Research On The Spatial And Temporal Variation Of Slopes Using Discrete Element Method And Deep Learning

China is a country with frequent geological disasters.Landslides are one of the most important disasters.In order to ensure the safety of people and property,it is necessary to carry out accurate safety analysis of slopes.However,due to the concealed slope engineering At present,the safety analysis of slopes is not comprehensive,the analysis of the spatial evolution of slopes is lacking,and the prediction of slope monitoring data is subjective and intelligent.Therefore,it is urgent to carry out a comprehensive study on the temporal and spatial evolution of slopes.This paper proposes a method for analyzing the laws of the temporal and spatial evolution of slopes based on discrete element and deep learning methods.The main research work of this paper is as follows:(1)Slope reliability analysis based on discrete element method.This paper establishes the slope discrete element model to analyze the influence of different material parameter changes on the slope stability.By analyzing the dynamics and energy changes during the slope instability,the strength reduction method suitable for the MatDEM discrete element method is proposed.in accordance with.Finally,the Monte Carlo method is used to analyze the reliability of the slope,and the probability of landslide of the slope is obtained.(2)Landslide risk assessment method based on discrete element method.For rainfalltype landslides,numerical simulation methods are used to analyze the depth of the slopes affected by different rainfall times,and then the dynamic model of the slope under rainfall conditions is established by the discrete element method,and its movement and accumulation characteristics are analyzed,and then the buildings are resistant to disasters.Strength,proposed a sub-regional building vulnerability assessment method.Finally,combining the slope probability of landslide and the probability of precipitation,the risk degree distribution of the slope is obtained.(3)Monitoring data preprocessing method based on time series method.Aiming at the characteristics of time series monitoring data,by comparing different outlier processing,interpolation and smoothing methods,a preprocessing method that adapts to different levels and different data characteristics is proposed.Then use auto-correlation and crosscorrelation analysis for different types of monitoring data to explore the relationship between the data.At the same time,the time series decomosition method is used to decompose the periodic data,which lays the foundation for subsequent analysis.(4)Prediction and early warning method of monitoring data based on DeepAR model.The DeepAR model deep learning probability prediction model is established,and the model data set division,hyperparameter selection and model accuracy evaluation method are analyzed.Then,the prediction effects of the single variable and the covariate model were compared,and the data prediction method of the combined DeepAR model suitable for slope monitoring data was further proposed,and the superiority of the method was verified.Finally,according to the results of probability prediction,an early warning method based on dynamic threshold of slope monitoring data is proposed.