Multi-objective optimization design method, system, storage medium and computer for pickup truck frames
By using a Gaussian process proxy model and a preference-driven mechanism, the problems of resource waste and inefficiency in the multi-objective optimization design of pickup truck frames are solved, achieving efficient and accurate frame design, focusing on the designer's preferred area, and improving design efficiency and reliability.
Patent Information
- Application Number
- CN202611288737.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing multi-objective optimization design methods for pickup truck frames are inefficient in incorporating designer preference-driven information into the optimization process, cannot effectively focus on the areas of concern to the designer, result in serious waste of computational resources, and lack effective evaluation of preferred areas by surrogate models.
By constructing a Gaussian process surrogate model, combined with Latin hypercube sampling and preference weight vector, the population is guided to efficiently generate candidate solutions within the preference region. Furthermore, a preference-enhancing Chebyshev scalarization function and preference dominance relationship are designed to optimize the concentration of search resources in the preference region. The surrogate model is then corrected using finite element simulation.
It significantly improves the efficiency of multi-objective optimization design of pickup truck frames, reduces the waste of computing resources, ensures that search resources are focused on the areas of concern to designers, and improves design accuracy and reliability.
Smart Images

Figure CN123116684A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of intelligent optimization design, surrogate model optimization and multi-objective evolutionary optimization, and particularly to a method, system, storage medium and computer for multi-objective optimization design of pickup truck frames. Background Technology
[0002] The pickup truck frame is the core structure for load-bearing the entire vehicle. As the mounting base for the powertrain, cab, cargo box, and suspension system, its structural performance directly determines the vehicle's load-bearing capacity. Noise, vibration, and acoustic roughness characteristics, crash safety, and fuel economy are all important considerations. As pickup trucks evolve towards multi-functionality and lightweight designs, chassis design must ensure sufficient bending stiffness, torsional stiffness, and good dynamic characteristics (modal characteristics) while maximizing weight reduction to decrease overall vehicle weight, reduce fuel consumption, and improve power performance. Different vehicle types (e.g., off-road, urban, and commercial) have significantly different weighting requirements for various performance indicators: off-road vehicles prioritize stiffness and strength, urban vehicles prioritize lightweighting and comfort, while commercial vehicles need to strike a balance between load-bearing capacity and lightweighting.
[0003] Traditional pickup truck frame design often relies on engineer experience or finite element simulation trial-and-error methods, typically optimizing a single performance index. This involves repeatedly adjusting parameters such as frame cross-sectional dimensions, wall thickness, and material grade, and then verifying the improved design through simulation. The frame structure involves more than ten continuous and discrete design variables, and these performance indices are interdependent. Existing methods still have the following shortcomings when dealing with preference-driven multi-objective optimization problems for pickup truck frames: First, traditional methods, such as... (Multi-objective evolutionary algorithms based on decomposition) typically search uniformly across the entire objective space, failing to effectively incorporate the designer's differentiated weighting preferences for different performance metrics into the search process. A significant amount of computational resources is wasted on non-preference regions that the designer is not concerned with. Secondly, existing preference fusion strategies are mostly posterior-based, meaning they are implemented only after obtaining the complete objective space. (Non-dominated solutions) The optimal frontier is then selected based on preferences, rather than guiding the search direction in real time during the optimization process, resulting in low computational efficiency; thirdly, the existing surrogate model's data collection function (such as the desired improvement) Confidence limit The first problem is that the design of a pickup truck frame multi-objective optimization method focuses only on the numerical improvement of the target value and lacks an effective mechanism to integrate preference region information into the selection of evaluation points. The second problem is that preference information is not systematically embedded in each stage of population initialization, sub-problem decomposition and final solution selection, resulting in insufficient synergy between preference-driven and search efficiency. Therefore, a multi-objective optimization design method that integrates prior preference knowledge into the entire optimization search process from initialization to final selection is needed. Summary of the Invention
[0004] To address the limitations of existing technologies, this invention provides a multi-objective optimization design method, system, storage medium, and computer for pickup truck frames. By integrating the designer's prior preference information into the entire optimization search process, search resources are concentrated in the preferred region of the target space, thereby improving the efficiency of multi-objective optimization design for pickup truck frames.
[0005] To achieve the above objectives, according to a first aspect of the present invention, a multi-objective optimization design method for pickup truck frames is provided, the method comprising the following steps:
[0006] In step S1, an initial optimization database is established based on the samples;
[0007] Using the length, width, height, and thickness parameters of the key cross-sections of the frame as continuous design parameters and the material parameters of the key cross-sections of the frame as discrete design parameters, based on the structural characteristics of the frame and combined with static load analysis, a simulation model and a multi-objective optimization mathematical model are constructed with bending stiffness, bending mode, and weight as optimization objectives and torsional stiffness and torsional mode as constraints. The target space preference region is defined according to the preference interval of each optimization objective, and an initial optimization database is established by combining Latin hypercube sampling with finite element simulation.
[0008] In step S2, the initial population is guided to generate efficiently within the preferred region;
[0009] A Gaussian process surrogate model is trained based on the initial optimization database. Reference points are sampled and generated within the preference region to construct the preference weight vector and decompose the sub-problems. The preference compliance score index is constructed using the surrogate model prediction information. Candidate solutions are selected from the design space to guide the initial population to be generated efficiently within the preference region.
[0010] In step S3, a preference enhancement function and preference dominance relationship are designed to guide the offspring's target values to concentrate in the preference region;
[0011] The design incorporates Chebyshev scalarization functions and preference dominance relationships to enhance the design preferences. Offspring are generated through crossover and mutation operations, and selection is driven by preference dominance relationships. A Gaussian process surrogate model is used to assist population evolution and guide the target values of offspring to concentrate in the preference region. At the same time, offspring solutions are selected for real finite element simulation to correct the surrogate model and ensure that search resources are focused on the preference region.
[0012] In step S4, preference-driven optimal solution selection is performed;
[0013] Based on the preference weight vector, a subproblem scalarization function is constructed, and the optimal solution is selected by preference-driven selection of the evolved population to obtain the set of candidate Pareto optimal solutions within the preference region of the current iteration;
[0014] In step S5, the candidate solution set is verified by finite element simulation. The simulation results are added to the optimization database and the Gaussian process surrogate model is updated. Then, the process returns to step S3 until the termination condition is met and the optimal solution is output.
[0015] Preferably, step S1 specifically includes:
[0016] S11 defines continuous design parameters and discrete design parameters; combined with the structural characteristics of the frame, the longitudinal beam section height, longitudinal beam section width, longitudinal beam web thickness, longitudinal beam flange thickness, crossbeam section height, crossbeam section width, crossbeam wall thickness, longitudinal position of the front crossbeam, longitudinal position of the middle crossbeam, frame length, frame width, and frame height are used as continuous design parameters; the longitudinal beam material grade, crossbeam material grade, longitudinal beam section type, crossbeam section type, and number of longitudinal beam stiffeners are used as discrete design parameters;
[0017] S12 constructs a three-dimensional parametric model of the chassis; based on the load conditions of the chassis structure and the materials selected for the chassis, a three-dimensional parametric model of the chassis is constructed using geometric modeling software.
[0018] Static load analysis using S13 finite element analysis software: The parametric model, the load conditions of the frame structure, and the key boundary conditions are imported into the finite element analysis software for static load analysis, resulting in simulation models of bending stiffness, bending mode, weight, torsional stiffness, and torsional mode.
[0019] S14 establishes a multi-objective optimization mathematical model for the chassis; the optimization objectives are to maximize bending stiffness, maximize bending modes, and minimize total weight, with torsional stiffness not lower than the design threshold. Torsional mode not lower than the design threshold To constrain the process, a multi-objective optimization mathematical model for the chassis is established, with the specific expression as follows: , ;
[0020] S15 sets the target preference range; designers specify preference ranges for each target based on the performance requirements of specific vehicle models (such as off-road, urban, and commercial); the preference area is defined as a hyperrectangle in the target space: ;
[0021] S16 performs finite element simulation evaluation on the initial sample; in the hybrid design space, Latin hypercube sampling is used to generate a scale of... The initial samples were used to perform finite element simulation evaluation on each sample, and their bending stiffness was calculated. Bending mode ,weight and torsional stiffness Torsional mode Construct an initial optimization database containing design variables, objective values, and constraint values. This provides a data foundation for the subsequent training of the Gaussian process surrogate model.
[0022] Preferably, step S2 specifically includes:
[0023] S21 decomposes the sub-problem; in the preference region Within, N uniformly distributed reference points are generated using Latin hypercube sampling. ;
[0024] S22 calculates the weight vector for the minimum target preference; the minimum value of each target in the current database is denoteed as the ideal point. For the k-th reference point, its corresponding preference weight vector is calculated using the following formula: ;
[0025] S23 performs a convex combination of the weight vector and the preference weight vector; to balance the ability to focus search within the preference region with the ability to explore the global space, an adaptive mixing coefficient is introduced. It dynamically adjusts based on the current distribution of the population within the preferred region. This coefficient is used to perform a convex combination of the uniformly distributed weight vector and the preferred weight vector. The calculation formula is as follows: ; ;
[0026] S24, train Gaussian process surrogate models for each objective using the initial optimization database; in the population initialization phase, first use the initial optimization database... Training Gaussian process surrogate models for each objective For any candidate solution , Give its first Predicted distribution of targets Define preference compliance score: ;
[0027] At the same time, utilizing the initial optimized database Training a Gaussian process surrogate model with various constraints For any candidate solution , Give its first Predicted distribution of each constraint Define the probability of constraint satisfaction. for and Joint probability: ;
[0028] S25, Define candidate solutions For subproblems The initial fitness function is defined; considering both preference compliance and convergence direction, candidate solutions are defined. For subproblems Initialization fitness function: ; .
[0029] Preferably, the preference-driven population evolution mechanism in step S3 specifically includes:
[0030] S31 defines the preference distance function; it measures any target vector. To preferred area Normalized weighted distance: ;
[0031] S32 constructs a preference-enhanced Chebyshev scalarization function: ;
[0032] S33, define the constraint-aware preference dominance relation; for any two solutions and First, the feasibility of the constraints is determined based on the predicted values of the GP model. The predicted mean of torsional stiffness is , The predicted mean for the torsional mode is A solution x is said to be constrained feasible if and only if and The constraint violation quantity for an infeasible solution is defined as: ;
[0033] S34 uses a Gaussian process surrogate model to predict the offspring target value; in each generation of evolution, for each subproblem k, a parent solution is randomly selected from its neighboring subproblems, and candidate solutions for the offspring are generated by simulating binary crossover and polynomial mutation. The Gaussian process surrogate model is then used to predict the offspring target value, and the calculation is performed. If the offspring candidate solutions favor the current solution or If the value is smaller, replace and update the current solution;
[0034] S35 updates the Gaussian process surrogate model for each objective and constraint; every interval Generation, selected from the current population Minimum value Each solution is evaluated using real finite element simulation. The simulation results are added to the optimization database, and the Gaussian process surrogate model for each objective and constraint is updated so that the surrogate model can maintain the prediction accuracy of the objective value in the preferred region in subsequent evolutionary generations.
[0035] Preferably, the preference-driven solution filtering in step S4 specifically includes:
[0036] S41 selects representative optimal solutions for each subproblem; after the current generation of evolution is completed, a preference confidence-guided selection function is constructed from the population. Select a subset of the population that simultaneously satisfies preference compliance and constraint feasibility. The process involves selecting representative optimal solutions for each subproblem; for each subproblem... The corresponding claims The second step generates a preference weight vector that directly points to the reference point within the preference region. The selection formula is as follows: ; ;
[0037] S42 performs deduplication on the representative optimal solutions of all subproblems; after deduplication on the representative optimal solutions of all subproblems, it outputs the candidate solutions within the preference region of the current iteration. Optimal solution set: ;
[0038] S43 passes the optimal solution set to step S5 for simulation verification; the deduplicated solution set is the candidate solution within the preference region obtained in the current iteration. The optimal solution set is then passed to step S5 for simulation verification.
[0039] Preferably, step S5 specifically includes:
[0040] S51 is the candidate solution set output in step S4. Each solution in the model is verified by a complete finite element simulation and then added to the optimization database.
[0041] S52 calculates the root mean square error of the prediction and retrains the Gaussian process surrogate model for each objective using a weighted method.
[0042] S53 Statistically analyzes the hypervolume covered by non-dominated solutions within the preferred region and makes a judgment; based on the updated database, it calculates the hypervolume index within the preferred region. It uses the upper bound vertex of the preferred region as a reference point and only counts the hypervolume covered by non-dominated solutions within the preferred region; if the current iteration and the previous iteration... Below the preset threshold Or the cumulative number of simulation evaluations reaches the preset limit. If convergence is achieved, the final frame parameter values and corresponding performance indicators are output, and the process is terminated. Otherwise, the updated proxy model is passed to step S3, and the process returns to execute the next round of preference-driven evolution.
[0043] Preferably, a pickup truck frame multi-objective optimization design system includes:
[0044] The model and database construction module is used to construct a simulation model and a multi-objective optimization mathematical model with bending stiffness, bending mode, and weight as optimization objectives and torsional stiffness and torsional mode as constraints, using the length, width, height and thickness parameters of the key cross-section of the frame as continuous design parameters and the material parameters of the key cross-section of the frame as discrete design parameters. The module defines the target space preference region according to the preference interval of each optimization objective and establishes an initial optimization database through Latin hypercube sampling combined with finite element simulation.
[0045] The preference-driven initialization module is used to train a Gaussian process surrogate model based on the initial optimization database, generate reference points in the preference region and construct a preference weight vector to decompose sub-problems, construct a preference compliance score index using the surrogate model prediction information, and screen candidate solutions from the design space to form an initial population.
[0046] The preference-driven evolution module is used to construct a preference-enhanced Chebyshev scalarization function and preference dominance relationship. It generates offspring through crossover and mutation, uses a Gaussian process surrogate model to assist population evolution and guide the target values of offspring to concentrate in the preference region, and selects the offspring solution for finite element simulation to correct the Gaussian process surrogate model.
[0047] The candidate solution selection module is used to construct a subproblem scaling function based on the preference weight vector and to select the optimal solution for the evolved population in a preference-driven manner to obtain a set of candidate Pareto optimal solutions within the preference region.
[0048] The simulation verification and iterative update module is used to perform finite element simulation verification on the candidate Pareto optimal solution set, add the simulation results to the optimization database and update the Gaussian process surrogate model, and output the optimal design parameters of the chassis when the termination condition is met; otherwise, it returns to the preference-driven evolution module to continue iterating.
[0049] Preferably, a computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the multi-objective optimization design method for pickup truck frames as described in any one of the claims.
[0050] Preferably, a computer includes a memory and a processor, the memory for storing a computer program and the processor for executing the computer program to implement the steps of the pickup truck frame multi-objective optimization design method described in any one of the following.
[0051] Compared with the prior art, the present invention has the following beneficial effects:
[0052] 1. By establishing a three-objective optimization model for bending stiffness, bending mode, and total weight, a Gaussian process proxy model is constructed to replace expensive finite element simulation. Based on a small number of simulation samples, the structural performance is accurately predicted, effectively balancing multiple performance indicators while significantly shortening the design cycle.
[0053] 2. Construct a preference weight vector to decompose the problem into subproblems, so that... The sub-questions densely cover the preference area; and a preference compliance score is constructed. By combining reference point information with the uncertainty predicted by the surrogate model, candidate solutions are screened from the design space to guide the efficient generation of the initial population within the preferred region, focusing on the preferred region from the search starting point.
[0054] 3. The traditional uniform domination criterion is replaced by a three-tiered progressive preference domination relationship, which causes the population to concentrate in the preferred area generation by generation, ensuring that the search resources are focused on the area of concern to the designer throughout the entire process.
[0055] 4. Effectively avoids screening bias caused by surrogate model prediction errors, and significantly improves the reliability of solutions within the preferred region. Attached Figure Description
[0056] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0057] Figure 1 This is a flowchart illustrating a preference-information-driven multi-objective optimization design method for pickup truck frames provided by the present invention. Detailed Implementation
[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and 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.
[0059] Example 1
[0060] This embodiment provides a multi-objective optimization design method for pickup truck frames, such as... Figure 1 As shown, the method is applicable to multi-objective optimization design of pickup truck frames based on the performance preferences of designers, specifically including steps S1 to S5.
[0061] In step S1, an initial optimization database is established based on the samples.
[0062] Using the length, width, height, and thickness parameters of the key cross-sections of the frame as continuous design parameters and the material parameters of the key cross-sections of the frame as discrete design parameters, based on the structural characteristics of the frame and combined with static load analysis, a simulation model and a multi-objective optimization mathematical model are constructed with bending stiffness, bending mode, and weight as optimization objectives and torsional stiffness and torsional mode as constraints. The target space preference region is defined according to the preference interval of each optimization objective, and an initial optimization database is established by combining Latin hypercube sampling with finite element simulation.
[0063] The specific steps of step S1 are as follows:
[0064] S11 defines continuous design parameters and discrete design parameters.
[0065] Based on the structural characteristics of the vehicle frame, the longitudinal beam section height, longitudinal beam section width, longitudinal beam web thickness, longitudinal beam flange thickness, crossbeam section height, crossbeam section width, crossbeam wall thickness, longitudinal position of the front crossbeam, longitudinal position of the middle crossbeam, frame length, frame width, and frame height are used as continuous design parameters; while the longitudinal beam material grade, crossbeam material grade, longitudinal beam section type, crossbeam section type, and number of longitudinal beam stiffeners are used as discrete design parameters.
[0066] S12 structural frame 3D parametric model
[0067] Based on the load conditions of the chassis structure and the materials selected for the chassis, a three-dimensional parametric model of the chassis is constructed using geometric modeling software.
[0068] Static load analysis using S13 finite element analysis software
[0069] The three-dimensional parametric model of the chassis, the load conditions of the chassis structure, and the key boundary conditions are imported into the finite element analysis software for static load analysis, resulting in simulation models of bending stiffness, bending mode, weight, torsional stiffness, and torsional mode.
[0070] S14 Establish a multi-objective optimization mathematical model for the chassis.
[0071] The optimization objectives are to maximize bending stiffness, maximize bending modes, and minimize total weight, while ensuring that the torsional stiffness is not lower than the design threshold. Torsional mode not lower than the design threshold To constrain the process, a multi-objective optimization mathematical model for the chassis is established, with the specific expression as follows:
[0072] ,
[0073]
[0074]
[0075] In the above formula, This indicates the search for the optimal solution to the design parameters. These represent the design parameters of the chassis, with the first 12 being consecutive design parameters. Indicates the height of the longitudinal beam section. Indicates the width of the longitudinal beam section. Indicates the thickness of the web of the longitudinal beam. Indicates the thickness of the longitudinal beam flange. Indicates the height of the beam section. Indicates the width of the beam cross section. Indicates the thickness of the beam wall. Indicates the longitudinal position of the front crossbeam. Indicates the longitudinal position of the middle crossbeam. Indicates the frame length. Indicates the width of the frame. This represents the frame height; the last five are discrete design variables. Indicates the material grade of the longitudinal beam. Indicates the material grade of the crossbeam. Indicates the cross-sectional shape of the longitudinal beam. Indicates the cross-sectional shape of the beam. Indicates the number of longitudinal beam reinforcement bars. This represents minimizing the negative values of bending stiffness, bending modes, and total weight. Indicates the frame in design variables The corresponding three-objective optimization objective function vector, where Indicates the frame's design parameters The corresponding negative value of bending stiffness at time Indicates the frame's design parameters The corresponding negative value of the bending mode, Indicates the frame's design parameters The corresponding total weight function, with the negative sign indicating that the maximization objective is transformed into a minimization form. This represents the constraints that need to be satisfied. Indicate design parameters The torsional stiffness of the frame must not be less than the minimum permissible torsional stiffness. Indicate design parameters The torsional modal frequency of the frame must not be lower than the minimum permissible torsional mode.
[0076] S15 sets the target preference range
[0077] Designers specify preference ranges for each target based on the performance requirements of specific vehicle types (such as off-road, urban, and commercial). The preference region is defined as a hyperrectangle within the target space.
[0078]
[0079] in, For preferred regions, To calculate its bending stiffness, For bending modes, For weight, This represents the preferred lower bound for bending stiffness. This represents the upper bound of the preference for bending stiffness. This is the lower bound for the preference of the bending mode. This represents the upper bound of the preference for bending modes. This is the lower bound for the preference for weight. This represents the upper bound of the weight preference;
[0080] S16 performs finite element simulation evaluation on the initial sample.
[0081] In a mixed-type design space, Latin hypercube sampling is used to generate a scale of The initial samples were used to perform finite element simulation evaluation on each sample, and their bending stiffness was calculated. Bending mode ,weight and torsional stiffness Torsional mode Construct an initial optimization database containing design variables, objective values, and constraint values. This provides a data foundation for the subsequent training of the Gaussian process surrogate model.
[0082] In step S2, the initial population is guided to generate efficiently within the preferred region.
[0083] A Gaussian process surrogate model is trained based on the initial optimized database. Reference points are sampled and generated within the preference region to construct the preference weight vector and decompose the sub-problems. The preference compliance score index is constructed using the surrogate model prediction information. Candidate solutions are selected from the design space to guide the initial population to be generated efficiently within the preference region.
[0084] The specific steps of step S2 are as follows:
[0085] S21 decomposes the subproblem.
[0086] In the preferred area Within, N uniformly distributed reference points are generated using Latin hypercube sampling. , where N is the preset number of subproblems;
[0087] S22 calculates the weight vector for the target minimum preference.
[0088] Let the minimum value of each objective in the current database be the ideal point. For the k-th reference point, its corresponding preference weight vector is calculated using the following formula:
[0089]
[0090] In the formula, To optimize the index number of the target, For bending stiffness, For bending modes, Total weight For the first The reference point at the ... Coordinate values in each target dimension For the ideal point at the th The components of the dimension, The coordinates of the reference point and the ideal point are at the th... Absolute distance in dimensional space, To prevent small constants with a denominator of zero, To from arrive Summation, where To calculate the summation index, a normalized denominator is calculated by iterating through bending stiffness, bending modes, and total weight. This formula ensures that the preference weight vector for each subproblem points to the corresponding reference point within the preference region. The individual questions densely cover the entire preference area;
[0091] S23 performs a convex combination of the weight vector and the preference weight vector.
[0092] To balance the ability to focus search within the preferred region with the ability to explore the global space, an adaptive mixing coefficient is introduced. It dynamically adjusts based on the current distribution of the population within the preferred region. This coefficient is used to perform a convex combination of the uniformly distributed weight vector and the preferred weight vector. The calculation formula is as follows:
[0093]
[0094]
[0095] In the formula, For the first Adaptive mixing coefficients during generational evolution This is the upper bound of the mixing coefficient. This is the lower bound of the mixing coefficient. For the first The target value falls into the preferred region in the generation population. The number of solutions, where N is the total number of subproblems. For the first Substitute all solutions within the preferred region Standard deviation of the value For the natural constant An exponential function with base 0, where the number of solutions within the preferred region... Increase and The distribution tends to be uniform. When the population size decreases, it indicates that the population has fully covered the preferred area. Automatically increasing the search resources accelerates the convergence of the search to the preferred region; conversely, when the solution is sparse or... When there are large fluctuations, Automatically reduce speed to retain more global exploration capabilities. The generated preference weight vector searches towards a reference point within the preference region. The weight vector is uniformly distributed in the target space and is used to preserve global exploration capabilities. This is the final search weight vector after convex combination;
[0096] S24, Train Gaussian process surrogate models for each objective using the initial optimized database.
[0097] During the population initialization phase, the initial optimization database is first utilized. Training Gaussian process surrogate models for each objective For any candidate solution , Give its first Predicted distribution of targets Define preference compliance score:
[0098]
[0099] In the formula, For candidate solutions Preference compliance rating The standard normal cumulative distribution function is... Gaussian process proxy model For candidate solutions In the The average of the predicted targets, Gaussian process proxy model For candidate solutions In the The standard deviation of the target prediction For the first The upper bound of the preference interval for each objective. For the first The lower bound of the preference interval for each objective. For the first The probability that the predicted value of a target value does not exceed the upper bound of the preference. For the first The probability that the predicted value of the nth objective value does not exceed the lower bound of the preference is the difference between the two. The probability of each target value falling within the preferred interval, the probabilities of bending stiffness, bending mode, and total weight are coupled in a product form, assuming that each target... Predict the joint probability that bending stiffness, bending mode, and total weight fall into their respective preferred intervals under independent conditions.
[0100] At the same time, utilizing the initial optimized database Training a Gaussian process surrogate model with various constraints For any candidate solution , Give its first Predicted distribution of each constraint Define the probability of constraint satisfaction. for and Joint probability:
[0101]
[0102] In the formula, , It also measures the joint probability of the three optimization objectives falling into the preference interval and the two constraints being satisfied, ensuring that the solutions in the initial population have both preference compliance and constraint feasibility.
[0103] S25, Define candidate solutions For subproblems Initialization fitness function
[0104] Taking into account both preference compliance and convergence direction, candidate solutions are defined. For subproblems Initialization fitness function:
[0105]
[0106]
[0107] In the formula, For candidate solutions For subproblems The initial fitness value is set, and a higher fitness value indicates that the solution is more suitable as a subproblem. The initial solution, For candidate solutions Given a preference weight vector and ideal point Scalar distance under given conditions To predict the mean and the ideal point at the th The strategy considers both the absolute distance in the dimension and the fact that the target value lies within the preferred region (by...). (driven) and approximating the ideal point along the preferred direction (by (Driven by) two objectives.
[0108] In step S3, a preference enhancement function and preference dominance relationship are designed to guide the offspring's target value to concentrate in the preference region.
[0109] The design incorporates Chebyshev scalarization functions and preference dominance relationships, generates offspring through crossover and mutation operations, and drives selection based on preference dominance relationships. A Gaussian process surrogate model is used to assist population evolution and guide the offspring's target values to concentrate in the preference region. Simultaneously, offspring solutions are selected for real finite element simulations to correct the surrogate model, ensuring that search resources are focused on the preference region.
[0110] The specific steps of step S3 are as follows:
[0111] S31 defines the preference distance function.
[0112] Define a preference distance function to measure any target vector. To preferred area Normalized weighted distance:
[0113]
[0114]
[0115] In the formula, target vector To preferred area Normalized weighted distance, To take the maximum of 0, the lower bound of preference minus the target value, and the target value minus the upper bound of preference, when the target value... Located in the preference range At that time, and ,thus A value of 0 indicates zero contribution from this item; when the target value is below the lower bound of the preference, , This positive value represents the degree of downward deviation; when the target value is higher than the upper bound of the preference, The positive value of max represents the degree of upward deviation. For the first The width of the preference interval for each target is used to normalize the deviation distance of targets with different dimensions. When the target vector is within the preference region, When deviating Furthermore, the greater the deviation, the larger the penalty value;
[0116] S32 constructs a preference-enhanced Chebyshev scalarization function.
[0117]
[0118] In the formula, To solve Given a mixed weight vector and ideal point The preference-enhancing Chebyshev scalar value under certain conditions; the smaller the value, the better the solution. For the first The target value on each target, For the target value and the ideal point at the th The absolute distance over the dimension, the first term is the standard Chebyshev scalar term after preference normalization, and the second term is the preference distance penalty term. This is the penalty coefficient;
[0119] S33 defines the constraint-perceived preference dominance relationship.
[0120] For any two solutions and First, the feasibility of the constraints is determined based on the predicted values of the GP model. The predicted mean of torsional stiffness is , The predicted mean for the torsional mode is A solution x is said to be constrained feasible if and only if and The constraint violation quantity for an infeasible solution is defined as:
[0121]
[0122] Based on this, it is called solution Preference Dominance Solution If and only if any of the following conditions is met: (a) The constraints are feasible. (b) The constraint is not feasible; All constraints are infeasible, and (c) All constraints are feasible, and one of the following sub-conditions is satisfied: (c1) The target vector of y falls within the preference region, while the target vector of y does not fall within the preference region; (c2) The target vectors are all within the preference region, and Dominance under traditional Pareto dominance (c3) The target vectors are all outside the preference region, and ,Right now It is closer to the preferred area.
[0123] This governing rule prioritizes constraint feasibility as the primary criterion, ensuring that all feasible solutions take precedence over infeasible solutions. For all feasible solutions, the regional solution relies on… The dominant selection mechanism, which compares the distances from the selected region to the preferred region, can guide the population to converge toward the decision-maker's preferred region while strictly satisfying the constraints.
[0124] S34 uses a Gaussian process surrogate model to predict the offspring target value.
[0125] In each generation of evolution, for each subproblem k, parent solutions are randomly selected from its neighboring subproblems. Candidate solutions for offspring are generated through simulated binary crossover and polynomial mutation. The target value for the offspring is predicted using a Gaussian process surrogate model, and the calculation... If the offspring candidate solutions favor the current solution or If the value is smaller, replace and update the current solution;
[0126] S35 updates the Gaussian process surrogate model for each objective and constraint.
[0127] Each interval Generation, selected from the current population Minimum value Each solution is evaluated using real finite element simulation. The simulation results are added to the optimization database, and the Gaussian process surrogate model for each objective and constraint is updated so that the surrogate model can maintain the prediction accuracy of the objective value in the preferred region in subsequent evolutionary generations.
[0128] In step S4, preference-driven optimal solution selection is performed.
[0129] Based on the preference weight vector, a subproblem scalarization function is constructed, and the optimal solution is selected by preference-driven selection of the evolved population to obtain the set of candidate Pareto optimal solutions within the preference region of the current iteration.
[0130] The specific steps of step S4 are as follows:
[0131] S41 selects representative optimal solutions for each subproblem.
[0132] After the previous generation of evolution has ended, a preference confidence-guided selection function is constructed to select from the population. Select a subset of the population that simultaneously satisfies preference compliance and constraint feasibility. The process involves selecting representative optimal solutions for each subproblem. For each subproblem... The corresponding claims The second step generates a preference weight vector that directly points to the reference point within the preference region. The selection formula is as follows:
[0133]
[0134]
[0135] In the formula, To solve Given a preference weight vector and ideal point The preference confidence-guided scalar value under the given conditions indicates that the smaller the value, the better the solution. To predict the mean and the ideal point at the th Absolute distance in dimensional space, Gaussian process proxy model Solution In the The standard deviation of the prediction over each objective represents the prediction uncertainty of the surrogate model at that point. The confidence penalty coefficient is... As a confidence modulator, when When the factor is small, it approaches 1, and the penalty effect is weak. When the value is large, this factor increases significantly, improving... The value makes this solution difficult to select. For in population subset Searching for the The solution that takes the minimum value To represent a population that satisfies the constraints and whose objective vector lies in the preference region. The set of all solutions within.
[0136] S42 performs deduplication on the representative optimal solutions of all subproblems.
[0137] The representative optimal solutions of all subproblems are deduplicated, and the candidate solutions within the preference region of the current iteration are output. Optimal solution set:
[0138]
[0139] In the formula, This is an operation to remove duplicate solutions.
[0140] S43 passes the optimal solution set to step S5 for simulation verification.
[0141] The deduplicated solution set is the candidate set within the preference region obtained in the current iteration. The optimal solution set is then passed to step S5 for simulation verification.
[0142] In step S5, the candidate solution set is verified by finite element simulation. The simulation results are added to the optimization database and the Gaussian process surrogate model is updated. Then, the process returns to step S3 until the termination condition is met and the optimal solution is output.
[0143] The specific steps of step S5 are as follows:
[0144] S51 is the candidate solution set output in step S4. Each solution is verified by a complete finite element simulation and added to the optimization database.
[0145] The candidate solution set output in step S4 Each solution is verified through complete finite element simulation to obtain the actual bending stiffness, bending mode, total weight, torsional stiffness, and torsional mode. The design variable vector of each solution, along with the actual target values and constraint values obtained from the simulation, are used as new samples and added to the optimization database.
[0146] S52 calculates the root mean square error of prediction and retrains the Gaussian process surrogate model for each objective using a weighted approach.
[0147] The actual target values obtained from simulation verification are compared with the predicted values from the Gaussian process surrogate model, and the root mean square error of the prediction is calculated. Simultaneously, the expanded database is used to select targets located in the preference region. Weights are assigned to samples within the sample. Samples located outside the preference region are assigned a weight of 1, and the Gaussian process surrogate model for each objective is retrained in a weighted manner, so that the surrogate model can obtain higher prediction accuracy within the preference region.
[0148] S53 Statistically analyze the hypervolume covered by non-dominated solutions located within the preferred region and make a judgment.
[0149] Based on the updated database, the hypervolume index within the preferred region is calculated. It uses the upper bound vertex of the preferred region as a reference point and only counts the hypervolume covered by non-dominated solutions within the preferred region. If the current iteration and the previous iteration... Below the preset threshold Or the cumulative number of simulation evaluations reaches the preset limit. If convergence is achieved, the final frame parameter values and corresponding performance indicators are output, and the process is terminated. Otherwise, the updated proxy model is passed to step S3, and the process returns to execute the next round of preference-driven evolution.
[0150] Furthermore, to verify the optimization performance of the method described in this embodiment, the three-objective constraint benchmark function DTLZ2 is used to verify the method. The DTLZ2 function has three optimization objectives and two constraints, and its expression is as follows:
[0151]
[0152]
[0153]
[0154]
[0155]
[0156] In the above formula, , and These are the first objective function, the second objective function, and the third objective function, respectively, and g(x) is the auxiliary constructor. This represents the first design parameter. This represents the second design parameter. These are the remaining design parameters within the design space, excluding the first two design parameters. It is the first A design parameter index, For the number of design parameters, These are the constraints that need to be satisfied. , These are two constraints.
[0157] The above benchmark functions are then subjected to the steps of a preference-information-driven multi-objective optimization method for pickup truck frames provided by this invention. To the steps The process was performed to obtain the experimental results.
[0158] To further illustrate this embodiment, a preference-information-driven multi-objective optimization design method for pickup truck frames, along with another classic and excellent non-dominated sorting genetic algorithm, is employed. For comparison, the maximum number of simulation evaluations in this embodiment is set to [number]. Next, the number of design parameters is set to... The experimental results are shown in the table. With the same number of simulation samples, the method in this embodiment achieves a better reverse generation distance (referred to as...). (value), and significantly better than The algorithm can be considered to have good performance in the multi-objective design optimization problem of pickup truck frame bending stiffness, bending mode and weight.
[0159] The method in this embodiment EMOA algorithm The obtained IGD value 0.4271 0.7577
[0160] Table 1. Comparison of optimization results of different methods
[0161] This invention provides a preference-information-driven multi-objective optimization design method for pickup truck frames. By constructing a Gaussian process proxy model to reduce the cost of finite element simulation, and designing preference-information-driven sub-problem decomposition, population initialization, population evolution, and confidence-guided solution selection mechanisms, it can fully integrate prior preference knowledge into the entire optimization search process according to the differentiated weight requirements of different vehicle models for bending stiffness, bending modes, and weight. While ensuring optimization accuracy, it avoids wasting computational resources in non-preference regions, providing a systematic solution for preference-driven multi-objective optimization design of pickup truck frame structures.
[0162] Example 2
[0163] This embodiment provides a multi-objective optimization design system for pickup truck frames, including a model and database construction module, a preference-driven initialization module, a preference-driven evolution module, a candidate solution screening module, and a simulation verification and iterative update module. Each module of the system corresponds to steps S1 to S5 in Embodiment 1.
[0164] The model and database construction module is used to determine the continuous and discrete design parameters of the chassis, establish a three-dimensional parametric model and a finite element simulation model of the chassis, construct a multi-objective optimization mathematical model of the chassis, define the target space preference region according to the optimization target preference interval, and generate an initial optimization database through Latin hypercube sampling and finite element simulation.
[0165] The preference-driven initialization module is used to train Gaussian process surrogate models corresponding to each optimization objective and each constraint based on the initial optimization database, generate reference points in the preference region and construct preference weight vectors, calculate preference compliance scores and constraint satisfaction probabilities based on the surrogate model's predicted mean and predicted standard deviation, and select candidate solutions from the design space to form an initial population.
[0166] The preference-driven evolution module is used to calculate the preference distance from the candidate solution to the preference region, construct the preference-enhanced Chebyshev scalarization function and the constraint-aware preference dominance relationship, generate offspring candidate solutions through crossover and mutation, and use the Gaussian process surrogate model to assist population evolution; at the same time, candidate solutions are selected according to preset conditions for real finite element simulation to update the Gaussian process surrogate model.
[0167] The candidate solution screening module is used to construct a preference confidence-guided screening function, select representative optimal solutions for each subproblem from candidate solutions that satisfy preference compliance and constraint feasibility, and obtain the set of candidate Pareto optimal solutions within the preference region of the current iteration after deduplication.
[0168] The simulation verification and iterative update module is used to perform a complete finite element simulation on the candidate Pareto optimal solution set, add the real target value and constraint value to the optimization database, perform weighted updates on the Gaussian process proxy model, and determine whether the termination condition is met based on the change in the hypervolume index of the preference region and / or the cumulative number of finite element simulations. If the termination condition is met, the optimal design parameters of the chassis and the corresponding performance index are output; otherwise, the module returns to the preference-driven evolution module to continue the next round of optimization.
[0169] Through data transfer and iterative coordination between the aforementioned modules, the system can continuously apply the prior preference information set by the designer for the target vehicle model to the multi-objective optimization process of the chassis, and reduce finite element simulation and search calculations in non-preference areas. The specific implementation methods and technical effects of each step in Example 1 are also applicable to this example.
[0170] Example 3
[0171] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the pickup truck frame multi-objective optimization design method described in Embodiment 1.
[0172] The computer-readable storage medium may include a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other media capable of storing computer-executable instructions. The specific implementation methods and technical effects of the method described in Embodiment 1 are also applicable to this embodiment.
[0173] Example 4
[0174] This embodiment provides a computer, including a memory and a processor. The memory is used to store computer programs, as well as chassis design parameters, target preference ranges, optimization databases, Gaussian process surrogate model parameters, and evolutionary population data; the processor is used to execute the computer program to implement the pickup truck chassis multi-objective optimization design method described in Embodiment 1.
[0175] The processor can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other processing units with data processing capabilities. The specific implementation methods and technical effects described in Example 1 are also applicable to this embodiment.
[0176] It should be noted that the memory can be volatile memory or non-volatile memory, or a combination of both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory of this invention is intended to include, but is not limited to, these and any other suitable types of memory.
[0177] A processor can be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method can be completed through integrated logic circuits in the processor's hardware or through software instructions. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute these functions.
[0178] This invention discloses various methods, steps, and logic block diagrams. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules within the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the aforementioned methods.
[0179] The method steps of this invention can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.
[0180] It should be noted that the explanation of the implementation method and beneficial effects of the above embodiment 1 also applies to this embodiment, and will not be elaborated in detail here to avoid redundancy.
[0181] The specific embodiments of the present invention have been described above. Based on the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention.
Claims
1. A multi-objective optimization design method for pickup truck frames, characterized in that, include: In step S1, an initial optimization database is established based on the samples; Using the length, width, height, and thickness parameters of the key cross-sections of the frame as continuous design parameters and the material parameters of the key cross-sections of the frame as discrete design parameters, based on the structural characteristics of the frame and combined with static load analysis, a simulation model and a multi-objective optimization mathematical model are constructed with bending stiffness, bending mode, and weight as optimization objectives and torsional stiffness and torsional mode as constraints. The target space preference region is defined according to the preference interval of each optimization objective, and an initial optimization database is established by combining Latin hypercube sampling with finite element simulation. In step S2, the initial population is guided to generate efficiently within the preferred region; A Gaussian process surrogate model is trained based on the initial optimization database. Reference points are sampled and generated within the preference region to construct the preference weight vector and decompose the sub-problems. The preference compliance score index is constructed using the surrogate model prediction information. Candidate solutions are selected from the design space to guide the initial population to be generated efficiently within the preference region. In step S3, a preference enhancement function and preference dominance relationship are designed to guide the offspring's target values to concentrate in the preference region; The design incorporates Chebyshev scalarization functions and preference dominance relationships to enhance the design preferences. Offspring are generated through crossover and mutation operations, and selection is driven by preference dominance relationships. A Gaussian process surrogate model is used to assist population evolution and guide the target values of offspring to concentrate in the preference region. At the same time, offspring solutions are selected for real finite element simulation to correct the surrogate model and ensure that search resources are focused on the preference region. In step S4, preference-driven optimal solution selection is performed; Based on the preference weight vector, a subproblem scalarization function is constructed, and the optimal solution is selected by preference-driven selection of the evolved population to obtain the set of candidate Pareto optimal solutions within the preference region of the current iteration; In step S5, the candidate solution set is verified by finite element simulation. The simulation results are added to the optimization database and the Gaussian process surrogate model is updated. Then, the process returns to step S3 until the termination condition is met and the optimal solution is output.
2. The method as described in claim 1, characterized in that, Step S1 specifically includes: S11 defines continuous design parameters and discrete design parameters; combined with the structural characteristics of the frame, the longitudinal beam section height, longitudinal beam section width, longitudinal beam web thickness, longitudinal beam flange thickness, crossbeam section height, crossbeam section width, crossbeam wall thickness, longitudinal position of the front crossbeam, longitudinal position of the middle crossbeam, frame length, frame width, and frame height are used as continuous design parameters; the longitudinal beam material grade, crossbeam material grade, longitudinal beam section type, crossbeam section type, and number of longitudinal beam stiffeners are used as discrete design parameters; S12 constructs a three-dimensional parametric model of the chassis; based on the load conditions of the chassis structure and the materials selected for the chassis, a three-dimensional parametric model of the chassis is constructed using geometric modeling software. Static load analysis using S13 finite element analysis software: The parametric model, the load conditions of the frame structure, and the key boundary conditions are imported into the finite element analysis software for static load analysis, resulting in simulation models of bending stiffness, bending mode, weight, torsional stiffness, and torsional mode. S14 establishes a multi-objective optimization mathematical model for the chassis; the optimization objectives are to maximize bending stiffness, maximize bending modes, and minimize total weight, with torsional stiffness not lower than the design threshold. Torsional mode not lower than the design threshold To constrain the process, a multi-objective optimization mathematical model for the chassis is established, with the specific expression as follows: , ; S15 sets the target preference range; designers specify preference ranges for each target based on the performance requirements of specific vehicle models (such as off-road, urban, and commercial); the preference area is defined as a hyperrectangle in the target space: ; S16 performs finite element simulation evaluation on the initial sample; in the hybrid design space, Latin hypercube sampling is used to generate a scale of... The initial samples were used to perform finite element simulation evaluation on each sample, and their bending stiffness was calculated. Bending mode ,weight and torsional stiffness Torsional mode Construct an initial optimization database containing design variables, objective values, and constraint values. This provides a data foundation for the subsequent training of the Gaussian process surrogate model.
3. The method as described in claim 1, characterized in that, Step S2 specifically includes: S21 decomposes the sub-problem; in the preference region Within, N uniformly distributed reference points are generated using Latin hypercube sampling. ; S22 calculates the weight vector for the minimum target preference; the minimum value of each target in the current database is denoteed as the ideal point. For the k-th reference point, its corresponding preference weight vector is calculated using the following formula: ; S23 performs a convex combination of the weight vector and the preference weight vector; to balance the ability to focus search within the preference region with the ability to explore the global space, an adaptive mixing coefficient is introduced. It dynamically adjusts based on the current distribution of the population within the preferred region. This coefficient is used to perform a convex combination of the uniformly distributed weight vector and the preferred weight vector. The calculation formula is as follows: ; ; S24, train Gaussian process surrogate models for each objective using the initial optimization database; in the population initialization phase, first use the initial optimization database... Training Gaussian process surrogate models for each objective For any candidate solution , Give its first Predicted distribution of targets Define preference compliance score: ; At the same time, utilizing the initial optimized database Training a Gaussian process surrogate model with various constraints For any candidate solution , Give its first Predicted distribution of each constraint Define the probability of constraint satisfaction. for and Joint probability: ; S25, Define candidate solutions For subproblems The initial fitness function is defined; considering both preference compliance and convergence direction, candidate solutions are defined. For subproblems Initialization fitness function: ; .
4. The method as described in claim 1, characterized in that, The preference-driven population evolution mechanism in step S3 specifically includes: S31 defines the preference distance function; it measures any target vector. To preferred area Normalized weighted distance: ; S32 constructs a preference-enhanced Chebyshev scalarization function: ; S33, define the constraint-aware preference dominance relation; for any two solutions and First, the feasibility of the constraints is determined based on the predicted values of the GP model. The predicted mean of torsional stiffness is , The predicted mean for the torsional mode is A solution x is said to be constrained feasible if and only if and The constraint violation quantity for an infeasible solution is defined as: ; S34 uses a Gaussian process surrogate model to predict the offspring target value; in each generation of evolution, for each subproblem k, a parent solution is randomly selected from its neighboring subproblems, and candidate solutions for the offspring are generated by simulating binary crossover and polynomial mutation. The Gaussian process surrogate model is then used to predict the offspring target value, and the calculation is performed. If the offspring candidate solutions favor the current solution or If the value is smaller, replace and update the current solution; S35 updates the Gaussian process surrogate model for each objective and constraint; every interval Generation, selected from the current population Minimum value Each solution is evaluated using real finite element simulation. The simulation results are added to the optimization database, and the Gaussian process surrogate model for each objective and constraint is updated so that the surrogate model can maintain the prediction accuracy of the objective value in the preferred region in subsequent evolutionary generations.
5. The method as described in claim 1, characterized in that, The preference-driven solution filtering in step S4 specifically includes: S41 selects representative optimal solutions for each subproblem; after the current generation of evolution is completed, a preference confidence-guided selection function is constructed from the population. Select a subset of the population that simultaneously satisfies preference compliance and constraint feasibility. The process involves selecting representative optimal solutions for each subproblem; for each subproblem... The corresponding claims The second step generates a preference weight vector that directly points to the reference point within the preference region. The selection formula is as follows: ; ; S42 performs deduplication on the representative optimal solutions of all subproblems; after deduplication on the representative optimal solutions of all subproblems, it outputs the candidate solutions within the preference region of the current iteration. Optimal solution set: ; S43 passes the optimal solution set to step S5 for simulation verification; the deduplicated solution set is the candidate solution within the preference region obtained in the current iteration. The optimal solution set is then passed to step S5 for simulation verification.
6. The method as described in claim 1, characterized in that, Step S5 specifically includes: S51 is the candidate solution set output in step S4. Each solution in the model is verified by a complete finite element simulation and then added to the optimization database. S52 calculates the root mean square error of the prediction and retrains the Gaussian process surrogate model for each objective using a weighted method. S53 Statistically analyzes the hypervolume covered by non-dominated solutions within the preferred region and makes a judgment; based on the updated database, it calculates the hypervolume index within the preferred region. It uses the upper bound vertex of the preferred region as a reference point and only counts the hypervolume covered by non-dominated solutions within the preferred region; if the current iteration and the previous iteration... Below the preset threshold Or the cumulative number of simulation evaluations reaches the preset limit. If convergence is achieved, the final frame parameter values and corresponding performance indicators are output, and the process is terminated. Otherwise, the updated proxy model is passed to step S3, and the process returns to execute the next round of preference-driven evolution.
7. A multi-objective optimization design system for pickup truck frames, characterized in that, include: The model and database construction module is used to construct a simulation model and a multi-objective optimization mathematical model with bending stiffness, bending mode, and weight as optimization objectives and torsional stiffness and torsional mode as constraints, using the length, width, height and thickness parameters of the key cross-section of the frame as continuous design parameters and the material parameters of the key cross-section of the frame as discrete design parameters. The module defines the target space preference region according to the preference interval of each optimization objective and establishes an initial optimization database through Latin hypercube sampling combined with finite element simulation. The preference-driven initialization module is used to train a Gaussian process surrogate model based on the initial optimization database, generate reference points in the preference region and construct a preference weight vector to decompose sub-problems, construct a preference compliance score index using the surrogate model prediction information, and screen candidate solutions from the design space to form an initial population. The preference-driven evolution module is used to construct a preference-enhanced Chebyshev scalarization function and preference dominance relationship. It generates offspring through crossover and mutation, uses a Gaussian process surrogate model to assist population evolution and guide the target values of offspring to concentrate in the preference region, and selects the offspring solution for finite element simulation to correct the Gaussian process surrogate model. The candidate solution selection module is used to construct a subproblem scaling function based on the preference weight vector and to select the optimal solution for the evolved population in a preference-driven manner to obtain a set of candidate Pareto optimal solutions within the preference region. The simulation verification and iterative update module is used to perform finite element simulation verification on the candidate Pareto optimal solution set, add the simulation results to the optimization database and update the Gaussian process surrogate model, and output the optimal design parameters of the chassis when the termination condition is met; otherwise, it returns to the preference-driven evolution module to continue iterating.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the pickup truck frame multi-objective optimization design method according to any one of claims 1-6.
9. A computer, characterized in that, It includes a memory and a processor, the memory being used to store a computer program and the processor being used to execute the computer program to implement the steps of the pickup truck frame multi-objective optimization design method according to any one of claims 1-6.
