摘要
针对经典算法BP网络存在的一些缺陷,提出了一种交替迭代算法神经网络,并用FORTRAN语言编制了该程序。在相同的初始条件下,将交替迭代算法神经网络和经典算法BP网络进行了比较,得出了交替迭代算法神经网络的特点和优点。在算例中,基于人工神经网络的非线性特点,在三维渗流有限元的基础上,结合该原理反演了水闸地基的渗透系数比值。与真实值的比较说明,反演结果是精确的,从而更进一步验证了该方法应用于反演分析中的可靠性。
Aimed at some limitation of the traditional BP Neural Network,a global optimal algorithm is put forward based on alternative and iterative algorithm,and the corresponding program of FORTRAN is developed. With the same initial condition,the Neural Network based on alternative and iterative algorithm is compared with that based on the traditional algorithm,and the characteristic and excellence of the former algorithm are explained. As an example,based on the nonlinear characteristic of ANN and 3D seepage flow FEM computation,the method combined with inversion principle is applied to the sluice foundation and the percolation parameter ratio is obtained. Compared with the actual value,the obtained result is accurate,which illuminates the reliability of the algorithm in back analysis.
出处
《岩石力学与工程学报》
EI
CAS
CSCD
北大核心
2004年第9期1470-1475,共6页
Chinese Journal of Rock Mechanics and Engineering
