基于偏最小二乘-径基网络模型的岩体变形分析  

Analysis of rock mass deformation based on radical basis function network model and partial least squares regression

作　　者：金永强[1] 张磊 

机构地区：[1]安徽省水利水电勘测设计院,安徽省合肥市230088 [2]河海大学水文水资源与水利工程科学国家重点实验室,南京210098 [3]河海大学水利水电工程学院,南京210098

出　　处：《西北水电》2011年第B09期38-40,共3页Northwest Hydropower

基　　金：973计划课题:多因素相互作用下地质工程系统的整体稳定性研究(2002CB412707);现代水工大体积混凝土裂缝机理与控制(50539010);国家自然科学基金雅砻江水电开发联合研究基金重点项目:岩石高边坡失稳的大型滑坡预警和防治(50539110);国家自然科学基金委员会二摊水电开发有限责任公司雅砻江水电开发联合基金项目(50539030);国家科技支撑计划课题基于风险的大坝安全评价技术开发(2006BAC14B03)

摘　　要：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,并能利用交叉有效性检验确定成分个数,在分析单因变量与多自变量间关系时结果令人满意;3层RBF神经网络模型具有自适应学习和记忆能力,因而被广泛应用。把这两者相关联,以岩体变形量为因变量,以6个影响因素为自变量,分析研究了实际工程的高边坡位移监测资料。工程应用实例分析研究表明,偏最小二乘-径基网络模型能有效克服各类因子变量间的相关性和与因变量的非线性关系,模型收敛速度快,求解稳定,对实测数据具有较好的拟合效果和预测精度,因而具有一定的实用价值。Partial least square regression （PLSR） method does not directly involve the regression of dependent variables and independent variables, but applies the new integral variables related to the system. Furthermore, it can verify and determine the number of components by application of cross validation test. Accordingly, the results are satisfactory when the relationship between the single dependent varia- ble and multi-independent-variables are analyzed. Three-layer radical basis function （RBF） network is with the ability of self-adaptive learning and remembering, so it is widely applied. By association of the two methods of PKSR and RBF ,with rock mass deformation being the dependent variable and six factors of rock mass being the independent variables ,the monitoring data on high-slope displacement of actual projects are analyzed and studied. It shows that the proposed method of RBF and PLSR can effectively overcome the effects between correlation among variables and nonlinearity of independent variables. The model has a good fitting accuracy and forecasting capability with a speedy and stable convergence, thus it has a certain practicability.

关 键 词：偏最小二乘回归 交叉有效性检验 径基网络 岩体变形 

分 类 号：TV642.45[水利工程—水利水电工程] TV698.11