A Generalized Ground Motion Model for selection of consistent Mainshock-Aftershock Ground Motion Sequences using Deep Neural Networks

Utilization of “consistent” mainshock (MS) - aftershock (AS) ground motions is desirable in practical engineering applications. Such consistency
in selecting MSAS sequences requires proper consideration of the correlations between and within the intensity measures of MS and AS ground
motions. This study proposes a generalized ground motion model (GGMM) to estimate consistent 30×1 vectors of intensity measures for
mainshocks (denoted as IMMS) and aftershocks (denoted as IMAS) using a framework of successive long-short-term-memory (LSTM) recurrent
neural network (RNN). The vectors of IMMS and IMAS consists of geometric means of significant duration (𝐷5−95,𝑔𝑒𝑜𝑚), Arias intensity
(𝐼𝑎,𝑔𝑒𝑜𝑚), cumulative absolute velocity (𝐶𝐴𝑉𝑔𝑒𝑜𝑚), peak ground velocity (𝑃𝐺𝑉𝑔𝑒𝑜𝑚), peak ground acceleration (𝑃𝐺𝐴𝑔𝑒𝑜𝑚) and RotD50 spectral
acceleration (𝑆𝑎(𝑇)) at 25 periods for both MS and AS ground motions. The proposed RNN-based framework is trained on a carefully selected
set of ~700 crustal and subduction recorded MSAS sequences. The inputs to the framework include a 5×1 vector of source and site parameters
for mainshock and aftershock recordings. The residuals of the trained LSTM-based RNNs are further used to develop empirical covariance
structures for IMMS and IMAS.