岩体变形模量偏最小二乘回归与神经网络关联性研究

STUDY ON PARTIAL LEAST SQUARE REGRESSION ASSOCIATED WITH NEURAL NETWORK FOR DEFORMATION MODULUS OF ROCK MASS

偏最小二乘回归不直接考虑因变量与自变量回归问题,而直接提取与系统有关的新的综合变量,并能利用交叉原理确定成分个数,尤其在分析单因变量与多自变量间关系上,其所得结果更为满意;人工神经网络具有自适应学习和记忆能力,尤其是3层BP网络模型。把这两者相关联,以岩体变形模量为因变量,岩石比重等7个因素为自变量,分析研究了焦作—晋城高速公路试验资料,所得拟合值与实测值误差最大为5.43%,较偏最小二乘法回归分析19.07%的误差好了许多。该方法为参数选取找到了一条新的途径。

Partial least square regression method does not directly consider the regression problem of dependent variables and independent variables, but extract the new integral variables with relation to the system. It can determine the number of components, and especially in analysis of the correlation between the single dependent variable and multi-independent-variables, the results are more satisfactory. The neural network has the ability of self adaptive learning and remembering, and the three-layer BP network has wide application. By associating the two methods, with the deformation modulus of rock mass being the dependent variable and seven factors of rock mass being the independent variables, the data from the observation of the Jiaozuo-Jincheng highway are analyzed and the maximum error between calculation and observation is 5.43%, which is better than the error of 19.07% by the partial least square regression method. So the association method gives a new way for the selection of parameters.