Predicting short-period spectral ordinates of hybrid ground shaking prediction tools: a comparative benchmark

Abstract

In this paper, a comparative perspective is provided on the efficiency and drawback of hybrid techniques to produce broad-band (BB) seismic time-histories, based on physics-based numerical simulations. The recently developed ANN2BB technique (Paolucci et al., 2018) is put under focus: it exploits Artificial Neural Networks (ANN) to predict short-period response spectra, feeding the algorithm with the low frequency outcome numerical simulations and producing hybrid broad-band (0-30 Hz) timehistories. The robustness of the methodology is argued by inputting correlated (from observations) and uncorrelated long-period spectral ordinates into the ANN predictive tool to draw an uncertainty map of possible predictions and test its sensitivity in relationship with the a priori accuracy numerical simulations.