Intelligent control of structural vibrations based on deep reinforcement learning

Abstract

This paper explores the application of Deep Reinforcement Learning (DRL) in structural vibration control, aiming to achieve effective control of the dynamic response of building structures during natural disasters such as earthquakes. A DRL-based control strategy is proposed, and dynamic interaction between the OpenSees environment and the deep reinforcement learning environment is realized. By adjusting the parameters in the reward function, the control preference of the DRL algorithm for different metrics can be effectively modified. Additionally, an intelligent structural vibration control platform based on DRL has been developed to simplify the design process of DRL algorithms. Case studies conducted on the platform demonstrate that DRL can effectively suppress structural responses in both single-layer and multi-layer complex structures. Meanwhile, comparisons with PID and LQR algorithms that are based on linear analysis design, reveal the stability advantages of DRL in handling structural dynamic responses characterized by high nonlinearity, time delay, and large actuator output intervals.