Research On Early Warning Model Of Railway Subsidence Disaster Based On Big Data Analysis And Mining

Land subsidence disaster is a common slowly-changing geological disaster,which has the characteristics of slow generation and difficult prevention and control.Severe land subsidence will do great harm to people’s economic life,especially the land subsidence along the railway.High-speed railways are particularly sensitive to ground subsidence.Uneven subsidence not only reduces the comfort of passengers,but also directly threatens the safety of trains.The topic of this thesis is to establish a railway ground subsidence prediction and early warning model through big data analysis and big data mining on the ground subsidence data along the railway.And the prediction and early warning model is applied to the railway land subsidence monitoring area for prediction and early warning.The main research work is as follows:1.Preprocess the collected satellite remote sensing images,and use the permanent scatterer synthetic aperture radar interferometry(PS-In SAR)technology to calculate the time series deformation value of the surface of the monitoring area.The obtained deformation value is used to monitor the changes in land subsidence along the railway.2.Perform big data analysis on the influencing factors of land subsidence disasters,and give quantitatively the influence of each influencing factor on land subsidence.Analyzing the law of ground subsidence time of railway monitoring sites,the results show that the ground subsidence value of tunnels and bridge sites generally changes periodically,and the subsidence speed of subgrade sites is first fast and then slow.The Aprioir algorithm is used to mine the association rules of the ground subsidence of various railway sites,and determine the relevant factors of the ground subsidence prediction.3.Compare and analyze multiple time series forecasting models,and choose long and short-term memory artificial neural network(LSTM)as the basic model for land subsidence prediction.Aiming at the problem of under-fitting caused by insufficient feature extraction ability of LSTM model,a CNN-LSTM combined model is proposed to improve the effect of feature extraction stage.According to the results of the CNN-LSTM prediction model,the temporal and spatial trend of land subsidence is predicted.The improvement effect of the CNN-LSTM prediction model is verified through comparative experiments.The results show that the model has a better fitting effect than a single LSTM prediction model.The average error rate in the three places of tunnel,bridge and roadbed is 0.84%,5.7 %,0.55%,which meets the requirements of railway land subsidence prediction.4.Combining the prediction information obtained by the prediction model and the risk indicators of railway subsidence disasters,three scenarios of railway land subsidence warning models are proposed.This model can be used for early warning of settlement risks in areas along the railway.The experimental results show that the accuracy rates of the tunnel,bridge,and roadbed settlement disaster risk early warning models proposed in this thesis are 76%,86%,and 84%,respectively,which meet the requirements of early warning.5.Based on Spring MVC,Bootstrap,My Batis and other frameworks to develop and realize the online warning application function of ground subsidence and geological disasters along the Sichuan-Tibet Railway.This application function visualizes big data on the early warning results of ground subsidence along the railway,helping railway ground subsidence monitors to take preventive measures.