An intelligent isolation system based-on long short-term memory module model for ground motion characteristics prediction

An intelligent control system based on a semi-active variable stiffness isolation system and the Long Short-Term Memory (LSTM) is proposed in this study. In the field of semi-active control, the design of control laws relies on the measurement and feedback of the structural response when an earthquake occurs. The optimal control parameters should be determined in real time to achieve the performance-oriented control target by identifying the ground motion characteristics, which have a great influence on the control performance in the early stage. To deploy the optimal control parameters, a LSTM-based module for predicting ground motion characteristics is first proposed. In addition, a sensitivity analysis using the world-wide ground motion database is also conducted, and the representative cases are selected to optimize the fuzzy inference surface for the Leverage-type Stiffness Controllable Isolation System (LSCIS). Numerical simulation indicates, by using the proposed GA-LSTM control system, prominent effect of suppressing the isolation displacement and superstructure acceleration for near-fault and far-field ground motions can be achieved respectively. Series of shaking table tests was also conducted to verify the actual control performance of the proposed system. Both numerical and experimental results have demonstrated that the proposed intelligent control system has high efficiency and reliability to protect the structures from earthquakes of all types of ground motion characteristics.