Multi-feature driven intelligent learning and prediction of deformation state of super-tall buildings
To predict the future state of structures using monitoring data,an intelligent learning and prediction method for estimating the deformation state of super-tall buildings driven by multiple features is proposed.Through self-adaptive signal decomposition,multi-dimensional analysis of response characteristics and sub-signal correlation analysis,structural response data reconstruction and combination are realized,and multi-channel learning and multi-step prediction are performed based on long short-term memory network.Using the measured data of the horizontal displacement response at the top of the Shanghai Tower under the influence of Typhoon Muifa,it is revealed that the displacement response data have superimposed characteristics of ultra-low-frequency quasi-static deformation,fundamental modal control vibration response and non-stationary characteristics in the time domain.Through reconstruction of the response data,three sub-signal combinations with different time scales,vibration amplitudes and stationary characteristics are formed.The displacement response of the next 1-60 time steps are successfully predicted based on 300 time steps of monitoring data via three independent LSTM models.The results show that under the condition of controlling the future prediction time step(in 60 time steps),the data-driven learning prediction model proposed in this study can fully learn and predict the known displacement response data characteristics and physical state with excellent accuracy,the normalization error can be controlled within 10％.The proposed method can predict the deformation state of super-tall buildings accurately in real time.
super-tall buildingstructural health monitoringdeformation stateLSTM modeltime series data
国家科技期刊平台
NSTL
万方数据
