Seismic reliability analysis of underground pipeline system using artificial neural networks

This paper proposes the use of the backpropogation algorithm-multilayer feedforward network (namely, BP network) to found the prediction model of underground pipeline response to earthquakes and connectivity model of pipeline system. According to the data sets from numerical computations, the trained neural networks can be obtained by means of offline training. Both of the models are able to perform real-time simulation by the trained networks. Better generalization capability of BP network is fully employed by both of the models to rapidly evaluate the performance state of underground pipeline after earthquakes. Neural networks are capable of overcoming the shortcoming of traditional methods requiring a great deal of computation time, and supply a new approach for the rapid evaluation of underground pipeline after earthquakes.