摘要
偏最小二乘回归方法能有效地提取对系统最佳解释能力的新综合变量,较好地克服自变量间的多重线性相关性,但不能有效处理因变量与自变量间复杂的非线性问题,而神经网络方法是解决非线性问题的有力工具,但由于输入数据的多重相关性使得网络的求解变得不稳定及收敛速度慢,将这2种方法结合起来对岩体力学参数进行优选.实例表明,结合方法更优越,通过偏最小二乘回归方法处理自变量,消除了自变量间的多重线性相关性,提取了2个新综合变量,使网络的输入层节点数目由4个减少到2个,简化了网络结构,增强了网络稳定性,网络计算次数小于500次,计算精度0.000 2,96%的数据拟合误差在20%以内,预测误差都小于1%.
Partial least-square regression method can effectively distill new synthesis variables with best explaining ability to the system; and it can preferably solve the problem of the many layers relativity among variables; but it can not ideally deal with the complicated problems of nonlinearity between variables and independent variables. The neural network model is the ideal tool to solve nonlinearity; but serious correlation of input data will make the network unsteady and very slow. The two methods are generated into selecting rock mass mechanical parameters. The results indicate that the combining method is better. The variables are dealt with to dispose of many layers relativity among variables by partial least-square regression when new synthesis variables of 2 are extracted, and the input layers of network are decreased from 4 to 2 with simplizing network construction and strengthening network stability. The concise is 0. 000 2 when calculating times is less than 500; and 96% of data fitting errors are less than 20%; and all predicting errors are less than 1%.
出处
《武汉大学学报(工学版)》
CAS
CSCD
北大核心
2008年第3期43-46,共4页
Engineering Journal of Wuhan University
基金
2005年度河南省高等学校创新人才培养工程
2005年度河南省高校杰出科研人才创新工程项目(编号:2005KYCX015)
华北水利水电学院青年基金项目(编号:HSQJ2008008)
关键词
岩体力学
偏最小二乘回归
小脑神经网络
力学参数
rock mechanics
partial least-square regression
CMAC neutral network
mechanical parameters