New methodology for wave-based probabilistic seismic hazard analysis

Fundamental study on probabilistic generative modeling of earthquake ground motion time histories using generative adversarial networks

“…The first 10 000 samples are discarded as burn-in, resulting in 100 000 y n samples in total. The distribution of observed records p y ð Þ and that of the generated ground motions q y ð Þ are calculated using Equations ( 10) and (11), and the KL divergence between the distributions is obtained by Equation (13). The change in the value of KL divergence for each epoch and z dim is shown in Figure 5.…”

“…The former approach is to select a suite of strong motion records from a compiled database of past earthquakes that are compatible with the conditions and response spectrum shape of the assumed earthquake scenarios. [7][8][9][10][11][12] This approach focuses only on a single or a few intensity measures, and the selected observed records do not always fully represent the variety of temporal or frequency characteristics of possible ground motions at a site.…”

“…There are two types of approaches for preparing a dataset of input ground motions to be used for response history analysis; one is to utilize observed strong motion records, and the other is to compute or synthesize ground motion data artificially using simulation. The former approach is to select a suite of strong motion records from a compiled database of past earthquakes that are compatible with the conditions and response spectrum shape of the assumed earthquake scenarios 7–12 . This approach focuses only on a single or a few intensity measures, and the selected observed records do not always fully represent the variety of temporal or frequency characteristics of possible ground motions at a site.…”