基于变分模态分解与GWO-MIC-SVR模型的滑坡位移预测研究

Displacement prediction of landslides based on variational mode decomposition and GWO-MIC-SVR model

针对目前滑坡位移预测研究中存在的随机性位移无法分解与预测、支持向量机(SVR)模型中输入向量权值无差异、最优训练组合无法确定等问题,基于时序分析理论和变分模态分解(VMD),结合灰狼优化算法(GWO)、最大信息系数(MIC)和SVR,提出一种新型滑坡位移预测模型。该模型首先采用时序分析法和VMD,将滑坡实际累积位移分解为趋势性位移、周期性位移和随机性位移。然后基于滑坡对各类影响因素的响应分析,为3种位移选择合适的影响因子,并采用GWO-MIC-SVR模型对各位移分量进行多数据驱动的动态单步预测。最终基于训练数据的时效性分析,确定最优训练组合,并应用时序加法模型将最优预测值进行叠加,实现对滑坡累积位移的预测。以三峡库区典型堆积层滑坡——白水河滑坡为例,选取监测点ZG93和ZG118从2004年1月~2013年4月的数据进行研究。结果表明,与以往的研究相比,该模型不仅有效预测时间较长,且预测精度较高,具有较高的实用性和推广应用价值。

This paper focused on some issues relating to the decomposition and prediction of stochastic displacement,the weight calculation of input vectors in support vector regression（SVR） and the determination of optimal training combination. A modified model of displacement prediction integrated with the gray wolf optimizer（GWO）,the maximum information coefficient（MIC） and the SVR,was proposed based on the time series theory and the variational mode decomposition（VMD）. In this model,the time series analysis and the VMD were firstly applied to decompose the cumulative landslide displacement into the trend displacement,the periodic displacement and the stochastic displacement. Subsequently,some reasonable inducing factors were selected according to the response analysis of landslide,and then the single step prediction supported by multiple data wasimplemented by using the GWO-MIC-SVR model. Finally,the optimal training combination was confirmed based on the timeliness analysis of training sets,and the optimal values were superposed to achieve the prediction of cumulative displacement. Baishuihe Landslide,a typical colluvial landslide in the area of Three Gorges Reservoir,was taken as an example. The monitoring data of ZG93 and ZG118 from January 2004 to April 2013 were analyzed. The results show that compared with previous studies,this model has longer period of effective prediction and higher accuracy in prediction.