基于优化时序分解和特征选择的滑坡位移预测模型

Landslide Displacement Prediction Model Based on Optimized Time Series Decomposition and Feature Selection

ES评分 0

DOI 10.14075/j.jgg.2025.08.293
刊名
Journal of Geodesy and Geodynamics
年,卷(期) 2026, 46(6)
作者
作者单位

1 长安大学地质工程与测绘学院, 西安市, 710054

摘要
针对时序分解模型难以准确区分诱发因子对不同位移分量的作用, 以及在气象数据存在不确定性下预测精度不足的问题, 提出一种基于优化时序分解与特征选择的滑坡位移预测网络模型。首先, 利用奇异谱分析(SSA)与遗传算法优化的变分模态分解(GA-VMD)方法对滑坡位移在诱发因子驱动下进行分解; 随后, 构建融合诱发因子的改进西原模型预测趋势项位移, 采用结合卷积神经网络与压缩和激励网络的门控循环单元(CNN-SE-GRU)组合网络建模预测周期项位移, 通过频域分析重构随机项位移; 最后, 结合核密度估计(KDE)和蒙特卡洛(Monte Carlo)模拟方法, 构建位移预测结果的概率区间。以甘肃省黑方台滑坡为例, 预测模型的RMSE和MAPE分别为1.52 mm和0.38%, 预测精度较传统的预测模型显著提升, 为滑坡预警提供了更可靠的技术支撑。
Abstract
Aiming at the problem that it is difficult for the temporal decomposition model to accurately distinguish the effects of induced factors on different displacement components, and the prediction accuracy is insufficient under the uncertainty of meteorological data, a landslide displacement prediction network model based on optimized time series decomposition and feature selection is proposed. Firstly, the variational modal decomposition (GA-VMD) method optimized by singular spectral analysis (SSA) and genetic algorithm is combined with induced factors to decompose the landslide displacement. Subsequently, an improved Nishihara model with fusion inducible factors is constructed to predict the trend term displacement, and the combined network of convolutional neural network and gated recurrent unit (CNN-SE-GRU) combined with compression and excitation network was used to model the period term displacement, and the random term displacement is reconstructed through frequency domain analysis. Finally, the probability interval of the displacement prediction results is constructed by combining kernel density estimation (KDE) and Monte Carlo simulation. Taking the Heifangtai landslide in Gansu province as an example, the RMSE and MAPE of the prediction model are 1.52 mm and 0.38%, respectively, and the prediction accuracy of the model is significantly improved compared with the traditional prediction model, providing more reliable technical support for landslide early warning.
关键词
滑坡位移预测; 时序分解; 特征选择; 神经网络; 误差传播建模
KeyWord
landslide displacement prediction; time series decomposition; feature selection; neural network; error propagation modeling
基金项目
页码 748-757
  • 参考文献
  • 相关文献
  • 引用本文

王东民, 赵丽华, 瞿伟, 杭资牧, 王利. 基于优化时序分解和特征选择的滑坡位移预测模型 [J]. 大地测量与地球动力学. 2026; 46; (6). 748 - 757.

  • 文献评论

相关学者

相关机构