Study On Stability Evaluation Of Loess Landslide Based On Integration Of Multi-Source Isomerous Data Constrained By External High-Precision Monitoring Information

China has the largest loess distribution area,the most complete loess and the only accumulating loess plateau in the world.Since loess itself has several special properties like strong water sensitivity,fragile structure,unique strength attenuation and complex deterioration process,the landslide risk is becoming increasingly fierce,thus posing serious threaten to human lives,properties and the environment.For landslide risk reduction and even prevention,landslide stability analysis evaluation cames into being and gradually becomes one of the hot topic both at home and broad.However,affected by interaction of multi-fields(deformation field,stress field,water field),mutlti-factors(landslide geological structures,human engineering activities,hydrological environment),the deformation of loess landlside has some characteristics like complexity,diversity,randomness and territoriality.As a result,it is difficult to obtain accurate evaluation results based on conventional landslide stability evaluation methods.Although sustainable achievement in loess landslide stability analysis have been obtained from constantly proposed and improved techniques,it is still difficult to find an efficient method that can accurately evaluate the stability of loess landslide.Firstly,although the displacement time series prediction method in geodesy has the advantages such as easy implementation and high precision,it is only a mathematical apparent model and cannot explain the mechanical cause of landslide deformation.In addition,although the numerical simulation method based on reliability theory,finite element theory or limit equilibrium theory in engineering geology is a physical and mechanical evaluation model,which can explain the disaster mechanism of landslide,there are many assumptions and restrictions in the numerical model,and the model parameters for actual landslide are difficult to be obtained accurately.Therefore,there still exists certain deficiencies when implementing slope stability analysis by the above-mentioned techniques.Accordingly,there is a need to establish a comprehensive evaluation model for landslide stability by coupling the landslide displacement evolution process with the dynamic disaster mechanism,which can ascertain the cause of landslide failure and the catastrophe mechanism,thus providing important reference for similar landslide prevention and mitigation in the future.Therefore,how to build a comprehensive evaluation method for landslide stability analysis by organically coupling landslide external high-precision monitoring data in geodesy with numerical simulation means in engineering geology to is still a topic that needs in-depth study.Accordingly,based on multi-source heterogeneous monitoring data such as displacement,rainfall and DEM,the author has studied high-precision prediction of landslide displacement and intelligent dynamic response surface research considering the uncertainty of groundwater level.Furthermore,combined with the fine three-dimensional geological model of landslide,the stability of landslide is further discussed constrained by external high-precision monitoring data.The high-precision deformation information outside the landslide is successfully connected with the internal deformation mechanism,and the ability to answer practical engineering problems is also improved.The main research contents and obtained results in this study are concluded as follows:(1)The conventional landslide displacement prediction model is systematically analyzed,and all kinds of conventional landslide stability evaluation methods are summarized.The basic principle of finite element difference method in landslide deterministic evaluation method is introduced emphatically,and the basic principle,advantages and disadvantages of landslide reliability analysis are summarized.The basic framework of Monte Carlo Simulation(MCS)response surface model based on artificial intelligence algorithm is also systematically discussed;(2)In order to solve the problem that the prediction results of landslide displacement by conventional Kalman filter are difficult to meet the accuracy requirements,the composition of dynamic coefficient noise array in conventional Kalman filter is derived with consideration of random disturbance of external landslide inducing factors(rainfall,temperature,etc.).Then an improved Kalman filter prediction model with consideration of many kinds of external induced factors is proposed,which can effectively overcome the defect of poor applicability of conventional Kalman filter model due to lack of induced factors.Experimental results show that the RMSE and MAE obtained from the proposed algorithm are 4.608 mm and 0.7%,respectively.Compared with the conventional Kalman filtering model,the prediction accuracy are improved by 58.95% and 65%,respectively,effectively improving the accuracy of landslide displacement prediction.Furthermore,in order to select more reasonable landslide inducing factors and meet the requirement of landslide displacement prediction accuracy,a new technique based on Variational Mode Decomposition(VMD)-Mutual Information Coefficient(MIC)-Multi-kernel Extreme Learning Machine(M-KELM)model is developed in this paper.And simultaneously,different inducing factors are given the same weight as MIC values after data normalization when implementing random displacement prediction.Based on the comparison between simulation and real data,it can be concluded that the inducing factors of model selection proposed in this paper are reasonable,and the introduction of MIC in random term prediction can effectively improve the overall precision of M-KELM model;(3)A multi-objective Gray Wolf Optimization(GWO)-mixed KELM(MKELM)intelligent response surface agent model based on intensity reduction method is developed in this paper.At present,most of the slope safety factor is obtained from the response surface model based on limit equilibrium method which needs to search the critical slip surface and the assumed the circular arc slip surface in advance.Also the existing models such as RBF and ANN have defects such as poor generalization ability with small samples.Furthermore,the optimal parameters may not be always achieved by the optimization algorithms with only one-single objective.To address the above-mentioned problems,first strength reduction method which is suitable for sliding surface of any shape without assuming the sliding surface to improve the scope of application of the model is applied to calculate the safety factor of all samples.Secondly,in order to improve the accuracy of the response surface model,MKELM model is constructed by combining the advantages of polynomial kernel function and RBF kernel function to enhance the learning and generalization ability of the model.Finally,the fitness of the model with minimum mean square error and maximum goodness of fit is used to obtain the optimal parameters of the developed model.Two classical landslide examples are applied for model verification.It is found that the results obtained are in good agreement with the existing research,indicating that the model’s application scope is enlarged,the model’s accuracy and calculation efficiency are also significantly improved.The newly developed approach is suitable for the stability evaluation of artificial and natural landslides with arbitrary position and shape;(4)A dynamic intelligent response surface model considering the uncertainty of groundwater level is developed.At present,since complex loess landslides cannot accurately obtain the groundwater level,the stability evaluation results may be too "optimistic".And the commonly used response surface models are static,which cannot guarantee the calculation accuracy.To address these issues,this paper first regards the groundwater level as a random variable.Then an intelligent response surface model of Whale Optimum Algorithm-Gauss Process Regression(WOA-GPR)is constructed based on uniform design with iteration.Based on two practical landslide examples,the slope instability probability considering the uncertainty of groundwater level is calculated,and the difference between the slope instability probability considering the groundwater level as a constant is also analyzed.It is found that with consideration of the uncertainty of groundwater level,a more reasonable slope failure probability is obtained from the proposed model.It proves that ignoring the uncertainty of groundwater level will "overestimate" slope stability.The developed approach can accurately assess the stability of complex landslide with unknown groundwater level,and can judge whether the landslide is in an unstable state in a period of time,which provides theoretical basis for accurate and effective evaluation of complex landslide reliability.(5)The comprehensive evaluation of landslide stability is carried out by integrating the external high-precision deformation monitoring data,rainfall data and irrigation data.Since most of the existing landslide stability models are only analyzed based on simplified two-dimensional models,the obtained results are quite different from the actual ones.In addition,the high-precision monitoring information outside the landslide is not combined with the internal deformation mechanism,which may reduce the reliability of the landslide stability evaluation.Accordingly,a fine three-dimensional geological model of landslide which can reflect the real nature characteristics is constructed by using Arcgis-Rhinoceros-Griddle based on Google image,high-precision multi-source monitoring information,engineering geological borehole information,groundwater level information and field investigation.Secondly,the stability of loess landslide under different influencing factors including irrigation,rainfall,groundwater level and fracture development is analyzed,and the mechanism of instability failure is also discussed.In addition,based on the detailed three-dimensional geological model,combined with irrigation,pore water pressure and high-precision monitoring information,the cause of instability and catastrophe mechanism of loess landslide in the study area are discussed.The results show that HF06/07 Beidou /GNSS monitoring site first becomes unstable,followed by HF09 Beidou /GNSS monitoring site,and HF05 Beidou /GNSS monitoring site is the last one that slides.It indicates that compared with the numerical simulation method of engineering geology,the results obtained by the developed technique in this chapter are better consistent with the actual monitoring situation.And the sliding sequence of landslide is also in high agreement with the actual sequence.All the obtained achievements in this paper demonstrate the organic coupling of the external high-precision monitoring information and the internal physical and mechanical evaluation model provides a new idea and perspective for the stability evaluation of loess landslide,and helps to well understand the deformation evolution mechanism of loess landslide.