Online prediction and intelligent control for structural vibration based on neural networks

Various semi-active control devices have been widely investigated to reduce the responses of structures under earthquake, including variable orifice dampers, and controllable fluid dampers (e.g. MR dampers). Due to the inherent nonlinear nature and uncertainty of the semi-active control system based on magnetorheological (MR) fluid damper, an improved back-propagation (BP) algorithm is proposed, which divides the actual structure model into a mechanism model part and a time-varying error model part. By means of off-line training and on-line modifying, the intelligent control system based on the improved BP algorithm are acquired. It can be used to predict the damping force of MR damper and eliminate the influence of time delay. Theoretical analysis and numerical simulations show that the intelligent control system can efficiently reduce the structure responses induced by earthquake, has good performance in adaptively tracking target and displays resistance to disturbances.