Selection of Ground Motions for Hazard‐Consistent Incremental Dynamic Analysis Using Genetic Algorithms

Hazard‐consistent incremental dynamic analysis (HC‐IDA) provides a computationally efficient alternative to multiple stripe analysis (MSA) by using multivariate regression to account for the effects of spectral shape and duration on structural collapse capacity. While HC‐IDA can be used with generic record sets, the statistical efficiency and hazard alignment of the resulting collapse fragility depend on how well the selected records sample the secondary intensity measure (IM) space. This paper develops a standardized, region‐specific ground‐motion record set optimized for Wellington, New Zealand, designed to improve the precision and hazard consistency of HC‐IDA without modifying its underlying probabilistic framework. Using a genetic algorithm (GA), a suite of 100 ground motions is selected from a database of 2322 candidate records to match target joint distributions of S a Ratio and D s 5−75 derived from site‐specific generalized conditional intensity measures (GCIM) analyses. The GA optimizes the set's “fitness” by ensuring uniformity and orthogonality across the target IM space, providing a robust basis for collapse regression. The proposed set is benchmarked against the federal emergency management agency (FEMA) P‐695, the pacific earthquake engineering research Center (PEER), and a previously developed HC‐IDA suites through IDA of 4‐storey and 12‐storey steel moment‐resisting frame buildings. Relative to the generic FEMA and PEER sets, the proposed set captures the damaging spectral shapes and long durations expected at the site, yielding hazard‐consistent collapse estimates, lower sampling error in estimated median collapse capacity, and a higher proportion of records within the low‐standard‐error region of the IM space. Relative to the previously developed set, which incorporates long‐duration records, the proposed set yields comparable median collapse intensity for the 4‐storey model but a higher median for the 12‐storey model, with comparable sampling precision, while achieving substantially lower record‐to‐record and total dispersion and a more concentrated, hazard‐aligned collapse‐risk deaggregation. These results indicate that GA‐based optimization to site‐specific conditional hazard enhances the statistical efficiency and hazard consistency of HC‐IDA‐based risk estimates while fully preserving the established HC‐IDA methodology.