基于RBF-FCM-OLS的边坡稳定性评价方法

Evaluation Method of Slope Stability Based on RBF-FCM-OLS

RBF神经网络具有很强的非线性并行处理能力和泛化能力,并且有很快的学习收敛速度,不易陷入局部极小,在边坡稳定性评价中已得到广泛的应用。但其过分依赖于隐含层数据中心的选取是否合适,故引入模糊-C均值聚类（FCM）算法对其进行优化。以122组边坡样本作为样本总体,其中1~114组为训练样本,115~122组为测试样本,运用FCM算法在边坡训练样本中初选多个RBF网络的数据中心,在此基础上运用正交最小二乘法（OLS）训练网络,利用训练后得到的回归矩阵信息在初选结果中重新选择RBF网络的数据中心,从而使数据中心得到优化,简化了RBF神经网络的结构。将优化后的RBF神经网络应用到边坡测试样本的安全系数的预测中,得到较高的预测精度。该方法加快了RBF神经网络的训练速度,提高了运算速率,与传统的BP网络进行比较,进一步证明RBF及其学习算法的优越性和实用性。

RBF neural network has strong nonlinear parallel processing ability and generalization ability,and has a quick learning convergence speed,it is not easy to fall into local minimum,and it has been widely used in slope stability evaluation. But its over-reliance on selecting hidden layer of the data center is appropriate or not. Therefore,the fuzzy-C-means（FCM） clustering algorithm is introduced for the optimization. Using the 122 groups of slope samples as the sample population,in which the 1 ~ 114 groups are the training samples,and the 115 ~ 122 groups are the test samples. Using the FCM algorithm,the multiple data centers of RBF networks are selected initially in the training sample of slope. Based on this,the orthogonal least squares（OLS） method is used to train the network. Using the obtained regression matrix information after training,the data centers of RBF network are chosen again from the primary results.So that the data centers are optimized,the structure of RBF neural network is simplified. The RBF neural network after optimization is applied to predict the safety factor of slope test samples,the higher prediction precision is got. This method accelerates the training speed of RBF neural network,and it improves the operation rate. Comparing with the traditional BP network,the superiority and practicability of RBF and its learning algorithm are proved further.