论文部分内容阅读
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 ability of nonlinear parallel processing and generalization, and has a fast learning convergence speed, not easy to fall into the local minimum, has been widely used in the slope stability evaluation. However, it is too dependent on the selection of hidden layer data center is appropriate, so the introduction of fuzzy-C means clustering (FCM) algorithm to optimize it. Based on 122 groups of slope samples as samples, 1 ~ 114 groups of training samples and 115 ~ 122 groups of samples as test samples, FCM algorithm was used to select data centers of multiple RBF networks in slope training samples Using OLS training network, the data center of RBF network is re-selected in the primary election results by using the regression matrix information obtained after training, which optimizes the data center and simplifies the structure of RBF neural network. The optimized RBF neural network is applied to the prediction of the safety factor of the slope test sample to obtain high prediction accuracy. This method accelerates the training speed of RBF neural network and improves the computing speed. Compared with the traditional BP network, this method further proves the superiority and practicability of RBF and its learning algorithm.