摘要
支持向量回归机(Support Vector Regression,SVR)在滑坡位移预测研究中已得到广泛应用,但SVR具有模型可解释性差的缺陷,即无法直接获得并筛选最佳预测变量,从而影响预测精度。为此,将较广泛应用于评价神经网络模型变量影响大小的平均影响值(Mean Impact Value,MIV)方法与SVR模型相结合,实现基于SVR-MIV的变量筛选,该方法不但能对所有预测模型初始变量影响大小进行排序,还可以进一步结合反向逐变量剔除分析实现变量筛选。为验证该方法的有效性,选择三峡库区两类典型水库滑坡代表的累积位移监测数据,在采用移动平均法将位移分解为趋势项和波动项的基础上,重点针对波动项位移,选择包括降雨及库水位变动特征在内的12项初始变量,采用SVR-MIV方法进行变量筛选分析。结果表明,该方法筛选出的变量理论上符合对应滑坡变形影响机理分析结论,且可以提高滑坡位移实际预测精度。
At present, Support Vector Regression（ SVR） is widely used in landslide displacement prediction. However,SVR has bad interpretability,which means the importance of each explanatory variable in SVR can’t be known,and then these significant variables can’t be selected to improve the prediction accuracy. The Mean Impact Value（ MIV） method had been broadly used to evaluate the effects of explanatory variables in neural network models,in this paper,the variable selection based on SVR-MIV method combining MIV with SVR is presented,which can not only sort all original variables according to their effects,but also filter out all important variables by using reverse-by-variable elimination. To check the effectiveness of SVR-MIV method,the cumulative displacement monitoring data of two typical reservoir landslides in Three Gorges reservoir area are used. Firstly,the moving average method is adopted to decompose the cumulative displacement into trend term and periodic term displacement. The periodic term displacement prediction accuracy become the key of an ideal total displacement prediction. Taking into account that the deformation of reservoir landslides is mainly affected by rainfall and reservoir water level fluctuation,12 original explanatory variables are picked to establish the SVR model,then the variable selection is carried out according to the SVR-MIV method. The results indicate that the selected variables conform to deformation impact mechanism of the two landslides,and can certainly improve the displacement prediction accuracy of landslide.
出处
《地下空间与工程学报》
CSCD
北大核心
2016年第1期213-219,共7页
Chinese Journal of Underground Space and Engineering
基金
国家自然科学基金(41302260)
湖北省自然科学基金创新群体资助项目(2015CFA025)
湖北省科技支撑计划(2015BCE070)
水电工程智能视觉监测湖北省重点实验室开放基金(2014KLA11)
关键词
边坡工程
滑坡
位移预测
变量筛选
支持向量回归机
平均影响值
slope engineering
landslide
displacement prediction
variable selection
support vector regression machine
mean impact value
