Research On Multi-input Parameter Surrogate-based Optimization Algorithm Based On Kriging Model

In the context of increasingly fierce market globalization and the rapid development of customized product consumption patterns,product design development and quality verification are increasingly relying on computer experimental methods.The high-precision simulation model based on the development of numerical calculation and computer modeling technology is widely used in product design and optimization,and the quality and reliability of traditional physical experimental design have been improved effectively.But for product design and system optimization problems involving multiple input parameters,high-precision simulation models face problems such as long optimization time and low optimization efficiency.Therefore,the use of surrogate-based model modeling instead of high-precision simulation modeling has gradually become a common practice in experimental design.Based on the Kriging model,this paper comprehensively uses the weighted expectation filling criterion and the partial least square transformation method to research the multi-input parameter surrogatebased optimization algorithm based on the Kriging model.The specific research content is as follows:（1）Surrogate-based optimization algorithm base on weighted expectation filling criterion and Kriging model.The traditional surrogate-based optimization algorithm uses the EI criterion to guide the addition of new experimental points to update the Kriging model,without considering the local optimal problem caused by the greedy characteristic of the classic EI criterion.This paper uses the weighted function with distance characteristics to adjust the new experimental points obtained by the classic EI criterion,so that it can balance the global exploration and local exploitation capabilities while making the new experimental points have distance characteristics.The experimental results show the proposed algorithm is efficient,and the proposed algorithm has good robustness and accuracy.（2）Multi-parameter surrogate-based optimization algorithm base on partial least squares and Kriging model.Aiming at the problems of low modeling efficiency and long training time when traditional surrogate-based optimization algorithms deal with high-dimensional optimization problems,a surrogate-based optimization algorithm that applies the partial least squares method to the construction of Kriging kernel function is proposed.The algorithm builds a partial least squares kernel function based on the dimensionality reduction characteristics of the partial least squares method to reduce the amount of hyper-parameter calculations,and uses CEI criteria and particle swarm optimization algorithm to search for new experimental points.The experimental results show that the algorithm has a better improvement in modeling efficiency and solution accuracy.（3）Multi-parameter Surrogate-based Optimization Algorithm based on PLS Weighted Expectation Filling Criterion and Kriging Model.In order to explore the effectiveness of the partial least squares method in dealing with multi-parameter problems,this chapter replaces the CEI criterion with the weighted expectation filling criterion,and then constructs the partial least squares weighted expectation filling criterion to guide the addition of points to update surrogate-based model.The experimental results show that the Kriging multi-parameter surrogate-based optimization algorithm based on PLS weighted expectation filling criterion has a better improvement in modeling efficiency,and it also shows that the partial least squares method to deal with multi-parameter problems has good performance under the action of different filling criteria.Finally,it summarizes the research content of this article,and points out the difficulties in the future research of the agent optimization algorithm and the direction that can be extended in the future.