A Reinforcement Learning–Controlled Novel Mode‐Switching Hybrid Mass Damper on a Real Tower

This work proposes a novel passive, semiactive and active mode‐switching hybrid mass damper (MSHMD) solution for the control of an actual high‐rise tower. For the mode‐switching control of the considered building system, a reinforcement learning algorithm, Deep Q‐Network (DQN), was implemented. The performance of the system acting solely as a passive‐tuned mass damper (PTMD/TMD), a semiactive‐tuned mass damper (SATMD) or an active‐tuned mass damper (ATMD) was also studied independently. For the control of the ATMD and the SATMD, the well‐established algorithms, linear quadratic regulator (LQR) and clipped‐LQR were utilised, respectively. Simulations, under natural wind excitation, showed that the proposed MSHMD had a similar performance to the sole ATMD while achieving a drastic decrease in the feed energy demand. More specifically, the MSHMD achieved a top‐floor dynamic response reduced by up to 26.7% when compared to the uncontrolled case (vs. 30% dynamic reduction for the ATMD over the uncontrolled case). Notably, the MSHMD managed to decrease both the peak power consumption and the cumulative energy consumption of the active constituent by up to 41% when compared to the ATMD alone, with only a marginal performance compromise. As such, the proposed MSHMD is deemed extremely efficient in controlling the dynamic responses of the studied tower in an energy‐optimum fashion, paving the way towards visionary net‐zero‐energy active vibration mitigation appendices.