Anlysis,Prediction And Application For Slope Displacement Time Series

There is a vast territory,complex geological conditions in China.With faster and faster infrastructure construction,these will lead to frequent landslide disasters and slope engineering accidents,which greatly threaten the safety of people's lives and property.Therefore,the research on the prediction and prediction of slope accidents and landslide disasters has important practical significance and application value.The most direct manifestation of the slope failure process is the displacement of the slope surface.Most of the prediction and prediction of landslide disasters are based on the displacement data of the landslide body.In recent years,high-precision and automated monitoring equipment and means such as measuring robots and GNSS,have provided abundant data for the study of disaster prediction,all of which are characterized by time series.Time series analysis was first produced in the field of statistics,and has been widely used in all walks of life.Hence,this thesis uses time series analysis method to carry out prediction research on slope displacement monitoring data.Three different categories of time series modeling methods are utilized in this thesis,including classical time series analysis model ARIMA,traditional machine learning methods of artificial neural networks,and Long Short-Term Memory networks(LSTM),a deep learning model which is becoming popular in recent years.Besides,the feasibility of time series analysis method for slope deformation and displacement sequence data prediction is explored.The main research contents and conclusions are as follows:(1)Taking the displacement monitoring data of No.63 and No.73 slope in an open-pit mine in Xinjiang as an example,the traditional time series analysis method ARIMA model is used to carry out prediction analysis and research on the two monitoring data,and the prediction results are evaluated by MAE,RMSE and MAPE.The results show that the ARIMA model is not suitable for non-stationary and long-term time series prediction,but can be applied to the research of fitting the trend of time series change.(2)The displacement monitoring data of No.63 and No.73 points are also taken as an example,the BP neural network of traditional machine learning model and LSTM of deep learning model are built,and the displacement data prediction research is carried out.The results show that both BP neural network and LSTM can be used for slope displacement data fitting and prediction.Because of its unique memory unit and gating mechanism,LSTM has lower prediction error and better prediction effect than BP neural network.(3)Based on the slope displacement monitoring time series data of the mine,the single point prediction accuracy and multi-point prediction adaptability of the long and short term memory network in this project are discussed and analyzed,and the selection of some super parameters in the network structure is finally formed a visual image of slope displacement prediction.The results show that the prediction precision of single monitoring point and multi-point is very high,and the model has good engineering adaptability.In this study,when the initial learning rate is 0.001,the batch?size is 8 or16,and the window is 5,it is the best prediction model.The contour map of slope displacement based on RBF interpolation can well reflect the slope displacement development trend and provide reference for subsequent prevention and control work.(4)Based on the displacement monitoring data of highway slope in a mountainous area of Fuling District in Chongqing,the generalization performance of LSTM model is tested.The results show that the displacement prediction error of the highway slope by the LSTM model is slightly improved compared with the previous one,but within an acceptable range.The model basically predicts the slope displacement variation trend in the first 6 time points of the testing set,that is,the model has a certain generalization performance.