Research On Monitoring And Prediction Of Subsidence In Yan’an New Area Based On InSAR And Machine Learning

Due to the acceleration of urbanization in the Loess Plateau,there is insufficient space for urban development.Affected by factors such as topography and the easy excavation of the overlying loess,the Loess Plateau has become the main distribution area of " mountain excavation and valley infilling " in my country,such as Lanzhou New District,Shanxi Luliang Airport,and Yan’an New District.Large-scale filling and excavation have caused serious land subsidence problems due to changes in the original topography and geomorphology,geological and hydrological conditions.Therefore,long-term and regional subsidence monitoring is particularly important for the safety of urban construction in the Loess Plateau.Compared with traditional single-point land subsidence monitoring,In SAR(Interferometric Synthetic Aperture Radar technology)can provide large-scale,long-time scale,high-precision deformation data.The use of integrated remote sensing technology to carry out research on the impact of building height on settlement rate and the prediction of settlement rate based on machine learning is relatively rare.Based on this,this paper selects Yan ’an New Area as the research object,which is a typical case of " mountain excavation and valley infilling " on the Loess Plateau,this thesis is based on relevant theories in engineering geology,soil mechanics,remote sensing science,geographic information science,etc.,using existing research results,with the help of SBAS-In SAR technology,shadow altimetry,laser ranging,field engineering geological survey and other means,comprehensively analyze the influence of factors such as building height,construction time and precipitation on the land subsidence of the excavation and filling area in Yan’an New District.On this basis,machine learning methods are used to carry out research on the prediction of land subsidence rate at a single point in the area of excavation and filling,and to select a prediction model suitable for land subsidence in Yan’an New Area.Among them,the use of integrated remote sensing technology provides a new idea for the study of the influence of building height on settlement rate.The application of RBF radial basis network in settlement prediction is a new exploration of settlement prediction.The above research results can provide new ideas for the monitoring and prediction of land subsidence in the " mountain excavation and valley infilling " area of the Loess Plateau,and provide a reference for the later planning and construction of the excavation and filling area.This paper has achieved three main research results as follows:1)Settlement rule: In Yan’an New Area 2015-2019,the LOS-direction speed change rate was-52～12 mm/yr,the vertical direction deformation rate was-74～18mm/yr,and the maximum accumulated settlement reached 250 mm.The Yan’an New Area has a belt-shaped settlement center,and the shape of the settlement area basically coincides with the shape of the fill area.The settlement rate of the fill area changes showing a slow-fast-little fast change rule.According to this,it can be judged that the filling area is still in the stage of primary consolidation period.Uplift-settlement deformation appears in excavation area.2)Causes of settlement: Correlation analysis of profile data in the filling area shows that the thicker the filling soil is,the thinner the undisturbed soil is,and the greater the settlement amount is.The thicker the undisturbed soil is,the thinner the undisturbed soil is,and the smaller the settlement amount is,indicating that the thickness of the filling soil has a greater impact on settlement compared with the undisturbed soil.The influence of building load on settlement is different in the filling and excavation area.Tall buildings in the fill area have a small amount of settlement,because of difference of fill thickness and foundation treatment,the settlement of lowrise buildings higher than that of high-rise buildings;under the same excavation depth in excavation area,low-rise buildings gravity offset by lift up phenomenon,high-rise building suffered a small amount of subsidence.Compared with the precipitation and deformation,it is found that there is no subsidence in the excavated and fill area when the precipitation increases,but there is uplift deformation in the excavated area.3)Settlement prediction: BP neural network is used for time series settlement prediction.Among the three algorithms,Feedforward,Cascadeforward and Elman,cascadeforward algorithm is the most suitable for time series prediction,and it is predicted that the settlement will be 160 mm after 84 days of selecting the deformation point.BP network and RBF network are used for settlement single-point space prediction.Compared with BP neural network,RBF radial basis function network has higher prediction accuracy and is more suitable for spatial prediction.It is predicted that the settlement of random sampling points in the pure fill area is 194～228mm,and the deformation in the pure excavation area is-8～19mm.