| Landslide disasters in my country mainly occur in western regions such as Yunnan,Guizhou,Chongqing,Gansu,and Shaanxi.Landslide disasters seriously affect the safety of local people’s lives and property,and restrict the development of local economy and society.It is imperative to carry out effective landslide disaster prevention and mitigation work.With the continuous in-depth research on landslides,various monitoring methods such as GNSS monitoring technology,displacement meter monitoring technology,inclinometer monitoring technology,rainfall monitoring,soil temperature and humidity monitoring,etc.have been continuously applied to landslide monitoring.How to comprehensively use these monitoring data to effectively improve the accuracy of landslide stage discrimination and prediction analysis is a difficult and hot point of current research.Multi-source data fusion technology can comprehensively analyze the landslide monitoring data collected by multi-source sensors,and obtain a set of fusion data that accurately reflects the deformation stage of the landslide.Therefore,this paper takes the 7# landslide body in Heifangtai Dangchuan,Yongjing County,Linxia City,Gansu Province as an example.Through the analysis of landslide multi-source monitoring data,the data-level fusion algorithm and feature-level fusion algorithm are respectively applied to the landslide stage discrimination and Forecast research.The main research contents and results of this paper are as follows:(1)This paper introduces the basic theory of multi-source data fusion,mainly including the definition,principle and structure of multi-source data fusion,and then analyze the multisource data fusion in landslide monitoring,and summarize its characteristics and existing problems.(2)In this paper,the four commonly used landslide monitoring technologies are introduced,and the three commonly used data collection and transmission methods in landslide monitoring are introduced by taking surface monitoring as an example.Finally,combined with the comparison of the preprocessing of landslide monitoring data by related scholars,summarize the three most commonly used data preprocessing methods in landslide surface monitoring.(3)For landslide monitoring,a single displacement sensor data is used for stage discrimination,and there is a problem that the monitoring data is one-sided,leading to unreliable warning results.It is proposed to use the weighted fusion method to fuse the landslide surface deformation data collected by the global navigation satellite system receiver and the displacement meter,and use the fused data for the discriminant analysis of the landslide stage.Experimental results show that both the adaptive weighted estimation fusion algorithm and the improved particle swarm optimization adaptive weighted fusion algorithm can effectively fuse the landslide surface displacement data.The improved particle swarm optimization adaptive weighted fusion result in the discriminant analysis of the landslide stage Compared with the adaptive weighted fusion result,it is more reliable and accurate.(4)In view of the problem that most of the current landslide predictions and forecasts only analyze the deformation time series collected from a certain surface displacement monitoring point on the landslide body,and do not comprehensively use the influence factors around the monitoring location to predict the problem.A feature-level fusion model based on clusteringstepwise regression analysis of multi-source heterogeneous landslide monitoring data is proposed.The experimental results show that the clustering-stepwise regression analysis model can effectively integrate the multi-source heterogeneous monitoring data of landslides.Three prediction algorithms are used to compare and analyze the GNSS monitoring point HF06 data,BP neural network data fusion data,and cluster-step regression analysis fusion data.Clusterstepwise regression analysis fusion data prediction results are better,and the multi-factor LSTM model is used to cluster-stepwise regression analysis fusion data prediction accuracy is higher,MAE and MRE are 8.2mm and 2.82%,respectively.It can be seen that the feature-level fusion is performed Prediction can effectively improve the accuracy of landslide prediction. |
