Application Of Data Mining Technology In Landslide Monitoring And Early Warning

As a common geological disaster,landslide has the characteristics of wide range and high risk in China.Among all kinds of geological disasters,its quantity accounts for the highest proportion,which seriously threatens people’s life and property safety.How to effectively monitor,identify,early warning,prevent and control landslide and reduce disaster losses is the focus of management departments and relevant scholars.In the long-term landslide prevention and control process,a large amount of survey data,monitoring data,historical cases,prevention and control experience and expert knowledge have been accumulated.As the first-hand data of historical disaster prevention and control,it has important reference value for analyzing the temporal and spatial development law of geological disasters,determining the influencing factors of geological disasters,and establishing the index system of geological disaster identification,evaluation and early warning.Predecessors have carried out a lot of research work on the development characteristics,disaster causing mechanism and influencing factors,monitoring and early warning of landslide,but combined with historical cases,further research is needed in landslide monitoring and early warning.In order to explore the useful information of historical case data to assist landslide prevention and control,the research on the construction of landslide monitoring index system and early warning model can provide technical support for landslide prevention and control.Relying on the key technology research on the construction of community geological disaster avoidance guidance system(212102310389),a scientific and technological research project of Henan Province,based on the national typical historical landslide data,combined with the relevant data of regional geography,geology,meteorology,hydrology,transportation and land use types,and based on the construction of landslide monitoring index system in different regions,this paper introduces data mining technology to carry out the research on the construction of landslide monitoring and early warning model,The following achievements and understandings have been achieved:(1)The basic database of landslide monitoring and early warning is created.Collected and sorted out the data of 959 landslides in China from 2010 to 2021;2916983 daily rainfall and 2133717 hourly rainfall data from 779 meteorological stations in China;61825 lithology and geological age records,6073 landforms,32522 faults,6102 rivers and 94843 roads;In addition,the national land use data,DEM data and remote sensing images are collected.Based on the unified coordinate system and scale of the above data,dealing with missing and abnormal data,data coding and dimensionless,the basic data of landslide monitoring and early warning research is formed.(2)The data mining algorithm suitable for historical landslide cases is defined.Using the sklearn Library in Python and based on historical landslide data,the landslide prediction model is established by using logistic regression,support vector machine,random forest and limit gradient lifting algorithm.The super parameters of the four models are optimized by grid search method,the confusion matrix of the four models is calculated,and the ROC curve is drawn.The accuracy of the model is analyzed and compared by using model evaluation indicators such as accuracy,accuracy,recall and kappa coefficient,Random forest is determined as the data mining model of this study.(3)This paper attempts to build a dynamic index system of landslide monitoring suitable for different regions.Landslide related data are collected through various channels to form a multi-source data set.After preprocessing the data set,the monitoring data of 959 typical historical landslide points and 16 landslide impact factors in China are extracted.The importance of all impact factors is analyzed by using the random forest model.Finally,12 main factors are selected to construct the landslide monitoring index system in the whole country.The order of importance is NDVI,15 day average rainfall Slope,lithology,river density,daily rainfall,road density,elevation,fault density,land use type,geological age and slope position;12 main factors such as NDVI,daily rainfall,lithology,elevation,slope,river density,road density,plane curvature,fault density,15 day average rainfall,profile curvature and geological age are selected to construct the regional landslide monitoring index system in Sichuan Province.Comparing the two landslide monitoring index systems,the following conclusions are drawn: there are differences in landslide monitoring index systems in different regions,but they all include natural and human activities;Landform and river density are the decisive factors affecting the development of landslide;Rainfall is the direct inducement of landslide development;In the process of road construction,slope excavation is the key index to accelerate the occurrence of landslide;Landslide monitoring needs multi-source data such as natural geography,geology,meteorology,hydrology,transportation and remote sensing as support.When monitoring landslide in multiple regions,regional differences should be considered,which should not only pay attention to the daily rainfall,but also consider the average rainfall in recent 15 days.(4)Landslide monitoring and early warning models in different regions are established.Based on the landslide monitoring index system,the landslide monitoring and early warning model of national and Sichuan administrative units is established by integrating three aspects: the occurrence probability of regional landslide,the importance of landslide factors and the frequency ratio of landslide factors.The results show that the early warning models of national administrative unit and Sichuan administrative unit predict that all historical landslide points are within the early warning range.Among them,the results of the national administrative unit early warning model show that the proportion of landslide points in low-risk early warning areas is 13.97%,that in medium risk early warning areas is 27.22%,that in high-risk areas is 58.81%,and that in medium and high-risk areas is 86.03%;In the early warning model of administrative units in Sichuan Province,the proportion of landslide points in low-risk,medium risk and high-risk early warning areas is 17.11%,29.61% and 53.29%respectively,and the proportion of landslide points in medium and high-risk areas is82.9%.The results of landslide early warning and classification are highly consistent with the distribution of historical landslide points,which shows that the multi regional landslide monitoring and early warning model based on data mining technology can be used for the risk assessment of landslide development in a large range and different regions.There are 74 figures,43 tables and 125 references.