Multiphysics Spatio-Temporal Response,Monitoring,and Early Warning Of Landslides Under Rainfall Conditions

Landslides are the most widely distributed and hazardous geological hazards in China,causing huge economic losses and human casualties every year.Rainfall is one of the main factors triggering landslide instability,and monitoring and early warning of such landslides is an economical and effective means of prevention.However,the current research on landslide monitoring and early warning has shortcomings such as small sensing coverage,high cost and low utilisation of data acquisition,shallow depth of multi-source monitoring data mining,and immediate and accurate warning.This thesis addresses these shortcomings and focuses on the core scientific problem of "how to use multi-source monitoring data on the spatial and temporal response of landslides under rainfall conditions to dynamically evaluate and predict the evolution of landslide safety status for rapid,accurate and reliable early warning of landslides".The research work has been carried out in the areas of multi-physics landslide safety evaluation,multi-source monitoring data analysis and early warning algorithm and system optimization,etc.The research results are as follows:(1)Taking the Lijiazu landslide in Hunan Province as the research object,COMSOL software was used to numerically simulate the spatiotemporal response of the landslide seepage field,plastic strain field and displacement field during the rainfall of this landslide,and a selfprogrammed adaptive finite element DC method simulation program based on unstructured grid was used to realize the numerical simulation of the spatiotemporal response of the landslide geoelectric field during the rainfall,and through the comparison and analysis with the field measured data,the spatiotemporal response characteristics of the landslide multiphysical field under the rainfall conditions were summarized,which laid the foundation for the subsequent safety evaluation.(2)A dynamic safety evaluation algorithm of landslide rainfall process based on Electrical Resistivity Tomography(ERT)data and numerical simulation is proposed.Firstly,an incomplete Gaussian Newton inversion algorithm based on multiple constraints is used to obtain the resistivity distribution for ERT data.Then the Water-Electricity Correlation Curve(WECC)is converted into the saturation distribution.Finally,an elasto-plastic finite element analysis considering the unsaturated state is used to calculate its mechanical state,and then the safety factor of the landslide is obtained by the intensity reduction method to realize the dynamic safety evaluation of the landslide.In order to verify the performance of the proposed algorithm,12 sets of numerical simulation experiments are carried out.The experimental results show that the proposed algorithm can effectively and reliably evaluate the safety state of landslides under different acquisition devices,different apparent resistivity noise levels and WECC curve uncertainties,and the performance of the dipole device is better than other devices.(3)For the problem that the multi-source monitoring data obtained from conventional physical field monitoring is not deep enough in analysis and mining,a landslide displacement interval prediction algorithm based on Bootstrap-VMD-LSTNet is proposed.The algorithm firstly decomposes the cumulative displacement into trend displacement,periodic displacement and random displacement using Variational Mode Decomposition(VMD)algorithm with minimum sample entropy constraint,and decomposes the features into high frequency component and low frequency component.Then the trend displacement is predicted by an optimized polynomial function fitting method,and the feature factor components are predicted by LSTNet deep learning network for the periodic displacement and random displacement.Finally,the three prediction results are summed to achieve displacement point prediction.The Prediction Interval(PI)at different confidence levels is also constructed using Bootstrap algorithm to evaluate the uncertainty of prediction results.The results of the application of the algorithm to the Baishuihe landslide in the Three Gorges reservoir area show that the point prediction results of the proposed algorithm are better than the baseline models such as LSSVR,BP and LSTM,and the quality of the PI constructed at 90%,95% and 99% confidence levels is also better than that of the baseline model.(4)To address the challenge of how to mine short-term disaster patterns of landslides from real-time,massive multi-source monitoring data for landslide early warning.A PSO-Kmeans-Apriori based algorithm for association rule mining of landslide multi-source monitoring data is proposed.The algorithm first uses the sliding window method to extract features from the pre-processed single-source monitoring data,then uses the PSO-optimized Kmeans algorithm to cluster the data from each monitoring source,and finally uses the Apriori algorithm to mine the strong association rules related to the landslide high-speed deformation process for the clustering results to obtain the landslide short-term disaster patterns.The data mining results of this algorithm in the Lishanyuan landslide and the Lijiazu landslide in Hunan province show that the short-term deformation pattern of landslide mined by this algorithm can be used as the precursor of the high-speed deformation of landslide,which can effectively improve the accuracy and reliability of early warning and realize the association rule mining of hour-scale or even minute-scale monitoring data,which is more concerned with the short-term deformation pattern compared with the ultra-long-term monitoring data analysis of monthly scale and more It is conducive to short-term real-time early warning.(5)Aiming at the problem of inaccurate early warning caused by high frequency,high noise and high missing rate characteristics of automated monitoring data,a hybrid early warning algorithm is proposed,which combines the conventional early warning method based on cumulative displacement,velocity and acceleration and the critical early warning method based on normalized tangent angle according to different strategies according to different needs,and optimizes the calculation process by various improvement measures.The early warning results of the proposed hybrid warning algorithm on the landslide of Lishanyuan and the landslide of Lijiazu in the Hunan Province show that the hybrid warning algorithm can not only identify the landslide accelerated deformation process quickly,but also effectively reduce the number of false warnings and improve the accuracy and reliability of the warnings.(6)To address the problem of how to monitor and warn quickly,accurately and reliably in large-scale landslide monitoring and warning scenarios,we propose to design and develop a landslide monitoring and warning system using a microservice architecture,which enables the decentralization of system services and computational pressure through distributed load balancing to ensure that the system provides fast warning services and is fault-tolerant at the same time.The historical application of the developed landslide monitoring and early warning system shows that the microservice architecture enables the system to have excellent performance in terms of response efficiency,fault tolerance and stability,and ensures that the system can provide various services,including early warning services,quickly and stably even when large-scale landslides are accessed.