摘要
针对堆石坝系统的地震响应分析问题,提出了一种递归神经网络建模方法。该神经网络模型包含内部状态神经元的反馈并具有状态空间形式。借助于该网络模型的逼近能力和动态信息存储能力,从观测的结构动态系统输入输出数据中重构原系统的输入输出特性,并对新的输入信号做出相应的预测和响应。分别对理想的有限元响应数据和实测的响应数据进行了仿真。结果表明,所提出的神经网络方法较好地学习了这两组结构系统的动态特性,并显示出较好的预测效果。
A kind of recurrent neural network was applied to the earthquake response analysis for Rock-fill Dam, The neural network model contains the feedback of the state neurons and takes the form of state space representation. With the approximation capability and dynamic information storage capability, the neural network model was trained to reconstruct the input-output characters from the observed input-output data. The model trained could perform response analysis and make prediction for new earthquake wave, Two data sets were used to test the method, the one was from a finite element program, and the other was from a real-life experiment. The responses of Rock-fill dam are well modeled, and the simulation result for new data indicates the better prediction ability.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2005年第10期2533-2536,2540,共5页
Journal of System Simulation
基金
国家自然科学基金项目(60374064)
关键词
递归神经网络
辨识
非线性系统
堆石坝
recurrent neural network
identification
nonlinear system
rock-fill dam
