基于GA-BP神经网络的拱坝地震易损性分析

GA-BP artificial neural networks for predicting the seismic response of arch dams

拱坝在其生命周期内可能会承受强烈地震,其地震易损性引起了广泛的关注。一般而言,采用非线性有限元法进行拱坝的地震易损性分析,需要大量的计算工作量。该文提出了一种预测拱坝地震响应的方法——基于遗传算法(genetic algorithm, GA)的多层前馈(back propagation, BP)神经网络,该方法可以替代部分非线性有限元分析计算,显著减少计算成本。以大岗山拱坝的易损性分析为算例,基于已有的390个有限元非线性动力分析工况数据,将结构的响应设定为BP神经网络的输出,地震强度参数IM作为输入,进行BP神经网络的训练和验证。结果表明,该文提出的GA-BP神经网络采用390个有限元结果中的30%的数据进行训练,即可得到满足精度的预测结果,给出合理的拱坝地震易损性曲线,说明采用GA-BP神经网络后可节省70%的非线性有限元计算成本。

Arch dams may be subjected to strong earthquakes during their lifecycle and their seismic response has attracted extensive attention in dam engineering. Nonlinear finite element seismic response analyses of arch dams require large amounts of computational effort. This paper presents a back propagation(BP) genetic algorithm(GA) method for predict the seismic responses of arch dams which replaces some of the finite element analysis calculations and significantly reduces the computational cost compared with the finite element method. A BP neural network was trained and validated for the Dagangshan arch dam based on 390 nonlinear dynamic response cases calculated using the finite element method with the structural response as the BP neural network output and the seismic intensity parameter, IM, as the input. The results show that the GA-BP neural network can properly predict the dam seismic response and give reasonable seismic response curves using 30% of the 390 cases for training which shows that the GA-BP neural network can save 70% of the nonlinear finite element cost.