Application Of GRA-RF In Landslide Risk Assessment Along Railway

The railway line in the study area is located in the Ya’an Batang section in the west of Sichuan Province.The climate in this area is changeable,the topographic and geological conditions are complex,and landslide disasters occur frequently.With the construction of the Sichuan Tibet railway,the length of the railway line in Western Sichuan is increasing,and the safety of railway lines and train operations are vulnerable to landslide geological disasters.Therefore,carrying out landslide risk assessment in the study area and calculating the possibility of landslide disasters along the railway is of great significance for the safe operation of trains and the safety protection of railway lines.Firstly,based on the historical landslide data in the study area,the landslide evaluation factors are pre-selected based on analyzing the environmental characteristics of landslide disasters.Secondly,principal component analysis,rough set,and grey correlation analysis are selected to screen the evaluation factors,and three different combinations of evaluation factors are obtained.Then,the pre-selected evaluation factor combination and the screened three evaluation factor combinations are input into the three evaluation models of random forest,support vector machine,and logistic regression respectively,and a total of 12 combination methods are obtained.Finally,according to the prediction accuracy of the model corresponding to the 12 combination methods,the best evaluation factor combination and model are selected,and the landslide risk of the study area is evaluated.The main research contents are as follows:(1)Based on the disaster pregnant factors and causes of landslides in the study area and previous studies,the topographic and geomorphic evaluation factors,namely elevation,slope,slope direction,plane curvature,and profile curvature,are pre-selected;Meteorological and hydrological evaluation factors,i.e.rainfall and distance from river;Geological condition evaluation factor,i.e.engineering rock group,distance from fault;Vegetation cover evaluation factor,namely normalized vegetation index;There are 12 evaluation factors for human engineering activities,i.e.land use type and distance from the road.Then the evaluation factors are classified,and the distribution frequency of historical landslide disasters in different grades of each evaluation factor is counted.(2)The pre-selected evaluation factors are tested for collinearity by calculating tolerance and variance expansion factor to ensure that the evaluation factors are independent of each other.Then,principal component analysis,rough set,and grey correlation analysis are selected to analyze the correlation of evaluation factors,select the factors with large contribution rate or nuclear factors,and eliminate the factors with poor correlation,to obtain three different combinations of evaluation factors.(3)Four combinations of pre-selected and screened evaluation factors were input into three evaluation models: random forest,support vector machine,and logistic regression,and the performance of the model was evaluated by confusion matrix and subject working characteristic curve.The evaluation results show that the combination of evaluation factors screened by grey correlation analysis has the best performance when input into the corresponding model of random forest,and the accuracy of the confusion matrix and the area under the working characteristic curve of subjects are the largest,which are 0.875 and 0.8806 respectively.(4)The best evaluation combination and model are used to evaluate the landslide risk,calculate the landslide risk,and generate the landslide risk distribution map.Using the natural discontinuity method,the risk assessment results are divided into five risk levels: low,low,medium,high,and high.By comparing the landslide risk assessment results with the landslide verification points,the test accuracy is 86.5%.Therefore,this paper can provide a reference for the risk assessment of landslide disasters along the railway in the study area.