基于优化经验模态分解和最小二乘支持向量机的边坡位移预测

我国库岸滑坡灾害频发，采用高精度优化算法对边坡位移时间序列进行预测对防灾减灾具有重要意义。边坡位移时间序列通常表现出高度非线性特征，传统模型难以对其进行准确预测。为此，本文提出一种基于优化经验模态分解和最小二乘支持向量机的边坡位移时间序列预测模型。该模型采用基于软筛分停止准则的经验模态分解（SSSC-EMD），可自适应地将边坡位移时间序列分解为多个本征模态分量和1 个残余分量。将残余分量定义为趋势项；通过K-means 聚类方法对分量进行聚类，将其定义为周期项和随机项。采用最小二乘法对趋势项进行预测；建立最小二乘支持向量机回归（LSSVM）模型对周期项和随机项进行预测。将各预测值累加求和，即得到累计位移预测值。以山口岩大坝为例，采用SSSCEMD-LSSVM 模型对厂址边坡位移时间序列进行预测。结果表明：模型能够有效预测位移时间序列，精度优于传统BP 神经网络和LSSVM 模型。

Prediction of slope displacement based on optimized empirical mode decomposition and least squares support vector machine

Landslide disasters frequently occur along the banks of reservoirs in China, and it is of great significance to use high-precision optimization algorithms to predict the displacement time series of slopes for disaster prevention and reduction. The displacement time series of slopes usually exhibit highly nonlinear characteristics, and traditional models have difficulty accurately predicting them. Therefore, this paper proposes a slope displacement time series prediction model based on optimized empirical mode decomposition (EMD) and least squares support vector machine (LSSVM). The model adopts EMD based on the soft sifting stop criterion (SSSC-EMD), which can adaptively decompose the displacement time series of slopes into multiple intrinsic mode components and one residual component. The residual component is defined as a trend term; the components are clustered using the K-means clustering method and defined as periodic and random terms. The trend term is predicted using the least squares method; the periodic and random terms are predicted using an LSSVM regression model. The predicted values are accumulated and summed to obtain the cumulative displacement prediction value. Taking the Shankouyan dam as an example, the SSSC-EMD-LSSVM model is used to predict the displacement time series of the slope at the site. The results show that the model can effectively predict the displacement time series with higher accuracy than traditional BP neural networks and LSSVM.