Dynamic reliability analysis of high-ACCWDs under stochastic near-fault ground motions based on a stacked LSTM–GPDEM

Abstract

High-asphalt–concrete core wall dams (ACCWDs) may be constructed in near-fault regions, where their seismic safety is critically influenced by the complex characteristics of near-fault ground motions (NFGMs). Existing studies predominantly focus on deterministic NFGM scenarios, neglecting the critical uncertainties arising from varying conditions and the inherent stochastic nature of near-fault seismic environments. To bridge this gap, this study introduces a comprehensive nonstationary NFGM model based on a time-varying process with a filtered white noise and wavelet multiscale decomposition incorporation approach. Furthermore, an efficient framework is developed for evaluating the seismic behaviour and time-varying reliability of high-ACCWDs under NFGMs, explicitly accounting for the coupled effects of two-phase porous media. This framework models the data-driven structural seismic response and dependability by integrating the generalized probability density evolution method (GPDEM) with a stacked long short-term memory (LSTM) network. The results show that NFGMs with velocity pulses can cause significant amplification of acceleration, stress, and displacement responses. Moreover, the stacked LSTM and GPDEM, combined with the first exceedance probability principle, provide a safety assessment approach that effectively describes the time-varying structural reliability under stochastic NFGMs. In addition, compared with that of the NFGMs without a velocity pulse, the time-varying dynamic reliability of the dam under the NFGMs with a velocity pulse is highly risky. Given that dams are important lifeline projects, the impact of stochastic NFGMs should be reasonably considered during the seismic design and seismic safety assessment stage.