Reliability Analyses Of Slopes Considering Soil Spatial Variability Of Under Earthquake Ground Motions

Slope instability induced by earthquake will have an immeasurable impact on social and economic development.Accurate and reasonable evaluation of slope stability under earthquake action is of great practical significance for seismic design and disaster prevention and mitigation of geotechnical engineering structures.Many factors affecting slope stability,such as soil parameters and external loads,are uncertain.Although the reliability analysis method has been greatly developed because it can quantify the influence of various uncertain factors,the reliability evaluation of slope seismic stability is still a key technical problem due to the complexity,randomness and non-repeatability of slope soil and earthquake action.At present,there is no effective analysis method to evaluate the reliability of slope dynamic stability considering both the randomness of ground motion and the spatial variability of soil parameters.This paper has carried out relevant work on this,the specific work and conclusions are as follows(1)This paper systematically reviews the research progress of certainty and uncertainty evaluation methods for slope stability under earthquake action.The dynamic strength reduction method and dynamic time history analysis method for dynamic stability analysis of rock slopes are developed,and the corresponding calculation principles and steps are given.At the same time,the advantages and disadvantages of these two methods are compared and evaluated,which lays an important foundation for selecting objective and reasonable dynamic stability analysis methods for slope dynamic reliability evaluation.(2)A spatial variability slope reliability analysis method based on convolutional neural network(CNN)and sample expansion is proposed.Firstly,the Karhunen-Loève(K-L)series expansion method is used to discretize the non-Gaussian random field of soil parameters,and the finite difference strength reduction method is used to calculate the slope safety factor to obtain a small number of initial samples.On this basis,the sample expansion is not required for slope stability analysis,and more training samples are obtained.All training samples are converted into pixel-valued digital images,and then a CNN surrogate model is established to fit the implicit function relationship between slope safety factor and random field digital image features.Calculate the slope failure probability.Finally,the accuracy of the method is verified by the example of friction / cohesive soil slope and three-layer undrained cohesive soil slope,which shows that the method has good computational efficiency and provides an effective way to solve the reliability problem of complex low probability horizontal spatial variation slope.(3)A calculation method of slope dynamic time-varying reliability considering both seismic randomness and spatial variability of soil parameters is proposed.Firstly,the evolution spectrum representation method of non-stationary random process is used to simulate the earthquake ground motion process,and the K-L series expansion method is used to simulate the spatial variability of soil parameters.On this basis,the agent model of slope dynamic safety factor is established by means of deep learning algorithm to calculate the time-varying failure probability of slope.Finally,the accuracy of the method is verified by the friction / cohesive soil slope model,which shows that the method can greatly save the calculation cost and solve the complex dynamic time-varying reliability problem of the slope.In addition,the failure probability of slope is often underestimated without considering the spatial variability of soil under earthquake action.It is suggested that the failure probability time history of slope should be used to replace the traditional minimum reliability as the dynamic reliability evaluation index of slope under earthquake action.