Intelligent Control Method Research for High Rise Building Vibration by Integrating Genetic Algorithm and LSTM

Abstract

In order to solve the problem of performance degradation, such as local optimality, that may occur when shallow learning is used to predict the high-rise buildings seismic response under difficult conditions, a high-rise building vibration intelligent control method integrating genetic algorithms and long short-term memory networks is proposed. First, a structural response prediction model is constructed and combined with vibration control theory. Furthermore, an intelligent control algorithm using long short-term memory networks is designed. In conjunction with this algorithm, a centralized controller that integrates convolutional neural networks at different levels is designed. The structure of the centralized control system is improved, and genetic algorithms and Lyapunov stability theory are used to optimize thenetwork hyperparameters through deep learning. The results showed that this framework had high prediction accuracy, with the smallest relative difference in predicting C-library data at −0.0053 cm on average. The largest prediction error for B-library data was 0.015 cm on average. The long short-term memory network had the smallest prediction error and the best learning and prediction performance. When the degradation level of each layer stiffness in the benchmark model was between 10.2% and 20.5%, this intelligent controller achieved the best control effect, maintaining above 39.8%. Optimized using genetic algorithm, the optimal fitness value after 80 iterations represented controllerloss function value, which were $8.3\times 10.5$ , $2.3\times 10.4$ , $2.2\times 10.4$ , and $3.0\times 10.4$ , respectively, demonstrating good prediction results. Compared with traditional trial calculation methods, this algorithm has higher computational efficiency and accuracy. The fusion of genetic algorithms and long short-term memory networks with different structural forms shows good seismic reduction effects on the time responses of benchmark models. The research method has good prediction accuracy, high reliability, and flexible system design, providing new strategies for intelligent control of high-rise building structures under different conditions.

Keywords

• Genetic algorithm
• LSTM
• architecture
• structure
• intelligent control