An adaptive cruise safety self-evolution method, device, and system based on safety boundaries

Abstract

This invention belongs to the field of vehicle driving control technology, specifically a method, device, and system for adaptive cruise safety self-evolution based on safety boundaries. The method includes collecting and preprocessing vehicle driving data and environmental information to obtain a time-series dataset containing the vehicle's state, control inputs, and corresponding reward values ​​at each time step. Then, a safe distance between the vehicle and surrounding obstacles is dynamically evaluated using a control obstacle function to determine the safe zone, while simultaneously constructing and updating the vehicle's safe driving boundary function. The MPC controller uses the safe driving boundary function as a constraint, aiming to smoothly approach the target speed and maintain a safe following distance, to obtain a control input sequence. The vehicle executes corresponding operations according to the control parameters, and the weight matrix of the MPC is continuously adjusted through an online optimization algorithm to progressively optimize the control parameters. This invention achieves adaptive adjustment of MPC parameters to improve the vehicle's safety and adaptability in complex traffic environments.

Description

Technical Field

[0001] This invention belongs to the field of vehicle driving control technology, specifically relating to an adaptive cruise safety self-evolution method, device, and system based on safety boundaries. Background Technology

[0002] Adaptive cruise control, as an important component of intelligent driving, provides a more comfortable and safer driving experience by adjusting vehicle speed and following distance. However, existing adaptive cruise control systems still have limitations in complex traffic environments, especially when facing unexpected traffic situations, and the safety and reliability of the system still need further improvement.

[0003] To address the challenges of adaptive cruise control in real-world environments, a self-optimizing module with a data closed loop is needed. By continuously collecting and analyzing real-time data, the system can progressively adjust its safety boundaries, control strategies, and perception algorithms. The core of this process lies in the continuous accumulation and learning of data, allowing the system to adapt and optimize under different driving conditions. This data closed loop not only enhances the safety of intelligent driving systems but also enables their self-evolutionary capabilities in complex traffic scenarios. Furthermore, with the continuous development of intelligent driving technology, safety has become a crucial research direction. Through the data closed loop, the system can promptly identify potential risks and make corresponding adjustments, thereby maintaining a high level of safety performance under various environmental conditions.

[0004] Against this backdrop, several new technical solutions have emerged. Patent CN116300850A proposes a self-evolving method and system for human-like safety in autonomous driving based on data mechanism fusion. This method extracts real human driving characteristics to construct a driver's decision-making and planning model. However, this method may face some limitations in practical applications, including a high dependence on historical data and insufficient adaptability to complex traffic scenarios, making it less flexible in dynamic and changing driving environments. Patent CN117609099A provides a general self-evolving method and system for autonomous driving algorithms. Although it improves algorithm performance through adversarial scenario generation and interactive training, it may still encounter overfitting problems in complex environments; that is, the algorithm performs well in specific scenarios but cannot effectively cope with other unknown scenarios. Therefore, while it improves the intelligence and performance boundaries of the algorithm, it still cannot guarantee reliability in various dynamic and open traffic environments. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive cruise safety self-evolution method, device, and system based on safety boundaries. By combining the control obstacle function with reinforcement learning, and through real-time data feedback and dynamic optimization of the control strategy, the controller parameters of the model predictive control (MPC) are adaptively adjusted, thereby improving the safety and adaptability of the system in complex traffic environments.

[0006] To achieve the above objectives, this invention provides an adaptive cruise safety self-evolution method based on safety boundaries, comprising the following steps:

[0007] S1. Real-time collection and preprocessing of vehicle driving data and environmental information to obtain the vehicle's state x at each time step. t , control input sequence u t and the corresponding reward value R t Time series dataset D;

[0008] S2. Based on the time series dataset D, dynamically evaluate the safe distance between the vehicle and surrounding obstacles by controlling the obstacle function, determine the safe area, and at the same time construct and update the vehicle's safe driving boundary function h(x);

[0009] The S3 and MPC controllers use the safe driving boundary function h(x) as constraints and aim to smoothly approach the target speed and maintain a safe following distance, thus obtaining the control input sequence u. t ;

[0010] S4. The vehicle performs corresponding operations according to the control parameters. During the execution, an online optimization algorithm is used to maximize the reward value R. t The long-term cumulative value is the target, and the weight matrix of the MPC controller is continuously adjusted to gradually optimize the control parameters.

[0011] Furthermore, in step S1, the vehicle driving data includes the current vehicle speed v. t acceleration a t Distance d from the vehicle in front t The environmental information, including traffic signals, road type, road condition, traffic flow density, weather conditions, and the distribution and movement of surrounding dynamic obstacles, provides precise input for vehicle safety boundary modeling through comprehensive collection and analysis of this data, thereby enabling dynamic optimization of adaptive cruise control strategies and improved safety.

[0012] Furthermore, in step S2, the method for determining the safe area includes: firstly, constructing an initial control barrier function using an initial safety controller; then, further optimizing the construction of the safety controller based on the initial control barrier function by determining whether the vehicle exceeds the lane boundary line; and then regenerating a new control barrier function. Next, the system checks whether the current safe area can continue to expand. If so, the process is repeated to increase the safe area; if not, the expansion is stopped, and the final control barrier function result is output to ensure that the system maximizes the safe area while satisfying the safety boundary.

[0013] Furthermore, the constraint condition refers to the safe driving boundary function h(x)≥0; wherein, when the vehicle is within the safe area, h(x)≥0, and when it is not within the safe area, h(x)<0.

[0014] Furthermore, in step S3, the objective function of the MPC controller is:

[0015]

[0016] Among them, L(x(t),u(t))=‖x(t)-x target || 2 +R‖u(t)‖ 2 Let x be the stage cost function, representing the deviation between the vehicle's current state and the target state; target The expected cruise state preferably includes target speed and safe distance; R is the weight matrix of the control input; u(t) is the control input sequence;

[0017] Φ(x(t+N))=Q‖x(t+N)-x goal || 2 Let x be the terminal cost function, used to evaluate the final state of the system at the end of the prediction time domain; where x goal The ideal final state is represented by Q, which preferably includes maintaining a safe distance from the vehicle in front and the desired speed; Q is the weight matrix of the state vector, used to balance the importance of the final state.

[0018] Furthermore, in step S4, the online optimization algorithm employs deep Q-learning, based on the reward value R at each time step. t The long-term cumulative value of the reward is obtained, and the training is carried out with the goal of maximizing the long-term cumulative value of the reward so that the control parameters gradually converge to the optimal policy.

[0019] Among them, the reward value R t The formula is as follows:

[0020]

[0021] Furthermore, the state space S = {x} of the deep Q-learning t ,u t ,E}, action space A={u min ,u max}, where E represents information about the surrounding environment; u min U represents the minimum speed. max This represents the maximum speed; the iterative update rules used in the deep Q-learning are as follows:

[0022]

[0023] Among them, s t s represents the state at the current moment. t+1 Let a represent the state at the next moment. t This represents the action taken at the current moment, where R refers to the immediate reward, α is the learning rate, and γ is the discount factor.

[0024] The present invention also provides an adaptive cruise safety self-evolution device based on safety boundaries, employing any one of the adaptive cruise safety self-evolution methods described above, including:

[0025] The real-time data feedback and learning module is used to organize vehicle driving data and environmental information into a time series dataset D, and to continuously adjust the weight matrix of the MPC controller through an online optimization algorithm so that the control parameters are gradually optimized.

[0026] The safety boundary calculation module is used to dynamically evaluate the safe distance between the vehicle and surrounding obstacles by controlling the obstacle function based on the time series dataset D, and at the same time construct and update the vehicle's safe driving boundary function h(x);

[0027] The MPC controller module is used to transmit control parameters to the safety boundary calculation module for correction based on the current vehicle state and the prediction of the future state to obtain safe control parameters.

[0028] The execution module is used to execute the safe control parameters obtained by the MPC controller module to adjust the actual driving state of the vehicle.

[0029] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the adaptive cruise safety self-evolution method described in any of the above embodiments.

[0030] The present invention also provides an adaptive cruise safety self-evolution system based on safety boundaries, including a computer-readable storage medium and a processor;

[0031] The computer-readable storage medium is used to store executable instructions;

[0032] The processor is configured to read executable instructions stored in the computer-readable storage medium and execute any of the methods described above.

[0033] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0034] 1. The adaptive cruise safety self-evolution method based on safety boundaries provided by this invention addresses the shortcomings of existing adaptive cruise control systems in terms of safety and adaptability in dynamic traffic environments. It combines a control obstacle function with reinforcement learning, and through real-time data feedback and dynamic optimization of the control strategy, achieves adaptive adjustment of the model predictive control (MPC) controller parameters, thereby improving the vehicle's safety and adaptability in complex traffic environments. Compared with existing technologies, this invention utilizes real-time data feedback to continuously optimize and adjust the algorithm, enabling it to have higher adaptability in complex and changing traffic scenarios.

[0035] 2. This invention emphasizes the safety of the method. It calculates the safety boundary using the CBF (Control Barrier Function) to assess the vehicle's safety status in real time. The defined safety boundary function h(x) ensures the vehicle remains within a safe driving range, avoiding potential collision risks. By receiving and correcting the control parameters from the MPC (Multi-Process Control) system, the safety of the control parameters is ensured. Reinforcement learning is combined to optimize the MPC control parameters. Through continuous iteration and optimization, a rapid and safe response can be made in the face of emergencies, reducing accident risks and achieving intelligent control of the vehicle's driving state, ensuring safety under various driving conditions.

[0036] 3. The MPC controller of this invention is based on the vehicle's dynamic model, combined with safety boundary constraints and target state. It generates cruise commands by solving for the optimal control input u(t). The MPC controller uses stage cost functions and terminal cost functions to generate the control input sequence u(t), ensuring the vehicle smoothly approaches the target state while satisfying safety boundary constraints. Simultaneously, a real-time data feedback and learning module updates newly collected data to the historical dataset, forming a complete optimization chain and driving MPC parameters to adapt to dynamic environmental changes. Furthermore, it integrates a reinforcement learning (RL) algorithm, analyzing real-time feedback data to optimize the MPC controller's parameter settings, enabling the system to adapt to different traffic environments and driving conditions, continuously improving the safety and adaptability of the adaptive cruise system.

[0037] 4. The key to this invention lies in the data closed loop of the real-time data feedback and learning module. By continuously collecting and analyzing real-time data, the vehicle can quickly make safe responses under different driving conditions, reducing the risk of accidents. Through this module, this invention can effectively improve the reliability and safety performance of intelligent driving systems, providing a solid theoretical foundation and practical guarantee for the future development of autonomous driving technology. Attached Figure Description

[0038] Figure 1 A flowchart of adaptive cruise safety self-evolution based on safety boundaries provided by this invention;

[0039] Figure 2 The flowchart for calculating the safety boundary. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0041] Please see Figure 1 and 2 This invention provides a safety-based self-evolutionary method for adaptive cruise control, aiming to address the shortcomings in safety and adaptability of existing adaptive cruise systems in dynamic traffic environments. This method calculates the safety boundary by introducing a control barrier function (CBF) and combines reinforcement learning (RL) to optimize model predictive control (MPC) parameters, thereby ensuring safe and efficient adaptive cruise control under complex and variable driving conditions.

[0042] To achieve self-evolving safety in adaptive cruise control, this invention first provides an adaptive cruise control self-evolving safety system, mainly composed of four modules: a real-time data feedback and learning module, a safety boundary calculation module, an MPC controller, and an execution module. When the system enters adaptive cruise control mode, the real-time data feedback and learning module first collects real-time vehicle driving data (including speed, acceleration, distance to the vehicle ahead, traffic signals, etc.) and environmental information through sensors on the vehicle (such as radar, cameras, lidar, GPS, etc.). This data is organized into a time-series dataset D, containing the vehicle's state x at each time step. t Control input u t and the corresponding reward value R tThis data serves as input to the reinforcement learning (RL) algorithm. Simultaneously, the preprocessed data is input to the safety boundary calculation module, and the trained MPC control parameters, along with the preprocessed data, are transmitted to the MPC controller.

[0043] The safety boundary calculation module utilizes the Control Barrier Function (CBF) theory to dynamically assess the safe distance between the vehicle and surrounding obstacles based on received preprocessed data, constructing and updating the vehicle's safe driving boundary function h(x). This safety boundary function ensures that the vehicle always remains within a safe range during operation, avoiding potential collision risks. The safety boundary calculation module is also responsible for receiving and correcting the control parameters from the MPC to ensure their safety.

[0044] Subsequently, based on the current vehicle state and predictions of future states, the MPC controller, combined with safety boundary constraints (i.e., the control parameters are transmitted to the safety boundary calculation module for correction) and the target state, generates cruise commands by solving for the optimal control input u(t). The MPC controller uses stage cost functions and terminal cost functions to generate the control input sequence u(t), and ensures that the vehicle smoothly approaches the target speed and following distance while satisfying the safety boundary constraints. These control commands are then sent to the execution module, where they are executed by the vehicle's actuators (such as the accelerator and braking system) to adjust the vehicle's actual driving state.

[0045] During control command execution, the real-time data feedback and learning module continuously receives the vehicle's latest driving status. Utilizing this feedback, the module adjusts the weight matrices R and Q in the MPC controller through an online optimization algorithm, enabling the control strategy to better adapt to different traffic environments and driving conditions. With continuous algorithm iteration and updates, the real-time data feedback and learning module drives the MPC parameters to gradually converge to the optimal configuration, enhancing the system's adaptability in complex dynamic environments and ensuring safe evolution.

[0046] In MPC, the optimal control input sequence is calculated over a finite future timeframe by optimizing the objective function. At the current moment, based on the system state and control input, MPC uses a model to predict the system response over a future period, known as the prediction time domain or prediction interval. The weight matrices Q and R are key parameters in the MPC cost function, representing the weights on the state variables and control input, respectively, used to adjust control performance and robustness. In practical applications, the determination of the Q and R matrices typically depends on the system model and design requirements; optimization algorithms can be used to adjust the matrix weights to achieve the desired control performance.

[0047] The motion control algorithm of this invention provides a self-evolving module through dynamic adjustment of safety boundaries and real-time optimization of MPC, enabling the algorithm to adapt to different driving environments and traffic conditions. During adaptive cruise control, the algorithm can not only learn from historical data but also provide real-time feedback based on current traffic conditions to continuously improve the system's safety and responsiveness.

[0048] The adaptive cruise safety self-evolution method provided by this invention includes the following steps:

[0049] S1. Real-time collection and preprocessing of vehicle driving data and environmental information to obtain the vehicle's state x at each time step. t , control input sequence u t and the corresponding reward value R t Time series dataset D;

[0050] S2. Based on the time series dataset D, dynamically evaluate the safe distance between the vehicle and surrounding obstacles by controlling the obstacle function, determine the safe area, and at the same time construct and update the vehicle's safe driving boundary function h(x);

[0051] The S3 and MPC controllers use the safe driving boundary function h(x) as constraints and aim to smoothly approach the target speed and maintain a safe following distance, thus obtaining the control input sequence u. t ;

[0052] S4. The vehicle performs corresponding operations according to the control parameters. During the execution, an online optimization algorithm is used to maximize the reward value R. t The long-term cumulative value is the target, and the weight matrix of the MPC controller is continuously adjusted to gradually optimize the control parameters.

[0053] In step S1, the vehicle driving data includes the current vehicle speed v t acceleration a t Distance d from the vehicle in front t The environmental information, including traffic signals, road type, road condition, traffic flow density, weather conditions, and the distribution and movement of surrounding dynamic obstacles, provides precise input for vehicle safety boundary modeling through comprehensive collection and analysis of this data, thereby enabling dynamic optimization of adaptive cruise control strategies and improved safety.

[0054] Specifically, this invention utilizes reinforcement learning to optimize the parameters of the MPC controller, thereby improving the system's safety and adaptability. The real-time data feedback and learning module continuously collects vehicle driving data, including the current speed v. t acceleration a t And the distance d from the vehicle in front tDeep Q-learning (DQN) dynamically adjusts the weight matrices R and Q in the MPC controller to optimize the control strategy and improve the system's adaptability and response speed. The real-time data feedback and learning module continuously interacts with the environment to learn how to dynamically adjust the weight matrices R and Q in the MPC controller to adapt to different traffic scenarios and driving conditions. During training, DQN uses the reward signal R received by the system as a basis for... t The optimization strategy aims to maximize long-term cumulative rewards, thereby improving the overall performance of the system. The state space S and action space A of DQN are defined as: S = {x} t ,u t ,E},A={u min ,u max The state space S contains the current vehicle state x. t Control input u t And the surrounding environment information E, while the action space A includes the possible acceleration and deceleration range of the vehicle, u min U represents the minimum speed. max This represents the maximum speed. To guide the learning process, a reward function R is defined to reflect the system's safety and efficiency, and its specific form is as follows:

[0055]

[0056] Here, h(x) is the safety boundary function. h(x) ≥ 0 indicates that the vehicle is within the safe driving range, receiving a positive reward and enhancing system stability; while h(x) < 0 indicates that the vehicle is approaching an obstacle or entering an unsafe area, receiving a negative reward and prompting the reinforcement learning algorithm to quickly adjust its control strategy to avoid potential dangers. This reward module allows the reinforcement learning algorithm to focus more on optimizing the safe range and avoiding collisions or other dangerous situations. To improve learning efficiency, the system uses the Q-value update formula:

[0057]

[0058] Among them, s t s represents the state at the current moment. t+1 Let a represent the state at the next moment. t This represents the action taken at the current moment, where R refers to the immediate reward, α is the learning rate, and γ is the discount factor.

[0059] Through continuous iterative updates, the Q-value converges to the optimal strategy, thereby improving the vehicle's adaptive capabilities. At each time step, the system monitors the vehicle's state and environmental information through sensors, forming a new dataset D. This dataset records the state, actions, and corresponding rewards at each moment, aiding subsequent learning and optimization. Newly collected data is continuously updated into the historical dataset for use in subsequent learning, allowing the reinforcement learning algorithm to effectively utilize the latest data, thus improving the accuracy and efficiency of learning. The real-time data feedback and learning module ensures that the parameters of the MPC controller can be dynamically adjusted according to the real-time traffic environment and vehicle state, enabling the adaptive cruise control system to operate efficiently, safely, and comfortably in complex traffic scenarios. With continuous optimization of the reinforcement learning algorithm and dynamic adjustment of parameters, the system can gradually converge to the optimal strategy, significantly improving the intelligence level and self-evolution capability of adaptive cruise control.

[0060] Specifically, the safety boundary calculation module of this invention is based on a mathematical model established by CBF (Carrier Flow Factor). It can adaptively adjust according to real-time changes in the vehicle and its surrounding environment, thereby ensuring that the vehicle does not cross the boundary or collide during cruising, achieving a dynamic balance between safety and efficiency. The calculation process of the safety boundary is as follows: Figure 2 The safety boundary calculation module first constructs an initial safety boundary factor (CBF) using an initial safety controller. Then, based on the initial CBF, it further optimizes the safety controller by determining whether the vehicle exceeds the safety boundary line, and then regenerates a new CBF. Next, the system checks whether the current safe area can continue to expand. If so, the process is repeated to increase the safe area; otherwise, the expansion stops, and the final CBF result is output to ensure that the system maximizes the safe area while satisfying the safety boundary. The system state is x(t), and the safety boundary is defined as a differentiable function h(x). When h(x) ≥ 0, the system is in a safe state; when h(x) < 0, the system is in an unsafe state. The vehicle's safe state is calculated using CBF theory. The safety boundary considers not only the vehicle's motion state but also surrounding traffic environment factors, such as the position and speed of other vehicles, pedestrians, and obstacles. The formula for calculating the safety boundary is:

[0061]

[0062] in γ(h(x)) is a non-negative function, representing the derivative of the safety boundary, used to adjust the conservatism of the safety boundary. When traffic conditions change, γ(h(x)) increases the conservatism of the safety boundary, enabling the vehicle to make timely avoidance and deceleration decisions. Under stable driving conditions, the system reduces conservatism to ensure the vehicle maintains a higher cruising speed, thereby improving efficiency. After the safety boundary calculation is completed, the Safety Boundary Supervisory Model Predictive Control (MPC) module is used as the constraint condition for the MPC controller, and the control parameters of the MPC module are adjusted to ensure vehicle driving safety.

[0063] Specifically, after receiving the corrected control parameters from the safety boundary calculation module, the MPC controller module of this invention calculates the optimal control commands (such as acceleration and braking) based on the vehicle's dynamic model and current state to achieve the adaptive cruise objective. The core objective of the MPC controller is to ensure that the vehicle not only maintains the set target speed and safe distance but also achieves smooth acceleration and deceleration control under different driving scenarios by solving the following optimization problems:

[0064]

[0065] Wherein, the stage cost function L(x(t),u(t))=‖x(t)-x target || 2 +R‖u(t)‖ 2 This indicates the deviation of the vehicle's current state from the target state, and balances the control output to ensure smooth driving; where x target Let R be the expected cruise state (such as the set target speed and safe distance), and R be the weight matrix of the control inputs used to balance the use of control inputs. u(t) is the sequence of control inputs. In an adaptive cruise system, the objective of this function is to ensure that the vehicle approaches the target state without generating excessive control output, thus maintaining driving smoothness.

[0066] Terminal cost function Φ(x(t+N))=Q‖x(t+N)-x goal || 2 This is used to evaluate the final state of the system at the end of the prediction time domain, where x goalRepresenting the ideal final state, such as maintaining a safe distance from the vehicle ahead and the desired speed, Q is the weight matrix of the state vector, used to balance the importance of the final state. The MPC controller continuously adjusts the value of u(t) based on safety boundary information, thereby ensuring that the vehicle always maintains an appropriate target speed and safe distance in the dynamic environment. When the surrounding environment changes (such as the vehicle ahead decelerating or other obstacles appearing), the safety boundary calculation module updates the safety boundary in real time, and the MPC controller adjusts the control input accordingly to ensure that the vehicle achieves the optimal cruise state while maintaining safety. This real-time optimization design enables the adaptive cruise system to maintain efficient and safe driving performance under different road conditions, vehicle speeds, and traffic densities.

[0067] The execution module described in this invention, upon receiving the optimal control commands generated by the MPC controller, is responsible for translating these commands into actual vehicle actions, such as acceleration, braking, and steering. The execution module precisely executes the MPC controller's commands through the vehicle's actuators (such as the engine control unit, braking system, and steering system) to achieve the target state of adaptive cruise control. This module not only needs to ensure the accurate transmission of control commands but also needs to monitor feedback during execution in real time to ensure the effectiveness and response speed of the commands.

[0068] In summary, this invention addresses the shortcomings in safety and adaptability of existing adaptive cruise control systems in dynamic traffic environments by proposing a safety boundary-based self-evolutionary method for adaptive cruise control. This system integrates a real-time data feedback and learning module, a safety boundary calculation module, an MPC controller, and an execution module to achieve real-time monitoring, dynamic optimization, and intelligent control of vehicle driving status, thereby significantly improving the performance and reliability of the adaptive cruise control system in complex traffic environments.

[0069] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A self-evolving adaptive cruise safety method based on safety boundaries, characterized in that, include: S1. Real-time collection and preprocessing of vehicle driving data and environmental information to obtain the vehicle's state at each time step. Control input sequence and the corresponding reward value Time series dataset D; S2. Based on the time series dataset D, dynamically evaluate the safe distance between the vehicle and surrounding obstacles using a control obstacle function to determine the safe zone, and simultaneously construct and update the vehicle's safe driving boundary function. ; The method for determining the safe area includes: first, constructing an initial control barrier function using an initial safety controller; then, further optimizing the safety controller based on the initial control barrier function by determining whether the vehicle exceeds the lane boundary line; and finally, regenerating a new control barrier function. Next, the system checks whether the current safe area can continue to expand. If so, the process is repeated to increase the safe area; otherwise, the expansion stops, and the final control barrier function result is output to ensure that the system maximizes the safe area while satisfying the safety boundary. S3 and MPC controllers use safe driving boundary functions As constraints, with the target speed and following distance as objectives, the control input sequence is obtained. The objective function of the MPC controller is: in, This is the stage cost function, representing the deviation between the vehicle's current state and the target state; The expected cruise state includes target speed and safe distance; R is the weight matrix of the control input; u(t) is the control input sequence; Let be the terminal cost function, used to evaluate the final state of the system at the end of the prediction time domain; where, The ideal final state is represented by Q, which includes maintaining a safe distance from the vehicle in front and the desired speed; Q is the weight matrix of the state vector, used to balance the importance of the final state. S4. The vehicle operates according to the control input sequence. The corresponding operation is performed, and during the execution process, an online optimization algorithm is used to maximize the reward value. The cumulative value is the target, and the weight matrix of the MPC controller is continuously adjusted to make the control input sequence... Optimize gradually.

2. The adaptive cruise safety self-evolution method based on safety boundaries according to claim 1, characterized in that, In step S1, the vehicle driving data includes the current vehicle speed. acceleration Distance to the vehicle in front The environmental information includes traffic signals, road type, road condition, traffic flow density, weather conditions, and the distribution and movement of surrounding dynamic obstacles.

3. The adaptive cruise safety self-evolution method based on safety boundaries according to claim 1, characterized in that, The constraint condition refers to the safe driving boundary function. When the vehicle is within the safe area, 0, when not within the stated safe area 0.

4. The adaptive cruise safety self-evolution method based on safety boundaries according to claim 1, characterized in that, In step S4, the online optimization algorithm employs deep Q-learning, based on the reward value at each time step. The goal is to obtain the long-term cumulative value of the reward, and then train the system to maximize this long-term cumulative value, thereby controlling the input sequence. Gradually converge to the optimal strategy; Among them, reward value The formula is as follows: 。 5. The adaptive cruise safety self-evolution method based on safety boundaries according to claim 4, characterized in that, The state space of deep Q-learning Action space ,in, E Indicates information about the surrounding environment; Indicates the minimum speed. This represents the maximum speed; the iterative update rules used in the deep Q-learning are as follows: in, Indicates the current state. Indicates the state at the next moment. This refers to the action taken at the current moment, where R stands for immediate reward. For learning rate, This is the discount factor.

6. A self-evolving adaptive cruise safety device based on safety boundaries, characterized in that, The adaptive cruise safety self-evolution method according to any one of claims 1-5 includes: The real-time data feedback and learning module is used to organize vehicle driving data and environmental information into a time series dataset D, and to continuously adjust the weight matrix of the MPC controller through an online optimization algorithm so that the control input sequence is gradually optimized. The safety boundary calculation module is used to dynamically evaluate the safe distance between the vehicle and surrounding obstacles based on the time series dataset D by controlling the obstacle function, and simultaneously construct and update the vehicle's safe driving boundary function. ; The MPC controller module is used to transmit the control input sequence to the safety boundary calculation module for correction to obtain a safe control input sequence based on the current vehicle state and the prediction of the future state. The execution module is used to execute the safe control input sequence obtained by the MPC controller module to adjust the actual driving state of the vehicle.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the adaptive cruise safety self-evolution method as described in any one of claims 1-5.

8. An adaptive cruise safety self-evolution system based on safety boundaries, characterized in that, Includes computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the adaptive cruise safety self-evolution method as described in any one of claims 1-5.

Images