電腦模擬系統下離線最佳化架構中後設模型和搜尋演算法之應用

Abstract

本研究使用離線最佳化的架構去求解模擬模型參數最佳解，以利之後代入模擬模型的結果可以與物理實驗值相近。此架構包含了後設模型和搜尋演算法的建立。首先，利用多項式迴歸模型、高斯過程迴歸模型和類神經網絡模型來當作此研究的後設模型。第二，在已經建立完成的後設模型，使用梯度下降法和鄰近搜尋法找尋模擬模型參數的最佳解。最後，將使用實際的個案例子去解析離線最架化架構的好處。

This study proposes the framework to use the offline optimization solving for simulation parameters so that the simualated results fit the results of physical experiments. The framework includes the metamodeling and searching algorithm. The first task of the study uses the polynomial regression models, gaussian process regression models, and neural network models in metamodeling. The second task is to bulid the optimize model to search for the simulation parameters by either the gradient descent search or the neighbor search method. Finally, the actual case data is used to test the proposed framework to analyze the benefits of the proposed offline optimization method.