A Combined ICEEMDAN-CNN-SVR Landslide Displacement Prediction Model
Landslide displacement prediction is an important component of early warning systems for landslides.In response to the problem of low accuracy in landslide displacement prediction caused by insufficient depth of displacement decomposition and feature selection,an ICEEMDAN-CNN-SVR landslide displacement combination prediction model is proposed.To solve the problem of insufficient displacement decomposition,the ICEEMDAN decomposition model was first used to decompose the landslide displacement curve.The IMF curve with good smoothness and increasing trend was used as the trend term displacement,and the sum of other IMF curves with fluctuating trend is reconstructed as the periodic term displacement.In order to solve the problem of insufficient depth in feature selection,feature variable selection was carried out for different displacement characteristics.Through two-dimensional tiling and CNN feature extraction,deeper information about the feature variables was obtained.The extracted feature information was input into the SVR prediction model to achieve accurate prediction of trend displacement and periodic displacement.Taking the typical accumulation layer landslide,i.e.,Bazimen landslide,as an example,the horizontal displacement data of ZG110 and ZG111 monitoring points during the typical deformation stage from January 2007 to September 2012 were selected for research.The research results showed that the prediction evaluation indicators,R2,ERMSE,and EMAE of ZG110 and ZG111 monitoring points were 0.995 1 and 0.998 9,5.748 9 and 2.753 2,4.509 1 and 1.852 9,respectively,with good prediction effect.Comparing the model prediction results with the EEMD-CNN-SVR prediction model and CNN-SVR prediction model,the comparison results show that the new model has improved prediction accuracy compared to the other prediction models.
万方数据
