Landslide Hazard Assessment Based On Uncertain Multi-classification Support Vector Machine Method

Landslide hazard is one of the most destructive natural disasters,which moves fast,spreads widely and is affected by many factors.As it is more difficult to predict.Our country is an area where natural disasters happen frequently.Because of the landslide hazard,economic losses reach ten billons of yuan and the number of casualties reach thousands of people,which not only has caused serious harm to our country’s economic construction,but also has a great threat to people’s life security.Therefore,efficient and scientific method selected has important guiding significance and practical significance for the futuristic study of landslide hazard prediction and forecast.Currently,there are many methods of landslide forecasting based data mining classification technique.Support vector machine is one of them,which is used to solve problems about high dimension and non-linear classification,has a solid theoretical foundation and simple mathematical model.Then it is widely used in the prediction of landslide hazard.Since landslide hazard is affected by many factors,in addition to basic factors,there is inducing factor rainfall,which is uncertain attribute value in a range and not exactly a value.Traditional support vector machine based sequential minimal optimization is for determination data and can’t handle uncertain data well,leading to that the traditional classification method in landslide hazard prediction accuracy is low.To solve the above problems,the paper construct hyper-ellipsoidal structure based mean and variance of uncertain values,to treat uncertain data rainfall,combined with traditional support vector machine based on sequential minimal optimization,uncertain multi-classification support vector machine is introduced.Landslide hazard forecasting model based on uncertain multi-classification support vector machine is designed,which uncertain data rainfall and other determination data of evaluation factors are brought into,the hyper-hyper-ellipsoid of space becomes line segment.Then the kernel function is put into the model,combining the grid search and particle swarm optimization algorithm to determine the model parameters,through the iterative optimization of the space segment to find the optimal classification surface,and then predict the landslide hazard level.Finally,Baota district of Yan’an city is chosen to do the prediction experiment with uncertain multi-classification support vector machine and traditional support vector machine method based sequential minimal optimization.The experimental results show that uncertain multi-classification method can not only achieve the accuracy standard of landslide hazard forecast,but also has the higher prediction accuracy than the traditional support vector machine method.Moreover it has certain feasibility and practicality.