Seismic control of adaptive variable stiffness intelligent structures using fuzzy control strategy combined with LSTM

Abstract

A novel semi-active control algorithm for adaptive variable stiffness intelligent structures that combines the fuzzy control strategy with Long Short-Term Memory (LSTM) is proposed in this study, with the aim of mitigating structural responses under earthquake excitations. The combined Fuzzy-LSTM strategy utilizes an LSTM neural network to predict the displacement and velocity responses of the structure at the next moment, and uses the predicted results as input for fuzzy control to obtain active control forces. Due to the difference between the displacement-velocity response and the fuzzy control domain, the genetic algorithm is used to optimize scaling factors of displacement and velocity responses and the control force, which is implemented by exploiting stiffness variability and structural inter-shift. The Fuzzy-LSTM strategy can compensate for the time delay effect on active control caused by stiffness variations, algorithmic computations and sensor measurements, and therefore achieves a better control effect through prediction. The semi-active independently variable stiffness (SAIVS) device is used to implement the presented control algorithm, which is a diamond device consisting of four linear springs and an electromagnetic actuator. The SAIVS device can achieve a continuous and smooth stiffness variation and meet the application requirements of the developed algorithm. Therefore, active control forces can be obtained through structural stiffness variations and interlayer displacements under seismic excitation, to achieve an approximate active control effect through a semi-active method. A 10-story building is presented as a case study for numerical simulations. Results show that the Fuzzy-LSTM strategy can significantly improve the control effect compared to the conventional fuzzy control strategy, and therefore achieve a better seismic response mitigation performance.