| China has always been one of the countries with the most occurrence of geological disasters in the world,and among which landslide disaster is the most serious category.Over the years,geological disasters have caused numerous fatalities and significant property losses,and have caused a bad impact on the economic and social development of our country.The Three Gorges Reservoir area(TGRA)has always been an area prone to geological disasters,and many disastrous landslides have occurred in history.Especially since the water storage of the TRGA in 2003,the periodic changes of the water level have revived a large number of old landslides and induced new ones.For example,under the influence of the experimental water impoundment of the TRGA,the surface deformation of the Outang Landslide has been concentrated since 2008.The Huangtupo landslide in Badong County also appeared dangerous situation.After a standardized and effective risk assessment,Anping Town where the Outang landslide is located and Badong Old County have been relocated to avoid the landslide risk.In this context,the evaluation and analysis of landslide hazard risk from regional to individual scale,the landslide susceptibility mapping,and the displacement prediction have been carried out successively in the TRGA.This paper takes the landslide disasters in Wanzhou District of the TGRA as the main research object.The development law of landslide disaster in Wanzhou District is summarized on the basis of collecting the landslide investigation data.With analyzing the landslide monitoring data,the cumulative displacement-time curve shapes of landslide disasters in Wanzhou District were classified.Based on mathematical statistics,the rainfall thresholds of landslide disaster in Wanzhou District were studied and analyzed.The hydrometeorological thresholds of typical landslide disasters in Wanzhou District were studied based on clustering algorithms and data mining algorithms.With combining two machine learning algorithms including deep learning algorithms,the displacement prediction of typical landslide disasters in Wanzhou District was carried out.Through the above research,the paper has achieved the following results:(1)The development law of landslide disaster in Wanzhou District were summarized and analyzed based on the collected geological disaster investigation data and other literatures.It is found that the temporal distribution law of landslide disaster is highly correlated with precipitation and reservoir fluctuation in Wanzhou District.The spatial distribution law of landslide disasters in Wanzhou District is highly correlated with geological structure,topography,geomorphology,and water systems distribution.The landslide hazard type in Wanzhou District is mainly colluvial landslide,and the sliding surface is mainly the soil/bedrock contact surface.The landslide hazard mainly distributes in the strata of upper Shaximiao Formation of the Middle Jurassic.There are many landslides distribute in near horizontal strata,and the scale of landslide disasters is mainly medium volume and shallow landslides.In addition,rainstorm is the main triggering factor of landslide disaster.(2)Based on the water level data of the TGRA from June 2003 to May 2020,the reservoir water operating time can be divided into: i.June 2003 to September 2006,ii.September 2006 to September 2008,iii.September 2008 to June 2010 and iv.June 2010 to May 2020.The characteristic water levels of the four stages are 130 m,145 m,171 m and 175 m respectively.Taking the month as the statistical unit,it can be divided into:period A,period B,period C,and other periods.The period A represent the periods during the rapid decline of water level,the period B represent the periods in flood season,and the period C represent the periods during the rapid increase of water level.The specific time of these three characteristic periods refers to May,June to August,and September.On the basis of analyzing the monitoring data of groundwater level of typical landslides,it is found that the groundwater level of landslides is controlled by many factors,such as reservoir water level,rainfall,topography,landform,and the properties of sliding bodies.The correlation between the groundwater level change curve and the reservoir water level change curve is better when the monitoring point is closer to the shoreline of the reservoir area.According to the statistics of the surface GPS displacement monitoring data of typical landslides,it is found that the displacement near 28 surface GPS monitoring points in 11 typical landslides is all very slow.Among them,4 landslides are in relative high displacement rate in period A,2 landslides are in relative high displacement rate in period B,and 5 landslides are in relative high displacement rate in period C.Therefore,the period of rapid landslide displacement is mainly concentrated in the period A and C when the reservoir water level changes rapidly.Moreover,in the period A of 2007 and 2015,in the period B of 2007,2014 and 2015,and in the period C of 2011,2012 and 2013,more landslides are in relative high displacement rate.(3)Based on the monitoring data and previous research,the classification of the cumulative displacement--time curves in Wanzhou District can be divided into: the Straight line type represented by the Sharentian landslide;the Curve line type represented by the Fujiayan landslide;the Convergence line type represented by the Jinjinzi landslide;the Fall line type represented by the Rangduchangbei collapse landslide;the step—like type represented by the Huayuanyangjichang landslide;the shock type represented by the Yangjiaba landslide and the compound type represented by the Tangjiao No.2 landslide.(4)The rainfall thresholds of landslide disaster in the reservoir bank section of specific strata and lithology in Wanzhou District under specific reservoir water level fluctuation stage and specific warning level was studied based on the collected monitoring data.Regarding the landslide hazard in Wanzhou distribute in the strata lithology of sand shale interbed library class of reservoir coastal and structure of slope in nearly horizontal bedded,when the forecast is expected as " some temporary buildings and old houses are damaged and can be reinforced by simple remedial methods at any time ",its effective rainfall threshold for 2 days is 45 mm in the rapid A,50 mm in period B and 35 mm in period C.The threshold values of the daily rainfall are: 15 mm in period A;20mm in period B;25mm in period C.(5)Based on the monitoring data of the Sifangbei landslide in Wanzhou District,the deformation evolution association rules system was established,and the meteorological and hydrological threshold criterion of Sifangbei landslide was constructed.In this paper,the clustering algorithms and the Apriori algorithm are applied in the Python language environment to dig out 15 association rules of deformation evolution of the Sifangbei landslide.The results of association rules show that the surface displacement near the monitoring point WZ02-03 is affected by the combined action of the reservoir fluctuation and rainfall,and the reservoir fluctuation is the main inducing factor for the surface displacement and deformation.In addition,based on C5.0 decision tree algorithm,7hydrometeorological threshold criterias of the Sifangbei landslide deformation states are established in this paper.It is found that the threshold of monthly accumulative rainfall is144.6mm,and the monthly fluctuation threshold of reservoir water level is-2.065 m which means the reservoir water level decreases at a rate of 2.065 m per month when the Sifangbei landslide deformed in medium speed.(6)An ensemble prediction model based on the linear weight theory,the Long short term memory network(LSTM)and the Support vector regression(SVR)algorithms was proposed in this paper.The proposed ensemble model is used to study the displacement prediction of the Sifangbei landslide in Wanzhou District in the TGRA.The results of the proposed ensemble prediction model are compared with the single model such as LSTM model and SVR model,and the results are obtained as good.The model is extended to the displacement prediction of the Shengjibao landslide in Fengjie County of the TGRA to verify the adaptability of the model.It is found that,overall,the LSTM model has better performance than the SVR model,but the results of the LSTM model are not closer to the original value than the results of the SVR model at each time steps of the predicting data set.Meanwhile,the model has a good adaptability in the application of the Shengjibao landslide displacement prediction.The proposed ensemble model combines the advantages of the LSTM algorithm and the SVR algorithm,and its prediction performance is better than that of the LSTM model and the SVR model. |
PDF Full Text Request