Body-in-white structure optimization method, device, equipment, medium, structure and vehicle

Abstract

The embodiment of the invention provides a body-in-white structure optimization method and device, equipment, a medium, a structure and a vehicle, and the method comprises the steps: carrying out the iterative updating of a current optimization model of a body-in-white structure through a current optimization algorithm, and determining a target optimization model; a plurality of optimization algorithms are adopted to carry out iterative optimization on the target optimization model, adaptability scores corresponding to the plurality of optimization algorithms are determined, the adaptability scores are determined based on the plurality of performance index scores, and the optimization algorithms comprise a current optimization algorithm and candidate optimization algorithms; and when the adaptability score corresponding to the current optimization algorithm is not the maximum adaptability score, updating the target optimization model to the current optimization model, updating the candidate optimization algorithm with the maximum adaptability score to the current optimization algorithm, and repeating the steps. According to the scheme, the adaptive score real-time switching optimization algorithm determined based on multiple performance index scores can be realized, so that the corresponding solution scheme can be determined more efficiently and flexibly.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of automobile manufacturing engineering design, and in particular to a body-in-white structure optimization method, device, equipment, medium, structure and vehicle. BACKGROUND

[0002] Body structure design is crucial to the safety, stability and performance of the automobile, which can not only improve the overall performance of the vehicle, but also effectively reduce the weight of the body and improve fuel efficiency. Conceptual design is a crucial link in the process of body structure design, and the separate optimization strategy has the problem of inaccurate solution in the process of high-dimensional multi-constraint optimization of body-in-white. In the prior art, the mixed optimization strategy plays an obvious role in the optimization of body-in-white structure. However, the existing mixed optimization strategy mainly relies on the pre-set analysis model for switching, lacks adaptability to dynamic changes in the design process, and cannot cope with complex design scenarios. SUMMARY

[0003] Embodiments of the present application provide a body-in-white structure optimization method, device, equipment, medium, structure and vehicle, aiming to improve the problem of lack of dynamic adaptability of the existing mixed optimization strategy.

[0004] A body-in-white structure optimization method comprises: iteratively updating a current optimization model of a body-in-white structure by using a current optimization algorithm to determine a target optimization model; iteratively optimizing the target optimization model by using a plurality of optimization algorithms to determine a plurality of adaptability scores corresponding to the optimization algorithms, wherein the adaptability scores are determined based on a plurality of performance index scores, and the optimization algorithms include the current optimization algorithm and a candidate optimization algorithm; when the adaptability score corresponding to the current optimization algorithm is not the maximum adaptability score, updating the target optimization model to the current optimization model, updating the candidate optimization algorithm with the maximum adaptability score to the current optimization algorithm, and repeatedly executing the iteratively updating the current optimization model of the body-in-white structure by using the current optimization algorithm to determine the target optimization model.

[0005] In the embodiment, the current optimization algorithm is used to update and optimize the current optimization model of the body-in-white structure, the target optimization model is determined, multiple optimization algorithms are used to iteratively optimize the target optimization model, the adaptive scores corresponding to the multiple optimization algorithms are determined based on the multiple performance index scores, so that the adaptive scores can quantify the overall performance of the multiple optimization algorithms in iteratively optimizing the target optimization model; when the adaptive score of the current optimization algorithm is not the maximum adaptive score, it is determined that the overall performance of the current optimization algorithm is lower than the overall performance of the candidate optimization algorithm corresponding to the maximum adaptive score, and the candidate optimization algorithm corresponding to the maximum adaptive score is switched to optimize the updated current optimization model, so as to achieve real-time switching of the optimization algorithm based on the adaptive scores determined based on the multiple performance index scores, provide a more efficient and flexible solution for complex design problems in the body-in-white structure, and have a strong application prospect.

[0006] In an embodiment, after the current optimization algorithm is used to iteratively update the current optimization model of the body-in-white structure to determine the target optimization model, the method further comprises: updating the cooling countdown corresponding to the current optimization algorithm; when the cooling countdown is not zero, updating the target optimization model to the current optimization model, and repeatedly executing the step of iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model; when the cooling countdown is zero, executing the step of iteratively optimizing the target optimization model using multiple optimization algorithms to determine the adaptive scores corresponding to the multiple optimization algorithms.

[0007] In the embodiment, when the cooling countdown corresponding to the current optimization algorithm is not zero, the target optimization model is updated to the current optimization model without comparing the adaptive scores of the multiple optimization algorithms, and the updated current optimization model is still iteratively updated using the current optimization algorithm, so that the current optimization algorithm has enough time to explore the solution space and ensure that it can gradually converge to the global optimal solution. When the cooling countdown corresponding to the current optimization algorithm is zero, the step of iteratively optimizing the target optimization model using multiple optimization algorithms is executed to determine the adaptive scores corresponding to the multiple optimization algorithms, so that when the adaptive score corresponding to the current optimization algorithm is not the maximum adaptive score, the candidate optimization algorithm corresponding to the maximum adaptive score is switched to the current optimization algorithm to iteratively update the updated current optimization model, so as to achieve real-time switching of the optimization algorithm based on the adaptive scores determined based on the multiple performance index scores, and provide an efficient and flexible solution for complex design problems in the body-in-white structure.

[0008] In an embodiment, the method further comprises: When the cooling countdown is zero and the fitness score corresponding to the current optimization algorithm is the maximum fitness score, an optimal design scheme of the body-in-white structure is determined based on the target optimization model.

[0009] In the embodiment, when the cooling countdown corresponding to the current optimization algorithm is zero and the fitness score corresponding to the current optimization algorithm is the maximum fitness score, it is indicated that the current optimization algorithm has basically found the global optimal solution of the current optimization model, and the overall performance of the optimization algorithm is better than that of other candidate optimization algorithms. At this time, the subsequent optimization operation is not needed, and the optimization process of the body-in-white structure is ended. The optimal design scheme of the body-in-white structure is directly determined based on the target optimization algorithm updated by the last iteration of the current optimization algorithm.

[0010] In an embodiment, the method for determining the fitness score corresponding to each iteration of the optimization algorithm includes: The method for determining the fitness score corresponding to each iteration of the optimization algorithm includes: The method for determining the fitness score corresponding to each iteration of the optimization algorithm includes: The performance index scores include a preference utility score, a convergence speed score, and a solution diversity score. The preference utility score is used to represent the utility score of the preference prediction of the target preference model on the optimization design scheme.

[0011] In the embodiment, the multiple performance index scores, such as the preference utility score, the convergence speed score, and the solution diversity score, are introduced in the process of iteratively optimizing the target optimization model by using the multiple optimization algorithms. The fitness scores are comprehensively evaluated to achieve the purpose of dynamically switching the optimization algorithm according to the designer's preference, improving the optimization efficiency, and guaranteeing the overall performance of the optimal design scheme.

[0012] In an embodiment, the method for determining the fitness score corresponding to each iteration of the optimization algorithm includes: The method for determining the fitness score corresponding to each iteration of the optimization algorithm includes: Based on the target weight corresponding to each optimization algorithm, the multiple performance standard scores corresponding to each optimization design scheme corresponding to the same optimization algorithm are weighted and processed to determine the comprehensive performance score corresponding to each optimization design scheme. The average of the comprehensive performance scores of all the optimized design schemes obtained in each iteration of the same optimization algorithm is determined as the fitness score corresponding to each iteration of the optimization algorithm.

[0013] In this embodiment, the performance index scores corresponding to multiple optimization algorithms are first standardized to determine the corresponding performance standard scores, making the performance standard scores of multiple optimization algorithms comparable. Then, the performance standard scores corresponding to the same optimization design scheme are weighted so that the comprehensive performance score of each optimization design scheme can comprehensively reflect the overall performance of the optimization design scheme from dimensions such as designer preference, optimization efficiency, and optimization breadth. Finally, the maximum value of the multiple comprehensive performance scores corresponding to the same optimization algorithm is determined as the fitness score corresponding to the optimization algorithm, so that the fitness score can characterize the optimal performance that the optimization algorithm can achieve.

[0014] In one embodiment, before iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model, the body-in-white structure optimization method further includes: Based on the initial optimization model corresponding to the body-in-white structure, multiple training design schemes corresponding to the body-in-white structure are determined. Acquire multiple preference data, each of which represents the designer's preference for any two of the training design schemes; A Gaussian process model is used to process multiple preference data to determine the target preference model.

[0015] In this embodiment, multiple training design schemes are first determined based on the optimization model of the body-in-white structure. Based on these multiple training design schemes, the designer's preference data is collected, which reflects the designer's preference for different training design schemes. The multiple preference data are then analyzed based on a Gaussian process model to determine a target preference model that can effectively capture the designer's preferences. This allows the target preference model to accurately predict the preference probability of any design scheme, facilitating subsequent iterative optimization of the optimization algorithm.

[0016] In one embodiment, the step of processing multiple preference data using a Gaussian process model to determine a target preference model includes: Based on multiple training design schemes, the covariance matrix of multiple training design schemes is determined, and based on the covariance matrix, a zero-mean Gaussian process prior is constructed. Based on multiple preference data, determine the likelihood function for the multiple preference data; The Gaussian process posterior is determined based on the prior of the Gaussian process and the likelihood function of multiple preference data. Based on the Gaussian process posterior, the target preference model is determined.

[0017] In this embodiment, since multiple training designs and their preference data often exhibit nonlinear and non-monotonic relationships, Gaussian processes require kernel functions (covariance functions) to capture the relationships between different training designs and preferences. Therefore, a zero-mean Gaussian process prior can be constructed based on the determined covariance matrix. This process provides a powerful framework for Gaussian process modeling, enabling it to flexibly express the relationships between different training designs, making the subsequent optimization process more stable and efficient, and helping to ensure the reliability and accuracy of the output target preference model. Understandably, without a zero-mean Gaussian process prior, the constructed target preference model will lose its adaptability, especially when facing complex training designs. Based on the preference data, its likelihood function is determined. Using the Gaussian process prior and the likelihood function of the preference data, the Gaussian process posterior is derived using Bayes' theorem, so as to combine the Gaussian process prior with the predicted preferences in the preference data to obtain a more accurate inference of the latent function with greater uncertainty quantification capability. The Gaussian process posterior is then optimized to determine the final target preference model, enabling this target preference model to accurately and effectively determine the preference probability of any training design.

[0018] A body-in-white structure optimization device, comprising: The target optimization model determination module is used to iteratively update the current optimization model of the body-in-white structure using the current optimization algorithm, and determine the target optimization model. An adaptive score determination module is used to iteratively optimize the target optimization model using multiple optimization algorithms, and determine the adaptive scores corresponding to the multiple optimization algorithms. The adaptive scores are determined based on multiple performance index scores, and the optimization algorithms include the current optimization algorithm and candidate optimization algorithms. The optimization algorithm update module is used to update the target optimization model to the current optimization model when the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, update the candidate optimization algorithm with the largest maximum fitness score to the current optimization algorithm, and repeatedly execute the iterative update of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model.

[0019] An electronic device includes a processor and a memory, wherein, Memory, used to store computer programs; The processor is used to execute the program stored in the memory to implement the above-mentioned body-in-white structure optimization method.

[0020] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described body-in-white structure optimization method.

[0021] A body-in-white structure, wherein the design parameters of the body-in-white structure are determined based on the target optimization model determined by the above-mentioned body-in-white structure optimization method.

[0022] A vehicle comprising the aforementioned body-in-white structure. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for optimizing the body-in-white structure according to an embodiment of this application; Figure 2 yes Figure 1 A flowchart following step S101; Figure 3 yes Figure 1 A flowchart of step S102; Figure 4 yes Figure 3 A flowchart of step S302; Figure 5 yes Figure 1 A flowchart preceding step S101; Figure 6 yes Figure 5 A flowchart of step S503; Figure 7 This is another flowchart of the body-in-white structure optimization method in the embodiments of this application; Figure 8 This is a schematic diagram of the optimization process interface in an embodiment of this application; Figure 9 This is a schematic diagram of the preference collection interface in an embodiment of this application; Figure 10 This is a convergence curve diagram corresponding to different preference weights in the embodiments of this application; Figure 11 This is a structural diagram of the body-in-white structure optimization device provided in the embodiments of this application; Figure 12 This is a structural diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] This application provides a method for optimizing the body-in-white structure. This method is applicable to electronic devices with data processing capabilities and is used to optimize the design of the body-in-white structure to determine the optimal design scheme of the body-in-white structure. The optimal design scheme includes optimal parameter values ​​corresponding to multiple design parameters. The optimal parameter values ​​can be understood as parameter values ​​that enable the optimization objective to reach the optimal solution when the constraints are met.

[0026] The following explains some terms used in the embodiments of this application.

[0027] The body-in-white structure optimization model is a systematic framework that iteratively improves the performance (such as strength, stiffness, lightweighting, and safety) of the body structure based on modeling, using mathematical methods, simulation tools, and engineering constraints. Its core objective is to determine the parameter values ​​corresponding to multiple design parameters that satisfy optimization goals (such as weight reduction, improved collision safety, and reduced vibration and noise) while meeting the constraints corresponding to design requirements (such as regulations, cost, and manufacturing processes). The initial optimization model refers to the optimization model used for initial design. The current optimization model refers to the optimization model that needs to be optimized at present. The target optimization model refers to the optimization model determined by the current optimization algorithm after iteratively updating the current optimization model of the body-in-white structure.

[0028] An optimization algorithm is a series of rules and steps used to solve an optimization model and find the optimal solution (or near-optimal solution) to the optimization objective. Its core is the iterative process of handling the entire optimization problem (including the objective function, constraints, etc.). As an example, optimization algorithms can be, but are not limited to, genetic algorithms (GA), particle swarm optimization (PSO), simulated annealing (SA), and gradient descent (GD).

[0029] Here, the current optimization algorithm refers to the optimization algorithm used in the current iteration of the optimization process. This current optimization algorithm can be the initial design algorithm or the algorithm updated during the optimization model iteration process. Candidate optimization algorithms refer to other algorithms besides the current optimization algorithm. For example, when the current optimization algorithm is a genetic algorithm (GA), its corresponding candidate optimization algorithms can be, but are not limited to, particle swarm optimization (PSO), simulated annealing (SA), and gradient descent (GD).

[0030] This application provides a method for optimizing the body-in-white structure. The method is illustrated using an electronic device (hereinafter referred to as "device") as an example. Figure 1 As shown, the methods for optimizing the body-in-white structure include: S101: Iteratively update the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model; S102: Iteratively optimize the target optimization model using multiple optimization algorithms, determine the fitness scores corresponding to the multiple optimization algorithms, the fitness scores are determined based on the scores of multiple performance indicators, and the optimization algorithms include the current optimization algorithm and candidate optimization algorithms; S103: When the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, update the target optimization model to the current optimization model, update the candidate optimization algorithm with the maximum fitness score to the current optimization algorithm, and repeatedly perform iterative updates of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model.

[0031] The current optimization model refers to the optimization model that needs to be optimized at the moment. This current optimization model can be the initial design optimization model or the target optimization model after the last iteration update. The target optimization model refers to the optimization model determined by the current optimization algorithm after iteratively updating the current optimization model of the body-in-white structure. The target optimization model is generally the model after optimizing the parameter range of at least one design parameter in the current optimization model. That is to say, the current optimization model and the target optimization model have different parameter ranges for at least one design parameter.

[0032] As an example, in step S101, when the device iteratively updates the current optimization model of the body-in-white structure for the first time, it determines the initial design optimization model as the current optimization model, the initial design optimization algorithm as the current optimization algorithm, and uses the current optimization algorithm to iteratively update the current optimization model of the body-in-white structure to determine the target optimization model. Alternatively, when the device iteratively updates the current optimization model of the body-in-white structure after a previous iteration, it determines the target optimization model after the previous iteration update as the current optimization model, the candidate optimization algorithm determined in the previous iteration update as the current optimization algorithm, and uses the current optimization algorithm to iteratively update the current optimization model of the body-in-white structure to determine the target optimization model.

[0033] In this example, the current optimization algorithm is used to iteratively update the current optimization model of the body-in-white structure to determine the target optimization model. Specifically, under the premise of satisfying the optimization objective and constraints, the parameter ranges corresponding to multiple design parameters are updated so that the target optimization model can be iteratively optimized in the future, so that it can search for the optimal solution more quickly or accurately, and thus determine the optimal design scheme corresponding to the body-in-white structure.

[0034] In this example, the body-in-white structure can be, but is not limited to, a truss body structure, a monocoque body structure, and a space-frame body structure. The design parameters, optimization objectives, and constraints of the optimization model for different body-in-white architectures can be adaptively changed according to the specific circumstances.

[0035] Taking the truss-type body structure as an example, its corresponding optimization model is as follows: Where X represents the structural parameters and material type. The optimization objective is to minimize the vehicle body mass Mass(X). In the constraints, Disp(X) represents the maximum displacement of the structure; Stress(X) represents the maximum equivalent stress of the vehicle body. and These represent the lower and upper limits for the overall length, respectively, and are dynamically determined based on vehicle type, regulations, and target performance, requiring compliance with constraints under all operating conditions. To account for manufacturing process errors, a safety factor of 0.6-0.8 has been added to the yield strength of each material. The design parameters and their corresponding ranges are shown in Table 1 below.

[0036] Table 1 Design parameters and parameter ranges for the body-in-white structure Among them, the critical reinforcement structure is the key component that bears the main load, including but not limited to A / B pillars, side sills, and front and rear longitudinal beams. The basic reinforcement structure is the secondary component that transfers local loads and maintains rigidity, including but not limited to floor beams and front and rear subframe connecting beams. The general reinforcement structure consists of localized ribs or braces, including but not limited to plate flanges, support stiffeners, and angle iron supports. The side lengths and thicknesses of the critical reinforcement structure, basic reinforcement structure, and general reinforcement structure must meet the following requirements. This constraint sets the thickness of the high-level reinforcement structure to the side length of the low-level reinforcement structure, giving the constructed optimization model the following advantages: it can establish a simple linear relationship between graded dimensions; it ensures smooth stiffness gradient and reduces stress concentration; it limits the number of design variables and reduces the optimization dimension; and it facilitates the sharing of plate thickness in manufacturing processes and reduces mold change costs.

[0037] Among them, the performance index score refers to the evaluation score of the performance index. The performance index here can be, but is not limited to, used to characterize quality, efficiency, designer preferences, etc.

[0038] As an example, in step S102, after the device determines the target optimization model in each iteration update, it can use multiple optimization algorithms (including the current optimization algorithm and at least one candidate optimization algorithm) to perform iterative optimization processing on the target optimization model respectively, and determine the scores of multiple performance indicators corresponding to each optimization algorithm; then, the scores of multiple performance indicators corresponding to each optimization algorithm are integrated to determine the fitness score corresponding to the optimization algorithm. The fitness score can quantify the overall performance of different optimization algorithms in iteratively optimizing the target optimization model, so that the overall performance of multiple optimization algorithms is comparable.

[0039] As an example, in step S103, after obtaining the fitness scores corresponding to multiple optimization algorithms, the device needs to compare the multiple fitness scores to determine the maximum fitness score and evaluate whether the fitness score corresponding to the current optimization algorithm is the maximum fitness score. When the fitness score corresponding to the current optimization algorithm is the maximum fitness score, it can be determined that the current optimization algorithm has the best performance in solving the optimization model, and in this case, there is no need to switch optimization algorithms. Conversely, when the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, that is, there is at least one candidate optimization algorithm with a fitness score greater than the fitness score corresponding to the current optimization algorithm, that is, there is at least one candidate optimization algorithm with better performance in solving the optimization model than the current optimization algorithm, in this case, the target optimization model needs to be updated to the new current optimization model, and the candidate optimization algorithm corresponding to the maximum fitness score needs to be updated to the new current optimization algorithm. Step S101 is then repeated, that is, the current optimization model of the body-in-white structure is iteratively updated using the current optimization algorithm to determine the target optimization model.

[0040] In this embodiment, the current optimization algorithm is used to optimize and update the current optimization model of the body-in-white structure. To determine the target optimization model, multiple optimization algorithms are needed to iteratively optimize the target optimization model. The corresponding fitness score is determined based on multiple performance index scores so that the fitness score can quantify the overall performance of different optimization algorithms in iteratively optimizing the target optimization model. When the fitness score of the current optimization algorithm is not the maximum fitness score, it is determined that the overall performance of the current optimization algorithm is lower than the overall performance of the candidate optimization algorithm corresponding to the maximum fitness score. It is necessary to switch the candidate optimization algorithm corresponding to the maximum fitness score to optimize the updated current optimization model. This achieves real-time switching of optimization algorithms based on the fitness score determined by multiple performance index scores, providing a more efficient and flexible solution for complex design problems in the body-in-white structure, making it have strong application prospects.

[0041] In one embodiment, after step S101, i.e., after iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm and determining the target optimization model, the method further includes: S201: Update the cooldown countdown for the current optimization algorithm; S202: When the cooling countdown is not zero, update the target optimization model to the current optimization model, and repeatedly execute the iterative update of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model; S203: When the cooling countdown is zero, perform iterative optimization of the target optimization model using multiple optimization algorithms to determine the fitness scores corresponding to the multiple optimization algorithms.

[0042] The CoolDownCounter is a counter variable used by the optimization algorithm during the search process to control the decay rate or determine the duration of a certain state. Generally, after each iteration of optimization using an optimization algorithm, the CoolDownCounter needs to be updated to decrease by 1.

[0043] As an example, in step S201, after the device iteratively updates the current optimization model of the body-in-white structure using the current optimization algorithm and determines the target optimization model, it needs to update the cooling countdown N corresponding to the current optimization algorithm. Specifically, N = N - 1 is used to update the cooling countdown so that the cooling countdown N can represent the remaining number of iterations for the current optimization algorithm. Generally, after any optimization algorithm is determined as the current optimization algorithm, the cooling countdown needs to be initialized based on the target number of iterations for that optimization algorithm. This allows the cooling countdown to be decremented by 1 after each iteration update of the current optimization model by the current optimization algorithm, so that the current optimization algorithm can be evaluated as to whether the iteration update has ended based on the real-time updated cooling countdown.

[0044] As an example, in step S202, if the updated cooling countdown of the device is not zero, it indicates that the number of iterations of the current optimization model using the current optimization algorithm has not reached the target number of the initial configuration. At this time, it is necessary to update the target optimization model to the current optimization model and repeat the iteration update of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model, so that the current optimization algorithm has enough time to explore the solution space and gradually converge to the global optimal solution, thereby achieving the effect of gradually optimizing the parameter range of the current optimization model of the body-in-white structure.

[0045] As an example, in step S203, when the updated cooling countdown of the device is zero, it indicates that the number of times the current optimization algorithm has been used to iterate and optimize the current optimization model has reached the target number of the initial configuration. In other words, the current optimization algorithm has basically found the global optimal solution of the current optimization model. This global optimal solution is the result of the current optimization algorithm solving the current optimization model. However, this result may differ from the results solved by other candidate optimization algorithms and may not effectively reflect the true optimal solution of the current optimization model. Therefore, it is necessary to perform iterative optimization of the target optimization model using multiple optimization algorithms and determine the fitness scores corresponding to multiple optimization algorithms. Based on the fitness scores of multiple optimization algorithms, it is possible to evaluate whether it is necessary to switch optimization algorithms to iteratively optimize the current optimization model.

[0046] In this embodiment, when the cooling countdown corresponding to the current optimization algorithm is not zero, there is no need to compare the fitness scores of multiple optimization algorithms. The target optimization model is updated to the current optimization model, and the current optimization algorithm is still used to iteratively update the updated current optimization model. This allows the current optimization algorithm sufficient time to explore the solution space and ensures that it gradually converges to the global optimum. When the cooling countdown corresponding to the current optimization algorithm is zero, the device needs to perform iterative optimization of the target optimization model using multiple optimization algorithms to determine the fitness scores of multiple optimization algorithms. When the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, the candidate optimization algorithm corresponding to the maximum fitness score is switched as the current optimization algorithm, and the updated current optimization model is iteratively updated. This achieves real-time switching of optimization algorithms based on the fitness score determined by multiple performance index scores, providing an efficient and flexible solution for complex design problems in body-in-white structures.

[0047] In one embodiment, the method further includes: When the cooling countdown is zero and the fitness score corresponding to the current optimization algorithm is the maximum fitness score, the optimal design scheme of the body-in-white structure is determined based on the target optimization model.

[0048] As an example, when the cooling countdown corresponding to the current optimization algorithm is zero and the fitness score corresponding to the current optimization algorithm is the maximum fitness score, it means that the current optimization algorithm has basically found the global optimal solution of the current optimization model. Moreover, the overall performance of the current optimization algorithm in solving the optimization problem is better than that of other candidate optimization algorithms. At this point, no further optimization operation is needed, and the optimization process of the body-in-white structure can be ended. The optimal design scheme of the body-in-white structure can be determined directly based on the target optimization algorithm updated in the last iteration of the current optimization algorithm.

[0049] When using a hybrid optimization strategy in the optimization of the body-in-white structure, in addition to the lack of adaptability to dynamic changes in the design process and the inability to cope with complex design scenarios, there is also the problem that the optimization design process lacks consideration of the designer's subjective preferences, which may lead to the optimization results not being fully matched with actual needs and reducing the satisfaction with the design solution.

[0050] In one embodiment, such as Figure 3 As shown, step S102 involves iteratively optimizing the target optimization model using multiple optimization algorithms and determining the fitness scores corresponding to the multiple optimization algorithms, including: S301: Use an optimization algorithm to iteratively optimize the target optimization model, and determine the multiple optimization design schemes corresponding to each iteration of the optimization algorithm and the multiple performance index scores corresponding to each optimization design scheme. S302: Based on the performance index scores of all optimization design schemes corresponding to each iteration of the same optimization algorithm, determine the fitness score corresponding to each iteration of the optimization algorithm; Among them, several performance index scores include preference utility score, convergence speed score and solution diversity score. The preference utility score is used to characterize the utility score of the pre-trained target preference model for predicting the preference of the optimized design scheme.

[0051] The utility score is a quantitative indicator used to measure the usefulness or quality of an optimal design solution, representing the degree to which the optimal design solution meets the optimization objective. The preference utility score is determined based on preference predictions of the optimal design solution using a goal preference model.

[0052] Among them, the convergence speed score is used to characterize the speed at which the optimized design scheme approaches the optimal design scheme. Specifically, it reflects the speed at which the objective function value of the optimized design scheme approaches the objective function value of the optimal design scheme. This convergence speed score is used to reflect the efficiency of the optimization algorithm in finding the optimal design scheme.

[0053] The Solution Diversity score is used to characterize the degree of difference between any optimal design scheme and other optimal design schemes. It reflects the distribution range of the optimal design schemes and can ensure the breadth of the algorithm search, so as to avoid the algorithm getting stuck in local optima and to find the global optimum, thereby ensuring the overall performance of its corresponding optimal design scheme.

[0054] As an example, in step S301, when the device iteratively optimizes the target optimization model using each optimization algorithm, it outputs multiple optimization design schemes corresponding to each iteration. After the optimization algorithm completes its iterative optimization, it can predict the preferences of each optimization design scheme based on a pre-trained target preference model to determine the preference utility score corresponding to each optimization design scheme. The optimal design scheme is then determined from among the multiple optimization design schemes. Based on the optimization design scheme and the optimal design scheme, the convergence speed score corresponding to each optimization design scheme is determined. Finally, based on each optimization design scheme and other optimization design schemes output by the same optimization algorithm, the solution diversity score corresponding to the optimization design scheme is determined. In this example, determining the preference utility score, convergence speed score, and solution diversity score corresponding to each optimization design scheme allows for the evaluation of the performance of the optimization design scheme from dimensions such as designer preference, optimization efficiency, and optimization breadth.

[0055] As an example, in step S302, the device comprehensively analyzes and determines the fitness score of the optimization algorithm based on the preference utility score, convergence speed score, and solution diversity score of all optimization design schemes corresponding to each iteration of the same optimization algorithm. This allows the fitness score to comprehensively reflect the overall performance of the optimization algorithm from dimensions such as designer preference, optimization efficiency, and optimization breadth.

[0056] In this embodiment, during the iterative optimization of the target optimization model using multiple optimization algorithms, multiple performance index scores such as preference utility score, convergence speed score, and solution diversity score are introduced to comprehensively evaluate its adaptability score. This aims to achieve dynamic switching of optimization algorithms according to the designer's preferences, while also improving optimization efficiency and ensuring the overall performance of the optimal design scheme, thereby enhancing satisfaction with the design scheme.

[0057] In one embodiment, such as Figure 4 As shown, step S302, which involves determining the fitness score for each iteration of the optimization algorithm based on the performance index scores of all optimized design schemes for each iteration of the same optimization algorithm, includes: S401: Standardize the scores of all performance metrics corresponding to multiple optimization algorithms to determine multiple performance standard scores for each optimization design scheme. The performance standard scores include the preference utility standard score, the convergence speed standard score, and the solution diversity standard score. S402: Based on the target weight corresponding to each optimization algorithm, the multiple performance standard scores corresponding to each optimization design scheme of the same optimization algorithm are weighted to determine the comprehensive performance score corresponding to each optimization design scheme. S403: The average of the comprehensive performance scores of all optimized design schemes obtained in each iteration of the same optimization algorithm is determined as the fitness score of each iteration of the optimization algorithm.

[0058] As an example, during the iterative optimization of the target optimization model using each optimization algorithm, the device initializes the fitness matrix (AAM) and related data structures (such as the archive of best design solutions). Based on this, after determining multiple optimization design solutions for iterative optimization of the target optimization model using the optimization algorithm, the device also performs performance testing on each optimization design solution to determine multiple performance index scores corresponding to the optimization design solution. These performance index scores include preference utility score, convergence speed score, and solution diversity score. The performance index scores corresponding to each optimization design solution are updated to the fitness matrix (AAM) so that the fitness matrices corresponding to all optimization algorithms can be integrated after iterative optimization of all optimization algorithms.

[0059] In this example, the initialization formula for the fitness matrix is: in Indicates the first An optimization algorithm, These represent the preference utility score, convergence speed score, and solution diversity score of the optimization algorithm, respectively. It is the number of algorithms. It represents the number of iterations for optimization.

[0060] As an example, in step S401, to ensure the comparability of performance metrics for all optimization algorithms, it is necessary to standardize the scores of all performance metrics corresponding to multiple optimization algorithms to determine the multiple performance standard scores corresponding to each optimization design scheme. In this example, the standardization of the scores of all performance metrics corresponding to multiple optimization algorithms can be determined using the following formula: in It is the performance score of a certain optimization algorithm on a certain performance metric. and These are the minimum and maximum scores for this performance indicator across all performance indicators, respectively. Standardized performance standard score. Ensure all performance metrics are within the range Within this scope, it facilitates subsequent comprehensive evaluation.

[0061] As an example, in step S402, after determining the multiple performance standard scores corresponding to each optimized design scheme, the device can determine the target weight corresponding to each optimization algorithm based on the actual situation. This target weight is then used to weight the multiple performance standard scores corresponding to the same optimized design scheme for that optimization algorithm, thereby determining the comprehensive performance score for each optimized design scheme. In this example, multiple optimized design schemes corresponding to the same optimization algorithm are weighted using the same target weight, so that the calculated comprehensive performance score of the multiple optimized design schemes can reflect the overall performance differences of the multiple optimized design schemes obtained during the iterative optimization process of the optimization algorithm.

[0062] For example, the adaptability score of each optimized design scheme is determined by a weighted calculation based on the scores of multiple performance indicators corresponding to the optimized design scheme. Specifically, the target weight corresponding to each optimization algorithm is determined, for example, let... The weights for the performance metrics of preference utility, convergence speed, and solution diversity are respectively given, and the overall performance score for each optimization design scheme is calculated. The calculation formula is: in, These are preference weights, speed weights, and diversity weights, set according to the actual optimization objectives. They represent the first The optimization algorithm in the th ... In each iteration of optimization, the standard scores for preference utility, convergence speed, and solution diversity are determined for the optimal design scheme.

[0063] The preference parameters were determined through experimental setup. The experimental group aimed to analyze the impact of preference weights on optimization results and determine the optimal balance between designer preferences and other performance metrics. The influence of different preference weights in the scoring function on optimization performance was investigated. Specifically, preference weights were set to 10, 15, 30, and 40, while all other conditions remained constant. By systematically changing the preference weights, this group aimed to determine the optimal balance between user preference feedback and algorithm performance metrics. The research results are expected to reveal how preference weights affect the dynamic switching process and the overall optimization results.

[0064] The test results are as follows Figure 10 As shown, the convergence curves illustrate the optimization performance of different preference weights (represented by blue, orange, green, and red curves, respectively) on the body-in-white structure. By comparing the curves, it can be seen that higher preference weights significantly improve performance by balancing exploration and development, enabling the algorithm to reach the global optimum more efficiently at the same convergence speed. Conversely, lower preference weights lead to slower convergence speeds and suboptimal solutions, indicating reduced effectiveness in guiding the transition to high-potential regions.

[0065] As an example, in step S403, the device determines the comprehensive performance score of all optimized design schemes corresponding to each iteration of the same optimization algorithm. Then, it is necessary to calculate the overall performance score corresponding to all optimized design schemes. The average value is used as the fitness score for the current iteration of the optimization algorithm. This allows the fitness score to reflect the optimal performance of the optimization algorithm during iterative optimization, so that it can be compared with other fitness scores to assess whether a switch to a different optimization algorithm is needed.

[0066] In this example, the fitness score of the candidate optimization algorithm Exceeding the current optimization algorithm If the adaptive score is reached and the cooldown countdown (CoolDownCounter) has reached zero, an algorithm switch is triggered. At this time, the current optimization algorithm... It will be replaced with If the switching conditions are not met, the current optimization algorithm will continue to be maintained, and the cooldown countdown will be decremented by 1 until the cooldown countdown reaches zero, at which point the iterative optimization process will end.

[0067] The switching condition can be expressed as: in, It is the switching threshold. It is the fitness score corresponding to the current iteration of the optimization algorithm. It is the fitness score of the current optimization algorithm. Cooling countdown begins.

[0068] In this embodiment, the performance index scores of all optimization algorithms are first standardized to determine the corresponding performance standard scores, making the performance standard scores of multiple optimization algorithms comparable. Then, the performance standard scores of the same optimization design scheme are weighted so that the comprehensive performance score of each optimization design scheme can comprehensively reflect the overall performance of the optimization design scheme from dimensions such as designer preference, optimization efficiency, and optimization breadth. Finally, the maximum value of the multiple comprehensive performance scores corresponding to the same optimization algorithm is determined as the fitness score of the optimization algorithm, so that this fitness score can characterize the optimal performance that the optimization algorithm can achieve.

[0069] In one embodiment, such as Figure 5 As shown, before step S101, that is, before iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm and determining the target optimization model, the body-in-white structure optimization method further includes: S501: Based on the initial optimization model corresponding to the body-in-white structure, determine multiple training design schemes corresponding to the body-in-white structure; S502: Obtain multiple preference data, each of which is used to characterize the designer's preference for any two training design schemes; S503: Use a Gaussian process model to process multiple preference data to determine the target preference model.

[0070] The training design scheme is a design scheme for training the preference model, which is determined based on the initial optimization model corresponding to the body-in-white structure.

[0071] As an example, in step S501, the device can determine multiple training design schemes corresponding to the body-in-white structure based on the initial optimization model corresponding to the body-in-white structure generated by modeling. In this example, the initial optimization model corresponding to the body-in-white structure includes multiple design parameters, optimization objectives, and constraints, with each design parameter corresponding to a parameter range; based on the optimization objectives and constraints, the parameter value corresponding to each design parameter can be determined within the parameter range corresponding to each design parameter; and a training design scheme is formed based on the parameter values ​​corresponding to multiple design parameters.

[0072] As an example, in step S502, after acquiring multiple training design schemes, the device can display multiple training design schemes to collect preference data provided by different designers (DMs), and form a preference dataset based on all preference data. In this example, each preference data represents a designer's (DM's) preference for any two training design schemes, and the resulting preference dataset can be... express, For a preference data, including solution pairs and corresponding preferences Composition, in which and These are two training design schemes corresponding to the body-in-white structure. It reflects the designer's preference, which may include three situations: express Superior (Preferred) express and Equivalent (no difference) express Better (Second choice).

[0073] In this example, to facilitate the collection of designers' preference data, the device can display multiple training design schemes through an auxiliary information interface and collect designers' preference data through a preference collection interface.

[0074] like Figure 8As shown, the auxiliary information interface allows designers to track the optimization progress in real time, such as the parameter range of the solved design parameters and the correlation of key parameters, especially in complex optimization design problems. This interface mainly consists of three main areas: The scheme attribute evaluation area primarily enables a comprehensive assessment of the attributes of the selected training design schemes, including detailed design parameters, performance metrics, and 3D model visualization. The combination of numerical data and graphical representations allows designers to intuitively evaluate the structure and functionality of potential training designs, and also facilitates the ranking of multiple training designs, ensuring that the designer's preferences are effectively integrated into the optimization process.

[0075] The auxiliary information interaction area is designed to support designer feedback by visualizing key information, such as bubble charts of key variables in training designs and line graphs of variable history. These tools allow designers to track the evolution of key parameters and their interdependencies across iterations.

[0076] The optimization process display area primarily provides a dynamic overview of the optimization process, including a summary table of design parameters for different design schemes and a visual representation of the optimization trajectory. Furthermore, it offers real-time tracking of algorithm state transitions during iterations, enabling designers to transparently understand how hybrid optimization strategies adapt to constantly changing problem characteristics. This feature ensures that designers can monitor the interactions between different algorithms and assess their impact on the overall process.

[0077] like Figure 8 As shown, designers can analyze and evaluate the attributes and optimization process of the displayed training design schemes through the preference collection interface. The optimization process provides designers with selected parameter attributes of intermediate solutions and sketches of the vehicle body model. Furthermore, additional functions, such as parallel coordinate system visualization, stress simulation analysis, and material switching simulation, are available for intermediate solutions. These functions help users more accurately express their ranking preferences among multiple training design schemes.

[0078] Among them, the Gaussian Process (GP) model is a nonparametric probabilistic model based on the Bayesian framework. Its core is to use the Gaussian distribution to describe the uncertainty of the function and infer the posterior distribution of the function through observed data, thereby realizing modeling, prediction and uncertainty quantification.

[0079] As an example, in step S503, after acquiring multiple preference data, the device can analyze the multiple preference data based on a Gaussian process model to form a target preference model that can capture the designer's preferences. In this example, the Gaussian process model is used to process the multiple preference data to transform the designer's preferences into a modelable utility function. Based on this utility function, a target preference model is constructed so that preference prediction can be performed for any design scheme using this target preference model.

[0080] In this embodiment, multiple training design schemes are first determined based on the optimization model of the body-in-white structure. Based on these multiple training design schemes, the designer's preference data is collected, which reflects the designer's preference for different training design schemes. The multiple preference data are then analyzed based on a Gaussian process model to determine a target preference model that can effectively capture the designer's preferences. This allows the target preference model to accurately predict the preference probability of any design scheme, facilitating subsequent iterative optimization of the optimization algorithm.

[0081] In one embodiment, such as Figure 6 As shown, step S503, which involves processing multiple preference data using a Gaussian process model to determine the target preference model, includes: S601: Based on multiple training design schemes, determine the covariance matrix of multiple training design schemes, and construct a zero-mean Gaussian process prior based on the covariance matrix; S602: Determine the likelihood function of multiple preference data based on multiple preference data; S603: Determine the posterior of a Gaussian process based on the prior of the Gaussian process and the likelihood function of multiple preference data; S604: Determine the target preference model based on the Gaussian process posterior.

[0082] As an example, in step S601, the device can determine the covariance matrix of multiple training design schemes based on multiple training design schemes. covariance matrix elements Training Design Scheme and Covariance function between , used to characterize and The similarity. In this example, the covariance matrix This device is used to quantify the similarity between different training designs, providing support for subsequent model training and prediction. It determines the covariance matrix. Then, based on the covariance matrix It can model multiple preference data based on a Gaussian process model, specifically defining a zero-mean Gaussian process prior for each training design scheme corresponding to the body-in-white structure: in, It is a utility function used to represent each training design scheme. It is a training design scheme and The covariance function is used to measure the similarity between two training designs. Common covariance functions include radial basis functions (RBF) kernels. By setting an appropriate covariance function, the similarity and differences between training designs can be captured.

[0083] As an example, in step S602, the device can provide a preference dataset. Each preference data in Construct the corresponding likelihood function. In a Gaussian process, each preference data point is the result of perturbing its corresponding utility function value with Gaussian noise. Therefore, the standard normal cumulative distribution function (CDF) can be used to model the likelihood function: in It is the standard normal CDF, which represents the probability that a designer will choose a particular design given the difference in utility. and Two training design schemes are provided. According to the designer's preferences, express Superior (Preferred) express and Equivalent (no difference) express Better (Secondary option). In this example, the likelihood function is constructed based on the preference data, which encodes the designer's preferences into the Gaussian process model.

[0084] As an example, in step S603, the device can derive the Gaussian process posterior using Bayes' theorem based on the Gaussian process prior and the likelihood function of multiple preference data. : in, It is a utility function The prior distribution of the Gaussian process; It is a likelihood function based on preference data. It represents the marginal likelihood of the observed data. The Gaussian process posterior provides an updated belief in the utility function given designer preference information.

[0085] As an example, in step S604, in order to obtain the utility function that best matches the designer's preferences... We maximize the posterior of the Gaussian process and optimize the parameters of the Gaussian process model through maximum a posteriori estimation (MAP) to determine the optimal utility function. : This process is accomplished by minimizing the negative log-posterior probability, and its optimization problem can be expressed as: The first term is the negative log-likelihood, and the second term is a regularization term based on the Gaussian process prior. In this example, by optimizing the posterior distribution of the Gaussian process, the optimal utility function can be obtained. The optimal utility function It can reflect the designer's preferences as accurately as possible, and can be used to determine the target preference model for the final analysis.

[0086] In this embodiment, since multiple training designs and their preference data often exhibit nonlinear and non-monotonic relationships, Gaussian processes require kernel functions (covariance functions) to capture the relationships between different training designs and preferences. Therefore, a zero-mean Gaussian process prior can be constructed based on the determined covariance matrix. This process provides a powerful framework for Gaussian process modeling, enabling it to flexibly express the relationships between different training designs, making the subsequent optimization process more stable and efficient, and helping to ensure the reliability and accuracy of the output target preference model. Understandably, if a non-zero-mean Gaussian process prior is constructed, the constructed target preference model will lose its adaptability, especially when facing complex training designs. Based on the preference data, its likelihood function is determined. Using the Gaussian process prior and the likelihood function of the preference data, the Gaussian process posterior is derived using Bayes' theorem, so as to combine the Gaussian process prior with the predicted preferences in the preference data to obtain a more accurate inference of the latent function with greater uncertainty quantification capability. The Gaussian process posterior is then optimized to determine the final target preference model, enabling this target preference model to accurately and effectively determine the preference probability of any training design.

[0087] In one example, a method for optimizing the body-in-white structure is provided, including the following steps: S701: Modeling and optimization of the body-in-white structure, collection of preference datasets. The covariance matrix K is calculated, the Gaussian process prior and covariance function are set, the Gaussian process model is initialized, and the target preference model used to predict the preference for body-in-white structural design schemes is determined. As a probabilistic model, the Gaussian process model can effectively capture designer preferences and make real-time adjustments during the optimization process. Without accurate preference data and an initialized Gaussian process, the model will be unable to learn from designer feedback and accurately predict design utility, thus affecting the final optimization results.

[0088] S702: Initialize the fitness matrix (AAM) and related data structures (such as the best solution archive). Set the historical smoothing parameters. Select the initial optimization algorithm And set the cooldown time. And cooling countdown. During each round of body-in-white structure optimization, performance metrics for each optimization algorithm are collected (including preference utility score, convergence speed score, and solution diversity score based on Gaussian process model prediction). Initializing the fitness matrix and collecting performance data are fundamental to ensuring that the algorithm can adjust based on real-time feedback. By collecting performance data for each algorithm, they can be dynamically compared and evaluated, ensuring more efficient algorithm selection during the optimization process. Without the fitness matrix and performance data collection, the algorithm will not be able to make reasonable adjustments based on historical data, thereby reducing optimization efficiency and effectiveness.

[0089] S703: Calculate and compare the fitness scores of each optimization algorithm according to its fitness score formula, and determine the candidate optimization algorithm with the highest fitness score. If the maximum fitness score exceeds the fitness score of the current optimization algorithm and the cooldown countdown is zero, an algorithm switch is triggered. If a switch occurs, the cooldown countdown is reset. In this example, scoring, comparison, and switching decisions are key steps in selecting the candidate optimization algorithm with the best overall performance and dynamically switching algorithms based on user preferences. By calculating the fitness score, the device can select the most suitable optimization algorithm for optimization based on a comparison of the overall performance of multiple optimization algorithms. Without this step, the optimization algorithm will not be able to dynamically adjust according to real-time performance changes, potentially leading to the loss of the global optimum.

[0090] S704: When switching optimization algorithms, the body-in-white structural design scheme, population data, and related parameters (as shown in Table 2) from the previous iteration are passed to the newly selected candidate optimization algorithm to achieve state transition and knowledge transfer. This means updating the target optimization model from the previous iteration to the new current optimization model for subsequent iterations. State transition and knowledge transfer ensure that useful data accumulated during the optimization process can be inherited and reused, greatly improving the efficiency and accuracy of the optimization process. By passing the optimal solution, population data, and related parameters from the target optimization model of the previous iteration, the new optimization algorithm can inherit the existing optimization state, avoiding starting the calculation from scratch. Without state transition and knowledge transfer, algorithm switching will lose previous optimization results, affecting the overall optimization effect.

[0091] Table 2 Parameters involved in algorithm switching S705: Execute the selected optimization algorithm to further optimize the current body-in-white structure design. The process continues until the termination condition is met, at which point optimization stops and the final body-in-white structural design is output. In other words, the optimal design solution is output. Only by ensuring that the optimization algorithm executes according to the design objective and stops when the termination condition is met can we ensure that the final output design solution is optimal. If this execution and termination step is missing, the optimization process will not end, resulting in the inability to obtain the final optimization result.

[0092] In this embodiment, the designer can obtain auxiliary decision-making information related to the optimization model and process through the auxiliary information interface, and receive feedback in the form of ranking scheme preferences through the preference feedback interface. Then, a target preference model is constructed using a Gaussian process model to generate preference utility scores for intermediate optimization solutions. Simultaneously, other performance index scores of the optimization process, such as convergence speed scores and solution diversity scores, are integrated to construct an fitness matrix for multiple optimization algorithms, serving as the basis for dynamic algorithm switching. The switched algorithm continues optimization, generating new optimization design schemes and updating the auxiliary information interface provided to the user. This cycle continues until optimization converges or the user is satisfied, thus ending the optimization process. Overall, it has the following advantages: First, by incorporating designers' preference data, a Gaussian process-based target preference model was constructed, which accurately reflects designers' preferences for different body-in-white structure design schemes. This target preference model not only improves the decision-making quality during the optimization process but also effectively reduces ineffective iterations. Through a customized optimization experience, the optimization process better aligns with the designers' actual needs. Designers can adjust their preference inputs in real time according to different design requirements and constraints, enabling the model to dynamically adapt to changing needs and significantly improving the accuracy and efficiency of body-in-white structure optimization design.

[0093] Secondly, a mechanism is provided that can adjust the optimization algorithm strategy in real time based on multiple performance index scores (such as preference utility score, convergence speed score, and solution diversity score). This mechanism solves the problems of poor adaptability and low efficiency of existing optimization algorithms when facing complex design problems by calculating the fitness score of each optimization algorithm and dynamically switching between them. Especially in the optimization design of body-in-white structures, the optimization algorithm needs to handle complex problems with multiple dimensions and objectives. The dynamic switching mechanism can select the optimization algorithm based on real-time feedback, thereby ensuring that the algorithm always adapts to the current optimization environment. This mechanism provides a new solution and standard for adaptive hybrid optimization in complex engineering applications, improving the efficiency and robustness of the optimization process.

[0094] This application embodiment also provides a body-in-white structure optimization device 110, such as Figure 11 As shown, it includes: The target optimization model determination module 111 is used to iteratively update the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model; The fitness score determination module 112 is used to iteratively optimize the target optimization model using multiple optimization algorithms and determine the fitness scores corresponding to the multiple optimization algorithms. The fitness scores are determined based on the scores of multiple performance indicators. The optimization algorithms include the current optimization algorithm and candidate optimization algorithms. The optimization algorithm update module 113 is used to update the target optimization model to the current optimization model and update the candidate optimization algorithm with the largest maximum fitness score to the current optimization algorithm when the fitness score corresponding to the current optimization algorithm is not the maximum fitness score. The module repeatedly executes the iterative update of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model.

[0095] This application also provides an electronic device 120, such as... Figure 12 As shown, it includes a memory 121 and a processor 122, wherein the memory 121 is used to store computer programs; the processor 122 is used to execute the programs stored in the memory 121 to implement the body-in-white structure optimization method described in any embodiment of this application.

[0096] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements method A as described in any embodiment of this application.

[0097] This application also provides a body-in-white structure, the design parameters of which are determined based on the target optimization model determined by the above-described body-in-white structure optimization method.

[0098] This application also provides a vehicle including the above-described body-in-white structure.

[0099] In this application, "multiple" refers to two or more.

[0100] In this application, unless otherwise expressly defined, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0101] The terms “first,” “second,” “third,” “fourth,” etc., in this application (if present) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0102] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0103] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if the method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if the method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.

[0104] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for optimizing the structure of a body-in-white, characterized in that, include: The current optimization algorithm is used to iteratively update the current optimization model of the body-in-white structure to determine the target optimization model; The target optimization model is iteratively optimized using multiple optimization algorithms to determine the fitness scores corresponding to the multiple optimization algorithms. The fitness scores are determined based on multiple performance index scores. The optimization algorithms include the current optimization algorithm and candidate optimization algorithms. When the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, the target optimization model is updated to the current optimization model, the candidate optimization algorithm with the maximum fitness score is updated to the current optimization algorithm, and the iteration update of the current optimization model of the body-in-white structure using the current optimization algorithm is repeated to determine the target optimization model.

2. The method according to claim 1, characterized in that, After iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model, the method further includes: Update the cooldown countdown corresponding to the current optimization algorithm; When the cooling countdown is not zero, the target optimization model is updated to the current optimization model, and the process of iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm is repeated to determine the target optimization model. When the cooling countdown reaches zero, the target optimization model is iteratively optimized using multiple optimization algorithms to determine the fitness scores corresponding to the multiple optimization algorithms.

3. The method according to claim 2, characterized in that, The method further includes: When the cooling countdown is zero and the fitness score corresponding to the current optimization algorithm is the maximum fitness score, the optimal design scheme of the body-in-white structure is determined based on the target optimization model.

4. The method according to claim 1, characterized in that, The step of iteratively optimizing the target optimization model using multiple optimization algorithms and determining the fitness scores corresponding to the multiple optimization algorithms includes: The optimization algorithm is used to iteratively optimize the target optimization model, and to determine multiple optimization design schemes corresponding to each iteration of the optimization algorithm and multiple performance index scores corresponding to each optimization design scheme. Based on the performance index scores of all the optimization design schemes corresponding to each iteration of the same optimization algorithm, the fitness score corresponding to each iteration of the optimization algorithm is determined. Among them, the performance index scores include preference utility score, convergence speed score and solution diversity score. The preference utility score is used to characterize the utility score of the pre-trained target preference model for predicting the preference of the optimized design scheme.

5. The method according to claim 4, characterized in that, The fitness score for each iteration of the optimization algorithm is determined by analyzing the performance metrics scores of all optimized design schemes for each iteration, based on the same optimization algorithm. The performance index scores corresponding to the multiple optimization algorithms are standardized to determine multiple performance standard scores corresponding to each optimization design scheme. The performance standard scores include preference utility standard score, convergence speed standard score and solution diversity standard score. Based on the target weight corresponding to each optimization algorithm, the multiple performance standard scores corresponding to each optimization design scheme corresponding to the same optimization algorithm are weighted and processed to determine the comprehensive performance score corresponding to each optimization design scheme. The average of the comprehensive performance scores of all the optimized design schemes obtained in each iteration of the same optimization algorithm is determined as the fitness score corresponding to each iteration of the optimization algorithm.

6. The method according to claim 4, characterized in that, Before iteratively updating the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model, the body-in-white structure optimization method further includes: Based on the initial optimization model corresponding to the body-in-white structure, multiple training design schemes corresponding to the body-in-white structure are determined. Acquire multiple preference data, each of which represents the designer's preference for any two of the training design schemes; A Gaussian process model is used to process multiple preference data to determine the target preference model.

7. The method according to claim 6, characterized in that, The step of processing multiple preference data using a Gaussian process model to determine the target preference model includes: Based on multiple training design schemes, the covariance matrix of multiple training design schemes is determined, and based on the covariance matrix, a zero-mean Gaussian process prior is constructed. Based on multiple preference data, determine the likelihood function for the multiple preference data; The Gaussian process posterior is determined based on the prior of the Gaussian process and the likelihood function of multiple preference data. Based on the Gaussian process posterior, the target preference model is determined.

8. A white body structure optimization device, characterized in that, include: The target optimization model determination module is used to iteratively update the current optimization model of the body-in-white structure using the current optimization algorithm, and determine the target optimization model. An adaptive score determination module is used to iteratively optimize the target optimization model using multiple optimization algorithms, and determine the adaptive scores corresponding to the multiple optimization algorithms. The adaptive scores are determined based on multiple performance index scores, and the optimization algorithms include the current optimization algorithm and candidate optimization algorithms. The optimization algorithm update module is used to update the target optimization model to the current optimization model when the fitness score corresponding to the current optimization algorithm is not the maximum fitness score, update the candidate optimization algorithm with the largest maximum fitness score to the current optimization algorithm, and repeatedly execute the iterative update of the current optimization model of the body-in-white structure using the current optimization algorithm to determine the target optimization model.

9. An electronic device, characterized in that, Including processor and memory, among which, Memory, used to store computer programs; A processor is used to execute a program stored in a memory to implement the body-in-white structure optimization method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the body-in-white structure optimization method according to any one of claims 1-7.

11. A body-in-white structure, characterized in that, The design parameters of the body-in-white structure are determined based on the target optimization model determined by the body-in-white structure optimization method described in any one of claims 1-7.

12. A vehicle, characterized in that, Includes the body-in-white structure as described in claim 11.

Images