基于PCA-BP融合的边坡稳定性评价模型  

Evaluation model for slope-mining stability based on the PCA-BP integration

作　　者：高文华[1] 罗新辉[1] 胡泽涛[1] 

机构地区：[1]湖南科技大学岩土工程稳定控制与健康监测省重点实验室,湖南湘潭411201

出　　处：《安全与环境学报》2016年第2期6-10,共5页Journal of Safety and Environment

基　　金：国家自然科学基金项目(41272324);湖南省自然科学基金项目(14JJ7057)

摘　　要：针对边坡岩土体物理力学性质复杂、边坡稳定性影响因素众多等特点,提出将主成分分析(PCA)和BP神经网络结合起来进行边坡稳定性评价的方法。若BP神经网络训练误差一定,则网络信息容量与样本数成正比。当样本数较少时,就必须减少样本维数,以达到较好的匹配效果,为此,引入主成分分析法(PCA)对影响边坡稳定的众多变量进行降维处理,以消除输入数据间的相关性,有效地减少预测模型的输入量,优化网络的输入节点数,提高网络的运行效率。针对BP算法容易落入局部最小、收敛速度慢等缺点,引入粒子群优化算法(PSO)优化神经网络的连接权重与阀值,从而克服了BP神经网络的固有缺陷。在此基础上,建立基于PSO优化算法的PCA-BP融合的边坡稳定性评价模型。模型分为3个层次,第一层次为输入层,即经过PCA分析之后获得的主成分;第二层次为隐含层;第三层次为输出层,即安全系数。应用该评价模型进行算例分析,结果表明,安全系数的模型计算值与参考值的绝对误差均很小,相对误差均控制在6%以内,吻合程度较高。The paper is attempting to introduce an evaluation model developed by us for the slope-mining stability based on the PCA- BP integration. Due to the complex physical and mechanical features of the rock mass and the soil body,there exist a lot of factors that tend to influence the slope stability,which makes it necessary to develop a new method to evaluate the slope stability based on the combination of the principal component and BP artificial neuralsystem. If it is possible to keep the training errors of BP artificial-neural network samples changeless,it would be possible to keep the network information capacity in a well proportionate to the sampling numbers. Even if the sampling numbers are not enough,it is also possible to reduce the sampling dimensions properly so as to make it better matched between the sampling number and the sampling dimension. Therefore,it is necessary to apply the PCA method to the reduction of the dimension of variables that may affect the slope stability and eliminate the imbalance of the input data in the prediction model and optimize the input node numbers of the evaluation model so as to raise the efficiency of the network. Seeing the fact that the BP algorithm may fall into the local minimum context easily and slow down the convergence speed,we have introduced the method of the particleswarm-optimization（ PSO） in hoping to optimize the connective weight and the critical value of BP neural network so as to overcome the inherent defects of BP neural network,which may serve as the underground for us to set up the PCA- BP evaluation model of slope stability. And,practically speaking,the model of slope-mining stability can be divided into three layers. The first layer is known as the input layer or the layer of principal component,whereas the second layer is known as the hidden layer.And,the third layer can be regarded as the output layer or the safety coefficient. When we have made a detailed analysis of the example through practical application,the results may indicate that littl

关 键 词：安全工程 粒子群优化算法 BP神经网络 主成分分析 评价模型 

分 类 号：X93[环境科学与工程—安全科学] TU415.7[建筑科学—岩土工程]