Application Of Neural Network In Runoff Simulation

In recent years,the frequent occurrence of flood disasters has seriously threatened the safety people's lives and property,and the accurate prediction of runoff is an important basis for effectively reducing such disasters.With the development of data science,the prediction precision of runoff has been improved effectively in the gauged.However,the prediction of runoff in ungauged areas is still the focus and difficulty of research.At present,the transplantation of hydrological data between similar basins is a common method to solve runoff prediction in ungauged watershed.When judging sim ilar watersheds,this paper uses hydrological data in watersheds to judge similarity between Watersheds Based on the ideas and methods of statistics and data science.Then,improved the geomorphology-based artificial neural networks model by using wavelet function,and validates the model by using hydrological data in Luanchuan and Li Qingdian watersheds.Due to the ubiquity of uncertainty,there is a serious phenomenon of missing values in hydrological data.Therefore,we need to repair the missing value in the runoff of Luanchuan and Li Qingdian basins before building the model.Combined with the similarity evaluation results and the verification results of the model,we will transplant the runoff of the Luanchuan basin to the Tantou Basin(ungauged watershed),and apply the improved model to predict runoff of the Tantou watershed.At the same time,we need to evaluate and analysis the prediction effect of the runoff in ungauged areas.The main research contents and results of this article are as follows:(1)Using ArcGIS software to extract geomorphic parameters(grade of river network,number of river network,average length of each stream,average area of order sub-basins,river rate,area ratio and bifurcation rate)from the river network data,digital elevation model(DEM)data and land use data of Henan Province.And it provides a data base for the prediction of runoff in watershed.(2)Judging the similarity of Luanchuan,Li Qingdian and Tantou basins.This paper bases on the method of coefficient of variation in mathematical statistics,and calculates the similarity of the basins by using watershed's the average length of the river network,the average number of river networks,the types of land use and the average annual rainfall for many years in this paper.The results show that two of the three basins can transplant hydrological data,because the time point of the Li Qingdian watershed does not correspond to the other two watersheds.Therefore,we can transplant runoff between Luanchuan basin and Tantou basin.(3)Interpolation of missing runoff values in Luanchuan and Li Qingdian watersheds.The judgment of the missing mechanism of runoff in data set is the premise of choosing datainterpolation methods.However,the complexity of watershed geographic conditions and the influence of other human factors make it impossible to judge the missing mechanism of runoff in the basin.Therefore,we select k nearest neighbor imputation method and the random regression interpolation method to interpolate the missing runoff data under the missing completely at random and missing at random respectively.(4)Prediction of runoff in ungauged areas.This paper uses the wavelet basis function to improve the geomorphology-based artificial neural networks model,and validates the improved model by using the complete of interpolation rainfall-runoff data in Luanchuan and Li Qingdian watersheds.The results show that the Nash efficiency coefficient,peak error and mean error related to time to peak of the improved model are better than those of the original geomorphological artificial neural network model.Then,based on the results of similar basins and model validation,we transplant directly the runoff of Luanchuan watershed to Tantou watershed.And predict the runoff of Tantou basin by using its own path probabilities and improved model.At the same time,analysis and evaluate the forecast results.The results show that the method in this paper can effectively predict runoff in ungauged areas.