Research And Application Of Surrogate-based Optimization Theory Based On Kriging Model

Computer simulation technology is widely used in modern industrial engineering for it can replace physical experiment to obtain data.High precision simulation is accompanied by high time consumption.Surrogate-based optimization technology has been widely studied and applied because it can solve the computationally expensive and black-box engineering optimization problems efficiently with a small amount of simulation data.Based on the theories and methods of scholars at home and abroad,this paper further studies the related theories and applications of surrogate-based optimization based on Kriging model.For the purpose of saving the number of simulations and improving the optimization efficiency,the existing surrogate-based optimization algorithms will be improved and expanded from three aspects: constraint optimization,parallel optimization and multi-objective optimization.The detailed research contents are as follows:(1)An adaptive constrained surrogate-based optimization algorithm based on Kriging model using multiple criteria.The classical constrained surrogate-based optimization is inefficient for it uses the constrained expected improvement(CEI)which directly multiplies the expected improvement criterion(EI)and the probability of feasibility(Po F).To solve this problem,a new efficient constrained surrogate optimization algorithm is proposed.This algorithm uses the WB2 criterion and probability of feasibility to explore the global optimal and constraint boundary.It generates the Pareto sets by using the multi-objective optimization framework to balance different criteria,then selects a new sample to update the model from the Pareto set.Finally,the performance of the algorithm is verified by three different types of numerical and one engineering benchmarks.The results show that the proposed algorithm is more efficient in convergence and the solution is more precise and robust.(2)A parallel surrogate-based optimization algorithm based on Kriging model using an adaptive multi-phases strategy.In the case that the parallel simulation technology is more and more mature,efficient global optimization(EGO)adds only one sample point at each iteration of updating the surrogate model,which is inefficient.Aiming at this problem,a surrogate optimization algorithm that can add batch points is proposed.In the two phases of global exploration and local exploitation,the algorithm proposes two corresponding multi-points filling algorithms.And it designes an adaptive switching strategy for two phases.The results of benchmarks show that the algorithm can improve the optimization efficiency and accuracy for it can make full use of the advantages of different criteria in different phases.(3)A multi-objective surrogate-based optimization algorithm base on kriging model and physical programming.In order to make full use of the existing single-objective surrogate-based optimization algorithm,an algorithm which is capable of transforming multi-objective problems into single-objective is proposed.The algorithm first uses physical programming to convert multi-objective problems into single-objective problems,then uses the single-objective surrogate-based optimization algorithm to solve the converted single-objective problem.Finally,the overall performance of the algorithm is verified by some benchmarks.The applicability of the sub-optimization step and its impact on the overall efficiency of the algorithm are also explored.Finally,the research content of this paper is summarized and the future research directions of surrogate-based optimization are elaborated.