| Landslides,belongs to a high incidence and serious harm of geological disasters,landslide hazards causing huge economic losses and heavy casualties,and affects the stability of the society.Landslide disaster in our country has a wide range,such as Sichuan,Guizhou and other places of geological structure is complicated,is the high incidence of landslide disaster areas.In recent years,with the mass of people activity,geological hazards,such as collapse of landslide debris flow occurs frequently,disaster prevention is particularly important.Therefore,provide a more accurate method for prevention and control of landslide disaster is already a very urgent task.When disasters occur,the priority is to make correct and quick emergency decision-making,for disaster management work,how can the geological disaster occurrence anddevelopment of a rapid and accurate assessment of the work,is a problem to be solved.This article simply introduces the research significance of landslide,landslide research progress and present situation at home and abroad,and a cloud platform of knowledge and the basic theory of evaluation model.Random forests for evaluating model,based on the graphs parallel processing framework for parallel programming model,to authenticate the improved model.Experimental data is Shanxi Province since 2000,geotechnical engineering,construction,earthquake peak acceleration,slope,precipitation,the scale is 1:500000,and has set up a Hadoop experiment platform,using the MapReduce parallel programming framework,through the parallel computing framework for parallel model is designed,and verify the validity of the model.Get the following conclusion :1.On a single node of the improved model accuracy validation.Parallelize the improved random forest model accuracy compared with the traditional serial random forest model of high precision,has certain feasibility and practicability.2.On Hadoop platform,different number of the machine,compare algorithm's execution time.Elected to take an equal amount of landslide sample data,increase number of the platform machine,the algorithm execution time reduced.3.Considering the total number of different samples,get the experimental results changing the number of the machine:(1)Data1 for sample data on a smaller scale,with the increase of number of machines,algorithm on the running time difference is not big.This is because the Hadoop platform parallel computing,the process of communication and data exchange between multiple devices is a big loss for time efficiency,thus the algorithm efficiency decline.(2)When large sample data,the single condition compared with one machine to participate in the operation found that the process curve slope is the largest,that is to say after the parallelization,the running time of random forest model is greatly reduced,the model efficiency improved obviously.(3)By comparing the number of the machine is 1or 2 or 3,found with the increase of number of machines,model running time is gradually decline,but the curve slope is also gradually decreased,and that the more the number of machines,algorithm efficiency is higher,but at the same time,time of the data communication between the equipments is also on the rise,this is why curve slope gradually become smaller.(4)When the number of machine is 2 and 3,algorithm running time of Data2 or Data3 or Data4 is less than Data1 sample data set.This phenomenon shows that parallel random forest model is more suitable for large-scale data,and the optimization effect is more evident.This article basically achieved the purpose of the paper,that is,through the evaluation model of parallel processing,the efficiency and accuracy of the assessment are improved,the realization of the aim of rapid assessment,providing a basis for improving the efficiency of disaster emergency decision-making for the future. |
PDF Full Text Request