基于多元自适应回归样条方法的RMR系统预测岩体变形模量

Prediction of Rock Deformation Modulus Using RMR System Based Multivariate Adaptive Regression Spline Method

在采用岩土分级系统RMR估计变形模量时,需假设组成系统的所有参数对变形模量有着完全相同的关系,这一内在假设降低了经验公式的精度。为此,引入多元自适应回归样条(MARS)构建模型。实例分析表明,MARS模型的适用性较好,各项指标均优于多元线性模型和多项式模型;MARS模型对变量的重要性分析表明,测量深度、完整岩石的单轴抗压强度(UCS)和岩石质量指标(RQD)对变形模量有更强的影响,而地下水情况(GW)与变形模量不存在相关性。

The assumption underlying the use of a geomechanical classification system regarding the estimation of the deformation modulus is that all of the parameters constituting the system have a correlation that are identical to those of the deformation modulus. This inherent assumption decreases the statistical accuracy of empirical correlations. To over- come this limitation, Multivariate Adaptive Regression Splines （MARS）have been applied to establish the correlation be- tween the parameters of RMR system and deformation modulus. The case study indicates that the applicability of MARS model is better. Compared with multiple and polynomial regression, the evaluation indicators of MARS model are better. The analysis of the importance of variables in MARS model show that measured depth, the uniaxial compressive strength （UCS） and Rock Quality Designation（RQD）have stronger influence on deformation modulus, and groundwater conditions is not related to the deformation modulus.