A Study of Data-Driven Seismic Response Analysis Based on the Identification of Temporal Evolutionary Law in Dynamic Systems

This study examines the application of dynamic mode decomposition (DMD), one of a data-driven time-series analysis, to seismic response analysis. In this study, a DMD data-driven seismic response analysis model was formulated based on the assumption to observe the seismic bedrock in earthquake engineering (SBEE) and surface layer waveforms. Additionally, we modeled the transfer function between SBEE and surface layer as the equation of motion for a single degree of freedom model (SDOF), which provides a physical meaning to the data-driven model. Then, the possibility of extracting the physical meaning of the ground is discussed by considering the relationship between the DMD model and SDOF through numerical experiments. First, we prepared five cases of soil layer configurations and 25 artificial input seismic waveforms, and seismic response analysis was performed using the one-dimensional equivalent linearized overlapping reflection theory SHAKE. Then, we examine whether the DMD learning model using SHAKE results is described the soil nonlinear response feature as the equivalent linearization parameters (i.e., convergence stiffness and convergence damping constant). Finally, considering the correspondence between the convergence stiffness and convergence damping constants obtained from the SHAKE calculation and DMD model, the possibility of developing a data-driven model that integrates engineering knowledge was discussed.

Keywords: mode decomposition, time series analysis, seismic response analysis, dynamic system