47 Application of Chaos-Radial Basis Function Neural Networks Coupling Model in Rainfall Forecasting

A coupling model which is based on the Chaos phase space reconstruct and the radial basis function neural network and its utility for rainfall prediction is developed. In general, the precipitation time series, while the time of precipitation processes from beginning to end, has a non-linear feature, which usually can be expressed through a monthly rainfall. Therefore, a new coupling model based on the Chaos theory is produced in this study. In according with the new model monthly rainfall time delay and embedding dimension was computed by autocorrelation function, saturated related dimension and CAO methods, respectively. Furthermore, phase space reconstruction is achieved after the time delay and embedding dimension are counted. Afterwards, the RBF neural networks based on chaotic time series were proposed. In this study, monthly average precipitation of Jianyang city of Sichuan Province, in southeast China, was carried out from 1951 to 2008 to investigate the Chaos phenomenon and simulation. The calculating results show that, the time delay for this time series is 3, the embedding dimension is 9 and the saturated correlation dimension exists. The RBF neural networks coupling with chaotic simulation results show that the proposed model has effective prediction outcome and high prediction precision for the simulated chaotic time sequence, as well as the final absolute error comes to 1.09e-l 1 and the relative errors are lower than 5% except several minimum data. In addition, this coupling model would be used to estimate an annual precipitation and adjust the pattern of planting and guarantee the harvest in agriculture, etc.