Deep reinforcement learning based performance optimization of hybrid system for base-isolated structure and shape memory alloy-inerter

Abstract

Seismicity in the regions of high seismic intensity poses substantial challenges to engineering structures, particularly to traditional base isolation systems (BIS). BIS will exhibit the excessive displacement of isolation layer during infrequent, high-magnitude seismic events. The system that integrates shape memory alloy inerter (SMAI) with BIS proposed by the authors of this article is one of the advanced control strategies to address the above-mentioned limitation of BIS. In order to well bolster the seismic resilience of the BIS+SMAI system, the present paper proposes an efficient optimization methodology of the BIS+SMAI system using different DRL algorithms, including deep Q-networks (DQN), deep deterministic policy gradient (DDPG), and soft actor-critic (SAC), which are adept at navigating complex, high-dimensional problems. Employing different DRL algorithms, a comprehensive evaluation on the optimization performance of the BIS+SMAI system has been made under near-field non-pulse, near-field pulse, and far-field seismic waves, respectively. The results demonstrate that in the optimization, DRL surpasses the particle swarm optimization (PSO) in efficiency, specifically with DDPG having swift convergence and SAC having stability, but with the discretization of DQN detracting from its optimization precise. In the optimization damping performance, the DRL-optimized BIS+SMAI system can significantly reduce displacement of isolation layer, and surpasses the effectiveness of traditional BIS and other hybrid systems [BIS+SMA and BIS+the tuned inerter damper (TID)]. In the meanwhile, the optimized BIS+SMAI system demonstrates consistent performance across a range of seismic events, including near-field non-pulse, near-field pulse, and far-field seismic waves. Therefore, the BIS+SMAI through DDPG and SAC optimization presents a promising approach for ensuring greater safety and stability of structures against seismic threats.