A Hybrid Intelligent Method For Slope Displacement Prediction

As the world second largest geological disaster just after the earthquake,landslides cause enormous human casualties and economic losses to our country every year,and it is unfavorable for the modernization of our state and society.The accurate prediction of slope deformation is the core component of an early warning system,which is one of the most effective measures to deal with landslides for us.Due to its nonlinear advantages,slope displacement prediction models based on intelligent algorithm are widely used in many practical projects at present.However,there are still some defects for this method need to be improved,such as high redundancy of database,algorithm defects,results lack of correction,etc.Focusing on the above deficiencies existing before,during,and after slope displacement prediction respectively,this paper proposed a hybrid intelligent method for slope displacement prediction with better preference based on the slope monitoring system built by our research group in Tuyang village of Shenzhen City,and established the corresponding multiparameter prediction and early warning mechanism,which is of great help for the disaster prevention and mitigation work of the similar soil slopes.The main research work and results are described as follows.The regularity of slope monitoring data and its deformation mechanism were analyzed and summarized.Based on the introduction of Tuyang slope monitoring system,the records during 26 th July to 26 th December of 2016 were analyzed for their regularity from deep displacement,surface displacement with its velocity,groundwater table with rainfall,and humidity with rainfall,on the basis of which the influencing factors of slope displacement were defined further.The related conclusions above were taken as the basis for the establishment of slope displacement prediction database.Owing to the fact that a lot of redundant information exists in the database and the data itself has high dimensional complex features,a combined method by using various data mining techniques was proposed and employed to obtain effective information and establish displacement prediction database.Attributes reduction of initial database was implemented based on Rough Set theory(RS),and its redundancy improvement effect was verified.Data dimensionality reduction was realized through information extraction by using Kernel principal component analysis(KPCA)based on reduced attributes,and a comprehensive index database was established on the basis of the result above.In order to obtain effective information,association rules mining is implemented on the the reduced database by using Aprori algorithm,and 14 prediction criteria for deformation stage with higher confidence obtained afterwards were consistent with the analysis results of slope deformation mechanism in previous part.Considering the main problems before,during and after slope displacement prediction,a hybrid intelligent prediction method was proposed based on the piecewise optimization thought.The comprehensive index database after RSKPCA process was ultilized to deal with database redundancy before prediction;Least squares support vector machine(LSSVM)with its parameters optimized by quantum particle swarm optimization(QPSO)was employed to realize reasonable selection and optimization of model building methods during prediction;Markov Chain(MC)model was applied to correct the forecasting values for the lack of amending after prediction.Based on the work above,the prediction effectiveness and precision of displacement prediction model at three different optimization stages was evaluated by combining other methods and measured,and its process and sphere of application were further summarized.Given the defects of existing slope early warning methods,a multiparameter prediction and early warning mechanism was proposed for Tuyang slope.Based on previous analysis results of meausred,the early warning threshold of rainfall,groundwater level and displacement rate were established at the same time,which overcomed false alarm problems in the early warning method with single index;High reliability prediction of dangerous slope instability time was achieved by introducing the inverse-velocity method,which made the available time of landslide avoidance clear;Combined the prediction criteria of deformation stage and the hybrid intelligent displacement prediction model,the multiparameter prediction and early warning mechanism was built.