Inversion of Soil Parameters Based on Optimized Multidimensional Support Vector Regression Model
To improve the accuracy of soil parameter inversion from displacement monitoring data,a method based on optimized multidimensional support vector regression model for soil parameter inversion is proposed.Taking the ultra-deep cutting slope project in loess region for an example,a two-dimensional slope model is established by using the finite difference software FLAC3D.An orthogonal experiment is conducted to analyze the multi-factor sensitivity of soil parameters,and determine the parameters to be inverted.Then,the three-dimensional excavation model of slope is established to reflect real excavation conditions.The simulated displacement values with different inversion parameters are calculated to obtain the training data.The optimal parameters of multidimensional support vector regression model are obtained by using the sparrow search algorithm to construct SSA-MSVR model,and the optimized model is trained by using the training data.Finally,the actual monitoring displacements are inputted into the trained model to obtain optimal inversion parameters.The correctness of inverted parameters are verified through forward analysis.The result indicates that the top 4 soil parameters in terms of sensitivity to slope stability coefficient are the internal friction angle of old loess,the internal friction angle of red clay,the cohesive force of old loess,and the elastic modulus of old loess.These 4 parameters are determined as the parameters to be inverted.When the excavation of ultra-deep cutting slope is completed,the settlement displacement occurs at the top of slope,while unloading rebound deformation appears at the bottom of slope.The relative errors between the displacement simulated values calculated with the inverted parameters and the actual monitoring values are all less than 10%,which proves that the SSA-MSVR model yields good results when applied to the inversion of soil parameters.It provides new ideas and methods for parameter inversion.
road engineeringparameter inversionsparrow search algorithmmultidimensional support vector regressionsensitivity analysisloess slope
NETL
NSTL
万方数据
