The Research And Application In Runoff Forecasting Based On SAGA Optimization Algorithm

Runoff forecasting has a very important application in modern hydrological field,especially in the effective management and rational utilization of water resources is of great significance.In the aspect of forecasting,domestic and foreign scholars have achieved favorable prediction effect by combining or mixing a variety of neural networks and intelligent algorithms.In this paper,the effect of intelligent algorithm optimization neural network model and its combination optimization model in actual prediction are studied by simulating the monthly average runoff and daily average runoff data.This paper proposes WNN,SVR,ELM and SAGA algorithms,and establishes the combination model of WNN,SVR and ELM by SAGA,which is applied to the prediction of monthly average runoff of Zamu River and the daily average runoff of Changmabao hydrological station.Firstly,the phase space reconstruction is applied to deal with the monthly average runoff data,and then using the SVR,WNN,GA-SVR,GA-WNN model respectively to forecast the processed data.Secondly,according to the seasonal characteristics of monthly runoff,the seasonal adjustment is carried out,then the WNN,SVR,GA-WNN and SAGA-WNN models are established respectively for the adjusted trend cycle sequences,and the ELM model was established for the irregular sequence.The empirical results show that the SEA-SAGAWNN-ELM model has high accuracy for monthly runoff prediction.Thirdly,considering the influence of precipitation,temperature and other factors on the runoff,the runoff forecasting model is established,which is based on precipitation and temperature,and is predicted by WNN,GA-WNN and SAGA-WNN respectively.The results of SAGA-WNN have better prediction accuracy compared with the results of the unmodified model.Finally,comparing the results of three different angles,it is necessary to do some analysis and processing of the data in the prediction,the hybrid model is superior to a single model,and the proposed SAGA-WNN model has higher accuracy than other single models.In addition,the combination optimization model based on ELM,WNN and SVR was established,which the weight coefficient was optimized by SAGA,and is used in the daily average runoff prediction of Changmabao hydrological station.The simulation results show that the model is superior to the single model in the prediction accuracy.In conclusion,the wavelet neural network model optimized by intelligent algorithm is superior to the not optimized model.The combined optimization model is better than the single prediction model,and the better the optimization performance,the higher the prediction accuracy.