Ensemble of long short-term memory prediction and variable universe fuzzy control for seismic response reduction using electromagnetic dynamic vibration absorber

This study proposes a robust and intelligent approach for seismic response reduction in nuclear plant equipment using an electromagnetic dynamic vibration absorber (EMDVA) that integrates linear and nonlinear electromagnetic springs. To overcome the inherent response time delays in intelligent vibration control systems, a control strategy is developed that combines variable universe fuzzy control (VUFC) with long short-term memory (LSTM) neural network-based time series prediction. LSTM networks with varying input patterns are evaluated for predicting seismic responses. Then EMDVA integrated with LSTM and VUFC (LV-EMDVA) is developed to investigate the time series prediction mechanism for addressing time delays compared with VUFC-based EMDVA (V-EMDVA). Furthermore, shaking table tests are conducted on an electrical cabinet equipped with the control systems. Results show that the dual-input pattern in LSTM prediction enhances the model's generalization capability and reduces prediction errors. Through simulations and experiments, it is verified that the time delay issues must be considered and addressed in the intelligent control systems. Without adequate compensation for time delays, the performance of the intelligent control systems may be inferior to that of passive dampers. As time delay increases, the performance of V-EMDVA degrades compared to EMDVA, whereas LV-EMDVA demonstrates significant improvements in both RMS and peak responses while maintaining control stability . The experimental results closely match simulation results, validating the effectiveness of the simulation methods for electrical cabinets with intelligent control systems. • Intelligent control using electromagnetic dynamic vibration absorber is proposed. • LSTM prediction-based variable universe fuzzy control strategy for damper is developed. • LSTM networks with varying input patterns to predict seismic responses are studied. • Intelligent control strategy for EMDVA is validated by experiment and simulation.