A generalized ground-motion model for consistent mainshock–aftershock intensity measures using successive recurrent neural networks

Abstract Several recent studies have investigated the risk posed to structures by earthquake sequences, proposing state-dependent fragility/vulnerability models for assets in damaged conditions. However, a critical component such efforts, i.e., ground-motion record selection, has received relatively minor consideration. Specifically, utilization of “consistent” mainshock (MS)–aftershock (AS) ground motions is desirable practical applications. Such consistency selecting MS–AS sequences requires proper consideration correlations between and within intensity measures MS AS motions. Most this domain utilize spectral accelerations as considered rely on empirical linear correlation developing, instance, selection approaches. This study proposes generalized model (GGMM) estimate consistent 30 × 1 vectors mainshocks (denoted IM ) aftershocks using framework successive long-short-term-memory (LSTM) recurrent neural network (RNN). The consist geometric means significant duration ( $$D_{5 - 95,geom}$$ D 5 - 95 , g e o m ), Arias $$I_{a,geom}$$ I a cumulative absolute velocity $$CAV_{geom}$$ C A V geom peak $$PGV_{geom}$$ P G acceleration $$PGA_{geom}$$ RotD50 $$S_{a} \left( T \right)$$ S T at 25 periods both proposed RNN-based GGMM trained carefully selected set ~ 700 crustal subduction recorded sequences. inputs include 5 vector source site parameters recordings. residuals LSTM-based RNN are further used develop covariance . finally illustrated select based multi-criteria objective function. motion then perform non-linear time-history analyses case-study two-spanned symmetric bridge structure. obtained engineering demand evaluated critically discussed.