Competitive Surrogate Assisted Optimization Method And Its Design Platform Construction

Surrogate-model assisted optimization is an optimization method proposed for the "black box problem".Although its related theories have been developed more maturely,they have not been promoted smoothly in domestic small and medium enterprises.On the one hand,this method is cumbersome to apply,and requires too much expertise,which leads to excessively high employment costs for enterprises.On the other hand,the relatively mature commercial optimization software on the market is too complicated in operation and only provide offline optimization methods assisted by surrogate models.Therefore,this paper uses the Qt Creator integrated development environment to develop a competitive surrogate-model assisted optimization design platform based on the C++ language,and applies it to the optimization design of two engineering examples.The main content and results of this paper are as follows:1.Research surrogate-model assisted optimization methods and related technologies deeply,based on operational simplicity principle of optimization design platform,an optimal design method based on competitive model is proposed in this paper.Then the basic architecture of the optimizzation platform is determined,which is divided into four main modules: design of experiment,sensitivity analysis,surrogate model and optimization algorithm.2.In order to facilitate the use of engineers,a graphical user interface has been developed.The layout of the interface follows the principle of simplicity and beauty,and the single page parameter setting is simple and not complicated.The operation flow of the interface strives to be simple and easy to use,and it is more natural for users to use.This is especially reflected in the fact that the surrogate model module has an automatic modeling function,and users do not need to set parameters.In addition,the more complicated parameter settings are all set to default values ??on the interface.3.The combination of the surrogate model and the optimization algorithm in the existing software is often through offline optimization,that is,a high-precision surrogate model is constructed in advance to replace the original "black box problem",and then the surrogate model is optimized.The biggest drawback of this method is that it relies too much on the predictive accuracy of the surrogate model.The optimization results only for reference,and need to be further verified.On this basis,this article adds an online optimization method,that is,a basic surrogate model is constructed from a small number of samples,and then sample points are added through the principle of sample filling,and then the model is continuously updated.The model is updated and optimized at the same time,and finally converges to the most excellent solution.Online optimization is a closed-loop optimization process,and its key technology is to achieve a complete closed-loop through a simulation interface.In this paper,the Abaqus interface is deployed to the optimization platform,which can realize script parametric modeling and submit it to Abaqus for background solution calculation,and then extract the calculation results to guide the next script file update,thereby forming a complete set of closed-loop optimization.4.This paper uses a series of numerical examples to test the functional modules of the optimization platform to prove its practicability.And the optimization platform is applied to the optimization of sheet metal stamping of automobile hoods and the optimization of composite lattice structure of high-speed train converter boxes.