Sequential Optimization Method Based On Kriging Surrogate Model And Its Application

Optimization of complex processes based on chemical process simulation models often requires a longer time to optimize, Considering that it is inefficient, this paper aims at constructing Kriging surrogate model for the process simulation model, how to improve the accuracy of Kriging surrogate model, and researching sequential optimization strategy based on the Kriging surrogate model. Under the premise of the model accuracy is guaranteed the calculation of the agent model is much smaller than the exact model. Using surrogate model can greatly reduce the amount of computation for the optimization process. Then it can improve the efficiency of engineering optimization. Following are the main work and innovation of this paper:First of all, this article briefly described the basic components of Kriging surrogate model, advantages and recent academic research on Kriging Model. The mechanism of Kriging surrogate model, its regression model, its relevant models had been elaborated. Then this paper introduced the sampling methods of Kriging surrogate model, its standards of accuracy test for Kriging surrogate model.Secondly, this paper introduced a method for improving the accuracy of Kriging model-Sequential Optimization Method, and described the basic process of Sequential Optimization Method. Then, considering the original Sequential Optimization Method——EGO’s (efficient global optimization) limitations and advantages of genetic algorithms, this paper constructed a new operational processes of Sequential Optimization Method based on genetic algorithm. It used a genetic algorithm to search for model’s updating interpolation points based on some certain criteria of plusing points; Based on the thought of weighted law for El (Expected Improvement), this paper proposed a new rule for plusing points-maximizing the value of DH. Then it improved modeling accuracy of kriging model by using genetic algorithms to find the model’s iterative interpolation pointsIn addition, this paper described the main principles and basic processes of NSGA-II algorithm, and successfully applied to the optimization of the operation of Kriging surrogate model. The method of maximizing the value of DH combined the prediction error and the mean square error of Kriging model. This paper chose the value of DHs to replace the evaluation function’s value of NSGA-II algorithm, and constructed a new multi-objective sequential optimization method--MODH. MODH algorithm used the genetic manipulation and Non-dominated Sorting operation of NSGA-II algorithm to search for the Pareto Solutions of Kriging model. It used this Pareto Solutions as iterative interpolation points to update the original Kriging model, which contributed to improving the accuracy of Kriging model. Then this paper proved the feasibility of above multi-objective sequential optimization method by using a test function of POL.Finally, this paper proposed a new algorithm called K-N algorithm, which combined Kriging model and NSGA-II algorithm. K-N algorithm used the Pareto Solutions of Kriging model as initial population of a new round of NSGA-II algorithm, which can guide the algorithm to research for solutions in the areas of the real Pareto Solutions. K-N algorithm combined the Kriging model’s advantage of high efficiency and process simulation model’s advantage of high accuracy. Then it successfully applied to the Operation optimization of PX oxidation process...