Application Research On Rainfall-Runoff Combinatorial Forecasting Theory

Runoff forecasting not only provides the evidence of decision-making, but also impacts greatly on economical operation of hydropower system, navigation, flood controlling and so on. The hydrologic systems, with their elements showing highly non-linear characteristic in space-time, were well known complicated. Apparently, the methods based on linear theories were difficult to make a great progress in hydrologic predictions. It was necessary to explore more new ways. In this paper, the combining forecast theory was introduced to predict runoff. First, in this thesis, a statistical model based on a model of rainfall-runoff and the time series analysis has been built. The basic idea of this model was that the prophase runoff and the nine phase of rainfall before the current and the mean of their rainfall affected the current runoff. According as the data of the model, the linear equations were built, and the parameters were fixed on using the least-square method. The results of this model were received. Then, in this thesis, a neural network model based on the nonlinear time series model has been built. The basic idea of this model was that the neural network model based on auto regress model was built. The three phases of runoff before the current phase affected the current runoff, and the neural network of three layers was built. The structure of the model is (3,7,1). The results of this model were received.At last, in this thesis, the combinatorial forecasting theory has been applied to the runoff forecasting. In turn, the nonlinear combinatorial forecasting model based on neural network and the nonlinear combinatorial forecasting model based on wavelet network were built. These basic ideas were that the two results from the former models were the inputs of the two models, then nonlinear combinatorial forecasting was put up. In view of the good of the wavelet network, wavelet network has been applied to the nonlinear combinatorial forecasting in runoff forecasting .For the sake of showing the superiority of the wavelet network and the combinatorial forecasting method, a nonlinear combinatorial forecasting model based on the neural network has been built. Compare to them, the result was that the nonlinear combinatorial forecasting method applying to the runoff forecasting based on the wavelet network can improve the forecast precision. The keystone of this paper was the fourth chapter and the fifth chapter, here the combinatorial forecasting theory was narrated and applied and the wavelet network was built.