Study On Surface Deformation Along Dongguan Subway Based On InSAR

Subway construction can not only save more and more ground traffic pressure,but also alleviate a lot of environmental protection.The subway is an important channel for the exchange of resources and population in the city.At present,Dongguan has two subway lines,of which line 1 is in the construction stage,and line 2 has been put into operation.The opening of Dongguan subway has strengthened the connection between Dongguan and Dawan District of "Guangdong,Hong Kong and Macao".Whether in the construction stage or in the operation stage,the subway will have different degrees of deformation,which may endanger the surrounding infrastructure.Therefore,it is necessary to study the surface deformation along the subway during construction and operation and its future development trend.Based on InSAR and deep learning principle,this paper studies the monitoring and prediction of ground deformation along the subway in Dongguan.The main research contents are as follows:(1)Using PS-InSAR technology to process sentinel-1a data covering 59 scenes in Dongguan,the deformation data of radar line of sight direction are obtained.The maximum deformation rate in the study area is-38.62mm/a,and the range of deformation rate is-38.62mm/a ～29.63mm/a,of which 82.64% of the regional deformation rate is-6.91～3.92mm/a,indicating that the overall deformation trend of Dongguan is stable.There is a settlement funnel in the central area of Dongguan.According to the data analysis,it is found that the settlement funnel is mainly related to the formation composition and the overexploitation of groundwater in this area.(2)The standard deviation ellipse is introduced to study the spatial and temporal evolution characteristics of the overall deformation in Dongguan,and the empirical orthogonal decomposition method is used to refine the spatial and temporal evolution of local deformation.The center of gravity of the overall deformation in Dongguan shifted to the southeast,and the long axis first decreased,then increased,and finally stabilized.The InSAR monitoring results of surface deformation along Dongguan Metro Line 1 are processed by empirical orthogonal decomposition method,and the time function and spatial component are obtained.The contribution rate of the first component is 60.92%,indicating that the first component is consistent with the time monitoring results.The combination of the first component space function and time function shows that the surface along line 1 shows a sinking trend.The results of the second component show that the area sinks first and then rises and gradually returns to the original state,and the third component shows that the area rises first and then sinks and gradually returns to the original state.(3)SOM-AE-BO-GRU model is constructed and applied to the training and prediction of surface deformation monitoring data along Dongguan metro lines 1 and 2 monitored by InSAR.Among them,the range of prediction results for line 1 is-46.302mm～50.609 mm,and the range of prediction results for line 2 is-50.638～55.422 mm.1.The prediction residual error of line 2is 92.3% and 92.1%,which is between-5-5mm,indicating that the error of the model prediction result is small.(4)Aiming at the problem of single evaluation index of single time series prediction results,the prediction accuracy of multi time series prediction model is evaluated from the perspectives of statistics and spatial consistency through P-P diagram,Q-Q diagram,standardized residual,autocorrelation coefficient and standard deviation ellipse.The results show that the prediction residuals of surface deformation along the two subways are white noise.Through the standard deviation ellipse,it is found that the prediction results are consistent with the spatial distribution of InSAR monitoring results.Combining the standard deviation ellipse and its related parameters to compare the prediction accuracy of the four prediction models,it is found that the prediction accuracy of the data is significantly improved after clustering and compression.Regardless of the spatial distribution or prediction residual,the prediction results of Gru model are better than LSTM,which shows that the SOM-AE-BO-GRU model proposed in this paper is accurate.The reliability of the evaluation index of the prediction results proposed in this paper is tested by mean absolute error and root mean square error.The results show that the results of the traditional index are consistent with the evaluation results of the index proposed in this paper.This paper evaluates the effectiveness of the multi time series prediction model from the two dimensions of residual error and spatial distribution.It provides a new method for the prediction of large-scale surface deformation along the subway and a new index for the effectiveness evaluation of multi time series prediction model.