| Aiming at the circulation stage of debris flow,this paper proposes a process based on the two-dimensional simulation results of existing debris flow simulation software,combined with GIS and An improved GAN model--Classification GAN.Based on the GIS,this process can further simulate the range and time relationship of the circulation stage of debris flow through a three-dimensional vector point model.Traditional debris flow simulation software displays results by mapping or rendering terrain grids,debris flow displayed in this way is usually a 2D result without 3D geographic information attributes and entity models,so it has no application in the 3D analysis of GIS and cannot be combined with other 3D vector models in GIS.In this paper,after simulating the debris flow through 3D vector points,the simulation results have 3D geographic information attributes and entity model attributes,which is of great significance to the 3D disaster prevention analysis of the GIS platform.The first half of the article mainly introduces how to obtain the data set for deep learning in debris flow simulation.First,obtain accurate topographic files of the study area through GIS interpolation.Secondly,obtain the front point and mud depth data of debris flow every 5seconds through simulation software.Then,using the water pattern analysis,the debris flow flow line representing the longitudinal trajectory of the debris flow is obtained.Further,through the proximity analysis,the closest position point set(the trajectory point of the flow line)to the front point of the debris flow every 5 seconds is obtained on the flow line.This is because the deviation between the front point of the debris flow every 5 seconds and the flow line is small,the trajectory point of the flow line can be approximately regarded as having the same time information as the front point of the debris flow every 5 seconds.Then,through the mud depth at the front end point of the debris flow every 5 seconds,terrain elevation and flow line trajectory points,the lateral range of the debris flow at each flow line trajectory point is generally screened,and the discrete distribution points of the lateral range of the debris flow are obtained.The time information of the distribution point in the horizontal range has the same time information as the trajectory point of the flow line.Finally,the coordinates of the distribution points in the horizontal range and the coordinates of the flow meridian track points are exported as the data set of the subsequent network.The second half of the article is about network establishment,parameter tuning,and vector model generation.First,according to the simulation needs,the data set is classified according to different time intervals.Each type of training data represents the training data of debris flow simulation in different time intervals,and the division principle of the training data is determined through experiments.Secondly,build a Classification GAN.Further by increasing the normalization,changing the number of neurons in the network,choosing the optimizer,and adjusting the learning rate,the Classification GAN has good training ability,and by mining the spatial distribution law in the training data,the generation of more samples to meet the sample size requirements of the simulation.Then,the vector point model of the debris flow is generated in GIS,and the elevation value of the vector point model is screened through the range of the maximum mud depth,so that the model is more in line with the elevation condition of the debris flow range.Finally,compared with the two parameters of length and width of debris flow in traditional simulation software,the Relative Differential Error in the width comparison is 11.8%,and the Relative Differential Error in the length comparison is 1.13%.It is proved that this method can accurately simulate the spreading range of debris flow under time conditions through the vector point model in GIS,which provides the basis for the 3D disaster prevention analysis of the GIS platform... |
