A new methodology for generating artificial earthquake accelerograms was developed in 1997, which uses the learning capabilities of neural networks to obtain the knowledge of the inverse mapping from the response spectra to earthquake accelerograms. Recently, the methodology has been further extended and enhanced. Due to the stochastic nature of the earthquake accelerograms, it was deemed more appropriate to use stochastic neural networks (SNNs). A new SNN has been developed capable of generating multiple earthquake accelerograms for a given response spectrum. The new stochastic features of the neural network is combined with a new strategy for data compression with replicator neural networks. The proposed methodology is more efficient in compressing earthquake accelerograms and extracting their characteristics and it produces a stochastic ensemble of earthquake accelerograms from the design response spectrum. An example is presented to demonstrate the performance of the stochastic neural network and its potential in future research.
展开▼
