Neural Network–Based Semiactive Control for Buildings Located in Indian Seismic Zones

Abstract

Adaptive intelligent data-driven controllers play a pivotal role in structural control systems due to their capacity to operate effectively under seismic forces without relying on detailed mathematical models. Their ability to quickly adapt and respond to changing conditions makes them more suitable for handling unpredictable structural scenarios compared to model-based controllers. In the present study, an adaptive intelligent control strategy is developed using neural network for a semiactive control system installed in a 10-story shear building frame located in Surat City, Gujarat, India. Response spectrum compatible time histories of ground motions are generated using SeismoArtif software for seismic zone III, considering various soil types according to IS 1893:2016. A magnetorheological (MR) damper is installed on each of first, second, and third floors. The linear quadratic regulator (LQR) algorithm, combined with the clipped optimal algorithm, is utilized to generate a training data set. Earthquake excitation, displacement, and velocity of all floors are provided as input to the neural network. Output from the neural network represents the desired control force generated by the LQR control algorithm. Findings of the study indicate that the proposed neural network-based semiactive control algorithm significantly reduces seismic response parameters, achieving results comparable to those of the LQR control algorithm. Furthermore, the proposed approach eliminates the need to define optimal weighting matrices by employing a neural network, thereby enhancing the overall performance of semiactive control systems.