Model Learning for Probabilistic Simulation on Rare Events and Scenarios
Abstract
This project established a new methodology for probabilistic inference and prediction of rare and special events and/or scenarios based on the simulation models and the observed data that rarely contain rate events, applied it to a rainfall flood risk analysis of Chikugo river, Japan, and showed that it can generate various rainfall scenario that causes a flood. Rainfall pattern that causes a flood is generated by Replica Exchange Monte Carlo algorithm, and covariant shift phenomenon was corrected by placing more weight on the flood region. This work gives a general framework to cope with the problem of handling a complex and/or large scale system where a complete set of possible events and scenarios is hardly obtained.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 06, 2015
- Accession Number
- ADA616937
Entities
People
- Takashi Washio
Organizations
- Osaka University