WA-BT-ELM耦合模型在黄土滑坡位移预测中的应用

发布时间：2018-08-28 20:26

【摘要】：黄土滑坡的变形演化过程往往受到多种因素的影响,呈现出非线性特征。基于小波分析函数(Wavelet Analysis,WA)、提升回归树(Boosting Regression Tree,BT),以及极限训练机(Extreme Learning Machine,ELM)方法,提出一种名为WA-BT-ELM的黄土滑坡位移预测新方法。该方法将非线性位移数据作为一时间序列,运用小波分析函数将监测点累积位移曲线分解为若干子小波;随后使用提升回归树对所有子小波进行重要度分析,剔除相关性不高的子小波以去掉冗杂信息;最后运用极限训练机,结合筛选得到的子小波对滑坡位移进行预测分析。基于该模型对甘肃省永靖县黑方台滑坡区的滑坡位移监测数据进行预测,得到了优于ANN,BPNN,SVM,ELM,以及WAELM预测模型的结果,故认为WA-BT-ELM模型是一种有效的黄土滑坡位移预测方法。
[Abstract]:The deformation and evolution process of loess landslide is often influenced by many factors, showing nonlinear characteristics. Based on wavelet analysis function (Wavelet Analysis,WA), lifting regression tree (Boosting Regression Tree,BT) and limit training machine (Extreme Learning Machine,ELM), a new method called WA-BT-ELM for predicting loess landslide displacement is proposed. In this method, the nonlinear displacement data is taken as a time series, and the cumulative displacement curve of monitoring points is decomposed into several subwavelets by wavelet analysis function, and then the importance of all subwavelets is analyzed by lifting regression tree. In order to remove the miscellaneous information, the subwavelet with low correlation is eliminated. Finally, the landslide displacement is predicted and analyzed by using the limit training machine and the selected subwavelet. Based on this model, the monitoring data of landslide displacement in Heifangtai landslide area of Yongjing County, Gansu Province are predicted, and the results are better than those of ANN,BPNN,SVM,ELM, and WAELM model. It is considered that WA-BT-ELM model is an effective method for prediction of loess landslide displacement.
【作者单位】：成都理工大学地质灾害防治与地质环境保护国家重点实验室;山东大学(威海)数学与统计学院;爱荷华大学智能系统研究实验室;
【基金】：国家重点基础研究计划(973计划)资助项目(2014CB744703) 国家杰出青年科学基金项目(41225011) 教育部“长江学者奖励计划”项目(T2011186)
【分类号】：P642.22

【相似文献】