Abstract
To address the limitations of traditional linear non-stationary random field models in handling the nonlinear variation of soil parameters in fill slopes, this paper proposes a non-stationary random field model employing piecewise linear functions as trend functions. The coefficient calculation method, discretization procedures, and structured computational workflow of this improved model are elaborated in detail, and its effectiveness and applicability are validated through comparative analysis of a loess fill slope example. Furthermore, since different discretization combinations of non-stationary random field models exhibit variations in simulation accuracy and computational efficiency, this paper analyzes the influence of different discretization combinations on simulation performance. The results demonstrate that the piecewise linear non-stationary random field model exhibits stronger adaptability and higher fitting accuracy compared to the traditional linear model. At various burial depths of fill slope soils, the mean and standard deviation of soil parameters simulated by the piecewise linear non-stationary random field model are significantly superior to those of the linear non-stationary random field model and closer to theoretical values. In terms of simulation accuracy: Discretization Combination 1 > Discretization Combination 2 > Discretization Combination 3; in terms of computational efficiency: Discretization Combination 1 < Discretization Combination 2 ≈ Discretization Combination 3. When discretizing non-stationary random fields, it is necessary to select appropriate discretization methods to enhance simulation accuracy and computational efficiency. Based on the aforementioned research, this paper recommends that Discretization Combination 2 achieves an optimal balance between simulation accuracy and computational efficiency, while Discretization Combination 1 ensures high adaptability and precision of the model.
Full Text
Preamble
A New Nonstationary Random Field Model and Its Application to Fill Slopes
Shuaihua Ye¹, Jiaxuan Zheng¹, Jingbang Li²
¹ School of Civil Engineering, Lanzhou University of Technology, Lanzhou 730050
² School of Civil Engineering, Lanzhou Institute of Technology, Lanzhou 730050
Abstract
To address the limitations of traditional linear nonstationary random field models in capturing the nonlinear variation of soil parameters in fill slopes, this paper proposes an improved nonstationary random field model employing piecewise linear functions as trend functions. The coefficient calculation methodology, discretization procedures, and structured computational framework of the proposed model are elaborated in detail. The validity and applicability of the improved model are verified through comparative analysis of a loess fill slope case study.
Furthermore, recognizing that different discretization combinations of nonstationary random field models yield varying simulation accuracy and computational efficiency, this paper investigates the influence of different discretization schemes on simulation performance. The results demonstrate that the piecewise linear nonstationary random field model exhibits stronger adaptability and higher fitting accuracy compared to the traditional linear model. At various depths within fill slope soils, the mean and standard deviation of soil parameters simulated by the piecewise linear model are significantly superior to those obtained from the linear nonstationary random field model and are closer to theoretical values. In terms of simulation accuracy, the ranking is: Discretization Combination 1 > Discretization Combination 2 > Discretization Combination 3; regarding computational efficiency, the relationship is: Discretization Combination 1 < Discretization Combination 2 ≈ Discretization Combination 3.
When discretizing nonstationary random fields, selecting appropriate discretization methods is essential for enhancing both simulation accuracy and computational efficiency. Based on these findings, this paper recommends Discretization Combination 2 for achieving the optimal balance between simulation accuracy and computational efficiency, while Discretization Combination 1 is recommended for ensuring high adaptability and precision of the model.
Keywords: Fill slope; Soil parameters; Trend function; Nonstationary random field; Mean and standard deviation