CN113935235A

CN113935235A - Engineering design optimization method and device based on genetic algorithm and agent model

Info

Publication number: CN113935235A
Application number: CN202111188161.4A
Authority: CN
Inventors: 陈科甲
Original assignee: Beijing Rope Is Systems Technology LLC
Current assignee: Beijing Rope Is Systems Technology LLC
Priority date: 2021-10-12
Filing date: 2021-10-12
Publication date: 2022-01-14

Abstract

The application discloses an engineering design optimization method and device based on a genetic algorithm and a proxy model, wherein the method comprises the following steps: constructing a target agent model corresponding to the target function and a constraint agent model corresponding to the constraint function based on the sample data set; solving the target agent model by adopting a genetic algorithm, wherein the mutation operation of the genetic algorithm comprises the following steps: obtaining a population S_nDetermining the variation disturbance value corresponding to each design variable according to the maximum value and the minimum value in the fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding the corresponding variation disturbance value to each design variable to obtain the variationThe posterior chromosome; if the simulation termination condition is not met, adding a sample and updating the target agent model and the constraint agent model; and if the simulation termination condition is met, taking the optimal solution obtained by solving as the optimization result of the design variable and outputting the optimization result.

Description

Engineering design optimization method and device based on genetic algorithm and agent model

Technical Field

The application relates to the technical field of simulation optimization, in particular to an engineering design optimization method and device based on a genetic algorithm and a proxy model.

Background

Aiming at the engineering optimization problem, generating initial sample point data of design variables by a test design method, performing real calculation on the sample points, and establishing a proxy model based on the sample points; then finding out an optimal result based on the agent model, and carrying out real calculation on the optimal result; and if the precision of the calculation result does not meet the requirement, reestablishing the proxy model for optimization based on the newly-added sample points until the optimal result output by the proxy model meets the precision requirement, and taking the optimal result output at the moment as the simulation optimization result of the design variable. However, the calculation amount for establishing and optimizing the proxy model is large, a large amount of time is consumed, and sometimes several weeks or months are consumed, so that the engineering practicability is greatly reduced by the design period, and the application field of the engineering optimization method is limited.

Disclosure of Invention

The embodiment of the application provides an engineering design optimization method, an engineering design optimization device, electronic equipment and a storage medium based on a genetic algorithm and a proxy model, and a variation method in the genetic algorithm is improved when the genetic algorithm is used for optimizing the proxy model, so that local optimization can be effectively skipped during optimization, and the optimization efficiency and the optimization effect of the proxy model are improved.

In one aspect, an embodiment of the present application provides an engineering design optimization method based on a genetic algorithm and a proxy model, including:

s201, obtaining input parameters required by engineering design optimization, wherein the input parameters comprise: at least two design variables to be optimized, and an objective function and a constraint function for the design variables;

s202, generating a sample data set, wherein each sample in the sample data set comprises: a set of values of the at least two design variables, and true response values of the objective function and the constraint function obtained based on the set of values;

s203, constructing a target agent model corresponding to the target function and a constraint agent model corresponding to the constraint function based on the sample data set;

s204, solving the target agent model by adopting a genetic algorithm to obtain an optimal solution of the target agent model under the constraint of the constraint agent model, wherein the optimal solution comprises optimal values of the at least two design variables; wherein, in the population S for each generation_nInheritance is performed to obtain a next generation population S_n+1Then, the population S is treated in the following manner_nPerforming mutation operations on each chromosome: obtaining a population S_nDetermining the maximum fitness and the minimum fitness from the obtained fitness, determining a variation disturbance value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding a corresponding variation disturbance value to each design variable of each chromosome to obtain a chromosome after variation; wherein, the population S_nComprising a predetermined number of chromosomes, each chromosome comprising: a set of values for the at least two design variables;

s205, judging whether a simulation termination condition is met, if so, executing a step S208, otherwise, executing a step S206;

s206, newly adding q samples in the sample data set;

s207, updating the target agent model and the constraint agent model based on the sample data set after the new sample is added, and returning to the step S204;

and S208, taking the optimal solution as an optimization result of the design variable and outputting the optimization result.

In one aspect, an embodiment of the present application provides an engineering design optimization apparatus based on a genetic algorithm and a proxy model, including:

an input module, configured to obtain input parameters required for engineering design optimization, where the input parameters include: at least two design variables to be optimized, and an objective function and a constraint function for the design variables;

an initial sample generation module, configured to generate a sample data set, where each sample in the sample data set includes: a set of values of the at least two design variables, and true response values of the objective function and the constraint function obtained based on the set of values;

the initial model building module is used for building a target agent model corresponding to the target function and a constraint agent model corresponding to the constraint function based on the sample data set;

the optimization module is used for solving the target agent model by adopting a genetic algorithm so as to obtain an optimal solution of the target agent model under the constraint of the constraint agent model, wherein the optimal solution comprises optimal values of the at least two design variables; wherein, in the population S for each generation_nInheritance is performed to obtain a next generation population S_n+1Then, the population S is treated in the following manner_nPerforming mutation operations on each chromosome: obtaining a population S_nDetermining the maximum fitness and the minimum fitness from the obtained fitness, determining a variation disturbance value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding a corresponding variation disturbance value to each design variable of each chromosome to obtain a chromosome after variation; wherein, the population S_nComprising a predetermined number of chromosomes, each chromosome comprising: a set of values for the at least two design variables;

the simulation termination judging module is used for judging whether the simulation termination condition is met, if so, executing the function of the output module, and otherwise, executing the function of the newly added sample module;

a new sample adding module for adding q samples in the sample data set;

the model updating module is used for updating the target agent model and the constraint agent model based on a sample data set after a sample is newly added and returning to execute the function of the optimizing module;

and the output module is used for taking the optimal solution as an optimization result of the design variable and outputting the optimization result.

In one aspect, an embodiment of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods when executing the computer program.

In one aspect, an embodiment of the present application provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, implement the steps of any of the above-described methods.

In one aspect, an embodiment of the present application provides a computer program product or a computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the method provided in any of the various alternative implementations of control of TCP transmission performance described above.

According to the engineering design optimization method, device, electronic equipment and storage medium based on the genetic algorithm and the proxy model, when the genetic algorithm is used for optimizing the proxy model, a variation method in the genetic algorithm is improved, so that local optimization can be effectively skipped, and the optimization efficiency and the optimization effect of the proxy model are improved. The method has the advantages that the design accuracy of global optimization is guaranteed, meanwhile, the calculation amount of the traditional simulation optimization method is reduced in magnitude, the optimization period of actual engineering design is greatly shortened, and a lot of design work which cannot be carried out becomes practical.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic diagram of an application scenario of an engineering design optimization method based on a genetic algorithm and a proxy model according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of an engineering design optimization method based on genetic algorithm and a proxy model according to an embodiment of the present application;

fig. 3 is a schematic flowchart of generating a sample data set according to an embodiment of the present application;

FIG. 4 is a schematic structural diagram of an engineering design optimization device based on a genetic algorithm and a proxy model according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

It should be noted that, in the case of no conflict, the features in the following embodiments and examples may be combined with each other; moreover, based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without any creative effort belong to the protection scope of the present application.

It is noted that various aspects of the embodiments are described below within the scope of the appended claims. It should be apparent that the aspects described herein may be embodied in a wide variety of forms and that any specific structure and/or function described herein is merely illustrative. Based on the present application, one skilled in the art should appreciate that one aspect described herein may be implemented independently of any other aspects and that two or more of these aspects may be combined in various ways. For example, an apparatus may be implemented and/or a method practiced using any number of the aspects set forth herein. Additionally, such an apparatus may be implemented and/or such a method may be practiced using other structure and/or functionality in addition to one or more of the aspects set forth herein.

For convenience of understanding, terms referred to in the embodiments of the present application are explained below:

the proxy model comprises the following steps: it is often referred to as an approximation mathematical model, also called a response surface model, approximation model or meta model, that can replace the more complex and time consuming numerical analysis during the analysis and optimization design process. For the simulation optimization design problem, a numerical analysis program or software can be regarded as an 'input-output' system, namely an objective function and a constraint function which are output quantities can be regarded as functions of design variables, and the optimal point can be rapidly predicted by establishing an agent model of the objective function and the constraint function on the design variables, so that the agent model method not only can greatly improve the optimization design efficiency, but also can reduce the optimization difficulty, and is beneficial to filtering numerical noise and realizing parallel optimization design. In the research aspect of the agent model method, various agent model methods including polynomial Response Surfaces (RSM), Kriging models, Radial Basis Functions (RBFs), Neural Networks (NN), Support Vector Regression (SVR), Multivariate Interpolation and Regression (MIR), polynomial chaotic development (PCE), and the like have been developed.

In the aspect of reasonably selecting and researching multidimensional space sample points, modern test design methods such as Latin Hypercube (LHS), orthogonal design, Uniform Design (UD) and the like suitable for computer numerical simulation experiments are developed besides a classical test design method (such as full factor, center combination, D-optimization and the like).

GA (genetic algorithm ): the search algorithm for solving the optimization in computational mathematics is one of the evolutionary algorithms.

Fitness Function (Fitness Function): and the evaluation function of the genetic algorithm is used for carrying out the advantages and disadvantages on the population. The higher the fitness function value, the easier it is to survive, otherwise it is easier to be eliminated.

Population S_n: representing the nth generation in the genetic process, and the population of each generation is mutated on the basis of the population of the previous generation. Population S_nMay be referred to as a population S_n+1Of the parent, correspondingly, the population S_n+1May be referred to as a population S_nThe progeny of (1).

Selecting operation: refers to selecting high-quality chromosomes from new chromosomes obtained by genetic evolution and adding the chromosomes into the next generation population S_n+1Thereby gradually optimizing chromosome quality in the population. Generally, the probability of chromosome selection inheritance to the next generation can be determined through a fitness function, i.e. chromosomes with higher fitness can be added to the next generation population S_n+1And eliminating chromosomes with low fitness. Alternatively, roulette may be used to randomly elect to join the next generation population S_n+1Dyeing ofThe proportion of each chromosome in the wheel can be determined according to the fitness of the chromosome, and the higher the fitness of the chromosome is, the higher the proportion in the wheel is, namely, the higher the probability of being selected is.

Mutation operation: is to perform genetic variation on a single chromosome to obtain a new chromosome. For example, for chromosome { p₃，q₁，x₃Mutation is carried out, i.e. p is respectively treated₃Carrying out a mutation, q₁Carrying out a mutation, x₃And carrying out mutation to obtain a variable combination which is a new individual.

And (3) cross operation: refers to a process of exchanging parts of genes of two chromosomes with each other in a certain way to form two new chromosomes. For example, the parent chromosomes are X and Y, the chromosome length of X is less than or equal to Y, the Y is reordered, each nearest neighbor gene in the corresponding X is found out, and then random point selection is carried out to carry out single-point crossing. For example, chromosome { p₃，q₁，x₃And chromosome { p }₅，q2，x₄Are crossed, p₃And p₂Cross, q₁And q is₂Cross, x₃And x₄And crossing, and combining the obtained results into a new individual.

Any number of elements in the drawings are by way of example and not by way of limitation, and any nomenclature is used solely for differentiation and not by way of limitation.

At present, the calculation amount of the optimization proxy model is large, a large amount of time is consumed, and sometimes weeks or months are consumed, so that the engineering practicability is greatly reduced by the design period, and the application field of the engineering optimization method is limited.

Therefore, the application provides an engineering design optimization method based on a genetic algorithm and a proxy model, and when the proxy model is optimized by using the genetic algorithm, a variation method in the genetic algorithm is improved, namely a variation disturbance value corresponding to each design variable in each chromosome is determined according to the maximum value and the minimum value in the chromosome fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and the corresponding variation disturbance value is added to each design variable of each chromosome to obtain the chromosome after variation, so that the diversity when the design variables vary is increased, and the local optimum is effectively skipped, thereby improving the optimization efficiency and the optimization effect of the proxy model.

After introducing the design concept of the embodiment of the present application, some simple descriptions are provided below for application scenarios to which the technical solution of the embodiment of the present application can be applied, and it should be noted that the application scenarios described below are only used for describing the embodiment of the present application and are not limited. In specific implementation, the technical scheme provided by the embodiment of the application can be flexibly applied according to actual needs.

Reference is made to fig. 1, which is a schematic view of an application scenario of an engineering design optimization method based on a genetic algorithm and a proxy model provided in an embodiment of the present application. The application scenario comprises a terminal device 101 and a server 102, and the terminal device 101 and the server 102 can be connected through a wireless or wired network. The terminal device 101 includes, but is not limited to, an electronic device such as a desktop computer, a mobile computer, a tablet computer, and the like, an application program for simulation optimization is installed inside the terminal device 101, the application program can display a responsive operation interface through the terminal device 101, a user configures simulation parameters, selects an optimization method and an agent model through the operation interface, and dynamically monitors a simulation process and a simulation result in real time. The server 102 mainly provides computing resources required in the simulation optimization process for the terminal device 101, and the server 102 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, middleware service, a domain name service, a security service, a CDN, a big data and artificial intelligence platform, and the like.

The engineering design optimization method based on the genetic algorithm and the agent model can be applied to the engineering design fields of pneumatic optimization design, structural optimization design, missile design, aerospace vehicle design, high-speed train appearance optimization design, multidisciplinary optimization design and the like so as to optimize various design variables in the engineering design, such as engineering design parameters of weight, size, performance and the like.

Of course, the method provided in the embodiment of the present application is not limited to be used in the application scenario shown in fig. 1, and may also be used in other possible application scenarios, and the embodiment of the present application is not limited. The functions that can be implemented by each device in the application scenario shown in fig. 1 will be described in the following method embodiments, and will not be described in detail herein.

To further illustrate the technical solutions provided by the embodiments of the present application, the following detailed description is made with reference to the accompanying drawings and the detailed description. Although the embodiments of the present application provide the method operation steps as shown in the following embodiments or figures, more or less operation steps may be included in the method based on the conventional or non-inventive labor. In steps where no necessary causal relationship exists logically, the order of execution of the steps is not limited to that provided by the embodiments of the present application.

Referring to fig. 2, an embodiment of the present application provides an engineering design optimization method based on a genetic algorithm and a proxy model, which is applicable to electronic devices such as a terminal device or a server, and includes the following steps:

s201, obtaining input parameters required by engineering design optimization, wherein the input parameters comprise: at least two design variables to be optimized, and an objective function and a constraint function for the design variables.

During specific implementation, a user can input various input parameters required for simulation optimization of specific engineering design problems through a setting interface provided by the terminal device, such as various design variables to be optimized, an objective function for describing the design variables, and a constraint function for describing the constraint relationship among the design variables, and can set one or more constraint functions according to the specific engineering design problems. For example, the design variables of the configuration include a₁、a₂、a₃And a₄The objective function is f (a)₁,a₂,a₃,a₄) The constraint function is g (a)₁,a₂,a₃) And h (a)₂,a₄)。

In practical applications, the user-configurable input parameters may further include: optimizing a target direction (such as optimizing a maximum value or a minimum value), optimizing method selection (such as a gradient optimizing algorithm like a genetic algorithm, a quasi-newton method, and the like), proxy model selection (such as a polynomial response surface, a Kriging model, a radial basis function, and the like), constraint condition method selection (such as a lagrange method), proxy model optimization iteration step number, proxy model convergence precision, design variable association parameters, a value range of design variables (including a design variable lower limit and a design variable upper limit), an initial sample point number, and the like, and specifically including functions provided by a configurable parameter visual application program and user selection determination, for example, when a user selects a genetic algorithm as an optimizing method, parameters required for executing the genetic algorithm can be further configured, including: population number, optimization algebra, lower limit of mutation probability, upper limit of cross probability, lower limit of cross probability and the like.

S202, generating a sample data set, wherein each sample in the sample data set comprises: a set of values of the at least two design variables, a true response value of the objective function and the constraint function obtained based on the set of values.

In specific implementation, a DOE test design method can be adopted, a certain number of sample points are randomly generated based on the design variables in the input parameters and the value range of the design variables, and the number of the sample points is determined by the number of the configured initial sample points. For example, the design variables of the configuration include a₁、a₂、a₃And a₄，a₁、a₂、a₃And a₄Respectively have the value ranges of (0,1), (1, 10), (2, 100) and (3, 5)]If the number of the initial sample points is 20, 20 sample points with different values are generated, and each sample point contains 4 design variables a₁～a₄For example, one of the sample points may be { a }₁＝0.1,a₂＝3,a₃＝20,a₄＝3}。

Then, inputting the generated sample points into the existing commercial software (such as IDE) for calculating the target function and the constraint function one by one, obtaining the real response value of the target function and the real response value of the constraint function corresponding to each sample point, and inputting all the sample points and the real response values thereofAnd sorting the corresponding real response values into a sample data set. Each sample in the sample data set may be represented as { a }₁,a₂,a₃,a₄F, g, h, wherein f represents the true response value of the objective function, and g and h represent the true response values of the constraint function.

In specific implementation, the sample needs to be deduplicated to ensure that there are no duplicate samples in the sample data set.

S203, based on the sample data set, a target agent model corresponding to the target function and a constraint agent model corresponding to the constraint function are constructed.

During specific implementation, a proxy model is established for the objective function to be optimized, and is recorded as a target proxy model, and a model file can be recorded as: sumo. Establishing an agent model for each constraint function, and recording the agent model as a constraint agent model, wherein model files are respectively as follows: restrant 1.sumo, restrant 2.sumo, and the like. The specific method for constructing the proxy model is determined by the type of the selected proxy model, and taking the kriging model as an example, the specific process for constructing the proxy model can refer to the paper "kriging model and proxy optimization algorithm research progress", and the specific construction method is the prior art and is not described any more.

S204, solving the target agent model by adopting a genetic algorithm to obtain an optimal solution of the target agent model under the constraint of the constraint agent model, wherein the optimal solution comprises optimal values of at least two design variables.

In specific implementation, a genetic algorithm can be adopted to call the target agent model to solve, the target minimum value is searched, and meanwhile, the constraint condition is considered. When there is constraint, different constraint proxy models are called to obtain the optimal solution x of the target proxy model_optAnd the optimal solution x is obtained_optInputting the target agent model to obtain the optimal target value (i.e. predicted value)

。

The process of solving the target agent model by adopting the genetic algorithm mainly comprises the following steps: initializing the first generation population S₀Cross selection, mutation selection, cross breedingPerforming fork operation and mutation operation, generating a progeny population, finding out the optimal chromosome and the like. Specifically, the solving process comprises the following steps:

first, generating an initialization population S of design variables₀Population S₀Comprising a predetermined number of chromosomes, each chromosome comprising: a set of values for at least two design variables;

second, the population S is treated_nPerforming mutation operation and crossover operation to obtain next generation population S_n+1Wherein the initial value of n is 0;

third, the slave population S_n+1Finding out the optimal chromosome;

fourthly, judging whether the genetic algorithm termination condition is met or not, if so, carrying out population S_n+1The optimal chromosome in (2) is used as the optimal solution of the target agent model, otherwise, the value of n is increased by 1 and the step (II) is returned.

In the second step described above, the population S is treated in the following manner_nPerforming mutation operations on each chromosome: obtaining a population S_nDetermining the maximum fitness and the minimum fitness from the obtained fitness, determining the variation disturbance value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding the corresponding variation disturbance value to each design variable of each chromosome to obtain the chromosome after variation.

Wherein n is a natural number, population S_nComprising a predetermined number of chromosomes, each chromosome comprising: a set of values for at least two design variables. When n is 0, the target agent model and the constraint agent model used herein are the target agent model and the constraint agent model constructed by step S203; when n is>0, the target agent model and the constraint agent model used here are the target agent model and the constraint agent model obtained by step S207. The mutation perturbation value is a value added to an original chromosome when a chromosome is mutated so as to change the chromosome within a reasonable range. The dyeing can be calculated by adopting any one of the existing fitness functionsAnd (4) fitness corresponding to the body.

Specifically, the deviation degree of the fitness corresponding to each chromosome may be determined according to the maximum fitness, the minimum fitness and the fitness of each chromosome in the chromosomes; then, for each design variable in the chromosome, based on the value of the design variable in the chromosome, the upper limit value and the lower limit value of the value range corresponding to the design variable, and the deviation corresponding to the chromosome, the variation disturbance value corresponding to the design variable in the chromosome is determined.

In specific practice, fit_kIs recorded as a population S_nFitness of the kth chromosome of (1), fit_minIs recorded as a population S_nMinimum fitness of (1), fit_maxIs recorded as a population S_nMaximum fitness of (1). For this purpose, step S2042 includes the steps of:

first, r and r are randomly selected from the range of (0,1)₁The value of (a).

Then, the fit is judged_minWhether it is equal to fit_max(ii) a When fit_max≠fit_minAccording to the formula

t₁＝t³And

obtaining a population S_nDeviation t of fitness of the kth chromosome₂(ii) a When fit_max＝fit_minThen, the population S_nDeviation t of fitness of the kth chromosome₂＝r³。

Then, r is determined₁The size of (2): if r₁Not less than 0.5, w corresponding to the ith design variable in the kth chromosome_i＝u_i+(v_i-u_i)*(1-t₂) Otherwise w_i＝u_i-u_i*(1-t₂) Wherein u is_i＝i_value-i_min，v_i＝i_max-i_min，i_valueFor the value of the ith design variable, i_minFor the lower limit of the value range of the ith design variable, i_maxAnd the upper limit value of the value range of the ith design variable.

And finally, determining a variation disturbance value W according to the specific value of W: if w_i<0, then the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i0; if 0 is less than or equal to w_iV is less than or equal to v, then the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i＝w_i(ii) a If w_i>v_iThen the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i＝v_i。

Based on this, assume the population S_nThe kth chromosome in (a)₁,a₂,a₃,a₄) A is obtained by the above steps₁、a₂、a₃、a₄Has a variation disturbance value of W₁、W₂、W₃、W₄Then the value after the chromosomal aberration is (a)₁+W₁,a₂+W₂,a₃+W₃,a₄+W₄)。

It should be noted that the embodiment of the present application mainly improves mutation operations in the genetic process, and other steps can be extended to processing modes in the existing genetic algorithm and are not described in detail.

S205, judging whether the simulation termination condition is met, if so, executing a step S208, otherwise, executing a step S206.

Wherein, the simulation termination condition comprises: and the error precision output by the agent model is smaller than a required value, or the iteration times reach an iteration time upper limit value.

In specific implementation, the currently obtained optimal solution x is used_optInputting the existing commercial software (such as IDE) for calculating the target function to obtain the real response value of the target function, and recording as y_opt. If y_optIf | is smaller than the preset threshold (e.g. 0.0001), the error precision of the proxy model output is

Otherwise the error precision of the proxy model output is

And S206, newly adding q samples in the sample data set.

The value of q may be set according to actual requirements, and is not limited here. The selection mode of the newly added sample can also be selected according to the actual requirement.

Specifically, an optimal solution x is selected_optCorresponding to y_optUsing proxy models optimization. sumo, restaint 1.sumo, restaint 2.sumo … …, the expectation of the objective function improvement for any sample point x was calculated: e [ I (x)]When E [ I (x)]Sample point x at maximum time_eSample point x_eSubstituting into IDE flow to complete calculation to obtain target value y_eSample point x_opt、y_optAnd sample point x_e、y_eAdding to the sample data set, if the optimization is to solve the maximum problem, then the sample point x_opt、-y_optAnd sample point x_e、-y_eAdding to the sample data set.

In practical applications, the sample points may be added by using an MSP (mixed-valued sampling) adding method, or by using an EI (improvement expectation criterion) adding method.

The MSP criterion minimization proxy model prediction criterion is the simplest, most direct and earliest adopted method, and the principle is to directly find the optimal solution x of the target function on the proxy model and obtain the optimal solution x of the target function predicted on the proxy model. And performing accurate numerical simulation analysis on x, adding the result serving as new sample data into the existing sample data set, and reestablishing the proxy model until the whole optimization process converges.

The improvement expectation criterion is also called an Efficient Global Optimization (EGO) method, and is only suitable for the case that the agent model solves the minimum problem, so that when the above genetic algorithm optimizes the agent model, the minimum problem is searched.

By finding x corresponding to the maximum value of E [ I (x) ], the point to be added is determined. Namely:

wherein: selecting the minimum response value as y in establishing the sample point set of the agent model used in the current optimization_opt. The selection method comprises the following steps: when no constraint exists, the minimum value of the sample response value is selected as y_opt. When there is a constraint, the minimum response value satisfying all the constraint conditions is selected as y_opt(ii) a If there are no sample points satisfying all constraints, the minimum response value satisfying the most constraint condition is preferentially selected as y_opt。

The predicted value of the proxy model is obtained for any x, and s is the standard deviation of optimation.

MSE＝s²(x)＝σ²{1-r^TR^-1r+(1-F^TR^-1r)²/F^TR^-1F}

φ is the probability density of the standard normal distribution:

Φ is the distribution function of the standard normal distribution:

using optimization algorithm (here, genetic algorithm) to obtain E [ I (x) within x value range]Is given as the maximum value of (a) to obtain x_e。

If the optimization problem is handled with constraints as follows: it is known from the foregoing that each constraint has a corresponding proxy model, namely, respaint 1.sumo and respaint 2.sumo … …. Calculating the standard deviation by adopting the formula to obtain corresponding s₁、s₂、s₃… … are provided. For any x, only for constraintsThe condition is that the proxy model of g (x) is less than or equal to 0, and the prediction values of the constraint obtained by calculating by adopting each constraint proxy model are g₁、g₂、g₃… … are provided. Assuming a total of N_cA constraint, such that the probability that any x satisfies the ith constraint is:

then, EI point addition with constraint is to use an optimization algorithm to solve the maximum value of the following formula in the value range of x to obtain x_e：

X obtained_eThe simulation calculation is completed (substituted into IDE flow) to obtain the result y_eI.e. the above-mentioned steps 10 and 11, and x is_e、y_eAdding to the sample data set.

And S207, updating the target agent model and the constraint agent model based on the sample data set after the new sample is added, and returning to the step S204.

According to the engineering design optimization method based on the genetic algorithm and the agent model, when the agent model is optimized by using the genetic algorithm, a variation method in the genetic algorithm is improved, the diversity during variation of design variables is increased, local optimization is effectively skipped, and therefore the optimization efficiency and the optimization effect of the agent model are improved. The evolutionary algorithms such as the genetic algorithm are adopted for optimization, the general calculated amount needs thousands of times, and the optimization method combining the agent model and the genetic algorithm is adopted, and the calculated amount is generally hundreds of times, so that the method provided by the application can be used for engineering design optimization, the calculated amount of the traditional simulation optimization method can be reduced in magnitude while the design accuracy of global optimization is ensured, the optimization cycle of actual engineering design is greatly shortened, and a lot of design work which cannot be carried out becomes practical.

On the basis of any one of the above embodiments, in the process of solving the target agent model by using the genetic algorithm, the initialization population S of the design variables is generated in the following way₀：

Step one, randomly generating sample points, wherein the sample points comprise a group of values of at least two design variables;

step two, carrying out numerical analysis on the generated sample points and obtaining corresponding output values, and if the output values are real response values, adding the sample points as a chromosome to the initialized population S₀Performing the following steps; if the output value is an abnormal value, deleting the sample point;

repeating the first step and the second step until the population S is initialized₀The number of chromosomes in the cell reaches a preset number.

In specific implementation, the sample points can be input into the existing commercial software (such as IDE) to obtain corresponding real response values, if the sample points are unreasonable, the commercial software returns abnormal values, and unreasonable sample points need to be removed.

In order to ensure that filial generations obtained through crossing and mutation are still discrete items, a binary coding mode is adopted to code discrete variables in a chromosome, a relationship between the binary and problem design variable discrete items is established, and the discrete items are converted into gray codes on the basis of binary coding, so that the probability of jumping out of local optimum is increased. The continuous design variables in the chromosome can be expressed by adopting a real number coding mode, and the operation efficiency can be improved by adopting the real number coding mode. Therefore, the optimization simulation method based on the genetic algorithm supports the simulation optimization problem of the continuous design variables and the simulation optimization problem of the discrete design variables, and solves the discrete and continuous mixed problem.

Specifically, when randomly generating sample points, the values of discrete design variables in the sample points are determined as follows: randomly selecting a numerical value in the value range of the discrete design variable, converting the selected numerical value into a binary code, and converting the binary code into a Gray code. The conversion between the binary code and the gray code is not described in detail in the prior art.

For example, the design variables are from 0 to b_i(i ═ 1,2,3 … n), and then a chromosome as a parent is combined. For example, the variable x₁Get p at random₃，x₂Variable random q₂If the corresponding gray codes are 011 and 01, respectively, the sequence of the parent individuals according to the design variables is { p }₃,q₂}。

For discrete design variables, each time a new chromosome is generated, there is a need for comparison with the already generated chromosomes, and no duplication of chromosomes can exist to ensure population diversity. Therefore, when genetic variation is performed on a chromosome including a discrete design variable, if a new chromosome obtained by the genetic variation is duplicated with a chromosome before the genetic variation, the genetic variation needs to be performed on the chromosome again. In specific implementation, whether to check the repeated chromosomes can be determined according to the solution space number and the population number, if the solution space number is more than 3 times of the population number, whether the repeated chromosomes exist needs to be checked, and otherwise, the repeated chromosomes do not need to be checked.

When performing the crossover operation on the chromosomes, the design variables in the chromosomes need to be subjected to the crossover operation one by one. For example, chromosome { p₃,q₂And chromosome { p }₅,q₁Is crossed, then p is added₃And p₅Cross over, q is₂And q is₁And crossing to obtain two new chromosomes. The process of interleaving the discrete design variables is as follows:

cross-over the first design variable in a chromosome (i.e., p)₃And p₅Cross), a location in the gray code is randomly selected, information behind the location is exchanged, and the location cannot be the last bit when attention is needed. For example, the design variable p₃Is 011, design variable p₅Is 110, the selected position is the second bit, p is swapped₃Third bit of (a) and p₅Third bit, the new variables produced by the crossover are: 111, and 010, where 111 is out of the design variable range, it needs to be dropped and the crossover operation is performed again. Cross-over (q) the second design variable in the chromosome₂And q is₁Crossover), the crossover process is as described above.

The variation process for discrete design variables is as follows: randomly selecting a position in Gray codes of the discrete design variables, wherein the position is changed into 1 if the position is 0, and is changed into 0 if the position is 1.

And if the new variable exceeds the value range, discarding the new variable and re-executing the crossover operation or mutation operation.

On the basis of any of the above embodiments, the termination condition of the genetic algorithm may include: in the process of solving the target agent model by adopting the genetic algorithm, the absolute value b is less than the threshold b for p times continuously_thWherein the absolute value b is the population S_n+1Average fitness and population S of_nThe absolute value of the difference between the average fitness of (2).

For this purpose, the population S is used_nObtaining next generation population S_n+1Thereafter, the population S is calculated_n+1Fitness of each chromosome in the population S is obtained_n+1Average fitness of (2); calculating population S_n+1Average fitness and population S of_nThe absolute value b of the difference between the average fitness values of (a); if the absolute values b obtained in the genetic process of the successive p generations are all less than the threshold value b_thTerminating the genetic algorithm and connecting the population S_n+1The optimal chromosome in (1) is used as the optimal solution of the target agent model, otherwise, the process is continued based on the population S_n+1And (4) carrying out heredity.

The termination condition of the genetic algorithm may further include: the inheritance frequency reaches a preset frequency. If the absolute values b obtained in the genetic process of the successive p generations are all less than the threshold value b_thOr when the genetic times reach the preset times, terminating the genetic algorithm and enabling the population S to be the same_n+1The optimal chromosome in (1) is used as the optimal solution of the target agent model, otherwise, the process is continued based on the population S_n+1And (4) carrying out heredity.

Wherein, p, b_thThe value of the preset times can be determined according to application requirements and by combining experience, and the embodiment of the application is not limited. For example, in some application scenarios, the fetching of pA value of 5, b_thThe value of (a) is 0.00001, and the value of the preset times is 1500.

By optimizing the convergence judgment standard (namely the termination condition of the genetic algorithm) of the genetic algorithm in the modeling process, the automatic judgment of convergence can be realized, the circulation can be stopped in time, and the time spent on invalid calculation is reduced, so that the modeling efficiency is optimized and improved. For example, the number of calculation iterations is set to 500, 500 loops are needed, and the optimal solution is stopped and output, whereas with the automatic determination convergence method provided in the present application, it is possible to meet the stop condition after 100 calculations, end the calculation and output the calculation result, and 400 invalid calculations can be reduced.

On the basis of any of the above embodiments, step S207 specifically includes: and for each agent model, if the agent model meets the preset condition corresponding to the incremental method, updating model parameters of the agent model based on the newly added q samples, and otherwise, reconstructing the agent model based on the sample data set after the newly added samples. Wherein the agent model comprises a target agent model and a constraint agent model.

Before each round of model iteration is started, an appropriate method is selected to update the proxy model according to the updating condition of the previous proxy model, when an increment method is selected, parameters of the proxy model are updated only on the basis of q newly added samples, the proxy model does not need to be reconstructed, the calculated amount can be greatly reduced on the premise of ensuring the precision of the proxy model, and the processing efficiency is improved; when the reconstruction method is selected, the proxy model can be reconstructed based on all samples in the sample data set, and the output precision of the proxy model is ensured. When the internal matrix scale of the proxy model is large, the reconstruction of the model takes hours or even more than ten hours, and the model updating time by adopting the incremental method only needs seconds, so that the calculation amount of the traditional simulation optimization method is reduced in magnitude while the design accuracy of global optimization is ensured, the optimization period of actual engineering design is greatly shortened, and a lot of design work which cannot be carried out becomes practical.

The reconstruction method in the embodiment of the application refers to: and training the proxy model based on all samples in the sample data set by means of maximum likelihood estimation or cross validation and the like to obtain the optimal model parameters corresponding to the proxy model. The specific implementation manner of the reconstruction method is similar to that in step S203, taking kriging as an example, reconstructing the proxy model, that is, finding the model parameter θ that maximizes L, and the specific process is the prior art and is not described again.

The incremental method in the embodiment of the application refers to: and on the basis of the existing agent model, the model parameter theta is unchanged, and q newly added samples are used for updating the model. Specifically, taking the kriging model as an example, the process of updating the proxy model by the incremental method mainly includes: calculating mathematical expected value beta related to model parameters of proxy model based on newly added q samples₀Increment of (a) based on₀Determining the updated model parameters theta' of the proxy model according to the increment and the existing model parameters theta of the proxy model.

Taking the kriging model as an example, the principle of the incremental method is as follows:

cholesky decomposition of the matrix R:

R＝LL^T

let F' be L^-1F，y_s'＝L^-1y_sWherein F and y_sSee proxy model component algorithm definition, L is a lower triangular matrix of n x n.

Carrying out QR decomposition on F': f' ═ QG^TWhere Q is a column vector of n x 1, G^TIf real, then: g^Tβ₀＝Q^Ty_s', can give beta₀。

Where | | | | is the norm (square root) of the solution, then squared.

In each iteration process in the agent model optimization, the newly added sample point matrix is x⁽ⁿ¹⁾,x⁽ⁿ²⁾...x^(nh)Wherein

m is the number of design variables. The calculated result of the sample point correspondence is y⁽ⁿ¹⁾,y⁽ⁿ²⁾...y^(nh)Finishing, in orderAs a matrix: Δ y_s＝[y⁽ⁿ¹⁾,y⁽ⁿ²⁾…y^(nh)]^T. At this time, the matrix R in the proxy model operation is:

wherein the content of the first and second substances,

for the R matrix in the mechanism model of the previous iteration, the R matrix is marked as R₀。

Reissue to order

Then it is determined that,

performing Cholesky decomposition on R:

order to

Then there are:

(namely, to

Cholesky decomposition to obtain L₃) Then, there are:

real response value y of sample used in previous iteration_s＝y_s0Y 'obtained from last iteration'_s＝y'_s0F ' ═ F ' obtained in the previous iteration '₀Corresponding QR is decomposed into

Then:

of the same wheel

Wherein:

Δy_sis the real response value vector of the newly added sample point.

Of the same wheel

Wherein:

F₀＝[1,1...1]^T，F₀n is 1; Δ F ═ 1,1.. 1]^TΔ F has h 1 s.

QR decomposition is carried out on the delta F':

it is known that Δ Q is obtained

Then is formed by

To obtain beta₀Increment of (a) delta beta₀

Therefore:

beta of this wheel₀Beta as the last round₀+Δβ₀

Of the same wheel

Y 'of the same wheel'_sF' of the wheel and beta of the wheel₀||²

Beta based on this wheel₀And σ²The updated model parameter theta can be calculated, and the concrete calculation formula refers to the study progress of the kriging model and the proxy optimization algorithm.

In a possible implementation manner, the preset condition corresponding to the incremental method may be: the number proportion of the samples which do not participate in the reconstruction agent model in the sample data set is not less than the proportion requirement.

For example, the preset conditions corresponding to the incremental method may be: a first number M_BAnd a second number M_AIs smaller than the preset value U. Wherein the first number M_BThe number of samples which do not participate in the reconstruction of the proxy model in the sample data set is determined; a second number M_AAnd (4) the number of samples participating in the reconstruction of the proxy model in the sample data set, namely the number of samples used when the proxy model is established by using the reconstruction method at the last time.

Suppose that the proxy model optimizes to add 2 samples in each loop, the number of samples in the initial sample data set is 10, and U is 0.25.

A first round: constructing a proxy model by using a reconstruction method based on 10 samples in the initial sample data set, wherein M is the time_A10. After the first round is finished, 2 samples are added in the sample data set.

And a second round: in the sample data set, if 10 samples participating in the reconstruction method and 2 samples not participating in the reconstruction method are present, M is_A＝10，M_B2, so M_B/M_A＝0.2<U, so the second round updates the proxy model using the incremental method. And after the second round is finished, adding 2 samples in the sample data set.

And a third round: since the reconstruction method is only used in the first round, 10 samples already participating in the reconstruction method in the sample data set at this time are obtained, and none of the newly added 4 samples participate in the reconstruction of the proxy model at the end of the first round and the second round, then M is obtained_A＝10，M_B4, thus M_B/M_A＝0.4>U, so the third round updates the proxy model using the reconstruction method, at which time 14 samples in the sample data set all participate in the reconstruction method. And after the third round is finished, adding 2 samples in the sample data set.

Fourth wheel: the third round was where 14 samples participated in the reconstruction method, then M_A14, the newly added 2 samples after the third round are not involved in the reconstruction method, then M_B2, so M_B/M_A＝2/14<U, so the fourth round updates the proxy model using the incremental method. And after the fourth round is finished, adding 2 samples in the sample data set. And the rest can be done in the same way until the simulation termination condition is met.

In another possible implementation, the preset condition corresponding to the incremental method may be: the output error of the proxy model constructed in the previous round is smaller than the error requirement.

For example, the preset conditions corresponding to the incremental method may be: u. of_sumoi≤u_THWherein, in the step (A),

u_THis a preset threshold value, y_iFor the true response value obtained based on the newly added sample,

for obtaining a prediction value s after inputting a newly added sample into the proxy model_sumoiThe mean square error of the proxy model is calculated based on the newly added samples. Wherein u is_THThe value of (a) can be determined according to actual requirements in combination with experience, e.g. u_THThe value of (A) can be 3,5, etc. In practical application, the newly added sample can be input into the existing commercial software (such as IDE) for calculating the objective function, so as to obtain the true response value y of the objective function_i。

Adding a new sample after each iteration is finished, and calculating the output error of the proxy model (comprising a target proxy model and a constraint proxy model) for each proxy model obtained in the current iteration before starting the next iteration:

if u is_sumoi≤u_THAnd updating the model parameters of the proxy model by adopting an incremental method, otherwise, reconstructing the proxy model by adopting a reconstruction method.

E.g. u_THIs 3. The parameter of the last round of target agent model f is beta₁And σ₁The parameter of the constraint proxy model g is beta₂And σ₂The parameter of the constraint proxy model h is beta₃And σ₃. After 2 samples are added, before the next iteration, the output error of the target agent model f on the first newly added sample is calculated to be u_{1 sample 1}The output error on the second new sample is u 2_{1 sample 2}1.8, both output errors are less than 3, so the target agent model is updated by adopting an incremental method; calculating the output error of the constraint proxy model g on the first newly added sample as u_{2 sample 1}The output error on the second new sample is u, 4_{2 sample 2}When the output error is larger than 3, a new constraint proxy model g is established by adopting a reconstruction method; calculating the output error of the constraint proxy model h on the first newly added sample as u_{3 sample 1}The output error on the second new sample is u, 6_{3 sample 2}And 6, both output errors are larger than 3, so a new constraint proxy model h is established by adopting a reconstruction method.

Providing two choices of a reconstruction method and an incremental method when the proxy model is reconstructed each time, and judging whether preset conditions corresponding to the incremental method are met or not based on relevant data of the proxy model constructed in the previous round; if the preset conditions are met, updating the proxy model by adopting an incremental method, namely updating the model parameters of the proxy model only based on the newly added samples without reconstructing the proxy model; and if the preset condition is not met, reconstructing the proxy model based on all samples in the sample data set after the new sample is added. Wherein, the preset condition may be: and the output error of the proxy model constructed in the previous round is smaller than the error requirement, or the number proportion of the samples which do not participate in reconstructing the proxy model in the sample data set is not smaller than the proportion requirement. Therefore, under the condition that the output error of the proxy model is small or the sample ratio not participating in the reconstruction of the proxy model is small, the model parameters of the proxy model are updated by selecting an incremental method, so that the calculated amount can be greatly reduced and the processing efficiency can be improved on the premise of ensuring the precision of the proxy model; and under the condition that the output error of the proxy model is larger or the sample ratio not participating in reconstructing the proxy model is larger, reconstructing the proxy model based on all samples in the sample data set, and ensuring the output precision of the proxy model.

On the basis of any of the above embodiments, referring to fig. 3, step S202 includes the following steps:

s301, dividing the value range corresponding to each design variable into a intervals, randomly taking a numerical value from each interval of each design variable, and further obtaining a sample point set containing a sample points, wherein each sample point comprises a group of values of at least two design variables.

For example, the design variables include a₁、a₂And a₃And the number of the sample points is 10, a is₁Is divided into 10 intervals, and a numerical value is taken from each interval to obtain a_1,1、a_1,2……a_1,10Obtaining a by the same method₂10 values of a_2,1、a_2,2……a_2,10And a₃10 values of a_3,1、a_3,2……a_3,10Then a is added₁、a₂And a₃The 10 values are randomly combined to obtain 10 sample points.

S302, aiming at any sample point in the sample point set, carrying out numerical analysis on the sample point and obtaining a corresponding output value, if the output value is a real response value, adding the sample point and the corresponding real response value into the sample data set as a sample, deleting the sample point in the sample point set, and if the output value is an abnormal value, deleting the sample point.

S303, judging whether the number b of the samples in the sample data set is equal to a, if so, executing the step S203, otherwise, executing the step S304.

S304, dividing the value range corresponding to each design variable into (a-b) intervals, randomly taking a numerical value from each interval of each design variable, further obtaining (a-b) sample points, adding the sample points into a sample point set, and executing the step S305.

S305, aiming at any sample point in the sample point set, carrying out numerical analysis on the any sample point and obtaining a corresponding output value, if the output value is a real response value, adding the any sample point and the corresponding real response value into the sample data set as a sample, deleting any sample point in the sample point set, and if the output value is an abnormal value, deleting any sample point.

S306, carrying out duplicate removal processing on the samples in the sample data set, and returning to the step S303.

When the sample points are reselected, the space domain is newly divided, and the diversity of the selected sample points is ensured.

As shown in fig. 4, based on the same inventive concept as the above engineering design optimization method based on the genetic algorithm and the proxy model, the embodiment of the present application further provides an engineering design optimization apparatus 40 based on the genetic algorithm and the proxy model, including:

an input module 401, configured to obtain input parameters required for engineering design optimization, where the input parameters include: at least two design variables to be optimized, and an objective function and a constraint function for the design variables;

an initial sample generating module 402, configured to generate a sample data set, where each sample in the sample data set includes: a set of values of the at least two design variables, and true response values of the objective function and the constraint function obtained based on the set of values;

an initial model building module 403, configured to build, based on the sample data set, a target agent model corresponding to the target function and a constraint agent model corresponding to the constraint function;

an optimization module 404, configured to solve the target agent model by using a genetic algorithm to obtain an optimal solution of the target agent model under the constraint of the constraint agent model, where the optimal solution includes optimal values of the at least two design variables; wherein, in the population S for each generation_nInheritance is performed to obtain a next generation population S_n+1Then, the population S is treated in the following manner_nPerforming mutation operations on each chromosome: obtaining a population S_nDetermining the maximum fitness and the minimum fitness from the obtained fitness, determining a variation disturbance value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding a corresponding variation disturbance value to each design variable of each chromosome to obtain a chromosome after variation; wherein, the population S_nComprising a predetermined number of chromosomes, each chromosome comprising: a set of values for the at least two design variables;

a simulation termination judging module 405, configured to judge whether a simulation termination condition is met, if yes, execute the function of the output module 408, and otherwise execute the function of the newly added sample module 406;

a sample adding module 406, configured to add q samples in the sample data set;

a model updating module 407, configured to update the target agent model and the constraint agent model based on the sample data set after the new sample is added, and return to execute the function of the optimizing module 404;

and the output module 408 is configured to output the optimal solution as an optimization result of the design variable.

In a possible implementation, the optimization module 404 is specifically configured to:

when fit_max≠fit_minAccording to the formula

t₁＝t³And t₂＝r^t1Obtaining a population S_nDeviation t of fitness of the kth chromosome₂Therein, fit_kAs a population S_nFitness of the kth chromosome of (1), fit_minAs a population S_nMinimum fitness of (1), fit_maxAs a population S_nMaximum fitness of (1);

if r₁Not less than 0.5, w corresponding to the ith design variable in the kth chromosome_i＝u_i+(v_i-u_i)*(1-t₂) Otherwise w_i＝u_i-u_i*(1-t₂) Wherein u is_i＝i_value-i_min，v_i＝i_max-i_min，i_valueFor the value of the ith design variable, i_minFor the lower limit of the value range of the ith design variable, i_maxUpper limit value of value range r and r for ith design variable₁Is a random number in (0, 1);

if w_i<0, then the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i＝0；

If 0 is less than or equal to w_iV is less than or equal to v, then the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i＝w_i；

If w_i>v_iThen the population S_nThe variation perturbation value W of the ith design variable of the kth chromosome_i＝v_i。

In one possible embodiment, when fit_max＝fit_minThen, the population S_nDeviation t of the kth chromosome in₂＝r³。

In a possible implementation, the optimization module 404 is further configured to: based on the population S_nObtaining next generation population S_n+1Thereafter, the population S is calculated_n+1Fitness of each chromosome in the population S is obtained_n+1Average fitness of (2); calculating population S_n+1Average fitness and population S of_nAbsolute difference of average fitness ofFor the value b; if the absolute values b obtained in the genetic process of the successive p generations are all less than the threshold value b_thTerminating the genetic algorithm and connecting the population S_n+1The optimal chromosome in (1) is used as the optimal solution of the target agent model, otherwise, the process is continued based on the population S_n+1And (4) carrying out heredity.

In a possible embodiment, the optimization module 404 is further configured to generate an initialization population S of design variables by₀：

Randomly generating sample points, wherein the sample points comprise a group of values of the at least two design variables;

step two, carrying out numerical analysis on the sample points and obtaining corresponding output values, and if the output values are real response values, adding the sample points as a chromosome to the initialized population S₀If the output value is an abnormal value, deleting the sample point;

repeating the first step and the second step until the initialization population S₀Up to said preset number.

In a possible implementation, the optimization module 404 is further configured to determine, when the sample points are randomly generated, values of discrete design variables in the sample points by: randomly selecting a numerical value in the value range of the discrete design variable, converting the selected numerical value into a binary code, and converting the binary code into a Gray code.

In a possible embodiment, the optimization module 404 is further configured to, when a genetic variation is performed on a chromosome containing a discrete design variable, if a new chromosome obtained by the genetic variation is duplicated with a chromosome before the genetic variation, perform the genetic variation on the chromosome again.

In a possible implementation manner, the model updating module 407 is specifically configured to update, for each proxy model, model parameters of the proxy model based on q newly added samples if the proxy model meets a preset condition corresponding to an incremental method, and return to execute the function of the optimizing module 404, otherwise, reconstruct the proxy model based on a sample data set after the newly added samples, and return to execute the function of the optimizing module 404; wherein the agent model comprises the target agent model and the constraint agent model;

in a possible implementation manner, the model updating module 407 is specifically configured to: calculating a mathematical expected value beta related to the model parameters of the proxy model based on the newly added q samples for each proxy model₀Increment of (a) based on₀Determining the updated model parameter theta' of the proxy model by the increment and the existing model parameter theta of the proxy model; wherein the agent model comprises the target agent model and the constraint agent model.

In a possible implementation manner, the preset conditions corresponding to the incremental method are as follows: the ratio of a first quantity to a second quantity is smaller than a preset value, wherein the first quantity is the quantity of samples which do not participate in the reconstruction of the proxy model in the sample data set, and the second quantity is the quantity of samples which participate in the reconstruction of the proxy model in the sample data set.

In a possible implementation manner, the preset conditions corresponding to the incremental method are as follows: u. of_sumoi≤u_THWherein, in the step (A),

for obtaining a prediction value s after inputting a newly added sample into the proxy model_sumoiThe mean square error of the proxy model is calculated based on the newly added samples.

In a possible implementation manner, the initial sample generating module 402 is specifically configured to:

s301, dividing a value range corresponding to each design variable into a intervals, randomly taking a numerical value from each interval of each design variable, and further obtaining a sample point set containing a sample points, wherein each sample point comprises a group of values of the at least two design variables;

s302, performing numerical analysis on any sample point in the sample point set and obtaining a corresponding output value, if the output value is a real response value, adding the sample point and the corresponding real response value into the sample point set as a sample, deleting the sample point in the sample point set, and if the output value is an abnormal value, deleting the sample point;

s303, judging whether the number b of the samples in the sample data set is equal to a, if so, executing a step S203, otherwise, executing a step S304;

s304, dividing the value range corresponding to each design variable into (a-b) intervals, randomly taking a numerical value from each interval of each design variable, further obtaining (a-b) sample points, adding the sample points into the sample point set, and executing the step S305;

s305, carrying out numerical analysis on any sample point in the sample point set and obtaining a corresponding output value, if the output value is a real response value, adding the sample point and the corresponding real response value into the sample data set as a sample, deleting the sample point in the sample point set, and if the output value is an abnormal value, deleting the sample point;

The engineering design optimization device based on the genetic algorithm and the proxy model and the engineering design optimization method based on the genetic algorithm and the proxy model adopt the same inventive concept, can obtain the same beneficial effects, and are not repeated herein.

Based on the same inventive concept as the engineering design optimization method based on the genetic algorithm and the agent model, the embodiment of the present application further provides an electronic device, which may be specifically (a control device or a control system inside the intelligent device, or an external device communicating with the intelligent device, for example) a desktop computer, a portable computer, a smart phone, a tablet computer, a Personal Digital Assistant (PDA), a server, and the like. As shown in fig. 5, the electronic device 50 may include a processor 501 and a memory 502.

The Processor 501 may be a general-purpose Processor, such as a Central Processing Unit (CPU), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, or a discrete hardware component, which may implement or execute the methods, steps, and logic blocks disclosed in the embodiments of the present Application. A general purpose processor may be a microprocessor or any conventional processor or the like. The steps of a method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware processor, or may be implemented by a combination of hardware and software modules in a processor.

Memory 502, which is a non-volatile computer-readable storage medium, may be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The Memory may include at least one type of storage medium, and may include, for example, a flash Memory, a hard disk, a multimedia card, a card-type Memory, a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Programmable Read Only Memory (PROM), a Read Only Memory (ROM), a charged Erasable Programmable Read Only Memory (EEPROM), a magnetic Memory, a magnetic disk, an optical disk, and so on. The memory is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory 502 in the embodiments of the present application may also be circuitry or any other device capable of performing a storage function for storing program instructions and/or data.

Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; the computer storage media may be any available media or data storage device that can be accessed by a computer, including but not limited to: various media that can store program codes include a removable Memory device, a Random Access Memory (RAM), a magnetic Memory (e.g., a flexible disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical Memory (e.g., a CD, a DVD, a BD, an HVD, etc.), and a semiconductor Memory (e.g., a ROM, an EPROM, an EEPROM, a nonvolatile Memory (NAND FLASH), a Solid State Disk (SSD)).

Alternatively, the integrated units described above in the present application may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as independent products. Based on such understanding, the technical solutions of the embodiments of the present application may be essentially implemented or portions thereof contributing to the prior art may be embodied in the form of a software product stored in a storage medium, and including several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media that can store program codes include a removable Memory device, a Random Access Memory (RAM), a magnetic Memory (e.g., a flexible disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical Memory (e.g., a CD, a DVD, a BD, an HVD, etc.), and a semiconductor Memory (e.g., a ROM, an EPROM, an EEPROM, a nonvolatile Memory (NAND FLASH), a Solid State Disk (SSD)).

The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An engineering design optimization method based on genetic algorithm and agent model is characterized by comprising the following steps:

s206, newly adding q samples in the sample data set;

2. The method of claim 1, wherein determining the variant perturbation value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome comprises:

when fit_max≠fit_minAccording to the formula

t₁＝t³And

obtaining a population S_nDeviation t of fitness of the kth chromosome₂Therein, fit_kAs a population S_nFitness of the kth chromosome of (1), fit_minAs a population S_nMinimum fitness of (1), fit_maxAs a population S_nMaximum fitness of (1);

if w_i<0, then the population S_nMutation of the ith design variable of the kth chromosomeDynamic value W_i＝0；

3. The method of claim 2, wherein when fit_max＝fit_minThen, the population S_nDeviation t of the kth chromosome in₂＝r³。

4. The method of claim 2, wherein the method is based on population S_nObtaining next generation population S_n+1Thereafter, the method further comprises:

calculating population S_n+1Fitness of each chromosome in the population S is obtained_n+1Average fitness of (2);

calculating population S_n+1Average fitness and population S of_nThe absolute value b of the difference between the average fitness values of (a);

if the absolute values b obtained in the genetic process of the successive p generations are all less than the threshold value b_thTerminating the genetic algorithm and connecting the population S_n+1The optimal chromosome in (1) is used as the optimal solution of the target agent model, otherwise, the process is continued based on the population S_n+1And (4) carrying out heredity.

5. Method according to any one of claims 1 to 4, characterized in that the initialisation population S of design variables is generated by₀：

6. The method of claim 5, wherein, when randomly generating sample points, the values of discrete design variables in the sample points are determined by:

randomly selecting a numerical value in the value range of the discrete design variable, converting the selected numerical value into a binary code, and converting the binary code into a Gray code.

7. The method of claim 6, further comprising:

when genetic variation is performed on a chromosome including a discrete design variable, if a new chromosome obtained by the genetic variation is duplicated with a chromosome before the genetic variation, the genetic variation is performed on the chromosome again.

8. An engineering design optimization device based on genetic algorithm and agent model, comprising:

an optimization module for solving the target agent model by using a genetic algorithm,obtaining an optimal solution of the target agent model under the constraint of the constraint agent model, wherein the optimal solution comprises optimal values of the at least two design variables; wherein, in the population S for each generation_nInheritance is performed to obtain a next generation population S_n+1Then, the population S is treated in the following manner_nPerforming mutation operations on each chromosome: obtaining a population S_nDetermining the maximum fitness and the minimum fitness from the obtained fitness, determining a variation disturbance value corresponding to each design variable in each chromosome according to the maximum fitness, the minimum fitness, the fitness of each chromosome and the value of each design variable in each chromosome, and adding a corresponding variation disturbance value to each design variable of each chromosome to obtain a chromosome after variation; wherein, the population S_nComprising a predetermined number of chromosomes, each chromosome comprising: a set of values for the at least two design variables;

a new sample adding module for adding q samples in the sample data set;

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the steps of the method of any of claims 1 to 7 are implemented when the computer program is executed by the processor.

10. A computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method of any one of claims 1 to 7.