CEEMDAN-LPF-RF方法在滑坡易发性预测中的应用研究

1.河南省第四地质勘查院有限公司,郑州 450001;2.河南省地质科学研究所有限公司,郑州 450001

易发性评价;完全集合经验模态分解;低通滤波器

Application of CEEMDAN-LPF-RF Method in Landslide Susceptibility Prediction
SUN Shuaitao1,SUN Xiaodan2

1.The Fourth Geological Exploration Institute of Henan Province Co., Ltd., Zhengzhou 450001, China;2.Henan Institute of Geological Sciences Co., Ltd., Zhengzhou 450001, China

Landslide susceptibility assessment; Complete ensemble empirical mode decomposition; Low-pass filter

DOI: 10.13512/j.hndz.2026.05.20

备注

滑坡易发性评价是滑坡灾害防治的基础工作,数据驱动模型在该领域应用广泛,然而,高频非线性噪声导致的模型精度劣化问题亟待解决。研究提出一种基于完全集合经验模态分解的低通滤波器去噪改进方法,通过自适应信号分解和噪声分离机制优化传统方法。以河南省灵宝市为例,选取15个致灾因子构建评价体系,采用随机森林模型进行滑坡易发性评价。通过ROC曲线对比完全集合经验模态分解—低通滤波器—随机森林(CEEMDAN-LPF-RF)、低通滤波器—随机森林(LPF-RF)和随机森林(RF)三种方法的精度。实验表明:提出的CEEMDAN-LPF-RF精度优于传统RF方法与LPF-RF方法。该模型提高了滑坡易发性预测的综合精度,可为构造活跃区、采矿扰动带等复杂地质环境下的滑坡风险精准防控提供参考。
Landslide susceptibility assessment serves as a fundamental task in landslide disaster prevention and con‐trol, where data-driven models have been widely applied. However, the problem of model accuracy degradation caused by high-frequency nonlinear noise urgently needs to be addressed urgently. This study proposes an improved denoising method based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) combined with a Low-Pass Filter (LPF), which optimizes traditional approaches through adaptive signal decomposi‐tion and noise separation mechanisms. Taking Lingbao City, Henan Province as the study area, 15 causative factors were selected to construct the evaluation system, and the Random Forest (RF) model was employed for landslide sus‐ceptibility assessment. The performance of three methods—CEEMDAN-LPF-RF, LPF-RF, and RF—was compara‐tively analyzed using ROC curves to evaluate their accuracy. Experiments show that the accuracy of CEEMDAN-LPF-RF proposed in this paper is better than that of traditional RF method and LPF-RF method. The model im‐proves the comprehensive accuracy of landslide susceptibility prediction, and can provide reference for precise pre‐vention and control of landslide risk in complex geological environments such as tectonic active areas and mining disturbance zones.
·