| Geological disasters are frequent in China,especially landslide disasters account for a relatively high percentage.Due to the perennial influence of rainfall,reservoir water fluctuation and geological activities,the Three Gorges reservoir area has become a high and vulnerable area for landslides.The comprehensive analysis and trend prediction of landslide deformation based on surface deformation monitoring data is an important basis for judging the evolution of deformation and stability of the bank slope in the Three Gorges reservoir area.Based on the non-contact measurement technology,the deformation data of the landslide reservoir water level change area can be supplemented,and the overall deformation characteristics of the landslide can be recognized,which is a basic research of data-driven landslide deformation monitoring and early warning forecasting,and has important theoretical and engineering significance for landslide disaster prediction and prevention.By analyzing the characteristics of rocky landslides in the Three Gorges reservoir area and taking a typical rocky landslide-Muyubao landslide as the research object,we systematically analyzed the data of 12 deformation monitoring points,proposed the deformation trend prediction method of self-optimizing multivariate dynamic neural network,and carried out the time-series deformation prediction for a single point of landslide;on the basis of the analysis of individual landslide surface monitoring data,we considered the difference in the elevation of measurement point settings and the dispersion of deformation data of different measurement points.Based on the analysis of individual landslide surface monitoring data,an improved elevation inverse distance weighting method considering the change of mountainous terrain is proposed to analyze the spatial deformation distribution characteristics of the non-water-related area of the landslide;to address the problem that the surface deformation monitoring data cannot cover the water level change area of the landslide reservoir,a spatial deformation analysis method based on point cloud information is proposed by using 3D laser scanning technology.Based on the above research contents,the main conclusions obtained are as follows:(1)The time trend prediction is carried out for the cumulative displacement curve of a single point in the Muyubao landslide hazard area.According to the feature that the nonlinear autoregressive curve can fit any curve,a neural network is used to solve the combined parameters of multiple curves for the cumulative displacement curve of the "step-type" landslide.In order to improve the training effect of the neural network,an optimization algorithm is used to find the optimal structure of the network,and to prevent the emergence of local optimal solutions,and to find the optimal solution for multiple fits as the best prediction value.Focusing on the time trend prediction at the point with the largest cumulative displacement of the landslide,the results show that the prediction of the cumulative displacement at a single point is well fitted by the self-optimizing dynamic neural network,and the conventional error can be controlled within 1%,but the error will increase to about 1% when the landslide undergoes step change or extreme weather change.(2)Based on the elevation distribution characteristics of measurement points on a typical landslide,considering the difference in elevation of measurement point settings and the discrete nature of deformation monitoring data of different measurement points,the spatial surface domain fitting of discrete monitoring points is carried out by improving the inverse distance weighting method,which reveals the key deformation area and potential deformation orientation of the landslide based on multiple discrete single-point monitoring data and surface domain characteristics of the whole landslide.The spatial surface domain characteristics of the Muyubao landslide show that the area with the largest deformation is in the southeast direction,i.e.,the right rear edge of the landslide,which is currently covered by one monitoring point,and the area with the smallest deformation is in the northwest direction,i.e.,the left middle of the landslide,which is currently covered by two monitoring points,and the overall deformation of the right side of the landslide is larger than that of the left side of the landslide,and the accumulated deformation shows a decreasing trend from the right rear to the left middle.(3)Using 3D laser scanning technology to obtain 3D point cloud coordinate information of the rock body,each point is based on the idea of undulation,and a certain range of point clouds near each point is fitted to the best-fit plane,and the distance from each point to the plane is calculated.The location of the observation point is introduced,the distance is marked positive and negative,and each point is colored according to the distance value.The single-phase point cloud map determines the surface features such as undulation,convexity and depression according to the color,and the multi-phase point cloud is aligned and differenced according to the color for deformation identification.The method was validated for the function model,concrete model and field subsidence zone,respectively.The final results show that the function model can correctly identify different undulations,and the concrete model can correctly identify the deformation location and deformation amount,which indicates that the method can very well identify the surface block changes of the rock mass,which well complements the reservoir water level change area that cannot be covered by the surface deformation monitoring data,and provides a technique and method to carry out the overall spatial deformation of landslide. |
PDF Full Text Request