Research on improved MLP model for seismic fragility analysis of rockfill dam by embedding spatial features of time-frequency
In view of the fact that the present ground motion indexes such as peak acceleration and peak velocity used in seismic fragility analysis do not fully reflect the complex spatial features of time-frequency of ground mo-tion,and the current seismic demand model based on the assumption of linear function in logarithmic space is diffi-cult to reveal the complex nonlinear relationship between ground motion index and seismic response,an improved multilayer perceptron(MLP)model for seismic fragility of earth-rockfill dams by embedding spatial features of time-frequency is proposed.The model makes use of the advantage that capsule network(CapsNet)can fully capture and characterize the spatial location distribution of target features,extract the deep features reflecting the spatial features of time-frequency distribution of ground motion from the Wavelet time-frequency diagram of ground mo-tion,and merge them with traditional features in the way of feature splicing to construct the ground motion fusion index.Furthermore,the super parameters such as the number of neurons and learning rate of MLP are optimized by Tree structured Parzen Estimator(TPE),and the seismic demand model of earth-rockfill dam based on TPE-MLP is proposed to reflect the complex nonlinear relationship between ground motion fusion index and seismic response,so as to realize the reliable analysis of seismic fragility of earth-rockfill dam.The case study shows that,compared with the traditional ground motion index,the MAE of the earth-rockfill dam seismic demand model based on the ground motion spatial features of time-frequency fusion index is reduced by 40.5％,indicating the reliability and superiority of the proposed model.
earth-rockfill damseismic fragilityspatial features of time-frequency of ground motionseismic demand modelCapsNetMLPwavelet transform
万方数据
维普
