Research On Stage-discharge Relationship In Dadu River Basin Based On Neural Network

Global hydrological change is of vital importance for global ecosystem changes.Therefore,hydrological simulation and estimation constitute a significant part of the field of hydrological change research.Particularly,the estimation accuracy of hydrological features such as water level and flow occupies an important place in compilation of hydrological and water resources data,operation of reservoir,prevention of flood,and engineering design and construction.Based on the measured water level and flow data of a certain river basin in Dadu River from 2007 to 2010,this paper aims to explore the internal connection between river water level and flow,establish various estimation models including least square flow estimation model,BP neural network flow estimation model,RBF neural network flow estimation model,Elman neural network flow estimation model,GA-Elman neural network flow estimation model and information entropy based GA-Elman neural network flow estimation model,and develop a flow estimation software.The main work and achievements are:(1)Analyzing the advantages and disadvantages of general data filling methods with missing and abnormal water level and flow data of a certain river basin in Dadu River.Center metric filling method was adopted to process missing and abnormal data to ensure the integrity of data and eliminate the influence of abnormal data on model.(2)Establishing neural network flow estimation models including BP model,RBF model and Elman model,which are capable of solving complex non-linear calculation and describing the intrinsic relationship between water level and flow,to enhance prediction accuracy.Using measured data to verify the model,the results indicate that Elman neural network had the best performance on river flow estimation.The root mean square error was reduced by 5.97% compared with classical analytical models and the absolute error of all points was well controlled within 180.(3)Establishing GA-Elman neural network flow estimation model and information entropy based GA-Elman neural network flow estimation model to further enhance prediction accuracy.Trained with measured data of a certain river basin in Dadu River from 2007 to2009 and tested with measured data of the same field at 2010,the information entropy based GA-Elman neural network flow estimation model had the best performance on prediction accuracy among all the models and its root mean square error was reduced by 18.23%compared with classical analytical models.(4)Developing a ML-based water flow estimation software with assistance of GUI technology of MATLAB to promote the practical application of the studied algorithm.This software combines various neural network water flow estimation models including least squares model,BP model,RBF model,Elman model,GA-Elman model and information entropy based GA-Elman model and contains three modules including login module,data processing module and estimation module,laying the foundation for subsequent research.