Evaluation Of Railway Debris Risk Assessment Based On Support Vector Machine

Debris-flow is a widely distributed geo-hazard in mountainous area. And due to its complicated terrain and geological structure, debris-flow is especially very common in K178-K728 of ChengKun railway which runs through Dadu river valley, Niuri river valley, Sunshui river valley, Wajimuliangzi mountainous area and Anning river valley, greatly threatening the safety of railway in the research area, so it is needed to evaluate the risk and hazard of debris-flow in this area. Support Vector Machine (SVM) is a new machine learning theory based on statistics of machine, seeking the nonlinear relationship between input variables and output variables with the help of sample training, thus judging the unknown input target more accurately.In this paper,53 debris-flow samples have been selected to analyze and study the rules and characteristics of danger of debris-flow and geological and environmental conditions in research area by using SVM and samples learing, and SVM model suited to hazard assessment of debris-flow in research area has been raised in this paper and some research results are as follows:1) K178-K510 railway runs through high mountainous area, which are affected by Hanyuan-Ganluo fault, Shimian-Puxiong fault and Mishi syncline, and debris-flow can be especially easily formed in this rear because of its large longitudinal gradient of gully and steep terrain; There are a large number of loose solid material in research area due to K510-K728 railway running through Anning river fault, greatly influence the stability of bank slope of this area, which can be benefit for the formation of debris-flow;2) SVM model of debris-flow hazard assessment has been put forward, in which radial basis function (RBF) can be selected as kernel function;3)53 debris-flow samples have been selected as population to globally optimize penalty factor of SVM and kernel function and determine model parameter of SVM, in which penalty factor C=263.8 and kernel function G=0.0277;4) 39 debris-flow samples have been selected as training set to train SVM model and construct SVM debris-flow assessment model which is fit for geological and environmental conditions in research area; and 14 debris-flow samples have been selected as testing set to verify the SVM debris-flow assessment model, and the precision result is 92.8571%(13/14).