Processing And Analysis Of Deformation Monitoring Data Based On Deep Learning

Surface subsidence is a kind of dynamic and long-term geological phenomenon.During the development process of Nanhu area in Tangshan city,significant surface subsidence has appeared.In order to grasp the real-time data of surface subsidence in Nanhu area,fully understand the situation of surface subsidence and reduce the impact on society.The PSIn SAR technology is used to dynamically monitor the real-time surface subsidence in this area,and deep learning prediction model is introduced to predict the change trend of subsidence based on the subsidence factors.It is of great significance to the local disaster prevention and reduction,development and construction.Taking Nanhu Area in Tangshan city as an example,the surface subsidence information of the study area(from September 12,2017 to September 20,2020)was extracted by using PS-In SAR technology.Based on monitoring results,the main subsidence areas were analyzed in east-west direction and north-south direction respectively.The results were verified by combining the measured data of leveling and GPS observation.By constructing ARIMA model,SVM model and LSTM model,the law of deformation monitoring data is revealed to predict the subsidence in the monitoring area.Three evaluation indexes,RMSE,MAE and MAPE,were used to evaluate the accuracy of the prediction results.At the same time,the correlation analysis was carried out in the adjacent regions where subsidence mainly occurred in the study area.The results show that the correlation coefficient between the accumulated surface subsidence extracted by PS-In SAR and the measured data of third-order leveling is 0.9431.The correlation between monitored subsidence from PS-In SAR and GPS observation is0.9971.It is proved that the surface subsidence monitoring results from PS-In SAR have a strong consistency with the measured results and are relatively reliable.According to the monitoring results from PS-In SAR,there was obvious subsidence in six regions during the monitoring period: the Earthquake Site Park,Phoenix Terrace,Lavender Manor,Swan Lake Zoo,Aishang Manor and Koi Center.Three different models were used to predict the subsidence based on time series data of PS-In SAR.The prediction results of LSTM model are consistent with the PS-In SAR monitoring results according to the subsidence-time trend.Compared with the other two models,the RMSE,MAE and MAPE values of LSTM model are the minimum,and the prediction accuracy of LSTM model is high,which meets the demand of surface subsidence monitoring project in Nanhu Area and can be applied to actual production.By analyzing the adjacent areas with subsidence,the results show that the correlation coefficients between adjacent areas are all above 0.89,indicating the possibility of continuous subsidence between adjacent areas.Figure 39;Table 20;Reference 88...