Real-time prediction of near-future seismic excitation adapting AR model to preceding information

Abstract

This paper examines a real-time prediction method, aimed at application in active structural control. The examined method applies preceding seismic excitation information at a certain moment to a time-variant AutoRegressive (AR) model and uses it to predict near-future excitation information. The performances of this method and appropriate identification parameters are examined by numerical experiments. In fact, the results of these experiments show that a time-variant AR model with appropriate identification parameters has little change in low-frequency components despite change in AR coefficients. The performance of a fixed-coefficient AR model is thus examined. The results show that even a fixed-coefficient AR model can sufficiently predict 0.05-s-future excitation information.