Evolutionary Algorithms Based Uniform And Optimal Experimental Design And System Realization

In industrial design problems,the complexity of the problem brings heavy computational pressure,which affects the performance of the experimental design.The most difficult problem is how to do the balance between the calculation accuracy and the efficiency in the process of product design.What’s more,there are often multiple objectives while need to satisfy the constraints and optimize at the same time in the process of design,which brings new challenges to traditional experimental design.Aiming at the problems of high cost,low cycle and low efficiency in the steps of experimental design,this paper focuses on using experimental design algorithms based on evolutionary algorithms to solve related problems,including the analysis of previous traditional algorithms,the implementation of the proposed algorithms,and the development of the software which intergrated algorithms which are widely used.In the dimension of the algorithm,the objective problem is modeled as a multi-objective problem with constraints based on the evolutionary algorithms which provide robust and impressive global search ability,then implement related algorithms to solve it.In the dimension of software,the classics algorithms in recent years about experimental design are implemented and integrated in the software for users with different needs.The main research work is as follows:(1)Research of the uniform experiment design algorithms.Three main methods of uniform experimental design are discussed and the advantages and challenges of their respective representative algorithms are analyzed.Further,for the classic algorithm ToPDE based on heuristically search,it’s low convergence quality of the first step solution,and the high computation complexity of the second step search is revealed.In order to solve those problems,this paper propose to combine the VWH model distribution estimation with DE algorithm in the first step of search to improve the quality of the solution.What’s more,the iterative replacement algorithm with lower time complexity is replaced by the deletion algorithm in the second step.Finally,the performance of the improved algorithm is proved through the comparison with ToPDE experiment。(2)Research of optimal experimental design algorithms.Three main evolutionary algorithms based on surrogate model are discussed and the widely used methods of constructing surrogate model are analyzed.To solve problem of the high computation complexity of GP,this paper introduces the neural process and its variant ANP which have the lower computational complexity and promising accuracy.Furthermore,ANP-EDA algorithm is proposed to solve the expensive black box optimization problem which replace the Gaussian process with ANP.Finally the results of comparative experiment prove that the ANP-EDA algorithm improves the quality of the optimal solution based on the basis of reducing the computational complexity of the surrogate model.(3)Development and implementation of experimental design analysis software.Aiming at the problem that most of the existing experimental design software are developed by foreign companies while users have difficulties to operate these software in general.This paper developed a experimental design analysis software in Chinese with friendly interface and convenient operations.Besides,this paper introduced the overall structure of the software and showed the implementation process of each module.The algorithm proposed in this paper and the previous classical algorithms,benchmark test problems and real application problems are integrated into the experimental design analysis software,which is helpful to promote the transformation of experimental design methods from traditional methods to heuristic methods,so as to solve various complex experimental design problems and provide support for related application research.