Loess-like silty clay (LSC), widely distributed in major Yellow River Basin projects, exhibits high variability in dynamic parameters, increasing seismic subsidence risk. This study employed combined resonance column-cyclic triaxial tests to quantitatively reveal, for the first time, systematic deviations in existing codes regarding LSC stiffness degradation and damping under small-strain conditions, providing a critical basis for revising seismic response analyses. Bender element tests and in situ wave velocity measurements established a deterministic conversion relationship between laboratory and field shear wave velocities for loess-like soils. To address the challenge of limited LSC datasets, a robust PSO-CatBoost prediction framework was developed: Particle Swarm Optimization (PSO) was employed to globally optimize CatBoost hyperparameters, with model robustness rigorously validated through residual analysis, Durbin–Watson test, and ± 1% input perturbation. Combined with Sobol global sensitivity analysis, this research elucidated for the first time the differential control mechanisms of wave velocities in LSC: shear wave velocity <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$V_{s}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mi>s</mi> </msub> </math></EquationSource> </InlineEquation> is dominated by burial depth (S1 = 0.786), reflecting stress-structure control characteristics, while compression wave velocity <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$V_{p}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mi>p</mi> </msub> </math></EquationSource> </InlineEquation> is primarily controlled by the liquidity index (S1 = 0.345), highlighting the core influence of pore fluids and the fluid-plastic state. The optimized model achieved significantly higher prediction accuracy for both <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$V_{s}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mi>s</mi> </msub> </math></EquationSource> </InlineEquation> (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">$R^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> = 0.916, MAPE = 7.25%) and <InlineEquation ID="IEq5"> <EquationSource Format="TEX">$V_{p}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mi>p</mi> </msub> </math></EquationSource> </InlineEquation> (<InlineEquation ID="IEq6"> <EquationSource Format="TEX">$R^{2}$</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> = 0.924, MAPE = 7.77%) compared to benchmark methods. This advancement provides a reliable tool to substantially reduce exploration costs and enhance the reliability of seismic safety assessments in LSC regions.

Dynamic characteristics of loess-like silty clay and methods for predicting wave velocity

期刊论文
发表日期：

2026年4月19日

查看全文

Liu Zhenghao,
Lu Linhai,
Ma Xianfeng,
Chaojun Wu,
Qianwei Xu,
Jiecheng Sun,
Bolong Ma

摘要

Loess-like silty clay (LSC), widely distributed in major Yellow River Basin projects, exhibits high variability in dynamic parameters, increasing seismic subsidence risk. This study employed combined resonance column-cyclic triaxial tests to quantitatively reveal, for the first time, systematic deviations in existing codes regarding LSC stiffness degradation and damping under small-strain conditions, providing a critical basis for revising seismic response analyses. Bender element tests and in situ wave velocity measurements established a deterministic conversion relationship between laboratory and field shear wave velocities for loess-like soils. To address the challenge of limited LSC datasets, a robust PSO-CatBoost prediction framework was developed: Particle Swarm Optimization (PSO) was employed to globally optimize CatBoost hyperparameters, with model robustness rigorously validated through residual analysis, Durbin–Watson test, and ± 1% input perturbation. Combined with Sobol global sensitivity analysis, this research elucidated for the first time the differential control mechanisms of wave velocities in LSC: shear wave velocity $V_{s}$ $V_{s}$ is dominated by burial depth (S₁ = 0.786), reflecting stress-structure control characteristics, while compression wave velocity $V_{p}$ $V_{p}$ is primarily controlled by the liquidity index (S₁ = 0.345), highlighting the core influence of pore fluids and the fluid-plastic state. The optimized model achieved significantly higher prediction accuracy for both $V_{s}$ $V_{s}$ ( $R^{2}$ $R^{2}$ = 0.916, MAPE = 7.25%) and $V_{p}$ $V_{p}$ ( $R^{2}$ $R^{2}$ = 0.924, MAPE = 7.77%) compared to benchmark methods. This advancement provides a reliable tool to substantially reduce exploration costs and enhance the reliability of seismic safety assessments in LSC regions.