Research On Rock Slope Stability Analysis Method Based On Data Driv

There are numerous slopes widely distributed in the nature and engineering field,and their stabilities are related to the safety of human life and property.Rock slope stability is controlled by engineering rock mass characteristics such as the rock mass shear strength and the distribution law of discontinuities.Accurately grasping these characteristic parameters of engineering rock mass is a prerequisite for slope stability analysis.Therefore,aiming at the problems in rock shear strength parameter estimation,rock dominant discontinuities group division,discontinuity roughness and its shear strength determination,slope stability prediction and reliability analysis,this paper introduces the methods and techniques including Bayesian inference,selective ensemble learning,deep learning,D-S evidence theory,then proposes the data-driven rock slope stability analysis method,provides a method basis and beneficial reference for landslide disaster prevention and control.The main research work and results are as follows:(1)A method based on Bayesian inference for estimating the uncertainty of rock shear strength parameters is proposed to address the issue that previous methods for estimating rock shear strength parameters cannot reflect and quantify their uncertainty.Aiming at the problem that the function form is difficult to be determined in Bayesian inference,a gene expression programming method is introduced to construct the function expressions of rock shear strength parameters and uniaxial compressive strength,tensile strength.The implementation steps of estimating rock shear strength parameters based on Bayesian inference are clarified,and the uncertainty prediction model of rock shear strength parameters is established.The simulation experiment results show that Bayesian inference can obtain a posterior probability distribution of rock shear strength parameters,achieving the uncertainty estimation of rock shear strength parameters with probability significance,and quantify the degree of uncertainty of the model.The analysis results are more accurate,scientific,and interpretable than common machine learning methods.(2)In response to the problems of misjudgment,omission,and noise interference in the traditional method of dividing the dominant discontinuities of rock masses.The DBSCAN algorithm is introduced to construct the clustering model to avoid the subjective impact of the artificially grouping number and initial clustering center of common discontinuity clustering algorithms.The selective clustering ensemble technology is applied to the division of discontinuities for the first time,and rock dominant discontinuities grouping method based on DBSCAN selective clustering ensemble is proposed,which realizes the transition from the traditional single discontinuity clustering model to the ensemble model.The research results show that the clustering result of the DBSCAN selective clustering ensemble method of discontinuity is objective and reasonable,and the clustering effect is significantly superior to that of common clustering algorithms.It not only effectively identifies noise points and outliers,but also overcomes the deficiency of over-segmentation or under-segmentation for an individual model.(3)There is a problem with the lack of objectivity in the visual analogy method of standard profile and the difficulty of any characteristic parameters fully reflecting the surface morphology of discontinuity.Inspired by the technology of data signal processing and deep learning,the discontinuity profile is converted into a time-frequency spectrogram through a short-time Fourier transform to clearly and accurately characterize the undulation degree and position information of the discontinuity.Based on convolutional neural networks,a method for determining the roughness coefficient of rock discontinuity based on the deep learning of time-frequency spectrogram is proposed,where the difficult and tedious feature extraction process of discontinuity morphology is avoided.Then the method flow for estimating shear strength parameters of engineering rock mass discontinuities based on the JRC-JCS model is clarified.The simulation experimental results show that the deep learning method of time-frequency spectrogram can automatically perform feature extraction and model training,and has higher recognition accuracy and generalization ability than empirical methods and machine learning models.This method opens a new pattern of discontinuity roughness recognition from an experience-driven artificial feature paradigm to a data-driven self-learning paradigm,and provides a feasible new way to determine roughness and shear strength parameters of discontinuities.(4)Aiming at the difficulty in selecting slope stability prediction algorithms and the high misjudgment risk of an individual model,a slope stability evaluation method based on the improved D-S evidence theory selective ensemble is established.According to the main influential factors of slope stability,a large slope stability analysis dataset is constructed by the limit equilibrium method.The base classifier selection technology based on boundary distance minimization is introduced to improve the generalization ability of the selective ensemble model.An improved D-S evidence theory is proposed to fuse the information of the base classifiers,which reduces the uncertainty and fuzziness in the decision-making process of selective ensemble model,and solves the problems of easy misclassification and inconsistent results of the existing slope stability evaluation model.The simulation experimental results show that the improved D-S evidence theory selective ensemble method improves the accuracy,reliability and certainty of slope stability prediction,providing a new method for rapid,accurate and efficient evaluation of slope stability in a wide scale.(5)The uncertainty of joint spatial distribution has not been fully considered in the reliability analysis of existing rock slopes.Based on the systematic analysis of uncertainty factors affecting rock slope stability,the influence of the randomness of joint location distribution on the failure mode and bearing capacity of rock mass is studied,and the influence mechanism of the uncertainty of joint spatial distribution on the stability of rock slope is clarified.Based on the discrete fracture network model,the random variable of joint location is introduced,and a rock slope reliability analysis method considering the uncertainty of joint spatial distribution is proposed by combining the Latin hypercube sampling and strength reduction method,providing an effective way to solve the reliability problem of complex rock slopes.The numerical simulation results show that this method comprehensively considers the impact of the variability of rock mass strength parameters and joint geometric parameters,as well as the randomness of joint location distribution on the reliability of rock slopes,truly represents the uncertain factors affecting the stability of jointed rock slopes,and improves the objectivity and accuracy of the reliability analysis results of jointed rock slopes.