Grey - Application Of Artificial Neural Network Combined Forecasting Method Of Runoff Forecast

Water resource is one of the most basic elements that support human beings’life, the growth of economic and the sustainable development, but among these factors, the changes of the runoff dominate the changes of the whole system. These factors, just like the climate, the natural geographical, social development and human survival needs, which all lead to the comprehensive impact on the runoff, the runoff has complex characteristic and rules of the changes, mainly shows in the randomness, mutability, nonlinear, and so on. Therefore, the forecast of runoff, especially the medium and long term forecast, which is always a very difficult work and is still the hot spot and the difficulty of domestic and foreign research.Artificial Neural Network is a kind of commonly used artificial intelligence, which is mainly applied to the prediction of precipitation, runoff forecast, flood forecasting, prediction of water quality and quantity, the optimal allocation of water resources, the reservoir optimization scheduling, groundwater management, water and soil resources utilization planning of groundwater and other aspects in the field of water resources system. Because the complexity and the uncertainty of hydrological phenomenon, hydrological data information is insufficient too, the hydrological system has the characteristics of gray system. Therefore, gray system theory can be widely applied in the evaluation of water resources, water resources management and prediction.This paper combines the Artificial Neural Network and Grey System theory to study the medium-and long-term runoff forecast, the main research contents and results are as follows:(1) The creation of the prediction method. According to the rule of variation of runoff series, Gray-Artificial Neural Network prediction method is put forward, which uses weakening buffer operator in the grey system theory to weak local fluctuations in the composition in time series, after that new sequence can be gotten and used as training samples of Artificial Neural Network model, then use Matlab neural network toolbox to forecast. Its main purpose is to make the sequence became gentle by weakening buffer operator structure and weakening the extreme changes of runoff, thus the hidden rules can be more easily found. Therefore in the neural network training, the accuracy of forecasting can be improved by better fitting the complex mapping relation between the corresponding input and output.(2) The research for annual runoff forecasting model. Grey-Artificial Neural Network is applied to the instance analysis, comparing forecast results finds the forecast of annual runoff forecast accuracy is improved in the method, but because the use of weakening operator reduces runoff’s extreme changes, the error about annual runoff prediction is bigger in the year, which the runoff is too large or too small, so the method is more suitable for flat water year.(3) The research for monthly runoff prediction model. Such Grey-Artificial Neural Network is applied to the instance analysis, comparing forecast results finds the forecast of monthly runoff forecast accuracy is obviously improved in the method, and variation trend of each monthly runoff of every year from1992to1996can be better reflected, especially improves the result of the flood forecasting, which confirms the feasibility of this method has good application value.