一种大坝渗透系数分区反演新方法研究

基于人工神经网络的非线性映射特征，在渗流有限元计算的基础上，结合水头和渗流量等实测资料提出了大坝渗透系数的反演方法。为了克服经典神经网络存在的缺陷，提出了模拟退火的交替迭代算法神经网络新方法。在相同的初始条件下，用该新方法和经典网络进行了比较，得出前者的优越性和有效性。同时将该方法用于大坝的渗流反分析，利用反演出的渗透系数进行渗流场计算。数值计算结果表明，这种方法对大坝渗透系数反演问题具有较高的识别精度，反演结果可靠，可以用于实际工程。

A study on a new method for divisional inversion of dam seepage coefficient

Based on the nonlinear characteristic of ANN, and on the basis of water flow FEM computation, the inversion method for seepage coefficient of dam is presented together with the observation data of water and seepage flux. In order to overcome some disadvantage of the traditional neural network, a new method of alternation and iterative algorithm based on simulated annealing applied in Neural Network is put forward. On the same initial conditions, the comparison of the new method with the neural network based on the traditional algorithm is accomplished; and the result shows the superiority and efficiency of the former one. At the same time, this algorithm is applied to seepage inversion analysis of dam, by using the inversed seepage coefficient to calculate seepage flow field. Numerical computation indicates that the method has high precision of identification and reliable inversion result, and can be applied to the practical engineering.