利用修正的BP神经网络进行框架结构损伤诊断  

Structural damage detection using amendatory BP neural networks

作　　者：刘昭清[1] 刘红艳[2] 韩冬生[3] 

机构地区：[1]西南交通大学力学系,四川成都610031 [2]唐山热力总公司,河北唐山063000 [3]河北理工大学监理公司,河北唐山063000

出　　处：《河北理工学院学报》2005年第1期129-134,共6页Journal of Hebei Institute of Technology

摘　　要：提出了一种基于修正的BP神经网络方法的结构损伤诊断方法,该方法通过基于误差曲面调整权值以及引入动量因子算法来避免出现局部极小,并通过自适应调整学习速率来提高学习效率,从而解决了传统BP神经网络在实际应用中存在的收敛速度慢和存在局部极小的问题。算例表明,该方法能够不仅较为准确的检测出结构的不同程度损伤,而且提高了计算效率,同时具有较强的鲁棒性。A new amendatory back - propagation neural network has been proposed in this paper. This method can avoid local minima by modulating modulated values through error surface, and by adopting a dynamic momentum algorithm. And it can accelerate steepest descent by self - adapting to modulate the study efficiency. This method solves two practical problems encountered by the traditional back - propagation method ,i. e. ,slow learning progress and convergence to a false local minimum. Numerical results show that the method may enhance computational eff-ciency as well as detection accuracy, and it also has a good robustness

关 键 词：神经网络 修正的BP算法 结构损伤诊断 

分 类 号：TU311.3[建筑科学—结构工程]