Research On Daily Runoff Prediction Model Based On Wavelet Neural-network

In the process of development and utilization of water resources, real-time and accurate daily runoff forecast plays an important role in the fields such as disaster reduction, flood control and other aspects. In this paper, the wavelet neural network model which is consist of traditional wavelet analysis and artificial neural network is used for daily runoff forecast.First of all, during an experiment of wavelet transform we confirmed that an action of noise declining to hydrological time series in advance enhance the accuracy of forecast efficiently. Then we present a better sample data selecting method by analysis and comparison. The model takes into account not only seasonal factors, but also the relevance of the runoff time series. Through the wavelet transform the load sequence is decomposed into subsequences on different scales, then using appropriate artificial neural networks these sub-sequences are forecasted by appropriate artificial neural network respectively, finally after the synthesis of the forecasted results from the sub-sequences the final forecasting result of the load sequence is obtained.In order to verify the validity of the wavelet neural network model, we assessed the effectiveness in the follow two aspects: on one hand we compared the wavelet neural network model with other traditional ones, on the other hand the model is applied to different lead times. We confirm that it can enhance the accuracy and certified rate of forecast efficiently in an appropriate leading time. By studying the climate adaptability of the model, we inferred that the wavelet neural network model is a fitting daily runoff prediction model to dry area.