CN116467794A

CN116467794A - Multi-disciplinary design optimization method and device for sound insulation performance of automobile and electronic equipment

Info

Publication number: CN116467794A
Application number: CN202310409928.4A
Authority: CN
Inventors: 刘经奇; 李学亮; 陈鹏; 梁兴凯; 冯秋翰
Original assignee: Chongqing Changan Automobile Co Ltd
Current assignee: Chongqing Changan Automobile Co Ltd
Priority date: 2023-04-17
Filing date: 2023-04-17
Publication date: 2023-07-21

Abstract

The invention relates to a multidisciplinary design optimization method and device for sound insulation performance of an automobile and electronic equipment, wherein the method comprises the following steps: DOE sampling is carried out on a plurality of influence indexes of which the sound insulation performance of the automobile needs to be optimally designed, and a target agent model group formed by target agent models corresponding to all the influence indexes is determined based on accuracy; and determining an optimal point based on the target agent model group, if the difference between the actual index response corresponding to the optimal point and the optimization constraint condition is larger than a preset threshold value, updating the target agent model group by utilizing an encryption DOE training sample data set corresponding to the optimal point, wherein the encryption DOE training sample data set is constructed based on the optimal point and the value ranges of a plurality of design variables in related parameters until the difference between the actual index response and the optimization constraint condition is smaller than or equal to the preset threshold value, and outputting the finally updated target agent model group for optimizing the sound insulation performance of the automobile. According to the method and the device for optimizing the sound insulation performance of the automobile, the joint optimization of the sound insulation performance of the automobile can be achieved by improving the accuracy of the agent model near the optimal point and selecting the target agent model with the highest accuracy.

Description

Multi-disciplinary design optimization method and device for sound insulation performance of automobile and electronic equipment

Technical Field

The application relates to the field of multidisciplinary design optimization, in particular to a multidisciplinary design optimization method and device for sound insulation performance of an automobile and electronic equipment.

Background

Automobile design is a complex system engineering, relates to the field of multiple disciplines, and comprises collaborative design of multiple disciplines such as structure, collision, NVH, heat, battery, optics and the like, data are difficult to transfer among different disciplines, and an experience-based manual trial-and-error optimization process leads to long research and development period and high research and development cost. Multidisciplinary design optimization (Multidisciplinary Design Optimization, MDO), by exploring and utilizing synergistic mechanisms of interactions in the system, and utilizing multi-objective strategies and computer-aided techniques to design complex systems and subsystems, the design cycle can be effectively shortened, and overall optimal performance of the system can be obtained.

The current multidisciplinary design optimization flow generally comprises three steps of test design (Design of experiment, DOE), proxy model establishment and optimization design. And multidisciplinary design optimization is applied to the field of automobiles, so that the research and development period can be greatly shortened, and the research and development cost is reduced. However, in the conventional multi-disciplinary design optimization of the automobile, the final result of the multi-disciplinary design optimization cannot meet the design requirement in the actual production process.

Disclosure of Invention

In order to solve the technical problems or at least partially solve the technical problems, the application provides a multidisciplinary design optimization method and device for sound insulation performance of an automobile and electronic equipment.

In a first aspect, the present application provides a method for optimizing multidisciplinary design of sound insulation performance of an automobile, including:

acquiring relevant parameters of the multi-disciplinary design optimization of the sound insulation performance of the automobile, determining a plurality of influence indexes of the sound insulation performance of the automobile, which need to be optimally designed, performing DOE sampling of the plurality of influence indexes, and determining an initial DOE training sample data set and a DOE verification sample data set based on a DOE sampling result;

determining target agent models corresponding to each influence index in a plurality of to-be-selected agent model groups corresponding to each influence index based on accuracy, and obtaining target agent model groups formed by target agent models corresponding to all influence indexes, wherein the to-be-selected agent model groups are pre-trained based on the initial DOE training sample data set and the DOE verification sample data set;

global optimization is carried out based on the target agent model group to obtain an optimal point, and an actual index response is determined based on the optimal point;

and if the difference between the actual index response and the optimization constraint condition is greater than a preset threshold, updating the target agent model group by using an encryption DOE training sample data set corresponding to the optimal point, wherein the encryption DOE training sample data set is constructed based on the optimal point and the value ranges of a plurality of design variables in the related parameters, and performing global optimization by using the updated target agent model until the difference between the actual index response and the optimization constraint condition is less than or equal to the preset threshold, and outputting the final updated target agent model group for optimizing the sound insulation performance of the automobile.

Optionally, updating the set of target proxy models with the encrypted DOE training sample data set corresponding to the optimal point comprises:

obtaining an encryption DOE training sample data set for increasing local sampling density in a preset range of the optimal point;

and training each target agent model in the target agent model group by using the encryption DOE training sample data set to obtain the updated target agent model group.

Optionally, obtaining an encrypted DOE training sample dataset that increases local sampling density within a preset range of the optimal point includes:

constructing a local sampling space based on the optimal points and the value ranges of a plurality of design variables in the related parameters;

performing DOE sampling on the local sampling space to obtain a local DOE training sample point set;

analyzing and calculating the local DOE training sample point set to obtain a local DOE training sample data set;

and merging the local DOE training sample data set into an initial DOE training sample data set to obtain an encrypted DOE training sample data set.

Optionally, constructing a local sampling space based on the optimal point and the value ranges of the plurality of design variables in the related parameters includes:

Determining the value ranges of a plurality of design variables based on the related parameters;

calculating the size of an initial sampling space based on the value range of each design variable;

determining a radius of the local sampling space based on a size of the initial sampling space;

generating an initial local sampling space according to the optimal point and the radius of the local sampling space;

and determining a local sampling space based on the initial local sampling space and an initial design space.

Optionally, determining the local sampling space based on the initial local sampling space and the initial design space includes:

comparing the initial local sampling space with an initial design space, and determining whether the initial local sampling space exceeds the initial design space;

if the initial local sampling space exceeds the initial design space, determining an intersection of the initial local sampling space and the initial sampling space as a local sampling space of the iteration of the round;

and if the initial local sampling space does not exceed the initial design space, determining the initial local sampling space as the local sampling space of the iteration of the round.

Optionally, performing DOE sampling of a plurality of impact indicators, determining an initial DOE training sample dataset and a DOE validation sample dataset based on DOE sampling results, comprising:

DOE sampling is carried out on a plurality of influence indexes of the sound insulation performance of the automobile, so that a DOE training sample point set and a DOE verification sample point set are obtained;

and analyzing and calculating the sample points in the DOE training sample point set and the DOE verification sample point set to obtain an initial DOE training sample data set and a DOE verification sample data set.

Optionally, the method further comprises:

training a plurality of proxy models respectively using the initial DOE training sample dataset and the DOE verification sample dataset;

and determining the trained proxy model combination as the proxy model group to be selected.

Optionally, determining, based on accuracy, a target proxy model corresponding to each influence index from a plurality of to-be-selected proxy model groups corresponding to each influence index, to obtain a target proxy model group consisting of target proxy models corresponding to all influence indexes, including:

determining the accuracy of each influence index under each agent model in the agent model group to be selected;

determining the agent model with the highest accuracy corresponding to each influence index as a target agent model corresponding to the influence index;

and combining the target agent models corresponding to all the influence indexes to obtain the target agent model group.

Optionally, global optimization is performed based on the target agent model group to obtain an optimal point, and determining an actual index response based on the optimal point includes:

based on the target agent model group, performing global optimization on a plurality of influence indexes of the sound insulation performance of the automobile to obtain the optimal point;

an actual index response of the optimal point at the real model is determined based on the optimal point.

In a second aspect, the present application provides an automotive sound insulation performance multidisciplinary design optimization apparatus comprising:

the acquisition module is used for acquiring relevant parameters of the multi-disciplinary design optimization of the sound insulation performance of the automobile, determining a plurality of influence indexes of the sound insulation performance of the automobile, sampling DOE of the influence indexes, and determining an initial DOE training sample data set and a DOE verification sample data set based on a DOE sampling result;

the first determining module is used for determining a target agent model corresponding to each influence index in a plurality of to-be-selected agent model groups corresponding to each influence index based on accuracy, so as to obtain a target agent model group consisting of target agent models corresponding to all the influence indexes, wherein the to-be-selected agent model group is pre-trained based on the initial DOE training sample data set and the DOE verification sample data set;

The optimization module is used for carrying out global optimization based on the target agent model group to obtain an optimal point, and determining an actual index response based on the optimal point;

and the updating module is used for updating the target agent model group by utilizing an encryption DOE training sample data set corresponding to the optimal point if the difference between the actual index response and the optimization constraint condition is larger than a preset threshold value, wherein the encryption DOE training sample data set is constructed based on the optimal point and the value ranges of a plurality of design variables in the related parameters, and the updated target agent model is used for global optimization until the difference between the actual index response and the optimization constraint condition is smaller than or equal to the preset threshold value, and the finally updated target agent model group is output for optimizing the sound insulation performance of the automobile.

Optionally, the updating module includes:

the acquisition sub-module is used for acquiring an encryption DOE training sample data set for increasing local sampling density in a preset range of the optimal point;

and the training sub-module is used for training each target agent model in the target agent model group by utilizing the encryption DOE training sample data set to obtain the updated target agent model group.

Optionally, the acquiring submodule includes:

the construction unit is used for constructing a local sampling space based on the optimal points and the value ranges of a plurality of design variables in the related parameters;

the sampling unit is used for performing DOE sampling on the local sampling space to obtain a local DOE training sample point set;

the computing unit is used for analyzing and computing the local DOE training sample point set to obtain a local DOE training sample data set;

and the merging unit is used for merging the local DOE training sample data set into the initial DOE training sample data set to obtain an encrypted DOE training sample data set.

Optionally, the building unit comprises:

a first determining subunit, configured to determine a value range of a plurality of design variables based on the relevant parameters;

a calculating subunit, configured to calculate a size of the initial sampling space based on a value range of each design variable;

a second determining subunit configured to determine a radius of the local sampling space based on a size of the initial sampling space;

a generating subunit, configured to generate an initial local sampling space according to the optimal point and a radius of the local sampling space;

and a third determination subunit, configured to determine a local sampling space based on the initial local sampling space and an initial design space.

Optionally, the third determining subunit is further configured to:

Optionally, the acquiring module includes:

the sampling submodule is used for performing DOE sampling on a plurality of influence indexes of the sound insulation performance of the automobile to obtain a DOE training sample point set and a DOE verification sample point set;

and the analysis and calculation sub-module is used for analyzing and calculating the sample points in the DOE training sample point set and the DOE verification sample point set to obtain an initial DOE training sample data set and a DOE verification sample data set.

Optionally, the apparatus further comprises:

the training module is used for training a plurality of agent models by utilizing the initial DOE training sample data set and the DOE verification sample data set respectively;

And the second determining module is used for determining the trained agent model combination as the agent model group to be selected.

Optionally, the first determining module includes:

the first determining submodule is used for determining the accuracy of each influence index under each agent model in the agent model group to be selected according to each influence index;

the second determining submodule is used for determining the agent model with highest accuracy corresponding to each influence index as a target agent model corresponding to the influence index;

and the combination sub-module is used for combining the target agent models corresponding to all the influence indexes to obtain the target agent model group.

Optionally, the optimizing module includes:

the optimization sub-module is used for carrying out global optimization on a plurality of influence indexes of the sound insulation performance of the automobile based on the target agent model group to obtain the optimal point;

a third determination sub-module for determining an actual index response of the optimal point at the real model based on the optimal point.

In a third aspect, the present application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the communication bus;

A memory for storing a computer program;

and the processor is used for realizing the multidisciplinary design optimization method for the sound insulation performance of the automobile according to any one of the first aspect when executing the program stored in the memory.

In a fourth aspect, the present application provides a computer readable storage medium, where a program of the automotive sound insulation performance multi-disciplinary design optimization method is stored, where the program of the automotive sound insulation performance multi-disciplinary design optimization method is executed by a processor to implement the steps of any one of the automotive sound insulation performance multi-disciplinary design optimization methods of the first aspect.

Compared with the prior art, the technical scheme provided by the embodiment of the application has the following advantages:

according to the method and the device, the target agent model group is built based on the target agent models with the highest accuracy, the accuracy of the target agent model group is greatly improved, the accuracy of each influence index when the target agent model group is subjected to multidisciplinary design optimization is guaranteed, in addition, the target agent model group is updated through the encryption DOE training sample data set with the local sampling density near the optimal point, the accuracy of the updated target agent model group near the optimal point is improved, the global accuracy of the agent model is effectively improved, the situation that the scheme which meets the standard through the agent model is optimized is avoided, and the situation that the agent model DOEs not meet the standard is found in an actual experiment.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, and it will be obvious to a person skilled in the art that other drawings can be obtained from these drawings without inventive effort.

Fig. 1 is a flowchart of a multi-disciplinary design optimization method for sound insulation performance of an automobile according to an embodiment of the present application;

FIG. 2 is a flowchart of another method for optimizing the multidisciplinary design of sound insulation performance of an automobile according to an embodiment of the present application;

FIG. 3 is a schematic diagram of a comparison between an FDVGA algorithm and a typical conventional optimization algorithm according to an embodiment of the present application;

FIG. 4 is a schematic diagram of the results of partial impact indicators after the first round of optimization before updating the target agent model according to the embodiment of the present application

FIG. 5 is a schematic diagram of a comparison between a result of a partial impact index output by a proxy model and an actual reverse result after a first round of optimization before updating the target proxy model according to an embodiment of the present application;

FIG. 6 is a schematic diagram of the results of partial impact indicators after final optimization of the updated target agent model according to the embodiment of the present application;

fig. 7 is a schematic diagram showing comparison of effects of local DOE training according to an embodiment of the present application;

FIG. 8 is a block diagram of an apparatus for optimizing multiple disciplinary design of sound insulation performance of an automobile according to an embodiment of the present application;

fig. 9 is a block diagram of an electronic device according to an embodiment of the present application.

Detailed Description

For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present application based on the embodiments herein.

The current multidisciplinary design optimization flow generally comprises three steps of test design (Design of experiment, DOE), proxy model establishment and optimization design.

The DOE is based on probability theory and mathematical statistics, and tests or high-precision simulation analysis are carried out through a small amount of representative samples, so that the whole design space is quickly explored, and ideal test results are economically and scientifically obtained. The method comprises the steps that a proxy model is built, actual test sample data obtained by DOE is applied, an approximate model between design parameters and product performance is built through a proxy model algorithm, a test or high-precision simulation analysis model is replaced, a response value is predicted efficiently, and the method is particularly aimed at complex large-scale, high-dimensional, nonlinear and multi-polar expensive optimization problems; the application of the proxy model enables the implementation of optimization algorithms such as evolutionary algorithms that require a large number of trials. The optimization design process combines an optimization algorithm and a proxy model, and applies approximate calculation of the proxy model to solve the optimal solution of a single or a plurality of targets in the design space. In the optimization algorithm, the evolution algorithm does not need to rely on the nature of the problem like a mathematical optimization algorithm, so that the application problem range is wider, the problems of nonlinear, non-microincompressible function, non-convex function, discrete and combined optimization can be solved, and compared with a gradient optimization method, the evolution algorithm is not easy to fall into local optimization and is widely applied to multidisciplinary optimization.

DOE methods include full factor designs, partial factor designs, plackett-Burman designs, center complex designs, latin hypercube designs, mesh designs, and the like.

The automobile high-frequency noise is one of the main NVH (noise, vibration and harshness) problems of the automobile, particularly for the electric automobile which is an important development direction of the automobile in the future, the masking effect of the noise from the traditional internal combustion engine is lacked, the electromagnetic noise presents the high-frequency characteristic which is easier to be perceived by human ears, and the high-frequency noise insulation performance of the automobile is required.

And multidisciplinary design optimization is applied to the field of automobiles, so that the research and development period can be greatly shortened, and the research and development cost is reduced. The traditional optimization method for the high-frequency noise insulation performance of the automobile is to perform key part optimization based on weak point identification, enable the performance to reach the standard, and reduce excess performance redundancy according to experience to realize cost control. This approach often fails to achieve the lowest cost or weight optimization objectives. However, in the conventional multi-disciplinary design optimization of the automobile, the final result of the multi-disciplinary design optimization cannot meet the design requirement in the actual production process.

Therefore, the embodiment of the application provides a multidisciplinary design optimization method and device for sound insulation performance of an automobile and electronic equipment. The design objective of this embodiment is to achieve the objective of lowest cost on the basis of optimizing a plurality of high-frequency impact indexes.

Based on the above, as shown in fig. 1 and fig. 2, the multi-disciplinary design optimization method for sound insulation performance of an automobile provided in the embodiment of the application may include the following steps:

step S101, acquiring relevant parameters of the multi-disciplinary design optimization of the sound insulation performance of the automobile, determining a plurality of influence indexes of the sound insulation performance of the automobile, which need to be optimally designed, sampling the DOE of the plurality of influence indexes, and determining an initial DOE training sample data set and a DOE verification sample data set based on the DOE sampling result;

in this embodiment of the application, relevant parameters for multidisciplinary design optimization of sound insulation performance of an automobile include: design variable, design space A _{design_space} DOE sampling methods, sampling scale and target and performance constraints, etc.

Illustratively, the sound insulation performance of the whole vehicle is affected by a plurality of factors, including: the number of the obtained design variables is 205 through statistics, and the number of the design variables is reduced to 101 through correlation analysis processing. Design space may implement space determination based on physical constraints and actual engineering. Considering the uniformity of sampling, the Latin hypercube sampling method is adopted in this example. And carrying out automatic calculation on 101 design variables through coding application acoustic performance solving software, and finally finishing to obtain quantized influence indexes 78.

In the embodiment of the application, the objective of the multidisciplinary design optimization of the sound insulation performance of the automobile is the lowest cost, and the performance constraint condition is set based on the requirement of the influence index. Combining the dimensional and trial time cost considerations of the design space, determining the sample size for DOE training may be determined according to the following: the number of sampling points satisfies > 5 x design dimensions (number of design variables), and for example, the sampling scale may be set to 1000 groups of sampling points, the DOE is verified to sample 20 groups of sampling points, DOE sampling of a plurality of impact indicators is performed, and an initial DOE training sample dataset and a DOE verification sample dataset are determined based on the sampled results.

Step S102, determining a target agent model corresponding to each influence index in a plurality of to-be-selected agent model groups corresponding to each influence index based on accuracy, and obtaining a target agent model group consisting of target agent models corresponding to all the influence indexes;

in this embodiment of the present application, the initial DOE training sample data set and the DOE verification sample data set may be used in advance to train a plurality of proxy models respectively, and a combination of trained proxy models is determined as the candidate proxy model group.

The plurality of proxy models may be different types of proxy model algorithms, and by way of example, the candidate algorithms in this embodiment may be Least Square (Least Square), kriging, marginal Gaussian Process (MGP), partial Least Square based Kriging proxy model (KPLS) or support vector regression (SVM regression), or the plurality of proxy models may be proxy model algorithms with partially or completely identical types.

The least square method has the advantages of high training speed and low time cost, and can be almost used as one of candidate algorithms of the agent model on any problem; the classical Kriging algorithm in the common algorithm adopts a KPLS algorithm which is more suitable for large-scale problems to replace because the training time aiming at the problems of high dimension and large sample is too long. The algorithm adopts a PLS (partial least squares) method to construct a covariance kernel, reduces the dimensionality of the internal parameters of the algorithm, and can remarkably improve the training speed of the proxy model.

In this embodiment of the present application, each proxy model group to be selected includes a plurality of proxy models, and the form of the proxy model group to be selected is shown in formula 1:

wherein the first subscript of f represents the serial number of the dependent variable (generally comprising an optimization index and a constraint index) to be subjected to the agent model prediction, wherein the dependent variable is an influence index, the 2 nd subscript is the serial number of the agent model, k is the total number of the dependent variables to be subjected to the agent model prediction, and l is the number of the agent models;

In this step, the accuracy of the impact index under each agent model in the agent model group to be selected may be determined for each impact index; then, determining the agent model with highest accuracy corresponding to each influence index as a target agent model corresponding to the influence index; and combining the target agent models corresponding to all the influence indexes to obtain the target agent model group.

That is, the accuracy of each influence index under different proxy models is calculated respectively, the proxy model with the highest accuracy (the smallest error) is determined as the target proxy model corresponding to the influence index, the target proxy models corresponding to all the influence indexes are combined to obtain the target proxy model group, and the situation that only a single proxy model is used in single multi-disciplinary design optimization design engineering is avoided, so that each specific influence index cannot be guaranteed to have enough accuracy.

The form of the target proxy model group is shown in formula 2, wherein for any ith optimal proxy model, the proxy model with the smallest error for the dependent variable (i.e. the influence index), i.e. the existence of the common formula, is the one with the smallest error

The relationship shown in equation 3.

Where ε () represents the error indicator value of the bracketed target proxy model.

The accuracy calculation may use mean absolute value error (MAE), mean Absolute Percentage Error (MAPE), or complex correlation coefficient (R ² ) The accuracy index is evaluated, and the expression of the accuracy index is shown in the formula 4-6:

where n is the number of sample points,representing the predicted value of the mth sample point in the proxy model, ym represents the experimental real response value of the mth sample point,>the mean value of the true responses of all test samples.

In one embodiment of the present application, the mean absolute percentage error of the KPLS algorithm is smaller than that of the SVM regression algorithm and the partial least squares algorithm, i.e. the accuracy is the highest, and the accuracy of some impact indexes under different agent models is shown in table 1. Thus, the target agent model set of this embodiment is a candidate agent model set of the KPLS algorithm for 78 impact indicators. For more common cases, the target proxy models of the impact indicators are different, and the target proxy model group should be combined according to the proxy model with the optimal actual accuracy of each impact indicator.

TABLE 1 Whole vehicle Sound insulation Performance Global optimization of the accuracy of different proxy models

Step S103, global optimization is carried out based on the target agent model group, the optimal point is obtained, and the actual index response is determined based on the optimal point;

In the step, global optimization can be performed on a plurality of influence indexes of the sound insulation performance of the automobile based on the target agent model group to obtain the optimal point; an actual index response of the optimal point at the real model is determined based on the optimal point.

In the step, a target agent model group is used for replacing high-frequency simulation software, and global optimization is performed by using a global optimization algorithm with the lowest cost as a target to obtain an approximate optimal solution aiming at the characteristics of more decision variables and high target dimension in the embodiment of the applicationThe near optimal solution corresponds to the optimal design variable in the iteration round>I.e. the optimal point. The embodiment of the application can realize the prevention of sinking into the local optimal point by using a global optimization algorithm.

Global optimization algorithms such as: non-dominant ordered genetic algorithm (NSGA-II), non-dominant ordered genetic algorithm (NSGA-III), population-based random optimization technique algorithm (PSO), differential evolution algorithm (DE), multi-objective evolution algorithm (MOED/D), fuzzy decision variable-based genetic algorithm (FDVGA), etc.

Compared with other algorithms, the FDVGA algorithm is more suitable for the large-scale high-dimensional problem of decision variables, as shown in FIG. 3, the test result of a standard test function DTLZ3 with 500 design dimensions shows that the solution set of the algorithm is uniformly distributed on an optimal plane, and the traditional algorithms NSGA-II and MOED/D sink into local optimization, so that the solution set is not converged.

Finally, the optimal point can be substituted into high-frequency simulation software for analysis and calculation to obtain the actual index response y 'of the current optimal point' _opt 。

Step S104, if the difference between the actual index response and the optimization constraint condition is greater than a preset threshold, updating the target agent model set by using an encrypted DOE training sample data set corresponding to the optimal point, wherein the encrypted DOE training sample data set is constructed based on the optimal point and the value ranges of a plurality of design variables in the related parameters, and performing global optimization by using the updated target agent model until the difference between the actual index response and the optimization constraint condition is less than or equal to the preset threshold, and outputting the final updated target agent model set for optimizing the sound insulation performance of the automobile.

Judging that the difference value between all the current actual index responses and the optimization constraint conditions is smaller than or equal to a preset threshold value (for example, the difference value is smaller than or equal to 1 dB), and then the actual index response of the current optimal point reaches the optimization target, the optimization is completed, and the optimal solution y is obtained _opt ＝y′ _opt Optimum point x _opt ＝x′ _opt 。

According to the method and the device, the relative error between the response of the actual index and the constraint condition is based, namely the actual index meets the standard and the redundancy is smaller as the condition of exiting iteration, the relative precision can be given according to engineering specifications or experience, the implementation is easier, the efficiency is higher, the optimization target is attached to the optimization target, the problem of new engineering is avoided, and when whether the iteration exits or not is judged in a mode of comparing the predicted value of the agent model with the response precision of the actual index, the precision requirement meeting the actual requirement of the engineering is difficult to be determined intuitively.

As shown in fig. 4, the partial impact indexes obtained in the step S104 show that the actual index response of each impact index after optimization has the situation that the performance constraint indexes do not reach the standards at the performance constraint conditions Y24 to Y34. As shown in FIG. 5, comparing the actual simulation results of the target agent model shows that the influence index output by the target agent model reaches the standard, which indicates that the local precision of the target agent model is insufficient at the optimal point. The optimal point at present does not reach the optimization requirement, and the optimization needs to be continued.

In the conventional multi-disciplinary optimization design of the automobile, the sampling randomness and uniformity of the whole sample in the whole design interval are often concerned in the DOE link, so that the global precision of a follow-up establishment agent model can be effectively improved, and convergence to a local optimal point in the follow-up optimization process is avoided. However, the disadvantage is that the global high precision does not mean that there is enough precision near the optimal point, and a scheme that is optimized to reach the standard through a proxy model often occurs, and the situation that the scheme does not reach the standard is found in an actual test.

Therefore, in the embodiment of the application, when the difference between the actual index response of the optimal point and the optimization constraint condition is greater than the preset threshold, the encrypted DOE training sample data set corresponding to the optimal point is obtained to update the target proxy model set, and the encrypted DOE training sample data set corresponding to the optimal point is the DOE training sample data set with the local DOE sampling density increased within the preset range near the optimal point, so that the accuracy near the optimal point is improved through the encrypted DOE training sample data set.

Illustratively, the target agent model set formed by the KPLS algorithm may be retrained using the encrypted DOE training sample dataset of 1050 sets of data to obtain an updated target agent model set.

And then performing global optimization by using the updated target agent model until the difference between the actual index response and the optimization constraint condition is smaller than or equal to a preset threshold value, and outputting a final updated target agent model group for optimizing the sound insulation performance of the automobile.

For example, when the next iteration after updating the target agent model set is performed, the updated target agent model set can be used, the lowest cost is used as a target, the global optimization is performed by applying the FDVGA evolution algorithm to obtain a new optimal point, the new optimal point is substituted into the high-frequency simulation software to perform analysis and calculation to obtain a new actual index response, after the inspection and comparison, all the influence index values in the obtained optimal point are found to reach the standard and have smaller redundancy, part of indexes are shown in FIG. 6, and the optimization is completed.

In yet another embodiment of the present application, step S104 updates the set of target proxy models with the encrypted DOE training sample data set corresponding to the optimal point, comprising:

step S201, obtaining an encryption DOE training sample data set for increasing local sampling density in a preset range of the optimal point;

in this step, a preset range may be first determined near the optimal point, and then the local sampling density may be increased within the preset range, which is equivalent to increasing the local sampling density near the optimal point in the initial DOE training sample dataset, to obtain the encrypted DOE training sample dataset.

Step S202, training each target agent model in the target agent model group by using the encryption DOE training sample data set to obtain the updated target agent model group.

And training each target agent model in the target agent model group by using an encryption DOE training sample data set with local sampling density increased near the optimal point to obtain the updated target agent model group.

According to the method, the encryption DOE training sample data set can be obtained by increasing the local sampling density near the optimal point, the encryption DOE training sample data set is utilized to update the target agent model group, the accuracy of the updated target agent model group near the optimal point is improved, the global accuracy of the agent model is effectively improved, the situation that the scheme which is optimized to reach the standard through the agent model DOEs not reach the standard in an actual test is avoided.

In yet another embodiment of the present application, step S201 obtains an encrypted DOE training sample data set that increases the local sampling density within a preset range of the optimal point, including:

step S301, constructing a local sampling space based on the optimal point and the value ranges of a plurality of design variables in the related parameters;

in this step, one can at the optimum point x' _opt A local sampling space is determined nearby.

Step S302, DOE sampling is carried out on the local sampling space, and a local DOE training sample point set is obtained;

in this step, a set of local DOE training sample points may be constructed using the selected sampling method in the local sampling space, e.g., local sampling space A _{opt_space} A Latin hypercube sampling method is applied in space, the number of sampling points in the example is 5% of the number of initial DOE training sampling points, namely 50 groups of sampling points are used for forming a local DOE training sample point set.

Step S303, analyzing and calculating the local DOE training sample point set to obtain a local DOE training sample data set;

and (3) performing test or high-precision analysis model (such as high-frequency simulation software) calculation on the local DOE training sample point set to obtain a local DOE training sample data set.

And step S304, merging the local DOE training sample data set into an initial DOE training sample data set to obtain an encrypted DOE training sample data set.

Because the local DOE training sample data set is obtained by locally increasing the sampling density of the local sampling space near the optimal point, after the local DOE training sample data set is combined into the initial DOE training sample data set, the obtained encrypted DOE training sample data set is a data set with the sampling density increased near the optimal point, and further, the encrypted DOE training sample data set can be used for updating the target agent model group, so that the accuracy of the updated target agent model group near the optimal point is improved, the global accuracy of the agent model is effectively improved, and the accuracy near the existing optimal solution can be remarkably improved while the stability of the whole agent model is maintained, as shown in fig. 7. Because the precision of the whole agent model is stable, even if the previous round of optimization before updating falls into local optimum due to the problem of the precision of the agent model, the updated round of optimization can be optimized to the vicinity of the global optimum point, so that the situation that the proposal which is optimized to reach the standard through the agent model does not reach the standard in the actual test is avoided.

In yet another embodiment of the present application, step S301 constructs a local sampling space based on the optimal point and the range of values of the plurality of design variables in the related parameters, including:

Step S401, determining the value ranges of a plurality of design variables based on the related parameters;

step S402, calculating the size of an initial sampling space based on the value range of each design variable;

the value ranges of each design variable may be combined to obtain the size of the initial sampling space, as shown in equation 7:

x _range ＝[x _1max -x _1min ,x _2max -x _2min ,…,x _pmax -x _pmin ] 7)

where p is the number of design variables, x _pmin For the lower bound of the design variable, x _pmax For the design variable x _p Is a lower bound of (c).

Step S403, determining a radius of the local sampling space based on the size of the initial sampling space;

the radius r of the local sampling space is determined based on the size of the initial sampling space, and is obtained by equation 8:

r＝α·x _range 0＜α＜0.5 8)

where α is a constant, an empirical value may be taken according to an actual problem of sound insulation performance of the automobile, and for example, α may be set to 0.1.

Because the value ranges of the design variables are preset, the size of the initial sampling space is determined based on the value ranges of the design variables, and then the radius of the local sampling space is determined to be stable, that is, the stable sampling radius is selected for optimizing the sound insulation performance of the automobile only based on the size of the initial sampling space and engineering experience, the method has better robustness, the size of the local sampling space is prevented from being defined in an initial sampling point area, and the problem of overfitting is avoided.

Step S404, generating an initial local sampling space according to the optimal point and the radius of the local sampling space;

in this step, the initial local sample space can be generated as in equation 9, i.e., with the existing optimum point x' _opt Taking r as an upper boundary and a lower boundary as a center, generating an initial local sampling space as shown in the following formula:

A′ _{opt_space} ＝x′ _opt ±r 9)

for example, the existing optimum point x 'can be used' _opt For the center, the upper bound increases the size of the initial sampling space by 0.1 times on the basis, and the lower bound decreases the size of the initial sampling space by 0.1 times on the basis, so as to generate the initial local sampling space.

Step S405, determining a local sampling space based on the initial local sampling space and the initial design space.

In one embodiment of the present application, the initial local sampling space may be compared with an initial design space to determine whether the initial local sampling space exceeds the initial design space; if the initial local sampling space exceeds the initial design space, determining an intersection of the initial local sampling space and the initial sampling space as a local sampling space of the iteration of the round; and if the initial local sampling space does not exceed the initial design space, determining the initial local sampling space as the local sampling space of the iteration of the round.

That is, the initial local sampling space can be compared with the initial design space, and whether the initial local sampling space exceeds the initial design space or not can be checked, and if so, the initial local sampling space is reduced to the initial design spaceInside (e.g. x _k Upper bound > x in initial local sampling space _kmax Then x is _k The upper bound of (2) is modified to x _kmax ) I.e. taking the intersection of the initial local sampling space and the design space as the local sampling space A of the iteration of the round _{opt_space} . As shown in the following formula 10:

A _{opt_space} ＝A′ _{opt_space} ∩A _{design_space} 10)

according to the method and the device, the size of the initial sampling space is calculated based on the value range of each design variable, the radius of the local sampling space is determined based on the size of the initial sampling space, the initial local sampling space is further determined based on the optimal point and the radius of the local sampling space, the local sampling space is further determined based on the initial local sampling space and the initial design space, the local sampling space is built based on the value range of each influence index and the optimal point, the built local sampling space is near the optimal point, and the local sampling space near the optimal point is further conveniently subjected to local increasing sampling density.

In yet another embodiment of the present application, step S101 performs DOE sampling of a plurality of impact indicators, determining an initial DOE training sample data set and a DOE verification sample data set based on the DOE sampling results, including:

Step S501, DOE sampling is carried out on a plurality of influence indexes of the sound insulation performance of the automobile, so as to obtain a DOE training sample point set and a DOE verification sample point set;

in the step, DOE sampling is carried out on a plurality of influence indexes of the sound insulation performance of the automobile, and a DOE training sample point set and a DOE verification sample point set are generated after sampling.

Step S502, performing analysis and calculation on the sample points in the DOE training sample point set and the DOE verification sample point set to obtain an initial DOE training sample data set and a DOE verification sample data set.

And performing test or high-precision analysis model calculation on the DOE training sample point set and the DOE verification sample point set to obtain a DOE training sample data set and a DOE verification sample data set containing accurate test results.

In practical application, as the problem of large-scale high-dimensional optimization is solved, the real vehicle test cost is too high to be realized, the high-frequency simulation software is applied, the DOE training sample point set and the DOE verification sample point set are substituted for analysis and calculation, and the whole process is automatically carried out through multi-disciplinary optimization business software, so that an initial DOE training sample data set and a DOE verification sample data set are obtained.

In still another embodiment of the present application, there is further provided an apparatus for optimizing the design of multiple subjects for sound insulation performance of an automobile, as shown in fig. 8, comprising:

The acquiring module 11 is configured to acquire relevant parameters of the multi-disciplinary design optimization of the sound insulation performance of the automobile, determine a plurality of impact indicators of the sound insulation performance of the automobile that need to be optimally designed, perform DOE sampling of the plurality of impact indicators, and determine an initial DOE training sample data set and a DOE verification sample data set based on a DOE sampling result;

the first determining module 12 is configured to determine, based on accuracy, a target proxy model corresponding to each impact indicator from a plurality of candidate proxy model groups corresponding to each impact indicator, so as to obtain a target proxy model group formed by target proxy models corresponding to all the impact indicators;

the optimization module 13 is used for performing global optimization based on the target agent model group to obtain an optimal point, and determining an actual index response based on the optimal point;

and the updating module 14 is configured to update the target agent model set by using an encrypted DOE training sample data set corresponding to the optimal point if the difference between the actual index response and the optimization constraint condition is greater than a preset threshold, wherein the encrypted DOE training sample data set is constructed based on the optimal point and the value ranges of a plurality of design variables in the relevant parameters, and perform global optimization by using the updated target agent model until the difference between the actual index response and the optimization constraint condition is less than or equal to the preset threshold, and output the final updated target agent model set for optimizing the sound insulation performance of the automobile.

Optionally, the updating module includes:

Optionally, the acquiring submodule includes:

Optionally, the building unit comprises:

Optionally, the third determining subunit is further configured to:

Optionally, the acquiring module includes:

Optionally, the apparatus further comprises:

Optionally, the first determining module includes:

Optionally, the optimizing module includes:

In yet another embodiment of the present application, there is provided an electronic device including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

a memory for storing a computer program;

and the processor is used for realizing the multidisciplinary design optimization method for the sound insulation performance of the automobile according to any method embodiment when executing the program stored in the memory.

According to the electronic equipment provided by the embodiment of the invention, the processor constructs the target agent model group through executing the program stored in the memory based on the target agent models with highest precision, so that the precision of the target agent model group is greatly improved, the precision of each influence index when the target agent model group is subjected to multi-disciplinary design optimization is ensured, moreover, the target agent model group is updated through utilizing the encryption DOE training sample data set for increasing the local sampling density near the optimal point, the precision of the updated target agent model group near the optimal point is improved, the global precision of the agent model is effectively improved, the situation that the scheme which is optimized to reach standards through the agent model is avoided, and the situation that the scheme which is not reached to standards is found in practical experiments is avoided.

The communication bus 1140 mentioned above for the electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industrial Standard Architecture (EISA) bus, etc. The communication bus 1140 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown in fig. 9, but not only one bus or one type of bus.

The communication interface 1120 is used for communication between the electronic device and other devices described above.

The memory 1130 may include Random Access Memory (RAM) or non-volatile memory (non-volatile memory), such as at least one magnetic disk memory. Optionally, the memory may also be at least one memory device located remotely from the aforementioned processor.

The processor 1110 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSP), application Specific Integrated Circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

In yet another embodiment of the present application, there is further provided a computer readable storage medium having stored thereon a program of an automotive sound insulation performance multidisciplinary design optimization method, which when executed by a processor, implements the steps of the automotive sound insulation performance multidisciplinary design optimization method described in any of the foregoing method embodiments.

It should be noted that in this document, relational terms such as "first" and "second" and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multidisciplinary design optimization method for sound insulation performance of an automobile is characterized by comprising the following steps:

2. The method of multidisciplinary design optimization of sound insulation performance of an automobile of claim 1, wherein updating the set of target agent models with the encrypted DOE training sample data set corresponding to the optimal point comprises:

3. The method of multidisciplinary design optimization of sound insulation performance of an automobile according to claim 2, wherein obtaining an encrypted DOE training sample dataset that increases local sampling density within a preset range of the optimal point comprises:

4. The multi-disciplinary design optimization method for sound insulation performance of an automobile according to claim 3, wherein constructing a local sampling space based on the optimal point and the range of values of a plurality of design variables in the related parameters comprises:

5. The method of multidisciplinary design optimization of sound insulation performance of an automobile of claim 4, wherein determining a local sampling space based on the initial local sampling space and an initial design space comprises:

6. The method of multidisciplinary design optimization of sound insulation performance of an automobile of claim 1, wherein performing DOE sampling of a plurality of impact indicators, determining an initial DOE training sample dataset and a DOE verification sample dataset based on DOE sampling results, comprises:

7. The method for multidisciplinary design optimization of sound insulation performance of an automobile of claim 6, further comprising:

8. The method for multidisciplinary design optimization of sound insulation performance of an automobile according to claim 1, wherein determining a target agent model corresponding to each influence index in a plurality of agent model groups to be selected respectively corresponding to each influence index based on accuracy, and obtaining a target agent model group composed of target agent models corresponding to all influence indexes comprises:

9. The method for multidisciplinary design optimization of sound insulation performance of an automobile according to claim 1, wherein global optimization is performed based on a target agent model group to obtain an optimal point, and determining an actual index response based on the optimal point comprises:

10. The utility model provides a car sound insulation performance multidisciplinary design optimizing device which characterized in that includes:

11. The electronic equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus;

a memory for storing a computer program;

and the processor is used for realizing the multidisciplinary design optimization method for the sound insulation performance of the automobile according to any one of claims 1 to 9 when executing the program stored in the memory.

12. A computer-readable storage medium, wherein a program of the vehicle sound insulation performance multidisciplinary design optimization method is stored on the computer-readable storage medium, and when the program of the vehicle sound insulation performance multidisciplinary design optimization method is executed by a processor, the steps of the vehicle sound insulation performance multidisciplinary design optimization method according to any one of claims 1 to 9 are implemented.