Study On Hazard Assessment Of Landslide In Sichuan Province Based On BP Neural Network And Optimized Algorithm

There are frequent geological disasters occurring in China every year.These geological disasters are mainly distributed in mountainous and rainy area,such as Sichuan Province.Sichuan Province is located at southwest China,where the terrain is very complex.After several severe earthquakes,the soil becomes more and more loose,and lots of deposits are generated.Therefore,the hidden dangers of geological disasters in Sichuan Province are getting larger.The area needed to be assessed is increasing.Landslide is the main geological disaster in Sichuan Province,and the primary work is to prevent the hazard of landslide.The hazard assessment of landslide in regional scale is an effective method to reduce landslide risk to people,and it can provide the guidance for the design,construction and operation of engineering projects,as well as government decisions on disaster prevention and mitigation.The hazard assessment of landslide in regional scale can be carried out based on many different mathematical models.Among them,the BP neural network is famous for its generalization and capability to solve nonlinear problems,and it is the most widely used model.Recently,worldwide researches have used the BP neural network to do hazard assessment of landslide,and achieve some results.However,the BP neural network will be troubled by accuracy and computing speed reducing when the area to be evaluated is getting larger,which limits the development of the BP neural networks.To establish a landslide hazard assessment method that is suitable for Sichuan Province,this study introduced the BP neural network and optimizes it to get higher accuracy.The content and results of this study are as follows:(1)100 landslide samples were collected in the study.Based on the results of field surveys,literature summaries and expert experience,this paper analyzed the distribution and the main influencing factors of landslides in Sichuan Province.And then,a landslide hazard assessment system with elevation,vegetation index,slope angle,average annual rainfall,surface cutting density and overburden soil type was established.(2)To meet the demand of input data for the BP neural network,this paper proposed the data processing standard that is suitable for the landslide hazard assessment in Sichuan Province.Based on the data processing standards,the data of 100 landslide samples was normalized,and the database was well prepared for the BP neural network.(3)The landslide hazard assessment system in Sichuan Province was carried out based on the BP neural network,and to optimize the accuracy of the existing models for a large region,the genetic algorithm(GA)and particle swarm optimization(PSO)are,respectively,coupled with the backpropagation(BP)neural network to determine the initial weights and thresholds in the BP neural network,which can be called GA-BP model and PSO-BP model.This paper analyzed the applicability of these three models.Results showed that the numerical accuracy and classification accuracy of the GA-BP model were respectively 22.6%and 44.3% higher than the BP model.The numerical accuracy and classification accuracy of the PSO-BP model were respectively 5.1% and 15.4% higher than the BP model,and the time was reduced.The GA-BP model was more suitable for use cases where accuracy is priority,and the PSO-BP model was more suitable for use cases where computing speed is priority.(4)This paper drew the landslide hazard zoning maps of the whole Sichuan Province and the area along Chuanzhu Temple-Jiuzhaigou highway.The maps matched well with the results of field surveys,which illustrated that the mapping performance of GA-BP model for hazard zoning under different areas and different scales is accurate enough.