Development of a Regional Seismic Response Model Using Building Clustering

Abstract

This paper presents a framework for regional seismic response estimations that uses machine learning driven building clustering and the relatively novel concept of indicator buildings. A robust database of buildings is required to provide detailed structural and site information to develop typologically similar building clusters and allow for an informed selection of the indicator buildings that are used to estimate the seismic response of all buildings in a cluster. Here, the framework is applied to a database of buildings consisting of over 400 buildings in the central business district of Wellington, New Zealand. First, key structural and site parameters are extracted from the building database and building clusters are generated using the k-prototype clustering methodology. Next, indicator buildings within each cluster are selected and modelled using ETABS. To provide more accurate estimations of drift across all buildings in the cluster, supplementary models were generated by modifying the stiffness of the base indicator building models to represent the structural period range across all buildings in each cluster. A case study was undertaken using the 2016 Kaikoura earthquake, and the response of the models are utilized to propose a linear regression model estimating seismic response of all buildings in the database.