| In recent years, the analysis of nonlinear response of structure has become an open problem in the fields of Earthquake Engineering and the technology of structural healthy diagnose has developed into the hot topic for the experts and scholars inside and outside China. The conventional integral method of structural nonlinear seismic response involves such puzzles as building of structure models, material nonlinearity, hysteretic models of structural members and integral methods while neural network (i.e. NN) is featured by cosmically parallelism, faulty-tolerant, self-organizing and self-learning, which makes it one of the efficient approaches to solving the nonlinear problem of structure. Combination with the existing research achievements, the thesis aims to some exploration and research on simulating and predicting structural seismic response based on MATLAB NN toolbox. The main conclusions are as follows: â‘ The simulating result of NN, which can conduct structural linear seismic response, is relatively similar to the result of time-history within the range of engineering; â‘¡Some suggestions are made such as network topology, node number of neuron, transfer function, collection of training samples, and training methods, which are used for simulating seismic structure response based on NN; the influence between node number of neuron and training result is analyzed in detail; â‘¢A NSIN neural network is built in the thesis, to simulate a plan RC frame and a 3D RC frame nonlinear seismic response. Moreover, some conclusions on collecting of training samples are drawn; Generally speaking, the accurate and simple method of simulating structural nonlinear seismic responses based on NN is more effective. It can see a fairly good perspective for it directly processes the input and output, avoiding the complex problems, like building of structure model, material nonlinearity, and hysteretic models of structural members and simplified processing of integral method. |
PDF Full Text Request