An Integrated Method for Simultaneous Calibration and Parameter Selection in Computer Models
| Autor: | Jun Yuan, Szu Hui Ng |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
021103 operations research
Computer science Bayesian probability 0211 other engineering and technologies Building energy 02 engineering and technology computer.software_genre 01 natural sciences Computer Science Applications Set (abstract data type) 010104 statistics & probability symbols.namesake Sequential analysis Modeling and Simulation symbols Calibration Normal approximation Data mining 0101 mathematics Gaussian process computer Selection (genetic algorithm) |
| Zdroj: | ACM Transactions on Modeling and Computer Simulation. 30:1-23 |
| ISSN: | 1558-1195 1049-3301 |
| DOI: | 10.1145/3364217 |
| Popis: | For many large and complex computer models, there usually exist a large number of unknown parameters. To improve the computer model's predictive performance for more reliable and confident decision making, two important issues have to be addressed. The first is to calibrate the computer model and the second is to select the most influential set of parameters. However, these two issues are often addressed independently, which may waste computational effort. In this article, a Gaussian process-based Bayesian method is first proposed to simultaneously calibrate and select parameters in stochastic computer models. The whole procedure can be conducted more efficient by sharing the data information between these two steps. To further ease the computational burden, an approximation approach based on a weighted normal approximation is proposed to evaluate the posteriors in the proposed Bayesian method. A sequential approximation procedure is further proposed to improve the approximation accuracy by allocating the sequential design points more appropriately. The efficiency and accuracy of the proposed approaches are compared in a building energy model and a pandemic influenza simulation model. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|