基于FLAC及神经网络的初始地应力场反演

Back Analysis of Mechanical Parameters and Initial Stress Field of Rock Masses Based on Finite Difference Method and RBF Neural Network

分别采用RBF网络和BP网络,利用FLAC进行正分析计算,依据测点的应力数据反演了计算区域的初始地应力场。结果表明,在样本数量相同的情况下,RBF神经网络反演分析的精度以及学习、收敛速度均优于采用BP网络的反演算法。

Using RBF NN and BP NN respectively authors of this paper identified initial stressesaccording to measured normal stresses of some specific points. Direct computations based on Fast LagrangianAna lysisof Continuum (FLAC)wereperformed to get enough trai ning samples for RBF NN and BP NN. An example shows that combination of RBF NN has faster convergence and is more effective than application of BP NN.