Probabilistic parameter estimation and reliability assessment of nonlinear structures based on generative adversarial networks with Gaussian process regression

Probabilistic parameter estimation and reliability assessment of nonlinear structures with multiple sources of uncertainty presents a significant challenge in the field of civil engineering. In this study, a novel method that combines Deep Convolution Generative Adversarial Network (DCGAN) with Gaussian Process Regression (GPR), denoted as GPR-DCGAN, is developed for precise probabilistic parameter estimation and reliability assessment of nonlinear structures subjected to seismic excitations. To enhance computational efficiency, GPR-DCGAN introduces a Gaussian Process Regression surrogate model to establish the relationship between structural parameters and dynamic response. Sample sets for GPR-DCGAN training are constructed using structural acceleration response and the parameters to be estimated. The trained GPR-DCGAN is capable of providing estimates of posterior parameters samples based on input acceleration response. Furthermore, the estimated parameters can be applied to investigate the failure probability and effectively assess the structural reliability. To validate the accuracy and feasibility of the proposed method, both numerical simulations and experimental shake table tests are conducted on a scaled steel–concrete bridge tower structure subjected to seismic excitations. The results affirm the accuracy and effectiveness of the GPR-DCGAN method for probabilistic parameter estimation of nonlinear structures, demonstrating the robust performance of the method against significant noise interference.