Recognition And Prediction Analysis Of Mining Area Surface Subsidence Based On InSAR And Deep Learning

The mining and blasting operations in mining areas cause ground subsidence,leading to a series of geological disasters,which pose a serious threat to the lives and property of workers in the mining areas and nearby residents.Therefore,scientific monitoring,prediction,and prevention of the subsidence in mining areas has become a prerequisite for ensuring safe production in these areas.The time-series InSAR technology has been widely recognized and applied in the field of mining subsidence monitoring due to its advantages of rich data sources and high efficiency.However,the current mining subsidence monitoring and prediction work still faces problems such as insufficient number of monitoring points and low prediction accuracy.Although the common prediction methods can reflect the future deformation trend of local areas to a certain extent,they have limited effect on revealing the overall deformation pattern and local deformation difference characteristics of mining area surface.In this study,the SBAS-InSAR technology was used to carry out dynamic monitoring of subsidence in Dexing mining area,and the subsidence results were predicted and analyzed based on TSOLSTM model.At the same time,similar hyperparameter optimization algorithm and Kmedoids-SVR algorithm were selected to compare and analyze the prediction accuracy of the model,The main research work and results are as follows.(1)Based on SBAS-InSAR technology,the subsidence dynamic monitoring was carried out in Dexing mining area,and the vertical subsidence time series of the study area was extracted.Based on the SBAS-InSAR technology,50 Sentinel-1A orbital ascent data covering the research area were used to obtain the vertical cumulative subsidence time series and the deformation average rate distribution of Dexing mining area.Three major subsidence areas are identified in the study area,which are located in the southwest of the copper plant,between Duyang Lake and the copper plant,and the southwest of Fujiawu stope.According to statistics,the overall average subsidence rate of the mining area is concentrated in-5～10mm/a,and the maximum subsidence rate is-372.75 mm /a.(2)The formation law of monitoring results was analyzed from two dimensions of spatial distribution and time evolution.From the perspective of spatial distribution of settlement results,the settlement levels in the three main settlement areas are obvious.The subsidence area A exhibits a subsidence pattern that spreads from the center to the surrounding areas,with two obvious subsidence centers in the south and north.The subsidence area B shows a trend of gradually intensifying subsidence from east to west,and the subsidence area C demonstrates a trend of gradually increasing subsidence rate from north to south.From the perspective of time evolution characteristics,the three main subsidence areas show a trend of fluctuating subsidence during the monitoring period.(3)The TSO-LSTM model was constructed and applied to the training and prediction of surface subsidence data in mining areas.A deep learning prediction method based on Tuna Swarm Optimization(TSO)Long Short-Term Memory(LSTM)network model hyperparameters was proposed.This method optimizes the input sample length,hidden layer depth,hidden layer neuron number and initial learning rate of LSTM model by TSO algorithm to improve the prediction performance of the model.The results show that the prediction performance of the optimized model is obviously improved,and the overall prediction accuracy of the deformation area meets the requirements of mining area deformation prediction.The model has certain practicability and application prospect.(4)Three models,GWO-LSTM,POS-LSTM and K-medoids-SVR,were selected for prediction analysis,the prediction accuracy was calculated and the model prediction performance was compared.The root mean square error,average absolute error and average prediction accuracy were used as the evaluation indexes of prediction accuracy,and the prediction accuracy of the above three prediction models was compared with the TSO-LSTM model to verify the prediction performance of the TSO-LSTM model.