Improved Hybrid Model for Structural Vibration Control Using MOGA, LSTM, and Digital Twin Technology

Abstract

With the acceleration of urbanization, the vibration control of high-rise buildings is becoming increasingly prominent. To address the limitations of traditional methods on convergence and robustness against noise, this study proposes a hybrid intelligent control model that integrates a genetic algorithm with a long short-term memory neural network, while incorporating an improved sparrow search algorithm for parameter refinement and a digital twin framework for bidirectional data-driven control. Numerical experiments using seismic records from the PEER Ground Motion Database demonstrated that the proposed model achieved approximately 20% faster convergence compared with particle swarm optimization, whale optimization, and artificial fish swarm algorithms. The root mean square error of vibration prediction was reduced to 0.0180, the identified stiffness error of each floor remained below 1%, and the total control energy was reduced by about 15%. These results confirm the advantages of the proposed method on prediction accuracy, control efficiency, and stability, and highlight its potential applications in seismic design of high-rise structures and smart construction practices.