Research On The Prediction Of Surface Deformation Around The Exit Section Of Sejila Mountain Tunnel Based On MT-InSAR And GWO-VMD Decompositio

The CZ Railway is a key project of the 13 th Five-Year Plan in China,attracting significant attention from various sectors of society.The newly constructed section of the railway between Ya’an and Linzhi is predominantly composed of tunnels and is located in a complex high mountain and gorge region,which poses a high risk of chain disasters.In recent years,global climate warming has accelerated,leading to increased occurrence of extreme weather events and geological hazards in the CZ Railway region.Surface deformation is a crucial indicator for identifying and monitoring geological hazards such as landslides and debris flows.Currently,the monitoring of surface deformation around tunnel construction sites mainly relies on costly methods such as leveling,GNSS deformation monitoring,and drone tilt measurements.These methods are not well-suited for the harsh conditions of high-altitude areas and can only provide limited monitoring coverage near the tunnel construction sites,making it difficult to address the chain disasters under the current and future climate change scenarios.Synthetic Aperture Radar Interferometry(InSAR)has become the preferred method for monitoring surface deformation in complex terrain due to its high accuracy,large coverage,and long temporal resolution.In this study,the 2019-2022 ascending-descending Sentinel-1A satellite single-look complex(SLC)images covering the research area of the tunnel project were used.The Small Baseline Subset(SBAS-InSAR)technique was applied to analyze the distribution of the two-dimensional deformation field and the temporal deformation information in the research area.Additionally,the Variational Mode Decomposition(VMD)algorithm,enhanced with intelligent algorithms,was employed to preprocess and extract features from the surface deformation time series data,resulting in deformation components related to trend,periodicity,and random variations under different geomorphological conditions.Based on the time series decomposition,prediction models using BP neural networks,LSTM neural networks,and Kernel Extreme Learning Machines(KELM)were constructed for surface deformation prediction.To avoid the interference of empirical parameter settings on the prediction performance,the Sparrow Search Algorithm(SSA)was used for parameter optimization of the prediction models.Finally,by comparing the prediction results of trend,periodicity,and random deformation components using different models,a combined prediction model was established.The following conclusions were drawn based on the research results:(1)The distribution of the two-dimensional deformation field obtained by SBAS-InSAR shows that the surface deformation around the tunnel engineering in the research area mainly concentrates on the top of the slopes on both sides of the valley and the ice lake dense area upstream of the valley,which is a certain distance from the construction site.Due to environmental factors such as terrain and landforms,there is a risk of gully-chain disasters in the deformation zone.Time-series deformation information shows that the vertical annual average deformation of the slopes on both sides of the middle and lower reaches of the valley is large,which may affect the stability of the tunnel engineering in the future.(2)The results of time series analysis show that the slope surface at the top of the slope on both sides of the middle and lower reaches of the valley and around the glacial lake group on the upper reaches of the valley show a continuous sliding trend,and the regularity of annual periodic movement.The deformation trend of surface particles slows down from March to July every year,and accelerates from August to February the next year.The position of the branch gully at the bottom of the middle and lower gully shows a trend of continuous uplift.Through the comparison with precipitation and other meteorological data,it is found that the surface deformation law in the study area is consistent with the seasonal precipitation and temperature variation law in the region.(3)Construct a machine learning combinatorial prediction model based on Sparrow algorithm parameter optimization.The first 70% of the time series decomposition results of vertical SBAS-InSAR deformation sequences were selected as the training set and the last 30% as the test set.The determination coefficient R2,root mean square error RMSE and absolute mean error MAE were used to evaluate the accuracy of the prediction results of different models.It was found that SSA-BP,SSA-LSTM and SSA-KELM had the best prediction effect on the deformation components of periodic,trend and random terms in the time series surface deformation data,respectively.Therefore,the combined prediction model of BP-LSTM-KELM under the optimization of SSA parameters was constructed,which has good applicability to the surface deformation data monitored by SBAS-InSAR in the study area.It can provide some reference for monitoring and early warning of surface deformation around the tunnel project along the Sichuan-Tibet Railway.