摘要
提出了模拟退火的 Gauss-Newton 算法的神经网络,克服了经典 BP 网络存在的一些缺陷。并以正弦函数的迭代收敛为例,证明了该方法的正确性,有效性和优越性。同时将该方法用于同乐坪大坝的渗流反分析,利用反演出的渗透系数进行渗流场计算。得到的水头预报值与观测值相吻合,可知反演结果是正确的,说明该方法用于实践工程的渗流参数识别是可行的。
Simulated Annealing Gauss-Newton algorithm applied to neural network is put forward. It overcomes some limitation of the traditional BP neural network. Taking iterative convergence of sine function for example, the correction, efficiency and superiority of the algorithm are proved. At the same time, this algorithm is applied to seepage back analysis of Tongleping dam, using seepage coefficient to calculate seepage flow field. And forecasted water heads approach to the observed values, which illuminates the back analysis result is correct and the algorithm is feasible in the practical seepage parameters identification.
出处
《岩土力学》
EI
CAS
CSCD
北大核心
2005年第3期404-408,414,共6页
Rock and Soil Mechanics
关键词
模拟退火的Gauss-Newton算法
反分析
渗透系数
Algorithms
Backpropagation
Dams
Geotechnical engineering
Identification (control systems)
Iterative methods
Neural networks
