Research On The Construction And Optimization Algorithm For Kriging Model

Surrogate model is an effective method for solving engineering optimization problems. Because it can use the approximate model to fit the relationship between design variables and design target in the optimization process, and then replaces the complex and time-consuming high precision numerical simulation program and simulation software, so the optimization method based on surrogate model has attracted many researchers' attention. Compared with other models, the Kriging model has relatively high prediction accuracy, so it is the most widely used model in computer experiments. The optimization of relevant model parameters is a core problem in the process of constructing Kriging model. At the same time, the objective function of a large number of engineering optimization problems is usually not explicitly expressed, but got through simulation calculation. However, the simulation calculation requires expensive cost in many cases, so it is one of the urgent problem to be solved how to optimize the system global optimization through the surrogate model in this field. In this paper, we study the construction method of Kriging model and improve the EGO algorithm in the framework of efficient global optimization(EGO) algorithm. The main work is summarized as follows:(1) The ideal of alternating direction method is used to obtain the gradient of the objective function concerning the model parameters, then optimization algorithm of Kriging model parameters and the improved efficient global optimization algorithm are put forward based on an active set conjugate gradient algorithm. Numerical experiments show that the problem of premature convergence of efficient global optimization algorithm is improved by using the improved global optimization algorithm. Finally, the improved EGO algorithm is used to solve the optimization design of the analog integrated circuit. Taking the optimal design of the band-gap voltage reference as an example, the global optimization of the circuit system is carried out.(2) The corresponding improved projection simplex gradient method is put forward by updating the strategy of simplex to deal with line search failure problem, and then projection simplex gradient method is used to optimize the parameters of the model, namely the simplex gradient is employed to approximate the gradient of the objective function, which avoids effectively the problem that is difficult to solve the gradient of the objective function; meanwhile, the simplex projection gradient method is used to improve EGO algorithm. Numerical experiment results show that the improved EGO algorithm is effective.