Dynamic analysis of soil-structure interaction using the neural networks and the support vector machines

Abstract

An approach to a soil-structure interaction problem is using data-based methods (DBM's) that benefit from developing mathematical models on the experimental data. A mathematical model for the seismic analysis of soil-pile-structure (SPS) systems is built in the neural networks environment based on the existing experimental data. A network consisting of two hidden layers is proved to be the most efficient among other choices. Three sets of data are utilized for training, testing, and validation of the ANN model to avoid over fitting by cross-validation. The accuracy of the neural networks to predict the seismic behavior is enhanced by the parallel vectorial analysis technique of the support vector machines. It is shown that the model can predict the dynamic characteristics and seismic response of the soil-structure system with good accuracy in much less time compared with the finite element method. This research sets out the practical importance of trying to produce more experimental data and using DBM's in solving the complex problem of dynamic analysis of SPS systems in which due to various unknowns, enough accuracy cannot be gained with conventional analytical approaches.