Runoff And Long-term Forecasting Model Based On Wavelet Analysis

Runoff forecast is the basic content of water resources management, its accuracy and foreseeable period are directly related to the use of water resources planning, program evaluation, policy-making and operational management's effectiveness and success. Accurate runoff forecast can maximize coordinate conflicts of water using arising from comprehensive utilization of water resources, it can also provide decision-making basis for scheduling and control of the right to water distribution.As the runoff formation process is not only affected by the certainty factors but also by the uncertainty factors, so the changes of runoff are very complicated. Their future is very difficult to describe accurately. Long-term runoff forecast's research is still at the development stage, compared with short-term hydrological forecast, it seriously lags behind actual production needs. Recently there are many medium - and long-term runoff forecast methods, but if runoff time series has large interannual changes, it's difficult to forecast and analysis by using conventional methods, its forecasting accuracy is poor.In this paper, wavelet analysis is combined with Artificial Neural Network and Adaptive Neuro-Fuzzy Inference System's powerful functional approximation. Zhaopingtai reservoir as an example, six coupling models are established for monthly runoff forecast.First, considering that in the nonlinear time series prediction, the model's order is difficult to determine, we use wavelet decomposition and wavelet transform to search for the major components of the runoff, then we use them as the input of the model, next we establish the nonlinear relationship between runoff and the input. In accordance with the established mapping between the different methods used, we establish Wavelet-Artificial Neural Network model and Wavelet-Adaptive Neural Fuzzy Inference System model for the runoff.Second, in order to solve the problem of years of runoff interannual variations is very large and single method is difficult to solve this problem, we apply wavelet decomposition and wavelet reconstruction .First we decomposed runoff series on different information space, then we can receive different frequency signals. Second we use different methods to forecast each signals, then by wavelet reconstruction we can receive the forecasting result of the monthly runoff. Wavelet analysis simplified the predicting model's structure, and is conducive to improve model's accuracy. According different methods on signal decomposition model, we establish Wavelet Decomposition-ANN model and Wavelet Decomposition-Adaptive Neuron Fuzzy Inference System model on monthly runoff forecast.Lastly,for the runoff series's random variation characteristics ,we use wavelet denoising technology to denoise the original runoff sequence, to get rid of some of the noises which is difficult to forecast; On this basis, we establish Wavelet Decomposition-ANN and Wavelet Decomposition-Adaptive Neuron Fuzzy Inference System model on monthly runoff forecast.Based on comparative analysis of every models' modeling and forecasting results, we find that Wavelet Decomposition - Adaptive Neuron Fuzzy Inference System model can simulate and predict monthly runoff series perfectly, it can be the recommended model of Zhaopingtai reservoir's runoff forecasting.