Predictive Model for the Seismic Capacity of High-Rise Building Using Deep Learning Algorithm

Abstract

Seismic capacity demands continuous detailing in response to its focal function in structural health monitoring. While modern methods in engineering technology and artificial intelligence have been meeting those demands, the unpredictability of natural events, such as earthquakes, requisites progressive preparations. Although instruments and models that have demonstrated their capacity to estimate structural integrity already exist, the question remains as to whether or not it produces accurate and effective real-time results. In this study, through the application of Deep Learning algorithms, an RNN-LSTM model was derived as an alternative approach to achieve more accurate predictions. Datasets collected from accelerometer sensors in two 15-storey high-rise buildings were used to measure peak ground acceleration (PGA), peak ground velocity (PGV), and peak ground displacement (PGD). The model was further validated to evaluate its capacity using performance metrics such as mean square error (MSE), mean absolute error (MAE), and root mean square error (RMSE). The produced RNN-LSTM predictive model has produced satisfactory results in comparison of the actual and predicted values as well as in the existing methods for seismic analysis. However, flaws in the approach, specifically the inadequacy of hyperparameters in sequence length, epoch, and data size, were seen as the limits of the produced model.