New Method of The Earthquake Response of The Dams by Recurrent Neural Network

摘要

Owing to the nonlinear characteristics of materials, and the uncertainty of mechanical and physical parameters, it is difficult to describe a real situation of a dam by an ordinary analysis method. The purpose of this paper is to modify the aseismatic design method of dams by developing the self-training and self-adjusting characteristics of neural network, using abundance information in the input and output, and advancing the precision of method. (1) Use earthquake simulating equipment to do physical model experiment, and collect reaction signal records for the different kinds of rock-filling dam models and the different kinds of random earthquake wave input signal. (2) Build a proper neural network model, and confirm the input and output of network. (3) Use observed input and output signal records as teaching signal to train the network model, and identify dynamic response of rock-filling dams. (4) check up the generalization ability of it, make the model more practical.