Study On Intelligent Prediction And Early Warning Method Of Landslide Based On Multi-point Information

China is a country with abundant water energy resources.The construction of many water conservancy projects also brings the problem of reservoir bank landslide.The landslide studied in this paper is a typical reservoir bank landslide,so this paper studies the accumulation body of Donglingxin landslide on the right bank of the upstream of Sanbanxi hydropower station by referring to the research ideas of "step type" landslide in the Three Gorges reservoir area.Firstly,the monitoring data of several measuring points on the typical profile of the landslide were collected,and a method of gross error detection and de-noising smoothing was proposed to preprocess the data.Then,based on the two inductive factors of rainfall and reservoir water level fluctuation and the introduction of time series decomposition method,a dynamic neural network prediction model based on NAR/NARX was constructed to predict the displacement.Finally,landslide early warning was realized based on the predicted results.The main research results of this paper are as follows:(1)Based on the analysis of the monitoring data of the accumulation body of Donglingxin landslide,a method for preprocessing the monitoring data is put forward.The method includes three parts:data equal-spacing,de-noising and smoothing,and time series decomposition.First,the discontinuous or even missing monitoring data were equally spaced by cubic spline interpolation method.Then,singular spectrum analysis is used to smooth and denoise the sequence after peer interval processing.Finally,CEEMDAN theory is applied to decompose the time series of the above-mentioned processed monitoring data to extract trend items and period items.Results show that this method is successful to transform the original monitoring data for time interval sequence,such as singular spectrum analysis method and make the data more smooth to eliminate occasional random noise,the final CEEMDAN decomposition method is to divide the original sequence is decomposed into low frequency to high school in the cycle and the gradual trend,can establish foundation for subsequent landslides prediction research.(2)Considering that there may be gross errors in the monitoring data,a new gross error detection method based on singular spectrum analysis(SSA)and density clustering algorithm(DBSCAN)is proposed.Combining the advantages of SSA in extracting signals and DBSCAN in distinguishing gross errors and outliers,this method is verified by introducing landslide monitoring sequence examples.The method proposed in this paper is compared with the median absolute deviation method(MAD)and Grubbs criterion method(Grubbs criterion).The results show that the SSA-DBSCAN gross error detection method proposed in this paper has excellent performance and low misjudgment rate compared with the above methods,which can lay a foundation for further research.(3)According to the characteristics that Donglingxin landslide belongs to reservoir bank landslide,monitoring data such as rainfall in reservoir area,reservoir water level and groundwater level near borehole are put forward as the input of the intelligent algorithm prediction model in this paper.Then,according to the internal relationship between the influencing factors of the above input and the historical deformation law of landslide,Finally,a landslide displacement prediction model based on intelligent algorithm is established which can effectively reflect the real evolution law of landslides.Firstly,Elman neural network,ELM extreme learning machine and NAR/NARX neural network are selected to explore the wave term displacement prediction method based on SSA-CEEMDAN.Then,polynomial fitting method is used to fit and predict the trend term displacement,and the total displacement prediction value is obtained by superposition of the two models.Finally,the accuracy of the model was evaluated,and NAR/NARX was selected as the wave term displacement prediction model.Thus,a new combined prediction model of landslide displacement was successfully established in this paper and applied to several measuring points.The results show that the SSA-CEEMDan-NAR/NARX model proposed in this paper has higher prediction accuracy than Elman or ELM model,which can provide an important reference for the displacement prediction of reservoir bank landslide.(4)On how to establish suitable for reservoir bank landslide early warning method of this problem,an improved displacement rate than landslide early warning method,this method not only inherited the displacement rate than have the prominent advantages of unified comparability,and combining with the practical characteristics of reservoir bank landslide has the nature of the step to the original method of promotion.Finally,combined with the early warning criterion,the landslide early warning with uniform comparability can be realized for the reservoir bank landslide.The study shows that the improved method proposed in this paper is suitable for reservoir bank landslide in hydraulic engineering,and has certain research significance and practical value.(5)Considering the single category of early warning indicators have limitations of this problem,to construct evaluation system of integrated multi-source factors of fuzzy comprehensive evaluation method was introduced,the landslide research field accordingly based on the analytic hierarchy process(AHP)and fuzzy comprehensive evaluation method of multi-source information early warning evaluation method,and the ridge east of letter landslide as an example for the preliminary exploration of the early warning research.The comprehensive early warning evaluation results show that the landslide is relatively safe,but has a tendency of continuous development and evolution.This result has been verified by science and is in line with the actual situation,which indicates that the idea and direction of the early warning exploration is correct and worthy of further study.