Research On Landslide Hazard Zoning Method Of Railway Based On Machine Learning

Landslide is one of the most significant geological hazards in China,which threatens resident safety and engineering construction.The construction of the XX railway faces extremely harsh geology,topography and natural environment.It is very significant to evaluate the landslide hazard risk in the line area for preventing landslide hazards.Based on the machine learning model and extenics method,the sectional landslide risk level of the XX railway is evaluated.The main steps are as follows:(1)According to literature research and the natural geological environment of eastern a certain province,9 landslide influencing factors,namely the elevation,slope,lithology,average annual rainfall,seismic peak ground acceleration(SPGA),distance to the fault zone,distance to the water system,distance to the highway and normalized vegetation coefficient(NDVI)are selected to construct an index system of the spatial landslide susceptibility evaluation.(2)Based on the historical landslides,the entropy-simulated annealing(ESA)algorithm is proposed to optimize the secondary factor interval division of some landslide influencing factors,namely the elevation,slope,NDVI and annual average rainfall.It reveals the internal correlation between the secondary factors and historical landslides.It indicates the three hazard factors with the most significant correlation between secondary factors and historical landslides are NDVI,elevation and average annual rainfall.(3)Based on the two datasets obtained using the natural breakpoint method and ESA algorithm,the landslide susceptibility mappings of the study area are obtained respectively using the three machine learning models,namely the Naive Bayesian Model(NBM),Classification and Regression Tree(CART)and Random Forest Model(RFM).The results show that the extremely high and high susceptibility area is mainly distributed in the northeastern part.The accuracy of the results is evaluated by the relative frequency density and ROC-AUC.The study concludes that the results obtained using the RFM model have the highest accuracy.(4)Three railway landslide risk factors,namely the landslide susceptibility of the railway section,the impact of historical landslides on the railway and the engineering type of railway section,are selected.The weight coefficients of the three factors are determined using the fuzzy analytic hierarchy process.The landslide risk of the railway section is comprehensively analyzed using the matter-element extension theory.The results indicate that the railway sections with the high risk and extremely high risk are DK920+900～DK923+814 and DK1006+100～DK1010+151.There are 28 bridge abutments and tunnel entrances in high-risk and extremely high-risk road sections.The results have been adopted by the XX railway design department,and the railway sections have taken the landslide prevention measures.20 figures,32 tables,and 73 references.