Study On Risk Assessment Of Landslide Based On Machine Learning

Landslides are a common geological hazard that threatens the environment in which people live,work,and study.There are many mountainous areas in Sichuan Province,with well-developed geological structures,complex topographical conditions,and wide distribution of landslide hazards.Therefore,study on landslide risk is of great significance to reduce the losses caused by landslide disasters,and it can also provide important decision support and scientific basis for landslide treatment.This thesis analyzes the influencing factors of landslide disasters,and extracts the characteristics of landslides,and processes the extracted characteristics.Using the machine learning methods,risk assessment models of landslide were established,and models were tuned.By verifying and comparing the assessment results,the model with the best performance was selected and applied to the landslide in Ganluo County.The main research content and results of the thesis are as follows:(1)Based on the collected landslide data,the influence factors of the landslide were analyzed,and the Pearson correlation coefficient was used to test the correlation between the influence factors and the risk of landslides.Then the landslide features were extracted from the landslide impact factors,and the multicollinearity test was performed on the extracted landslide features using the variance inflation factor.The results showed that there was multicollinearity between the features.The method of Principal Components Analysis(PCA)was used to eliminates multicollinearity between features.(2)Based on machine learning methods,using Support Vector Machine(SVM),Random Decision Forests(RF),Extreme Gradient Boosting(XGBoost),Light GBM(Light Gradient Boosting Machine)algorithm and Stacking Algorithm,established risk assessment models of landslide.Using grid search and cross-validation to optimize models.Using methods such as accuracy,confusion matrix,and Kappa coefficient to and evaluate the model.After testing and comparing all models,the results show that the accuracy of Stacking model is 0.95,and the Kappa coefficient is 0.93,which is better than other models.(3)The Stacking model is applied to the landslide risk assessment in Ganluo County.The results show that among the 97 landslides in the county,a total of 73 landslides are located in the high-risk areas of the landslide risk assessment,accounting for 75.26% of the total landslides.The landslide risk assessment results are consistent with the actual landslide distribution,indicating that the landslide risk assessment model is effective.(4)The results obtained by the landslide risk assessment model constructed in this thesis can make a clear judgment on the risk of landslides.In the prevention and treatment of landslides and emergency treatment of landslide disasters,the results of this thesis can be used as the basis for disaster prevention work and the reference for decision-making of treatment projects.