Using genetic algorithms to select ground motions for conducting HC-IDA

Abstract

This study applies genetic algorithms to select a set of generic ground motions for conducting hazard-consistent incremental dynamic analysis (HC-IDA). HC-IDA is a recently developed analysis procedure that overcomes the primary drawback of traditional incremental dynamic analysis (IDA) by enabling the computation of a hazard-consistent collapse fragility curve. HC-IDA has been shown in previous studies to produce results comparable to multiple stripe analysis (MSA), while not requiring any selecting site or structure specific ground motion selection. The set of ground motions required to conduct HC-IDA should ideally cover the broad range of response spectral shapes and durations expected at the site of interest, and they should be uniformly distributed to minimise the uncertainty in predicting ground-motion collapse intensity. This study employs genetic algorithms, which imitates the process of natural selection, to select an optimal set of ground motions. The selection procedure starts with an initial guess of the optimal record set, obtained via Latin hypercube sampling. The fitness of the selected record set, which is quantified using the Kolmogorov-Smirnov test, is then optimized over successive generations by crossover and mutation operations. Using this procedure, a set of 100 ground motions covering the range of ground motion response spectral shapes and durations anticipated at Wellington is selected. The wide range of ground motion response spectral shapes and durations expected in Wellington ensures that this set is also usable for a wide range of other sites in New Zealand. The selected record set is demonstrated to perform significantly better than other commonly used record sets such as the FEMA P695 far field set.