Multi-step Prediction Of Time Histories Of Near-fault Ground Motions And Seismic Responses Of Building Structure

In recent years, the unique engineering characteristics and severe destruction action of near-fault ground motions have attracted intensive attention of scholars in earthquake engineering. For seismic controling of engineering structure's active and semi-active system in near-fault regions, realizing short-time prediction of ground motion time series and precisely predicting the structural seismic responses is the key issues which should be solved urgently. Ultilizing the emerging method in nonlinear time series prediction, one can perform the multi-step forecasting of ground motions and their seismic responses, which promptly conducts real-time control to minimize the structural seismic responses and eventually reduce the loss of lives and properties. This thesis proposes a new multi-step prediction method based on Empirical Mode Decomposition (EMD)-Extreme Learning Machine (ELM), and several near-fault ground motions time history, seismic responses of SDOF systems and seismic responses of elasto-plastic high-rise frame stucture are selected to predict respectively. The detailed works are presented as follows:Firstly, considering the strong nonlinear and nonstationary property of earthquake ground motions and seismic responses, the nonstationary signal is decomposed to several intrinsic mode functions which possess lower degree of complexity and irregularity by Empirical Mode Decomposition. Note that there are the advantages of strong self-learning and nonlinear mapping capability for Extreme Learning Machine, then ELM method is utilized to predict each intrinsic mode function, and each predicted values are superimposed to obtain original time series prediction. Therefore, this study proposes a new EMD-ELM method that is applicable for multi-step prediction for near-fault earthquake ground motions and seismic responses. Verification of typical time series signals demonstrates that multi-step prediction method is suitable for nonlinear and nonstationary time series prediction.Secondly, the performance of two multi-step prediction methods with Empirical Mode Decomposition (EMD) and without EMD is compared, and the results illustrate that, the forecasting accuracy of combined mehod using EMD-ELM is obviously superior to that mehod just using ELM. Then multi-step prediction for acceleration time histories of near-fault ground motions is carried out. Results show the susessful prediction for one-step and five-step, ten-step prediction also achieves high accuracy. Morever, this method presents a good property of robustness.Finally, the multi-step prediction for seismic response of a single degree of freedom system and elasto-plastic high-rise frame structure subject to near-fault ground motions is conducted. The prediction of the acceleration, velocity and displacement seismic responses of SDOF system with different representative periods present preferable performance. In addition, the acceleration and displacement responses of elasto-plastic high-rise frame structure subject to near-fault ground motions are predicted. The forecasting results illustrate that the proposed EMD-ELM method is effective and feasible for one-step and multi-step prediction. Short-term prediction of seismic response of building structures can provide fairly accurate dynamic response time histories for active control system, thus facilitate the real-time online vibration-reduction control of engineering structures.