Application of Pool-Based Active Learning Methodology in Ground Motion Selection

Session: Back to the Future: Innovative New Research with Legacy Seismic Data [Poster]
Type: Poster
Date: 4/28/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Performing response history analysis is a key step in deriving seismic fragility curves for seismic risk assessment. Using a limited number of ground motions for performing response history analysis is unavoidable because it is computationally demanding to perform a great deal of response history analyses, especially in case of sophisticated numerical models. Therefore, we develop a step by step procedure based on the concept of active learning method to reduce the need for unnecessary response history analyses when deriving fragility curves. Specifically, we use a pool-based query by committee active learning methodology with the results of multiple strip analyses to efficiently and accurately estimate the parameters of fragility curves. We implement the structural responses of an 8-story steel moment-resisting frame building subjected to a set of ground motions selected based on the generalized conditional intensity measure framework as the database. The statistical results using the implemented database explain that using active learning can significantly reduce the number of required response history analyses while improves the accuracy of the structural responses. In addition, active learning provides a better performance in comparison to the passive learning while using the same number of ground motions for performing response history analyses.

Presenting Author: Jalal Kiani

Authors