Slope Stability Analysis Based On Convolution Neural Network

In recent years,slope accidents caused by slope instability have brought serious economic losses and casualties.Whether it is a natural slope or an artificial slope,the stability analysis of the slope is required at various stages before,during and after construction.In the traditional analysis methods,most of them are manual detection.Subjective factors can easily lead to judgment errors,which cause errors in the judgment of stability results.Secondly,stability analysis involves a lot of formula calculations,which is time-consuming and labor-intensive.In order to complete the slope stability analysis efficiently and accurately,this thesis proposes a Convolutional Neural Network(CNN)method for slope stability analysis.The main research contents include:(1)Data sample amplification.In order to solve the problem of the large number of samples required for CNN training,this thesis first collects 40 groups of representative soil slopes.Then,based on the limit equilibrium method,the shape data and physical parameter data of the slope are randomly changed within a given range,and the generated slope satisfies:the stress relationship between different soil layers;different shapes;different physical properties;different safety and stability results.Finally,40 groups of actual slopes were expanded,and 20,000 groups of slope data with different shapes and physical parameters were generated,which solved the problem of insufficient CNN training samples.(2)Build a network model.Using Matlab software to build the CNN model,by comparing different numbers of convolutional layers,it is determined that the prediction effect of the four convolutional layers is the best;by comparing and analyzing the influence of pooling,it is determined that the features can be preserved without pooling;by comparing the batch size,it is determined that a value of 30 can achieve the best prediction effect.By comparing the influence of neural network parameters such as convolution layer,pooling layer,batch processing,etc.on the slope stability evaluation results,the network parameters are finally determined,and the optimal model is constructed.This model is used to predict the slope stability,and the prediction results are compared with the limit equilibrium method,and the accuracy and efficiency of the CNN prediction results are obtained.(3)Analysis of rock slope.By collecting 30 groups of representative rock slopes,the safety and stability of the slopes are analyzed by the limit equilibrium method,the stability of the rock slopes is predicted by the CNN model,and the same slope data is used for the stereographic projection analysis.Finally,the prediction results of CNN are compared with the results of the stereographic projection analysis,and a higher prediction accuracy is obtained,and the preliminary feasibility of the CNN model for the analysis of rock slope safety and stability is obtained.(4)Comparison of intelligent algorithms.Comparing the trained CNN network with the current intelligent algorithm,from the results,CNN has the advantage of saving time in processing data and accurate prediction results.By using CNN in the safety and stability evaluation of actual slopes in recent years,it is concluded that the network has practical engineering value in evaluating slope stability.(5)Comprehensive slope analysis.Aiming at the complicated characteristics of the actual slope stability detection,a comprehensive stability detection method is studied from three aspects: soil slope,rock slope and soil-rock mixed slope.Through the trained network model,the stability state of various slopes is analyzed.The results show that the prediction accuracy of the comprehensive slope is 95%,which is relatively accurate,and has certain reference significance for further scientific research and engineering applications.