基于RS-SVM模型的滑坡易发性评价因子选择方法研究

Research on the Method to Select Landslide Susceptibility Evaluation Factors Based on RS-SVM Model

为提高滑坡易发性评价的精度和准确性,利用粗糙集理论和支持向量机模型进行滑坡易发性评价因子选择。以三峡库区秭归到巴东段为研究区,首先通过相关分析和主成分变换对初始滑坡评价因子进行筛减与合并,将剩余的评价因子分为控制因素和影响因素,采用粗糙集理论分别对其进行属性约简,生成核因子集。然后,利用核因子集对支持向量机模型进行训练和预测,得到滑坡易发性评价结果,其预测精度和曲线下面积分别为76.9%和0.939。最后,将两种传统的对所有滑坡评价因子一同进行筛选的方法用于该研究区的滑坡易发性评价,并将传统方法与该方法的评价结果进行比较,结果表明,该文所提出的滑坡评价因子选择方法的精度和准确性均优于传统方法。

The evaluation of regional landslide susceptibility is very important for the medium and long-term prediction and forecast of the landslide hazard.Whether the evaluation results are in line with the actual situation depends on the scientific and reasonable determination of evaluation factors.In order to improve the precision and accuracy of evaluation result,this paper proposed a method to select evaluation factors of landslide susceptibility based on Rough Set theory（RS）and Support Vector Machine model（SVM）.Zigui to Badong section in the Three Gorges Reservoir Area is used as a study area.Firstly,the original evaluation factors obtained from remote sensing image and digital elevation model are screened and combined by correlation analysis and Principal Components Analysis（PCA）.The remaining evaluation factors were divided into control factors and influence factors,and attribute reduction is carried out on the control factors and the influencing factors by RS respectively to generate core factor set.Then,SVM model is used to train and predict the core factor set selected to get the results of the regional landslide susceptibility evaluation,the precision and accuracy of which are 76.9% and 0.939 respectively.In order to make a comparison between the proposed method to select valuation factor and those traditional ones which screen all environmental factors at the same time,the existing two traditional methods are also used for the landslide susceptibility evaluation in this study area,and the precision and accuracy of which are 53.6%,0.912 and 72.8%,0.932 respectively.The results show that the proposed method to select landslide susceptibility evaluation factors in this paper is better than the traditional methods.