Research On Landslide Prediction Technology Based On Resistivity Imaging Technology And Multivariate Data Fusion

Landslide deformation evolution law is complex,and quantitative evaluation of its stability is an important basis for landslide deformation damage mechanism research and monitoring and early warning.The resistivity imaging technology has the characteristics of rapid and multi-dimensional profile imaging according to the difference of geotechnical composition and structure and the difference of electrical properties between strata,which can effectively distinguish the spatial and temporal distribution range of water content of landslides and the location of potential slip surface.In this paper,we construct a multi-data neural network with time-space-attribute characteristics of resistivity imaging data and related monitoring items’ neighborhoods,and analyze the resistivity data layer by layer and recursive regression by layer to solve the problem that the resolution of resistivity imaging technology decreases rapidly as the monitoring range and depth increase.By constructing a unified expression matrix of the relationship between the neighboring domain features of each monitoring term,a simplified interactive multi-model based on threshold discrimination and a traceless Kalman filter prediction algorithm are proposed to achieve rapid convergence of the spatial weights of the main parameters in the proslip phase of the numerical simulation,which effectively improves the real-time performance of the early warning model for geopower monitoring.The main results achieved are:(1)A resistivity imaging monitoring system and electrode network applicable to landslide monitoring were designed.On the basis of the ECUTRES-III distributed electrophotometer studied in the previous stage,modules such as arbitrarily expandable electrical measurement sub-stations,intelligent electrodes and wireless data transmission were designed to meet the needs of long-term monitoring in the field;and various electrode networks were designed according to the needs of large area of geological disaster hidden site identification and variable density and range monitoring in key areas.The structural and compositional characteristics of the mounded layer landslide,which accounts for the largest proportion of all landslide types and the most serious hazard in China,are analyzed.Based on the characteristics of this type of landslide such as large differences in structural composition and obvious differences in electrical characteristics among layers,combined with the advantage of higher sensitivity of resistivity imaging technology to respond to the differences in electrical characteristics caused by water content,the scope of resistivity imaging monitoring technology applied to the field of landslide monitoring and prediction was determined as mounded layer landslides.(2)An effective method to improve the resolution of the imaging monitoring data of the internal structural electrical information of landslides is studied.A nonlinear regression analysis algorithm based on the deep neural network of resistivity imaging technology is proposed to construct a "spatial-temporal-attribute" relationship between various types of monitoring data,shallow sample points of resistivity imaging monitoring data and deep fitting points of resistivity imaging monitoring data by adding "attribute" correlation information.The spatial-temporal-attribute" relationship neural network is constructed to obtain the unified expression matrix of spatio-temporal attribute relationship in the neighboring domain,and the neighboring domain weight depth neural network is constructed with this matrix as the input to obtain the spatial weight matrix of all data,and finally the fused output data of the fitted points of resistivity imaging monitoring data are obtained by multivariate nonlinear regression analysis.And the electrical characteristic model of multivariate data fusion is constructed to accurately quantify important parameters such as slip surface boundary change,rainfall infiltration boundary,head boundary and flow boundary during the slope deformation.The algorithm effectively improves the accuracy of resistivity imaging deep monitoring data and provides reliable input data for the subsequent prediction algorithm of landslide deformation process.(3)The prediction algorithm of landslide internal structure imaging based on resistivity imaging technology is studied.The prediction algorithm of landslide internal structure deformation based on simplified interactive multi-model and traceless Kalman filter is proposed.According to the composition of landslide material and geological structure,as well as the sensitivity difference of various acquisition devices of resistivity imaging technology,the number and type of models of devices involved in the interaction are selected reasonably.The Markov transfer probability of each device in the model set is gradually adjusted,and the instability state of the landslide is discerned by the threshold.When the landslide is in the stable stage or creep-slip stage,a single device model is used for the trace-free Kalman filter prediction.When the landslide is found to be in the accelerated deformation or proslide stage,the interactive output of multiple models is used to improve the prediction accuracy.With the widespread application of resistivity imaging monitoring technology in the field of landslide monitoring,the accuracy of prediction results is ensured while the real-time prediction is effectively improved.Based on the advantages of the numerical model of landslide geoelectricity characteristics,this study integrates the monitoring data of soil water content,rainfall and macroscopic deformation of slope body,solves the key problems of multi-solution of geoelectricity information model and the gradual decrease of resolution with the increase of detection depth,and realizes the fast and accurate response to the sudden change signal of underground water migration during the evolution of disaster.The research results have wide application prospects in the fields of landslide evolution mechanism research,deformation process monitoring and imaging prediction,and provide reliable data support for the study of landslide damage theory and actual disaster prevention and mitigation work.