Multi-agent model auxiliary optimization method and system for optimizing side protection performance of vehicle

Abstract

The invention relates to the technical field of vehicle intelligent design, and provides a multi-agent model auxiliary optimization method and system for optimizing the side protection performance of a vehicle, and the method comprises the following steps: initializing a population, and carrying out the collision simulation; according to a collision simulation result, obtaining a correlation relationship between an optimization target and an optimization constraint condition; generating filial generation individuals by using an enhanced filial generation strategy, determining the population size based on a correlation relation between an optimization target and an optimization constraint condition, and forming a new filial generation population; for the filial generation population, sampling by adopting a dynamic sampling criterion according to an evaluation index determined by the constructed neural network proxy model based on the radial basis function to obtain an optimal solution of the iteration; and updating the parent population of the next round of iteration to carry out the next round of iteration until a termination condition is met, and obtaining an optimal solution. According to the method, the contradiction between the side protection performance of the vehicle and the lightweight of the vehicle body is effectively solved through auxiliary optimization of the multi-agent model.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to vehicle intelligent design, and in particular, to a multi-agent model-assisted optimization method and system for optimizing the side protection performance of an automobile. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the continuous expansion of the global automobile market and the increasing demand for vehicle safety performance by consumers, automobile manufacturers are facing unprecedented challenges. Vehicle design not only needs to meet increasingly stringent internal and international safety standards, such as side impact test standards, but also needs to take into account the dual needs of environmental protection and energy efficiency. The side impact test implemented since 1999 has comprehensively evaluated the vehicle's crashworthiness by simulating real side impact scenarios; among them, the survival distance, as a key indicator, directly reflects the vehicle's ability to protect occupants. The implementation of relevant standards has prompted automobile manufacturers to continuously optimize the body structure to improve side impact performance. However, improving side impact performance is often accompanied by an increase in body weight, which is contrary to the current trend of energy conservation and emission reduction and achieving vehicle lightweighting. In addition, there are also strict regulations on the fuel consumption limit of passenger cars, which further exacerbates this contradiction. Therefore, how to achieve lightweight body while ensuring vehicle safety performance has become a key issue that needs to be solved in the field of automobile design.

[0004] To solve the above problems, computer crashworthiness analysis technology has gradually become an indispensable tool in the automotive design process. With the rapid improvement of computer performance and the maturity of nonlinear structural dynamics simulation technology, numerical collision analysis based on the finite element method (FEM) has shown great potential in vehicle collision performance evaluation and optimization with its high efficiency and economic advantages. However, although the FEM model can highly restore the actual collision process, it has the problems of high computational complexity and long simulation time, especially when dealing with large-scale models, the computing resources consumption is huge.

[0005] In order to overcome this bottleneck, artificial intelligence technology, especially optimization methods such as evolutionary algorithms, have been introduced into the field of automobile side collision performance optimization. These algorithms can search for optimal solutions in complex design spaces by simulating the natural evolution process, providing new ideas for automobile design. However, when facing the highly nonlinear, non-convex and highly coupled optimization problem of automobile side collision, existing methods still have limitations such as insufficient optimization ability and easy to fall into local optimality, making it difficult to meet diverse design needs. Summary of the invention

[0006] In order to solve the above problems, the present disclosure proposes a multi-agent model-assisted optimization method and system for optimizing the side protection performance of vehicles, which makes full use of historical individual information and constructs radial basis function neural network models as evaluation agent models for the mass size and each constraint violation corresponding to each solution. In order to make more reasonable use of limited computing simulation resources, the dynamic sampling mechanism helps the population jump out of the local optimum and quickly search for the global optimal solution while balancing convergence, diversity and feasibility.

[0007] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0008] One or more embodiments provide a multi-agent model-assisted optimization method for optimizing vehicle side protection performance, comprising the following steps:

[0009] Initialize a population and perform collision simulation on individuals in the population; the individuals include body structure parameters and material property information of the vehicle;

[0010] According to the collision simulation results, the correlation relationship between the optimization target and the optimization constraint conditions is obtained;

[0011] The enhanced offspring generation strategy is used to generate offspring individuals, and the population size is determined based on the correlation between the optimization goal and the optimization constraints to form a new offspring population.

[0012] For the offspring population, according to the evaluation index determined by the constructed radial basis function neural network proxy model, dynamic sampling criteria are used to obtain the optimal solution of this iteration; the parent population of the next iteration is updated for the next iteration until the termination condition is met and the optimal solution is obtained.

[0013] The above scheme effectively solves the contradiction between vehicle side protection performance and body lightweight through multi-agent model-assisted optimization. By optimizing the body structure and material properties, the weight reduction goal can be achieved while ensuring vehicle safety. The enhanced offspring generation strategy and dynamic sampling criteria based on the agent model improve the global search capability of the algorithm, overcome the problem that traditional optimization methods are prone to fall into local optimality, and significantly improve the optimization efficiency and reliability of the results.

[0014] One or more embodiments provide a multi-agent model-assisted optimization system for optimizing vehicle side protection performance, including:

[0015] Initialization module: configured to initialize the population and perform collision simulation on individuals in the population; the individuals include body structure parameters and material property information of the vehicle;

[0016] Correlation calculation module: configured to obtain the correlation relationship between the optimization target and the optimization constraint condition according to the collision simulation result;

[0017] Offspring population generation module: configured to generate offspring individuals using an enhanced offspring generation strategy, determine the population size based on the correlation relationship between the optimization target and the optimization constraint conditions, and form a new offspring population;

[0018] Optimization sampling module: It is configured to sample the offspring population according to the evaluation index determined by the constructed radial basis function neural network proxy model, adopt dynamic sampling criteria to obtain the optimal solution of this iteration; update the parent population of the next iteration for the next iteration until the termination condition is met to obtain the optimal solution.

[0019] An electronic device includes a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned multi-agent model-assisted optimization method for optimizing vehicle side protection performance are completed.

[0020] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned multi-agent model-assisted optimization method for optimizing vehicle side protection performance are completed.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The disclosed method effectively solves the contradiction between the vehicle side protection performance and the lightweight of the vehicle body through multi-agent model-assisted optimization. By optimizing the vehicle body structure and material properties, the weight reduction goal can be achieved while ensuring vehicle safety. The enhanced offspring generation strategy and the dynamic sampling criteria based on the agent model improve the global search capability of the algorithm, overcome the problem that traditional optimization methods are prone to fall into local optimality, and significantly improve the optimization efficiency and the reliability of the results.

[0023] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0025] Figure 1 is a flow chart of the multi-agent model-assisted optimization method of Embodiment 1 of the present disclosure;

[0026] Figure 2 It is a block diagram of the multi-agent model-assisted optimization system of Example 2 of the present disclosure. DETAILED DESCRIPTION

[0027] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0029] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0030] Example 1

[0031] In the technical solutions disclosed in one or more embodiments, Figure 1 As shown, a multi-agent model-assisted optimization method for optimizing vehicle side protection performance includes the following steps:

[0032] S1, initialize the population, perform collision simulation on the individuals in the population, and obtain the vehicle collision performance index of each individual;

[0033] Among them, each individual in the population contains relevant parameter information such as vehicle body structure parameters and material properties;

[0034] Optionally, the body structure parameters are design parameters of the vehicle, including the size and structure of each part, such as the geometric dimensions, material selection and thickness of key parts including doors, B-pillars, side beams, etc.;

[0035] Material properties can include the density, elastic modulus, yield strength and ductility of the selected material, which will affect the collision energy absorption capacity and weight of the vehicle body;

[0036] Specifically, collision simulation can use LS-DYNA simulation software;

[0037] LS-DYNA is used to simulate the actual collision process, including the vehicle's deformation, stress distribution, energy absorption, occupant protection, and vehicle structure weight under the design scheme;

[0038] Specifically, the constraints that the collision performance needs to meet may include collision survival distance, energy absorption, degree of structural deformation, etc., and the target indicator of the solution is structural quality.

[0039] S2, according to the collision simulation results, determine the correlation relationship between the optimization target and the optimization constraint conditions, that is, obtain the relationship between the target and the constraint;

[0040] Specifically, in this embodiment, the optimization goal of optimizing the side protection performance of the vehicle is to minimize the structural mass, and the constraint is to meet the safety protection requirements of the side collision of the vehicle;

[0041] Optionally, the constraints for optimizing the vehicle side protection performance may include:

[0042] 1) The side impact survival distance meets or exceeds the set minimum standard to ensure the safety of occupants in side impacts;

[0043] The collision survival distance refers to the space in the passenger compartment that remains unintruded or compressed when the vehicle is involved in a side collision.

[0044] 2) Structural deformation limitation: the deformation of key structures after collision is less than the set maximum deformation to prevent serious invasion of the passenger compartment; key structures include B-pillars, doors, side beams, etc.;

[0045] 3) The energy absorption meets the set requirements. The collision energy must be effectively absorbed within the set energy absorption area to reduce the energy transferred to the passenger compartment;

[0046] 4) Material strength and yield stress meet minimum strength requirements to avoid destructive failure during collision.

[0047] 5) After a side collision, the vehicle door should maintain a certain degree of opening ability to ensure that the occupants can be rescued quickly or escape on their own.

[0048] Furthermore, the correlation relationship between the optimization target and the optimization constraint condition includes the following steps:

[0049] S21: According to the target index and constraint index data obtained from the individual vehicle collision simulation, the individuals in the population are sorted according to the degree to which the constraint conditions and the optimization target are satisfied, and a sorting order is obtained;

[0050] Specifically, the individuals in the population are sorted according to the following feasibility criteria, which is recorded as sorting 1. The feasibility criteria are as follows:

[0051] 21.1) When individual A satisfies all constraints and individual B cannot satisfy all constraints, then individual A is better than individual B;

[0052] 21.2) When individuals A and B satisfy all constraints, if the mass of individual A is less than the mass of individual B, then individual A is better than individual B, otherwise, individual B is better than individual B;

[0053] 21.2) When both individual A and individual B cannot satisfy all constraints, and the degree of violation of the constraints by individual A is less than that of individual B, then individual A is better than individual B, and vice versa, individual B is better than individual A;

[0054] The degree to which individual A violates the constraint is the constraint violation index and is calculated using the following formula:

[0055]

[0056] Among them, g j (x) The constraint value of the jth constraint, g j (x)≤0 means that the jth constraint is satisfied, g j (x)>0 means that the jth constraint is not satisfied; m represents the number of constraints;

[0057] It should be noted that constraints have numerical values. For example, in the simulation software, the collision survival distance is 5, and the required survival distance is greater than 10, then the violation of the constraint is 5. Similarly, each constraint has a numerical value, and the larger the sum, the greater the degree of violation of the constraint.

[0058] According to the above judgment criteria, the individuals are ranked according to their excellence;

[0059] This embodiment adopts a feasibility criterion judgment method, which can effectively and quickly screen out individuals with greater potential in a complex optimization process. Through clear rules and standards, it is possible to reduce the waste of computing resources and accelerate the optimization process. Especially in early iterations, this method can eliminate inappropriate design solutions as early as possible, thereby improving the overall optimization efficiency.

[0060] S22: According to the target optimal criterion, the individuals in the population are sorted from small to large according to the structural weight of the vehicle under the corresponding solution, which is recorded as sorting 2;

[0061] In this step, by minimizing mass as the target optimal criterion, the direction of optimization can be guaranteed, and ultimately the goal of lightweighting the vehicle body can be achieved while meeting safety requirements.

[0062] S23: Calculate the similarity of the two groups of data, sorted one and sorted two, to obtain a correlation coefficient;

[0063] Optionally, the similarity calculation may be performed using Pearson correlation analysis to obtain a correlation coefficient μ;

[0064] In step S2, by sorting and calculating the similarity of individuals in the population, this method can accurately capture the relationship between the optimization goal and the constraints, which helps to balance the vehicle safety performance and lightweight goals during the optimization process. This method can clearly identify the individuals with the best weight under the premise of satisfying the constraints, thus providing a clear direction for the optimization strategy. This method reduces unnecessary computational overhead and improves optimization efficiency, especially when dealing with high-dimensional and complex constraints, showing good robustness and reliability.

[0065] S3, based on the individual information in the population and its collision simulation results, a radial basis function neural network proxy model is constructed for each constraint and goal;

[0066] S31: Obtain a set of solutions that have been simulated in the past, and construct a target database by combining the solution information of these individuals with the target information (structural quality) obtained by the corresponding simulation, and construct a constraint database by combining the solution information of these individuals with the information of each constraint obtained by the corresponding simulation (including collision survival distance, rib deflection, etc.);

[0067] Among them, the simulated solutions include both individuals composed of design parameters that can meet the collision requirements and individuals composed of design parameters that cannot meet the collision requirements;

[0068] Among them, the simulation can be carried out in the LS-DYNA simulation software for collision simulation;

[0069] S32: According to the target (structural quality) database, a radial basis function neural network model of quality is constructed as a proxy model of the target; and used to predict the weight of the vehicle according to the individual information;

[0070] S33: According to the constraint database, a radial basis function neural network model is constructed for each constraint condition as a proxy model of each constraint condition, which is used to predict the degree to which the individual satisfies the constraint condition, the input data is the individual, that is, a solution, or a design plan, and the output is the condition that satisfies the constraint;

[0071] In step S3 of this embodiment, by using a proxy model based on a radial basis function neural network, it is possible to replace LS-DYNA simulation to a certain extent, thereby greatly reducing the complexity and time consumption of direct simulation calculations. In this embodiment, after the offspring is generated in step S4, simulation analysis is used to estimate the target and constraint performance of the offspring individuals. This method uses existing simulation data to build a model, which can efficiently and accurately predict the performance of the new design scheme, thereby replacing part of the actual simulation. This not only shortens the optimization cycle, but also can quickly screen out potential high-quality design schemes at an early stage. At the same time, multiple proxy models constructed for different constraints can finely evaluate the degree of satisfaction of individuals with each constraint, thereby improving the accuracy and robustness of the optimization results.

[0072] S4, using the enhanced offspring generation strategy to generate offspring individuals, determining the population size based on the correlation between the optimization goal and the optimization constraint conditions, and forming a new offspring population;

[0073] S41: Pair NP parent individuals in pairs to perform simulated binary crossover and polynomial mutation to generate NP intermediate offspring Offspring1;

[0074] S42: according to the feasibility criteria in step 21, the optimal individual after the collision simulation is retrieved, which is recorded as optg;

[0075] S43: according to the target optimal criterion of step S22, the best individual after the collision simulation is retrieved, which is recorded as optf;

[0076] S44: If there are no individuals that meet all constraints at present, the two best individuals under the target optimal criterion are selected to form population 1, and the NP-2 individuals with the best feasibility criterion are selected to form population 2;

[0077] S45: If there are individuals that meet all constraints, and optg is equal to optf, then the best NP-2 individuals retrieved by the target optimal criterion are taken to form population 1, and the two individuals with the best feasibility criterion are taken to form population 2;

[0078] On the contrary, if optg is not equal to optf, then take max(2,μ*NP) individuals to form population 1, and NP-max(2,μ*NP) individuals to form population 2; where μ is the correlation between the optimization objective and the optimization constraint.

[0079] S46: Population 1 and population 2 are optimized separately;

[0080] Optionally, population one performs particle swarm optimization under a target optimality criterion, and population two performs particle swarm optimization under a feasibility criterion;

[0081] S47: After reaching the maximum number of iterations, the offspring population consisting of the best solution of each individual is output, recorded as Offspring2;

[0082] S48: Merge Offspring1 and Offspring2 to form a new offspring population, denoted as Offspring;

[0083] S5, for the offspring population, according to the evaluation index determined by the constructed radial basis function neural network proxy model, a dynamic sampling criterion is used to perform sampling to obtain the optimal solution for this iteration;

[0084] S51: Determine the target index for offspring evaluation based on the radial basis function neural network proxy model;

[0085] Calculate the CV of each offspring nor ,OB nor ,FE nor ,CB nor ,SP nor Five indicator values;

[0086] CV nor : Constraint violation characteristic index, reflecting the degree to which a solution, i.e., individuals in the population, satisfy the constraints; it can be obtained by normalizing CV(x);

[0087] OB nor : Target characteristic index, which reflects the performance of the solution in terms of the optimization target. The optimization target of this embodiment is to minimize the mass. It is obtained by normalizing the mass of the vehicle predicted by the radial basis function neural network model of the mass.

[0088] In this embodiment, the vehicle weight is the vehicle mass;

[0089] FE nor : Feasibility characteristic index, which indicates the performance of the solution based on the feasibility criterion; obtained by normalizing the feasibility ranking of individuals;

[0090] CB nor : Constraint boundary characteristic index, indicating the closeness of the solution to the constraint boundary. It is obtained by calculating the CB(x) of each individual and then normalizing it. The calculation formula of CB(x) is as follows:

[0091]

[0092] in, represents the proxy prediction value of the nth constraint, and m represents the number of constraints.

[0093] SP nor: Sparsity characteristic index, which represents the distance between an individual and a solution that has been simulated in the decision space. It is obtained by normalizing the individual's sparse distance SP(x). The calculation formula of sparse distance SP(x) is as follows:

[0094]

[0095] Among them, Archive represents the collection of all simulated solutions, and distance represents the Euclidean distance between two individuals.

[0096] S52: Determine the optimization stage according to the degree to which the new individual satisfies the constraint conditions and the degree of closeness to the optimization target;

[0097] According to the Archive analysis, the current optimization stage is in the first stage when no individual (i.e. solution) that satisfies all constraints is found. When a solution that satisfies all constraints is found and the goal is still improving, it is in the second stage. When a solution that satisfies all constraints is found but the goal cannot be further improved, it is in the third stage.

[0098] Among them, Archive saves all simulated individuals and their corresponding goals and constraints information;

[0099] S53: According to different optimization stages, different goals are set based on the correlation between the optimization goal and the optimization constraints, and the Pareto dominance method is used to select non-dominated individuals as sampling solutions; specifically:

[0100] In the first stage, SP nor 、CV nor 、CV nor +OB nor As three objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions;

[0101] In the second stage, SP nor +FE nor , OB nor As two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions;

[0102] In the third stage, if μ ≥ 0, CB nor +SP nor 、μ*OB nor +(1-μ)*FE nor As two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions; otherwise, μ≤0, FE nor +SP nor 、CV norAs two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions, and the final sampling solution is recorded as O pareto ;

[0103] In the above scheme, through this multi-stage dynamic sampling criterion, dynamic monitoring and adaptive adjustment of the optimization progress can be achieved to ensure that the sampled solutions of each iteration have high quality and diversity. The first stage focuses on reducing the degree of constraint violation, thereby gradually guiding the population into the feasible domain; the second stage focuses on improving the target performance to accelerate the achievement of the optimal solution; the third stage, when the target performance stagnates, by adjusting the optimization target combination, the exploration ability of diverse solutions is improved to avoid falling into the local optimum. Overall, this method significantly enhances the global search capability of the optimization algorithm and the robustness of the results, which helps to achieve efficient and accurate multi-objective optimization.

[0104] S54: Retrieve the best individuals under the feasibility criteria and the target optimal criteria in Archive, and use the two individuals as the starting point to solve the optimal solution using the sequential quadratic programming method with the assistance of the multi-agent model, denoted as O star .

[0105] Optionally, a sequential quadratic programming method is used to solve the optimal solution with the assistance of a multi-agent model, including the following steps:

[0106] 54.1) For the starting point x0, first initialize the Lagrange multiplier λ0 and the matrix H0;

[0107] Among them, the matrix H0 is usually an estimate of the identity matrix or the initial Hessian matrix; x0 is the starting individual;

[0108] 54.2) Then iterate, for the kth iteration, based on the current iteration x k , construct and solve a quadratic programming subproblem. This subproblem is a quadratic approximation of the original nonlinear problem. The objective function of the quadratic programming subproblem is usually in the following form:

[0109]

[0110] Where p is the value from x k To new point x k+1 The moving direction of g(x k ) is the constraint violation degree estimated by the constraint agent model, f(x k ) Structural quality data estimated by the target agent;

[0111] 54.3) Calculate the new iteration point x based on the solved p k+1 =x k +α k p;

[0112] Among them, αk is the step size, which can be determined by the line search method;

[0113] 54.4) Update the Lagrange multiplier λ k+1 ;

[0114] 54.5) Update the approximate Hessian matrix H k+1 ,Common update methods include BFGS, DFP, etc.

[0115] 54.6) Check whether the gradient of the optimization objective, the feasibility condition (the degree of satisfaction of the constraints) and the step size p meet the preset convergence criteria;

[0116] 54.7) If the convergence conditions are met, such as the objective function value changes very little, the gradient is close to zero, or the maximum number of iterations is reached, the iteration is stopped and the optimization result is output;

[0117] The feasibility criterion is as described in step S21, and the target optimality criterion is to sort the targets according to their quality;

[0118] S55: Merge sampling solution O pareto And the optimal solution O star As the final sampling solution of this iteration O sample , is the optimal solution for this iteration;

[0119] S56: In LS-DYNA sample Carry out simulation analysis on the scheme in

[0120] In step S5 of this embodiment, the dynamic sampling criterion can be used to adaptively adjust the strategy in the multi-stage optimization process, thereby improving the optimization efficiency and the accuracy of the results. By selecting non-dominated individuals through the Pareto dominance method, potential high-quality solutions can be effectively retained to avoid falling into the local optimum too early. In addition, the use of the sequential quadratic programming method can quickly approximate the optimal solution with the assistance of the proxy model, reducing the burden of direct simulation calculations. Overall, this method significantly improves the efficiency of the optimization process and the quality of the solution, and helps to better balance the vehicle side protection performance and lightweight requirements.

[0121] S6, based on the final sampling solution obtained and the original parent individuals, a new parent individual is selected according to the feasibility criterion;

[0122] S61: Combine the NP parent individuals and the final sampling solution O after simulation analysis sample Merge into one population Popular;

[0123] S62: Sort the population Popular under the feasibility criterion and select the first NP individuals as the new parents;

[0124] S7, judging whether the set maximum number of simulation times has been reached, if so, the algorithm terminates and outputs the optimal solution, i.e., the design scheme for optimizing the vehicle side protection performance, otherwise, jumps to S2 for the next iteration;

[0125] In order to verify the performance of the present invention on the automobile side collision optimization problem.

[0126] A classic automobile side collision optimization case, namely the automobile side collision test configuration of the European Enhanced Vehicle Safety Committee, was selected. The mass of the side deformation barrier in this case is 1500 kg, and it hits a stationary vehicle at a speed of 50 km / h. In addition, the total number of elements in the simulation model is about 90,000, and the number of nodes is about 96,000. The optimization goal is to minimize the design quality of the automobile side under the constraints of the test dummy, including head injury standards, abdominal load, pubic symphysis force, viscosity standards and rib deflection (upper, middle and lower). For detailed information of the case, please refer to the document "Gu L, Yang RJ, Tho CH, et al. Optimisation and robustness for crashworthiness of side impact [J]. International journal of vehicle design, 2001, 26 (4): 348-360.". In the method of this embodiment, the algorithm parameters are set as follows: the population P size is set to NP = 60, and the maximum number of simulations is set to 200.

[0127] The optimization method of this embodiment is compared with four methods currently available for automobile side collision (SA-TADE, GLoSDE, OCSPSO, SParEA). Among them:

[0128] SA-TADE is an algorithm disclosed in the paper "Liu Y, Liu J, Jin Y, et al. A surrogate-assisted two-stage differential evolution for expensive constrained optimization [J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2023, 7(3): 715-730.";

[0129] GLoSDE is an algorithm disclosed in the paper "Wang Y, Yin DQ, Yang S, et al. Global and local surrogate-assisted differential evolution for expensive constrained optimization problems with inequality constraints [J]. IEEE transactions on cybernetics, 2018, 49(5): 1642-1656."

[0130] OCSPSO is an algorithm disclosed in the paper Zhang Y, Ji XF, Gao XZ, et al. Objective-constraint mutual-guided surrogate-based particle swarm optimization for expensive constrained multimodal problems[J]. IEEE Transactions on Evolutionary Computation, 2022, 27(4): 908-922.

[0131] SParEA is an algorithm disclosed in the document "Rahi KH, Singh HK, Ray T. Partial evaluation strategies for expensive evolutionary constrained optimization[J]. IEEE Transactions on Evolutionary Computation, 2021, 25(6): 1103-1117."

[0132] The quality corresponding to the optimal solution obtained by each method, i.e., the target data, is shown in Table 1. Among them, the smaller the target value, the better the algorithm effect, and the more optimal solutions the algorithm can provide. If the algorithm cannot find a solution that satisfies all constraints during operation, the result is represented by "NaN". According to the data analysis of Table 1, the method of this embodiment achieves optimal performance and surpasses all comparison methods. This is mainly due to the proposed enhanced offspring generation strategy, which takes into account the excellent local development ability of the genetic algorithm and the global search ability of the particle swarm algorithm. In addition, the dynamic sampling criterion takes into account multiple factors such as feasibility, diversity, and convergence, implements different sampling preferences at different evolutionary stages, and efficiently utilizes limited computing simulation resources, thereby obtaining the best solution. In actual production operations, users can reasonably configure the test settings according to their own specific circumstances to obtain the most satisfactory design solution.

[0133] Table 1 Comparison results of this embodiment and other four advanced evolutionary algorithms for optimizing vehicle side collision performance

[0134]

[0135] Example 2

[0136] Based on Example 1, this embodiment provides a multi-agent model-assisted optimization system for optimizing vehicle side protection performance, such as Figure 2 As shown, including:

[0137] Initialization module: configured to initialize the population and perform collision simulation on individuals in the population; the individuals include body structure parameters and material property information of the vehicle;

[0138] Correlation calculation module: configured to obtain the correlation relationship between the optimization target and the optimization constraint condition according to the collision simulation result;

[0139] Offspring population generation module: configured to generate offspring individuals using an enhanced offspring generation strategy, determine the population size based on the correlation relationship between the optimization target and the optimization constraint conditions, and form a new offspring population;

[0140] Optimization sampling module: It is configured to sample the offspring population according to the evaluation index determined by the constructed radial basis function neural network proxy model, adopt dynamic sampling criteria to obtain the optimal solution of this iteration; update the parent population of the next iteration for the next iteration until the termination condition is met to obtain the optimal solution.

[0141] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0142] Example 3

[0143] Based on Example 1, this embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps of a multi-agent model-assisted optimization method for optimizing the vehicle side protection performance described in Example 1 are completed.

[0144] Example 4

[0145] Based on Example 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in a multi-agent model-assisted optimization method for optimizing vehicle side protection performance described in Example 1 are completed.

[0146] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0147] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance, characterized in that: The steps include: Initialize a population and perform collision simulation on individuals in the population; the individuals include body structure parameters and material property information of the vehicle; According to the collision simulation results, the correlation relationship between the optimization target and the optimization constraint conditions is obtained; The enhanced offspring generation strategy is used to generate offspring individuals, and the population size is determined based on the correlation between the optimization goal and the optimization constraints to form a new offspring population. For the offspring population, according to the evaluation index determined by the constructed radial basis function neural network proxy model, dynamic sampling criteria are used to obtain the optimal solution of this iteration; the parent population of the next iteration is updated for the next iteration until the termination condition is met and the optimal solution is obtained.

2. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 1, characterized in that: The method for determining the correlation relationship between the optimization target and the optimization constraint condition includes the following steps: According to the individual vehicle collision performance index data, the individuals in the population are sorted according to the degree of satisfying the constraint conditions and the optimization goal, and the sorting one is obtained; According to the target optimal criterion retrieval, the individuals in the population are sorted from small to large according to the weight of the vehicle, which is recorded as sorting 2; For the two groups of data, sorted one and sorted two, similarity calculation is performed to obtain the correlation coefficient between the optimization objective and the optimization constraint conditions.

3. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 2, characterized in that: The feasibility criteria are determined as follows: When individual A satisfies all constraints and individual B cannot satisfy all constraints, then individual A is better than individual B; When both individuals A and B meet all constraints, if the mass of individual A is less than that of individual B, then individual A is better than individual B. Otherwise, individual B is better than individual B. When both individual A and individual B cannot satisfy all constraints, and the degree to which individual A violates all constraints is less than that of individual B, then individual A is better than individual B, and vice versa, individual B is better than individual A; The optimal criterion for the target is that the smaller the weight of the vehicle, the better.

4. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 1, characterized in that: Based on the radial basis function neural network proxy model, the construction method includes the following steps: Obtain historical individual information, retrieve all individuals that have been simulated for collision, and determine the effect of each individual after collision, which is divided into a target database and a constraint database; According to the target database, a radial basis function neural network model of mass is constructed as a proxy model of the target; it is used to predict the weight of the vehicle based on individual information; According to the constraint database, a radial basis function neural network model is constructed for each constraint condition as a proxy model of each constraint condition, which is used to predict the degree to which the individual satisfies the constraint condition according to the individual collision simulation results; Alternatively, the method for constructing a new offspring population comprises the following steps: Pair NP parent individuals in pairs to perform simulated binary crossover and polynomial mutation to generate NP intermediate offspring Offspring1; According to the feasibility criterion, the optimal individual after the collision simulation is retrieved and recorded as optg; According to the target optimal criterion, the best individual after collision simulation is retrieved and recorded as optf; If there are no individuals that meet all constraints, the two best individuals under the target optimality criterion are selected to form population 1, and the NP-2 individuals with the best feasibility criterion are selected to form population 2; If there are individuals that meet all constraints, and optg is equal to optf, then the best NP-2 individuals retrieved by the optimal criterion are taken to form population 1, and the two individuals with the best feasibility criterion are taken to form population 2; when optg is not equal to optf, max(2,μ*NP) individuals are taken to form population 1, and NP-max(2,μ*NP) individuals are taken to form population 2; μ is the correlation between the optimization objective and the optimization constraints; Population 1 and population 2 are optimized separately. After reaching the maximum number of iterations, the offspring population composed of the best solution of each individual is output, which is recorded as Offspring2; Merge Offspring1 and Offspring2 to form a new offspring population.

5. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 1, characterized in that: For the offspring population, according to the evaluation index determined by the constructed radial basis function neural network proxy model, dynamic sampling criteria are used to perform sampling to obtain the optimal solution for this iteration, including the following steps: Based on the radial basis function neural network agent model, the target indicators for offspring evaluation are determined; Determine the optimization stage according to the degree to which the new individual satisfies the constraints and how close it is to the optimization goal; According to different optimization stages, different goals are set based on the correlation between the optimization goal and the optimization constraints, and the Pareto dominance method is used to select non-dominated individuals as sampling solutions; The best individuals under the feasibility criteria and the target optimal criteria are retrieved from the Archive. The optimal solution is solved by using the sequential quadratic programming method with the assistance of the multi-agent model. star ; Merge Sampling Solution pareto And the optimal solution O star As the final sampling solution of this iteration O sample .

6. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 5, characterized in that: When no individual that satisfies all constraints is found, it is in the first stage; When a solution that satisfies all constraints is found and the goal is still improving, it is in the second stage; When a solution that satisfies all constraints is found but the objective cannot be improved further, it is in the third stage; Alternatively, in the first stage, use SP nor 、CV nor 、CV nor +OB nor As three objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions; In the second stage, SP nor +FE nor , OB nor As two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions; In the third stage, if μ ≥ 0, CB nor +SP nor 、μ*OB nor +(1-μ)*FE nor As two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions; otherwise, μ≤0, FE nor +SP nor 、CV nor As two objectives, the Pareto dominance method is used to select non-dominated individuals as sampling solutions, and the final sampling solution is recorded as O pareto ; CV nor is the constraint violation characteristic indicator, OB nor is the target characteristic index, FE nor is the feasibility characteristic indicator, CB nor is the constraint boundary characteristic index, SP nor It is a sparse feature indicator.

7. A multi-agent model-assisted optimization method for optimizing vehicle side protection performance according to claim 1, characterized in that: Based on the final sampling solution obtained and the original parent individual merged, a new parent individual is selected according to the feasibility criterion, including the following steps: The NP parent individuals and the final sampling solution O after simulation analysis sample Merge into one population Popular; Sort the population Popular according to the feasibility criterion and select the first NP individuals as the new parents.

8. A multi-agent model-assisted optimization system for optimizing vehicle side protection performance, characterized in that: include: Initialization module: configured to initialize the population and perform collision simulation on individuals in the population; the individuals include body structure parameters and material property information of the vehicle; Correlation calculation module: configured to obtain the correlation relationship between the optimization target and the optimization constraint condition according to the collision simulation result; Offspring population generation module: configured to generate offspring individuals using an enhanced offspring generation strategy, determine the population size based on the correlation relationship between the optimization target and the optimization constraint conditions, and form a new offspring population; Optimization sampling module: It is configured to sample the offspring population according to the evaluation index determined by the constructed radial basis function neural network proxy model, adopt dynamic sampling criteria to obtain the optimal solution of this iteration; update the parent population of the next iteration for the next iteration until the termination condition is met to obtain the optimal solution.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the multi-agent model-assisted optimization method for optimizing the side protection performance of a vehicle as described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a multi-agent model-assisted optimization method for optimizing vehicle side protection performance as described in any one of claims 1 to 7.

Images