Historical evolutionary trend learning driven vehicle speed control dynamic multi-objective optimization method

Abstract

This invention discloses a dynamic multi-objective optimization method for vehicle speed control driven by historical evolution trend learning. This method first models the PID parameter optimization problem of vehicle speed control as a dynamic multi-objective optimization problem, and then designs a dynamic multi-objective optimization algorithm through a historical evolution trend learning mechanism to accurately solve for the Pareto optimal PID parameter solution set, providing a reliable parameter selection basis for vehicle speed control systems under dynamic operating conditions. The initial population is divided into regions, and the initial and final population regions of the previous environment are matched according to the nearest neighbor principle. The displacement direction of the region centroid is determined as the historical evolution trend. Combining the relative positional relationship between the initial population regions of the old and new environments, the historical evolution trend is mapped to the corresponding population region of the new environment. Finally, the learned evolution trend is used as the optimal search direction to guide individuals within the population region to generate efficient offspring, improving the search efficiency and solution accuracy of dynamic optimization.

Description

Technical Field

[0001] This invention relates to a dynamic multi-objective optimization method for vehicle speed control driven by historical evolutionary trend learning. Background Technology

[0002] In the field of intelligent connected vehicles and autonomous driving, vehicle speed control, as the core component of longitudinal motion control, is the underlying execution foundation for various core autonomous driving functions such as adaptive cruise control, traffic jam assist, and convoy cooperative driving. Its performance directly determines the control accuracy, stability, and overall operational performance during dynamic vehicle driving. The core task of vehicle speed control is to ensure that the actual vehicle speed accurately, quickly, and smoothly tracks the desired vehicle speed. The Proportional-Integral-Derivative (PID) controller, with its significant advantages such as simple structure, outstanding robustness, and low engineering implementation difficulty, has become the most widely used and technologically mature basic control algorithm in the speed control modules of mass-produced vehicles.

[0003] PID parameter optimization in vehicle speed control is a typical multi-objective optimization problem, which requires determining the appropriate proportional coefficient. Integral coefficient With differential coefficients The goal is to achieve three core optimization objectives: shorten the vehicle speed response rise time, allowing the actual vehicle speed to reach the desired effective speed range more quickly to improve response efficiency; reduce the maximum speed overshoot deviation, minimizing fluctuations in the actual vehicle speed exceeding the desired speed to ensure driving stability; and improve steady-state convergence speed, enabling the actual vehicle speed to quickly stabilize within the allowable error range of the desired speed to achieve precise steady-state control. The core characteristic of this type of optimization problem is that the vehicle's driving conditions and the system's own characteristics change dynamically over time, and the corresponding optimization objectives and optimal parameter solution sets also shift synchronously. Based on this, by incorporating dynamic time-varying characteristics into the optimization framework, the adaptive vehicle speed control problem in complex time-varying scenarios can be transformed into a dynamic multi-objective optimization problem centered on PID parameter optimization. The challenge lies in quickly locating the Pareto-optimal parameter set (PS) after environmental changes, providing an effective basis for parameter selection in real-time decision-making for the control system.

[0004] Multi-objective evolutionary algorithms, with their implicit parallelism based on population search and the advantage of not relying on problem gradient information, have become the mainstream technical paradigm for solving multi-objective optimization problems. However, these algorithms do not fully consider the dynamic characteristics of environmental changes, making it difficult to accurately locate the target population (PS) after environmental changes within a limited time when facing dynamic multi-objective optimization problems. Dynamic multi-objective evolutionary algorithms, by introducing dynamic response strategies to environmental changes and utilizing historical environmental information to generate high-quality initial populations, provide a better starting point for population search in new environments, and have become a research hotspot in this field. Although existing dynamic response strategies have shown significant advantages in initializing the population using historical environmental information, they generally overlook the potential guiding value of historical evolutionary experience for the subsequent search process of the population. How to effectively explore and utilize historical evolutionary experience to guide the population search direction in new environments and improve the solution performance of dynamic multi-objective optimization problems has become a key issue that urgently needs to be addressed. Summary of the Invention

[0005] To address the challenge of optimizing PID parameters for vehicle speed control under dynamic and time-varying driving conditions, this invention proposes a dynamic multi-objective optimization method for vehicle speed control driven by historical evolution trend learning. This method first models the PID parameter optimization problem of vehicle speed control as a dynamic multi-objective optimization problem, and then designs a dynamic multi-objective optimization algorithm through a historical evolution trend learning mechanism to accurately obtain the Pareto optimal PID parameter solution set, providing a reliable basis for parameter selection for the vehicle speed control system under dynamic conditions.

[0006] The core implementation idea of ​​this invention is as follows: If the current environment is not the initial driving environment, the intrinsic structure of the population distribution is first captured by the spectral clustering algorithm, and the initial population, final population of the previous historical environment, and initial population of the new environment are divided into regions respectively; then, the initial population region and final population region of the previous environment are matched according to the nearest neighbor principle, and the displacement direction of the centroid of the region is determined as the historical evolutionary trend; combined with the relative positional relationship of the initial population regions of the old and new environments, the historical evolutionary trend is mapped to the corresponding population region of the new environment; finally, the learned evolutionary trend is used as the high-quality search direction to guide individuals in the population region to generate efficient offspring, thereby improving the search efficiency and solution accuracy of dynamic optimization.

[0007] The specific implementation steps of the historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control proposed in this invention are as follows:

[0008] (Step 1) Model the optimization of the PID parameters used in vehicle speed control as a dynamic multi-objective optimization problem. Using the proportional gain... Integral coefficient Differential coefficients As decision variables, to minimize the system rise time. Maximum overshoot ), Setting Time To optimize performance, the system's response speed, driving smoothness, and steady-state convergence efficiency are all taken into account.

[0009] (Step 2) In the initialization phase, the random generation size is... The initial population is formed, and each individual in the population corresponds to a set of candidate solutions for PID parameters, which then enters the iterative optimization process.

[0010] (Step 3) At the beginning of each iteration, determine whether the driving environment has changed. If the environment has changed, call the dynamic response strategy to generate a new high-quality initial population. If the environment has not changed, continue the iteration of the current population directly.

[0011] (Step 4) For the initial population, final population, and initial population of the new environment, a consistent spectral clustering method is used to divide the population regions. For each population to be divided, a Gaussian kernel similarity matrix is ​​first constructed and the corresponding normalized Laplacian matrix is ​​solved. The low-dimensional manifold representation of the population is obtained through eigenvalue decomposition. Then, K-means is used to divide the low-dimensional representation into different population regions based on the latent distribution of the population on the low-dimensional manifold. Correspondingly, the population in the original decision space is also divided into multiple identical representative population regions.

[0012] (Step 5) Extract the historical evolutionary trends from different population regions in the previous environment and map them to the corresponding population regions in the new environment. Specifically, firstly, based on the nearest neighbor principle, match the initial population region and the final population region in the previous environment, and determine the historical evolutionary trend of that region as the displacement direction from the initial population region to the final population region after matching. Then, using the relative positional relationship between population regions in adjacent environments and the global centroid of the population, map the historical evolutionary trends extracted in the previous step to the initial population region of the new environment.

[0013] (Step 6) The first environment directly generates offspring through the genetic operator and then executes step 8; for non-first environments, step 7 is executed to combine the learned historical evolutionary trends to guide the offspring to generate in the direction of optimization potential, while adaptively adjusting the search step size and guidance probability to balance the use of historical experience and the adaptability of the new environment.

[0014] (Step 7) Based on the historical evolutionary trends extracted from the learning process, the offspring population is guided to search in directions with greater optimization potential to generate higher-quality offspring individuals. At the same time, in order to achieve a balance between utilizing historical evolutionary trends and adapting to the new environment, this method adaptively adjusts the search step size and guidance probability according to the actual survival status of the offspring individuals, ensuring that the offspring always complete iterative optimization along effective evolutionary trends, thereby further improving the quality of the offspring.

[0015] (Step 8) Merge the parent population (i.e. the current population) with the generated offspring population and perform environmental selection to obtain a new generation population.

[0016] (Step 9) Repeat steps 3 to 8 above until the termination condition is met. Finally, output the PS corresponding to each environment.

[0017] Compared with the prior art, the present invention has the following obvious advantages and beneficial effects:

[0018] 1) By leveraging the historical evolutionary trends in the previous environment, the population's search process in the new environment can be guided, which helps to enhance the population's search capabilities and thus improve its ability to solve dynamic multi-objective optimization problems.

[0019] 2) In order to effectively learn historical evolutionary trends from the previous environment, spectral clustering is used to capture the manifold structure of population distribution, divide the population into different regions, and learn the evolutionary trends of each region. These trends are then used to adaptively adjust the guidance for various population regions in the new environment, which helps to achieve differentiated guidance for populations in the new environment. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the method proposed in this invention.

[0021] Figure 2 This is a schematic diagram of a vehicle speed control system that optimizes PID parameters using the method of the present invention. Detailed Implementation

[0022] To more clearly illustrate the technical solution of the present invention, the present invention will be described in detail below through specific implementation examples.

[0023] Figure 1Figure 1 shows the overall flowchart of the method proposed in this invention. For dynamic multi-objective optimization problems, the execution flow of the proposed algorithm is as follows: First, the algorithm's relevant parameters are initialized, and an initial population is randomly generated. Then, the iterative optimization loop of the population begins. During the iteration process, it first checks whether the current optimization environment has changed: if the environment has not changed, subsequent evolutionary operations are directly executed; if the environment has changed, a dynamic response strategy is invoked to generate a high-quality initial population for the new environment. Then, the historical evolutionary trends from the previous environment are learned and mapped to the new environment to provide guidance for subsequent population evolution.

[0024] Subsequently, a temporary offspring population is generated using genetic operators, and it is determined whether the current environment is the first environment in which the algorithm runs. If it is the first environment, there is no historical evolutionary information available, and the temporary offspring are directly used as the final offspring for environment selection. If it is not the first environment, the learned historical evolutionary trends are used to enhance the existing offspring, guiding them to evolve towards more promising search directions, generating a high-potential offspring population. Finally, the parent and offspring populations are merged to perform environment selection, resulting in a new generation of population. If the algorithm does not meet the termination condition, it returns to the environment change detection step to enter the next iteration until the termination condition is met, at which point the corresponding PS for each environment is output.

[0025] To verify the performance of the proposed HETL-DMOEA method, it was compared with five representative algorithms (SET-MOEA / D, KPDCS, KTM-MOEA / D, DIP-DMOEA, and KGB) in a vehicle speed control task. Each algorithm was run independently 20 times, and the hypervolume (HV) index value of the PS obtained by each algorithm in each environment was recorded. Generally, the larger the HV, the better the performance of the method.

[0026] The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control proposed in this invention, HETL-DMOEA, is described in detail below.

[0027] (Step 1) The vehicle speed control task is described using the following mathematical model:

[0028]

[0029] in, It is the time required for the actual vehicle speed to first reach 90% of the target speed, minimized. This ensures the system's rapid response capability and shortens the transition process for vehicle speed tracking; This represents the maximum deviation of the vehicle speed from the target value, and is minimized. It can suppress speed fluctuations and improve the stability and comfort of vehicle driving; The total time required for the vehicle speed to stabilize within the allowable error band, minimized. This can accelerate the steady-state convergence speed of the system. The above three optimization objectives, as the time-varying system... The changes are subject to adjustment, and a controller can be used. and decision vector To determine, among which It includes three decision variables: the proportional coefficient. Integral coefficient and differential coefficients Time-varying systems and controller It can be represented in the following form:

[0030]

[0031]

[0032] in, and These are parameters that change over time. For example... Figure 2 As shown, the method proposed in this invention optimizes key parameters in a PID controller. , and This is to achieve vehicle speed control.

[0033] (Step 2) In the initialization phase, an initial population of 100 individuals is randomly generated, with each individual corresponding to a set of candidate solutions for parameters to be optimized. Then, the fitness of the individuals is evaluated using the objective function, and the iterative optimization process begins accordingly.

[0034] (Step 3) Before each iteration of optimization, the environment will be checked first. If the environment has changed, the dynamic response strategy will be called to regenerate the initial population for the new environment, and then steps 4 and 5 will be executed; if the environment has not changed, the process will proceed directly to steps 6 to 8 to continue iterative optimization of the current population.

[0035] (Step 4) For the initial population, final population of the previous environment, and initial population of the new environment, use a consistent spectral clustering method to divide the population regions: For each population to be divided First, a similarity matrix is ​​constructed using the following Gaussian kernel function formula. :

[0036]

[0037] in, and For any individual in the population to be divided; kernel bandwidth parameter Determined by the following formula:

[0038]

[0039] Then, based on the similarity matrix The normalized Laplace matrix is ​​calculated. And perform feature decomposition on it to extract the previous features. The feature matrix is ​​composed of the eigenvectors corresponding to the smallest non-zero eigenvalues, where each row represents the individual's eigenvalues. A low-dimensional representation in the 3D manifold space is obtained; finally, K-means clustering is performed on this low-dimensional representation to divide it into 1. Each cluster is thus divided into corresponding groups, thereby dividing the population in the original space into clusters. Different regions.

[0040] (Step 5) Extract the historical evolutionary trends from different population regions in the previous environment and map them to the corresponding population regions in the new environment to guide the subsequent search process of the population. Specifically, first, match the initial population region and the final population region in the previous environment based on the nearest neighbor principle. Define the displacement direction from the initial population region to the final population region after matching as the historical evolutionary trend of that region. The calculation method is as follows:

[0041]

[0042] in, The first representing the initial population of the previous environment The centroid of a region, Represents the previous environment and The final population region centroid is matched. Then, using the relative positions of population regions in adjacent environments with respect to the global population centroid, the extracted historical evolutionary trends are mapped to the initial population regions of the new environment. This is then applied to the first population region in the new environment. Each population region, and the corresponding historical population region index, is determined using the following formula:

[0043]

[0044] in, For the first in the new environment The relative position vector of each population region with respect to the global centroid of the new environment population. For the first in the previous environment The relative position vectors of each population region with respect to the global centroid of the population in the new environment. Finally, the position vectors of the first population region in the previous environment... The historical evolutionary trend corresponding to each population region is allocated to the new environment. Within each population region, this information is used as prior knowledge to guide subsequent population search processes.

[0045] (Step 6) Generate temporary offspring using the genetic GA operator. If the current environment is the first one, the temporary offspring is directly used as the final offspring, and the process jumps to step 8 to perform environment selection; otherwise, step 7 is executed to enhance and optimize the temporary offspring using the learned historical evolutionary trends, thereby generating offspring individuals of higher quality.

[0046] (Step 7) Based on the learned historical evolutionary trends, guide individuals within the corresponding population area in the new environment to search in directions with greater potential. Specifically, for temporary offspring individuals... First, determine the corresponding region index according to the following formula. :

[0047]

[0048] in, It is the first generation of the parent population The centroid of a region. After determining the region, the historical evolutionary trend corresponding to that region is used. Guided by promising search directions Search towards promising regions and generate new offspring according to the following formula. :

[0049]

[0050] In the formula, for The corresponding search step size is adaptively adjusted based on the offspring survival rate, and the calculation method is shown below.

[0051]

[0052] in, Representing a new environment The current number Generations of populations and These represent the sets of offspring generated in the previous generation of the current environment, guided and unguided by historical evolutionary trends, respectively. This represents the number of individuals in the set. This ratio reflects the trend-guided offspring of the previous generation in the current generation. The survival rate of offspring. A higher ratio indicates a better guidance effect, allowing for a larger step size to continue the search. Notably, to balance the utilization of historical information and the need to adapt to the new environment, an adaptive guidance probability is further defined to control whether offspring accept guidance from historical evolutionary trends. The calculation formula is as follows:

[0053]

[0054] in, and The term represents the number of surviving offspring from the previous generation in the current environment, generated through historical evolutionary trends and those not.

[0055] (Step 8) Perform environmental selection on the parent population and the generated offspring population to obtain a new generation population.

[0056] (Step 9) Repeat steps 3 to 8 above until the termination condition is met. Finally, output the PS corresponding to each environment.

[0057] The HETL-DMOEA method of this invention was compared with five representative methods, including SET-MOEA / D and KPDCS, and run independently 20 times in 20 continuous dynamic driving environments. Hypervolume (HV) was used as the performance evaluation index (a higher HV value indicates better optimization). Experimental results show that the method of this invention achieves the optimal HV value in 95% of the dynamic environments. Its PID parameter optimization accuracy and search efficiency under dynamic conditions are superior to other comparative methods, effectively supporting adaptive speed control in intelligent connected vehicles.

[0058] Table 1. Performance comparison of the proposed method with other methods in vehicle speed control.

[0059] environment SET-MOEA / D KPDCS KTM-MOEA / D DIP-DMOEA KGB HETL-DMOEA 1 0.5299 0.5293 0.5263 0.5242 0.5253 0.5366 2 0.6580 0.6189 0.6565 0.6567 0.6426 0.6579 3 0.6543 0.6222 0.6473 0.6392 0.6420 0.6639 4 0.6403 0.6329 0.6358 0.6225 0.6311 0.6511 5 0.6207 0.6073 0.6088 0.6024 0.6061 0.6246 6 0.5676 0.5543 0.5644 0.5466 0.5505 0.5831 7 0.5074 0.4935 0.5003 0.4839 0.4842 0.5213 8 0.4506 0.4223 0.4363 0.4240 0.4219 0.4666 9 0.3904 0.3727 0.3834 0.3603 0.3741 0.4159 10 0.3521 0.3184 0.3357 0.3256 0.3332 0.3776 11 0.3175 0.2955 0.2985 0.3083 0.3042 0.3531 12 0.2984 0.2764 0.2902 0.2842 0.2839 0.3277 13 0.2740 0.2540 0.2572 0.2618 0.2674 0.3073 14 0.2731 0.2560 0.2570 0.2646 0.2662 0.3031 15 0.2919 0.2661 0.2777 0.2868 0.2898 0.3253 16 0.3154 0.2887 0.3014 0.3060 0.3027 0.3502 17 0.3423 0.3140 0.3339 0.3382 0.3418 0.3797 18 0.3818 0.3734 0.3750 0.3828 0.3849 0.4221 19 0.4466 0.4204 0.4368 0.4370 0.4415 0.4705 20 0.5099 0.4882 0.4960 0.4989 0.5015 0.5265 Best / All 1 / 20 0 / 20 0 / 20 0 / 20 0 / 20 19 / 20

Claims

1. A dynamic multi-objective optimization method for vehicle speed control driven by historical evolutionary trend learning, characterized by: The specific implementation steps of this method are as follows: Step 1 optimizes the PID parameters used in vehicle speed control as a dynamic multi-objective optimization; using the proportional coefficient... Integral coefficient Differential coefficients As decision variables, to minimize the system rise time Maximum overshoot Adjusting time To optimize the target, the system's response speed, driving smoothness, and steady-state convergence efficiency must be taken into account. Step 2, in the initialization phase, the random generation size is... The initial population, each individual in the population corresponds to a set of candidate solutions for PID parameters, and then the iterative optimization process begins; Step 3: At the beginning of each iteration, determine whether the driving environment has changed. If the environment has changed, call the dynamic response strategy to generate a new high-quality initial population in the new environment. If the environment has not changed, continue iterating the current population directly. Step 4: For the initial population, final population of the previous environment, and initial population of the new environment, a consistent spectral clustering method is used to divide the population regions. For each population to be divided, a Gaussian kernel similarity matrix is ​​first constructed and the corresponding normalized Laplacian matrix is ​​solved. The low-dimensional manifold representation of the population is obtained through eigenvalue decomposition. Then, K-means is used to divide the low-dimensional representation into different population regions based on the latent distribution of the population on the low-dimensional manifold. The population in the original decision space is also divided into multiple identical representative population regions. Step 5: Extract the historical evolutionary trends from different population regions in the previous environment and map them to the corresponding population regions in the new environment. Specifically, firstly, based on the nearest neighbor principle, match the initial population region and the final population region in the previous environment, and determine the displacement direction from the initial population region to the final population region as the historical evolutionary trend of that region. Then, using the relative positional relationship between population regions in adjacent environments and the global centroid of the population, map the historical evolutionary trends extracted in the previous step to the initial population region of the new environment. Step 6: In the first environment, offspring are generated directly through the genetic operator, and then step 8 is executed; in non-first environments, step 7 is executed to combine the learned historical evolutionary trends to guide the offspring to generate in the direction of optimization potential, while adaptively adjusting the search step size and guidance probability to balance the use of historical experience and the adaptability to the new environment. Step 7: Based on the historical evolutionary trends extracted from the learning process, guide the offspring population to search in directions with greater optimization potential to generate higher-quality offspring individuals. At the same time, in order to achieve a balance between utilizing historical evolutionary trends and adapting to the new environment, this method adaptively adjusts the search step size and guidance probability according to the actual survival status of the offspring individuals, ensuring that the offspring always complete iterative optimization along effective evolutionary trends, thereby further improving the quality of the offspring. Step 8: Merge the parent population (current population) with the generated offspring population and perform environmental selection to obtain a new generation population; Step 9: Repeat steps 3 to 8 above until the termination condition is met; finally, output the PS corresponding to each environment.

2. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: Step 1 describes the vehicle speed control task using the following mathematical model: ； in, It is the time required for the actual vehicle speed to first reach 90% of the target speed, minimized. Ensure the system's rapid response capability and shorten the transition process for vehicle speed tracking; This represents the maximum deviation of the vehicle speed from the target value, and is minimized. It can suppress speed fluctuations and improve the stability and comfort of vehicle driving; The total time required for the vehicle speed to stabilize within the allowable error band, minimized. It can accelerate the steady-state convergence speed of the system; the three optimization objectives change with the time-varying system. The changes are adapted to the control system. and decision vector To determine, among which It includes three decision variables: the proportional coefficient. Integral coefficient and differential coefficients Time-varying systems and controller It can be represented in the following form: ； ； in, and These are parameters that change over time.

3. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 2, during the initialization phase, an initial population of 100 individuals is randomly generated, with each individual corresponding to a set of candidate solutions for parameters to be optimized. Then, the fitness of the individuals is evaluated using the objective function, and the iterative optimization process is initiated accordingly.

4. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 3, before each iteration of optimization, it is first checked whether the environment has changed; if the environment has changed, the dynamic response strategy is called to regenerate the initial population for the new environment, and then steps 4 and 5 are executed; if the environment has not changed, it directly proceeds to steps 6 to 8 to continue iterative optimization of the current population.

5. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 4, a consistent spectral clustering method is used to divide the population regions for the initial and final populations of the previous environment and the initial population of the new environment: for each population to be divided... First, a similarity matrix is ​​constructed using the following Gaussian kernel function formula. : ； in, and For any individual in the population to be divided; kernel bandwidth parameter Determined by the following formula: ； Then, based on the similarity matrix The normalized Laplace matrix is ​​calculated. And perform feature decomposition on it to extract the previous features. The feature matrix is ​​composed of the eigenvectors corresponding to the smallest non-zero eigenvalues, where each row represents the individual's eigenvalues. A low-dimensional representation in the 3D manifold space is obtained; finally, K-means clustering is performed on this low-dimensional representation to divide it into 1. Each cluster is thus divided into clusters, and the population in the original space is correspondingly divided into... Different regions.

6. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 5, the historical evolutionary trends from different population regions in the previous environment are extracted and mapped to the corresponding population regions in the new environment to guide the subsequent search process of the population. Based on the nearest neighbor principle, the initial population region and the final population region in the previous environment are matched, and the displacement direction from the initial population region to the final population region after matching is defined as the historical evolutionary trend of that region. The calculation method is as follows: ； in, The first representing the initial population of the previous environment The centroid of a region Represents the previous environment and The final population's regional centroid is matched; then, using the relative positional relationship between population regions in adjacent environments and the global population centroid, the extracted historical evolutionary trend is mapped to the initial population region of the new environment, targeting the first [evolutionary trend] in the new environment. Each population region, and the corresponding historical population region index, is determined using the following formula: ； in, For the first in the new environment The relative position vector of each population region with respect to the global centroid of the new environment population. For the first in the previous environment The relative position vectors of each population region with respect to the global centroid of the population in the new environment; finally, the first population region in the previous environment... The historical evolutionary trend corresponding to each population region is allocated to the new environment. Within each population region, this information is used as prior knowledge to guide subsequent population search processes.

7. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 6, the genetic GA operator is used to generate temporary offspring. If the current environment is the first environment, the temporary offspring is directly used as the final offspring, and the process jumps to step 8 to perform environment selection. Otherwise, step 7 is executed to enhance and optimize the temporary offspring using the learned historical evolutionary trends, thereby generating offspring individuals of higher quality.

8. The historical evolution trend learning-driven dynamic multi-objective optimization method for vehicle speed control according to claim 1, characterized in that: In step 7, based on the learned historical evolutionary trends, individuals within the corresponding population region are guided to search in directions with greater potential under the new environment; specifically, for temporary offspring individuals... First, determine the corresponding region index according to the following formula. : ； in, It is the first generation of the parent population The centroid of a region; once a region is determined, its corresponding historical evolutionary trend is used. Guided by promising search directions Search towards promising regions and generate new offspring according to the following formula. : ； In the formula, for The corresponding search step size is adaptively adjusted based on the offspring survival rate, and the calculation method is shown below; ； in, Representing a new environment The current number Generations of populations and These represent the sets of offspring generated in the previous generation of the current environment, guided and unguided by historical evolutionary trends, respectively. Represents the number of individuals in the set; defines the adaptive guidance probability that controls whether offspring accept the guidance of historical evolutionary trends. The calculation formula is as follows: ； in, and The term represents the number of surviving offspring from the previous generation in the current environment, generated through historical evolutionary trends and those not.

Images