Study On Prediction Model Of Single Debris Flow Risk Degree Based On SVM

Debris flow hazard is one of the major geo-hazards which often happen in the mountaineous areas,with the characteristics of recurrency,burstiness,massiveness,simultaneousness and transmit performance.It can destroy houses,factories,all kinds of transport facilities,block river,drown people and livestock,destroying the land even cause village crash disaster,which is a serious threat to people's life property safety.Therefore,reasonable and effective prediction of debris flow danger degree has important directive significance to disaster prevention and reduction work in mountain areas.This paper taked 37 debris flow ditches in Yunnan Province as study objects,28 debris flow ditches including Zhixi ditch,Shuimofang ditch,Xiaojing ditch etc.were chosen as study samples,9 debris flow ditches including Daqiaohe ditch,Heishui ditch,Xiaohai ditch etc.were chosen as test samples.Then rough set and principle component analysis were used to make reduction for 10 debris flow influencing indexes respectively,the newly generated data were taken as training sample of support vector machine.And the optimal SVM parameters was found by PSO.Finally the prediction model of debris flow danger degree based on support vector machine was established.And the research results were as follows:(1)Debris flow characteristics,classification,occurrence conditions,influence factors were discussed,at the same time,according to the selection principles of debris flow danger degree,10 indexes including the maximum volume of one-time debris flow,debris flow frequency,drainage area,main channel length,maximum relative height difference,valley incision density,vegetation cover within the watered,loose mass reserves,24 maximum rainfall,population decsity were used for debris flow danger degree prediction model.(2)In order to reduce the dimensions of debris flow danger degree indexes,rough set and principle component analysis were used to make reduction for original data.4 influencing indexes which including the maximum volume of one-time debris flow,debris flow frequency,main channel length,valley incision density were chosen as the core index set by attribute reduction,and the weight of core indexes were 0.7647,which can stand for the main information of whole debris flow danger degree indexes.And the cumulative contribution rate of 4 principal components determined by principal component analysis reached 84.221%,which included the main information of the10 influencing indexes and eliminated redundant information between variables,and had a significant dimension reduction effect.(3)RBF kernel function was selected as the kernel function of support vector machine,then particle swarm optimization algorithm was used to find the global optimum parameters of support vector machine,thus the optimal combination parameters of RS-PSO-SVM model were C=5.6392,g=1.6241,and the optimal combination parameters of PCA-PSO-SVM model were C=1.4817,g=1.1807.(4)RS-PSO-SVM model and PCA-PSO-SVM model were used to predict the test samples,the prediction accuracy of two models was 88.89%,and compared with the results of PSO-SVM,the results show that on the premise of the same prediction accuracy,RS-PSOSVM model and PCA-PSO-SVM model shortened the running time and improved the operation efficiency greatly.Moreover,It also proved that the support vector machine can express the complex nonlinear relationship between debris flow danger degree and the influencing indexes properly.