Research And Application Of Multi-constraint Surrogate Assisted Optimization Method Based On Parallel Strategy

A variety of research methods have emerged for the expensive "black-box" constraint optimization problems in multidisciplinary optimization and engineering applications.The surrogate-assisted global optimization algorithm can obtain an accurate approximate optimal through a small number of function evaluations and solve expensive “black box” function problems effectively,which has become the research focus of multidisciplinary and structural optimization algorithms.This method mainly replaces the original complex model by constructing a surrogate model and uses the infill sampling criterion to obtain the optimal,which is excellent in solving optimization problems based on expensive "black box" functions.As a well-known surrogate-assisted optimization algorithm,EGO can obtain an optimal solution and well balance the local search and global search by the EI infill sampling criterion.Therefore,the suggested multi-constraint parallel EGO optimization(MCP-EGO)algorithm has certain theoretical significance and engineering value by improving the effect and the efficiency of EGO.The specific research content and results are mainly reflected as follows:(1)Comparing the accuracy of various popular surrogate models deeply and selecting the LSSVR model to replace the Kriging model while the variance information is still obtained by Kriging.This method can improve the approximate accuracy of the MCP-EGO algorithm.(2)Suggesting a MCP-EGO algorithm that includes maximize probability of feasibility(MPF)criterion,variant probability of feasibility criterion(VPF)and parallel criterion which enhances the ability of the MCP-EGO algorithm to solve complex constraint optimization problems,the ability to solve multi-constrained optimization problems and the efficiency.(3)Applying the IGA method to the structural optimization analysis and combining the FEA method and MCP-EGO algorithm to form a fast closed-loop optimization system and develop a CAD/CAE integrated fast closed-loop optimization software based on MATLAB/GUI.Providing a simple and convenient operation platform for the terminal designers who are not researchers in the field.