Importance-weighted actor-learner framework for reinforcement learning vibration control of three coupled flexible beams

Abstract

Three-coupled Flexible Beam (TCFB) system is a kind of vibration structure characterized by low damping, close space modes, and coupled vibrations. To address these challenges, the control scheme of reinforcement learning (RL) algorithm based on Importance-weights Actor-Learner Architecture (IMPALA) is applied. Using wavelet transform and Peafowl Optimization Algorithm (POA) to identify the system model matrices, an accurate system model is obtained, used to train RL algorithm. After training, the effectiveness of the IMPALA controller is verified by experiments. Compared with the traditional large gain proportional-derivative (PD) controllers, the experimental results show that the IMPALA controller can enhance vibration suppression, especially in rapidly attenuating small amplitude vibration. The IMPALA controller shows excellent performance and proves robustness.