基于LSTM的自复位中心支撑钢框架地震响应预测OA

To predict the response of self-centering braced steel frame structures under seismic loading,this paper employs a deep learning algorithm to construct a Long Short-Term Memory(LSTM)neural network model.This model establishes a nonlinear mapping relationship between input seismic acceleration and output structure seismic drift responses,exploring the effects of data window-size and dataset splitting ratios on the predictive performance of the model.The research findings indicate that the proposed LSTM neural network model demonstrates robust predictive capabilities,particularly in predicting the roof drift of the structure,achieving a peak relative error of 2.36%and a correlation coefficient of 0.94.Increasing the data window-size appropriately can still accurately predict both the roof drift and residual inter-story drift of the structure while enhancing prediction efficiency,without significantly altering the peak phase distribution of the predicted responses.For scenarios with limited sample sizes,when the known dataset splitting ratio falls within the range of 2∶1 to 3∶1,the model exhibits satisfactory overall predictive performance.When the training and validation sets adequately cover various time-step types,the model's overall generalization ability is effectively improved.However,it displays slightly less sensitivity to micro-data.