Uncertainty Analysis And Modeling Of Multi-scale Rock Mass Fractures Of Hydraulic And Hydropower Engineering And Its Application

The geological conditions of hydraulic and hydroelectric engineering projects are usually inclement,and large and small fractures are intricately distributed in the geological bodies.Fractures not only destroy the integrity of rock mass but also are the weakest parts of a rock mass.Therefore,it is one of the most critical things to reasonably analyze and describe the fractures in hydraulic and hydroelectric engineering.However,due to the sparsity of geological exploration information and the fuzziness of description,there must be varying degrees of uncertainty in the analysis and modeling process of fractures.In addition,the insufficiency of the current uncertainty analysis methods in geological research also makes it difficult to fully understand the accurate interpretation of fractures.Aiming at this problem,the uncertainty characteristics of fracture and related key problems are fully analyzed in this research,and the underlying laws of the random state of fractures are deeply studied through various methods,such as the methods in structural geology and hydraulic rock mechanics,statistical analysis methods,machine learning,image processing,and 3D modeling techniques.Finally,a series of research results are obtained and several uncertainty modeling and analysis methods are put forward,as follows:(1)The basic pattern of describing and analyzing the multi-dimensional parameters of fractures based on Copula methods is presented;the feasibility of using generative intelligent models to establish the relationships among fracture parameters is studied,and a specific solution is put forward.Firstly,based on the Copula theory,the uncertainty relationship between dip direction and dip angle,trace length,and opening are analyzed respectively.The relationship between aperture and fracture size is then established through "pseudo trace" and a set of copula functions to solve the difficulty of aperture assignment in the Baecher disk method.Furtherly,a random oblate ellipsoid discrete fracture network model is then proposed,which can be a foundation of further research such as seepage and stability analysis.After that,the principles and characteristics of generative intelligent models(including variational autoencoder(VAE),generate adversarial network(GAN),and gaussian mixture model(GMM))are analyzed,and a complete solution for the uncertainty analysis of high-dimensional fracture parameters is established based on GMM algorithm.Finally,the applicability and advantages of this method in characterizing the complex relationship between parameters,data simulation,probability calculation,and visualization are proved by two engineering examples.(2)A new discrete fracture network(DFN)modeling method is proposed by replacing disks with polygons,the rationality of fracture shape is proved,and an iterative inversion algorithm for calculating the distributions of fracture size is given.The “control circle method” is used to generate random polygon fractures,the result of which is closer to the actual situation than the Baecher disk method;the iterative inversion algorithm is proposed to approximate the size distribution function of fractures to avoid the difficult mathematical derivation caused by the irregular shape of fractures.This method is an extension of the Baecher disk method and can describe the development of internal fractures in rock mass more accurately.(3)A graphical validation algorithm for 3D discrete fracture network models is presented.When implementing the algorithm,the actual trace map and the trace map of the corresponding part of a 3D DFN model are firstly standardized.Then the similarity of the two trace maps is measured from the following four aspects,including(1)overall gray,(2)gray grading curve,(3)direction characteristics,and(4)gray density curve.Finally,a comprehensive evaluation formula is obtained from the weighted average of the above features.This algorithm provides a more objective way for DFN verification.(4)A solution of the intelligent perception for uncertain geometric shape of large-scale fractures is established;and a chain representation of complex fault networks is proposed,which achieves semi-automatic modeling and uncertainty analysis of faults.The mixed density neural network is used as the core to perceive and characterize the uncertainty of single faults.The method system includes data preprocessing,intelligent model designing,intelligent model testing,quantitative simulation method of faults under specific confidence,and evaluation method of the uncertainty of fault model.Based on the above,the chain representation of complex fault networks is proposed by reference to the binary tree representation method,and the uncertainty analysis and semi-automatic modeling of fault networks are achieved.This solution is an implicit intelligent modeling method for large-scale fracture,which is more automatic than the traditional explicit modeling approaches and can describe the uncertainty of faults more accurately.Besides,it provides an important reference for the uncertainty analysis of large-scale geological structures such as strata and folds.