Research On Deformation Monitoring And Prediction And Warning Methods Of Landslides In Karst Mountainous Areas

The karst mountainous areas in southwestern China have a lot of rainfall,complex geological structures,and landslide disasters are fully developed and frequent,which has caused major threats to the lives and properties of local people.Real-time comprehensive monitoring,accurate prediction and early warning have become the actual needs and top priorities of emergency response,disaster prevention and reduction.With the continuous advancement of field monitoring,monitoring data has grown rapidly,analysis and processing difficulties have increased,server load intensity has increased,and efficient real-time processing,analysis,and release of information results are unsustainable.So it is urgent to explore new data computing frameworks and mine effective prediction and early warning models and advance in depth.For this reason,this paper takes the typical karst mountainous area demonstration site—Faer Town landslide in Guizhou as an example.First,apply the Hadoop computing platform for storage management and preprocessing of the raw data collected in the field;secondly,apply the time convolutional network model to the study of landslide deformation prediction and evaluate the accuracy;finally,the instability criterion of the landslide in this area is analyzed based on the slope deformation mechanism and actual monitoring data in the study area.The main research content and results of this paper are as follows:(1)The data integration and preprocessing work is completed based on the Hadoop platform.First,deploy GNSS and rain gauge equipment to continuously observe the landslide deformation information and meteorological data in the study area;secondly,use database synchronization technology and crawler technology to synchronize remote data to a local database,and then upload it to the distributed file system under the Hadoop computing platform In order to complete the integration of monitoring data;finally relying on the MapReduce calculation framework and related algorithms to achieve gross error elimination and interpolation of GNSS monitoring data of karst landslides.(2)Establishing a karst landslide deformation prediction model based on Temporal convolutional network and its applicability.The Temporal convolutional network model is introduced into the karst landslide deformation prediction and experiments are carried out with the FE04 and FE06 monitoring points as examples.The results show that the root mean square error and the mean absolute percentage error of the FE04 prediction by the model are 10.5mm and 0.1333%,respectively.The predicted root mean square error and mean absolute percentage error are 16.5mm and 0.066% respectively,compared with the long and short-term memory network model,the BP neural network model and the support vector regression model,the Temporal convolutional network model has higher accuracy and better applicability.(3)Analyze the landslide instability criterion in the study area.Based on the analysis of the landslide deformation mechanism in the study area and the actual changes in the cumulative displacement of each monitoring point,a segmented improved tangent angle model is proposed on the basis of the existing model and four warning levels are set to conduct corresponding experimental analysis on the data of each monitoring point.The results show that the improved model can issue a warning at the right time when the deformation of the landslide body accelerates and cancel the warning when the deformation trend eases.The warning result is consistent with the actual deformation of the slope body.