A neural network to predict spectral acceleration
Abstract
In this study, the main effort was evaluating the efficiency of artificial intelligence-based machine learning algorithms in the ground motion acceleration prediction (GMPE). To this end, a backpropagation neural networks (BPNN) is selected to build a data-driven model. This research evaluates the results of 25, 745 records provided by the Pacific Earthquake Engineering Research Center (PEER). A total of nine independent variables have been considered to describe ground motion acceleration. Linear regression is applied to the model as a benchmark. The effect of a number of hidden layers, different activation functions, and optimizers are also examined. The results declared that one-hidden layer BPNN with ‘RMSprop’ optimizer and ‘Softplus’ activation function performed as the best predictor.
Publication Title
Basics of Computational Geophysics
Recommended Citation
Kashani, A., Akhani, M., Camp, C., & Gandomi, A. (2020). A neural network to predict spectral acceleration. Basics of Computational Geophysics, 335-349. https://doi.org/10.1016/B978-0-12-820513-6.00006-0
