Adaptive Sampling for Accurate Surrogate Modeling of Level 3 PRA Code WinMACCS

Kurokawa, Ryogo

Permalink

https://hdl.handle.net/2142/121812 Copy

Description

Title

Adaptive Sampling for Accurate Surrogate Modeling of Level 3 PRA Code WinMACCS

Author(s)

Kurokawa, Ryogo

Issue Date

2023

Keyword(s)

• Surrogate model
• Machine learning
• Bayesian optimization
• WinMACCS

Abstract

Expensive simulations in complex nuclear engineering typically incur significant time costs. Surrogate models to approximate the expensive simulation models have gained attention as a means of reducing the computational burden. However, the accuracy of surrogate models highly depends on their sample points.Adaptive sampling is a sampling method that uses information from existing surrogate models and input parameter space. This approach preferentially samples from regions of interest where the accuracy of the surrogate model can be enhanced. Compared to existing design-of-experiments methods, adaptive sampling can yield more accurate surrogate models using fewer sample points. In this presentation, I introduce an adaptive sampling method that efficiently constructs surrogate models through Bayesian optimization and an ensemble of multiple machine learning algorithms. The method is applied to the Level 3 PRA code WinMACCS and demonstrates its applicability to models with high-dimensional input parameter spaces, such as WinMACCS. And the proposed method is compared with existing design-of- experiments techniques and other adaptive sampling methods, demonstrating its superiority.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121812

Owning Collections