SOFT ACTOR-CRITIC REINFORCEMENT LEARNING FOR ROBUST ACTIVE STRUCTURAL VIBRATION CONTROL
Active structural control systems are among the most effective technologies for suppressing vibrations in civil engineering structures subjected to dynamic loads such as earthquakes and wind. However, conventional active control methods often suffer from performance degradation due to time delays, measurement noise, and changing operational conditions. To overcome these limitations, data-driven control strategies based on reinforcement learning have emerged as promising alternatives. This study introduces advanced soft actor-critic (SAC)– based control approaches designed to enhance adaptability and robustness in complex real-world environments. Challenges in Traditional Active Control Systems Traditional controllers, such as linear quadratic Gaussian (LQG) control, rely on predefined system models and fixed parameters. In practical applications, uncertainties such as sensor noise, communication delays, and environmental variability can significantly reduce their effectiveness. Thes...