the processor, perform the method of any one of claims 1-7.

Description

Automatic lane change control method and system for vehicle and vehicle Technical Field The invention belongs to the technical field of vehicle lane change control, and particularly relates to a vehicle automatic lane change control method and system and a vehicle. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. With the evolution and large-scale application of automatic driving technology, the automatic lane changing function of a vehicle has become one of the core modules of an advanced driving assistance system. The traditional automatic lane changing technology still has key technical bottlenecks in complex scene application, namely firstly, a traditional decision system based on a fixed rule or a shallow machine learning model is easy to lose effectiveness, multiple target vehicle motion trends and environment dynamic changes cannot be accurately weighed, lane changing time misjudgment, process interruption or collision risk promotion are caused, secondly, a transverse control strategy is difficult to dynamically adapt to working conditions such as vehicle speed change and the like, vehicle transverse shaking and track tracking deviation are easy to exceed standards, thirdly, in a multiple target vehicle dynamic interaction scene, the traditional algorithm is high in computational complexity, data processing and instruction response delay are easy to generate, lane changing time is missed, fourthly, unified control parameters and decision logic are adopted, driver style preference is not considered, output and expected deviation are caused, and user trust and use experience are reduced. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides the automatic lane changing control method and system for the vehicle and the vehicle, which realize the high safety, strong working condition adaptability, efficiency and comfort of the automatic lane changing of the vehicle under a complex scene, and have high engineering practicability and easy landing. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect, the present invention provides a vehicle automatic lane change control method, including: Acquiring the current state, lane line information, surrounding vehicle states and environment information of a target vehicle; Based on a deep reinforcement learning model, taking the fused vehicle state, lane line information, surrounding vehicle state and environment information as input, and optimizing a decision strategy through a reward function, and learning and outputting the current lane changing intention and a target lane of a target vehicle, wherein the reward function consists of safe rewards, efficiency rewards, comfortable rewards and successful rewards; And generating a reference lane change track according to the lane change intention, controlling the vehicle to execute lane change based on the reference lane change track, re-planning the lane change track according to real-time surrounding environment change in the lane change process of the vehicle, and controlling the lane change track after the vehicle executes re-planning. The method comprises the steps of determining a safe reward, wherein the safe reward is a forward reward, zero reward or strong punishment according to the distance between a target vehicle and surrounding vehicles, the efficient reward is that lane changing is completed in a set first time under the premise of rewarding vehicle safety, punishment speed is lower than low-efficiency lane changing of the set speed, or long-term low-speed car following with speed lower than the set speed and lane changing time longer than the set time is achieved, the comfortable reward is determined based on transverse movement smoothness, and the successful reward is that the forward reward or punishment is given according to the result of lane changing success or lane changing interruption or lane changing failure. Further, training the deep reinforcement learning model by adopting a near-end strategy optimization algorithm, wherein a cutting objective function of the near-end strategy optimization algorithm is as follows: Wherein, the A security risk value for the current state; Is a security risk threshold; is a safety punishment coefficient; The desire for time step t; Policy ratio; Estimating the dominance of the time step t; is a clipping function of the PPO; Parameters are tailored for PPO. Further, generating a reference lane change track according to the lane change intention, specifically: Generating an initial reference lane change track through a polynomial curve according to the lane change intention, wherein the lane change start time and the target lane travel time are taken as key nodes, and boundary conditions for generating the initial reference lane change tr