基于优化经验模态分解和聚类分析的滑坡位移智能预测研究

Intelligent prediction of landslide displacements based on optimized empirical mode decomposition and K-Mean clustering

针对三峡库区"阶跃式"滑坡的变形特征,提出了一种新的滑坡位移预测方法。以白水河滑坡ZG118和XD-01监测点位移数据为例,采用基于软筛分停止准则的经验模态分解(SSSC-EMD)将累计位移-时间曲线和影响因子时间序列自适应地分解为多个固有模态函数(IMF),并采用K均值(K-Means)聚类法对其进行聚类累加,得到有物理含义的位移分量(趋势性位移、周期性位移以及随机性位移)和影响因子分量(高频影响因子和低频影响因子)。使用最小二乘法对趋势性位移进行拟合预测;采用果蝇优化-最小二乘支持向量机(FOA-LSSVM)模型对周期性位移和随机性位移进行预测。将各位移分量预测值进行叠加处理,实现滑坡累计位移的预测。研究结果表明,所提出的(SSSC-EMD)-K-Means-(FOA-LSSVM)模型能够预测"阶跃式"滑坡的位移变化规律,且预测精度高于传统的支持向量机回归(SVR)、最小二乘支持向量机(LSSVM)模型;并通过改变训练集长度,进行单因素分析,发现其与预测精度之间呈正相关关系。

According to the deformation characteristics of step-like landslides in the Three Gorges Reservoir area, a new method for predicting the landslide displacement is proposed. The monitoring displacements of points ZG118 and XD-01 in Baishuihe landslide are taken as example analysis. By using the empirical mode decomposition with soft screening stop criteria(SSSC-EMD), the cumulative displacement-time curves and the influencing factor time series are adaptively decomposed into multiple intrinsic mode functions(IMF). The K-Means clustering method is adopted to cluster and accumulate IMFs. The displacement components(including the trend, periodic and stochastic displacements) and the influence factor components(including high-frequency and low-frequency factors) which contain physical meanings are obtained. The trend displacements are fitted by the least square method. The periodic and stochastic displacements are predicted by combating fruit fly optimization and least squares support vector machines(FOA-LSSVM) model. Finally, the cumulative prediction displacement is found to be the addition of the three component prediction values. The results show that the proposed(SSSC-EMD)-K-Means-(FOA-LSSVM) model has the capability of predicting the displacement variation of step-like landslides. The prediction accuracy of this model is higher than those of traditional SVR and LSSVM models. Furthermore, the single factor analysis is performed by changing the length of the training, and it is positively correlated with the prediction accuracy.