Experimental verification of the support vector regression based structural identification method by using shaking table test data

The aim of this Study is to use the observed data from a shaking table test to verily experimentally a support vector regression (SVR) based method for Structural identification. SVR is a promising data processing technique. It employs a novel epsilon-insensitive loss function and the 'Max-Margin' idea, thus Shows an excellent performance for function estimation. The SVR-based structural identification method has been developed in previous research, and numerical examples have shown that it is able to identify structural parameters accurately and robustly, even when the I/O data are corrupt by different kinds and intensity noises. However, because SVR is a novel technique and has seldom been used in the structural identification filed, ail experimental validation is necessary to check how this method applied Successfully in the reality. For this Purpose, a five-floor shear-building shaking table test has been conducted to verify experimentally the efficiency of the SVR-based approach. Two cases, input excitations to the shaking table of the Kobe (NS 1995) earthquake and a sine wave with constant amplitude, are investigated. Copyright (C) 2007 John Wiley & Sons, Ltd.