Neural-network control of building structures by a force-matching trainingscheme

A method to generate an efficient control law for a neural-network controll er is presented to reduce the dynamic response of buildings exposed to eart hquake-induced ground excitations. The proposed training scheme for the neu ral-network controller does not rely on the emulation of the structure to b e controlled. The approach used for this work is based on a force-matching procedure, and it directly utilizes the dynamic data characterizing the str ucture response to generate an efficient training signal. The proposed cont roller has a feedback structure, utilizing a limited set of response quanti ties. A shear building actuated at its top by a tuned-mass damper is utiliz ed to demonstrate the effectiveness of the controller. For training purpose s, an ensemble of synthetically generated ground-motion time histories, wit h appropriate site spectrum characteristics, have ken used. The performance of the trained controller is then evaluated for two different historic gro und-acceleration records that do not belong to the training set of time his tories. The numerical simulations show the control effectiveness of the pro posed scheme with modest control requirements. Copyright (C) 1999 John Wile y & Sons Ltd.