Forecasting Runoff with Higher-Embedded Dimensions Using DMEP-Based Artificial Neural Networks

The runoff series always presents complex chaos phenomenon with higher embedded dimensions because of the influence of many complicated facts. Generally, it is an effective method to combine phase space restructures theory with artificial neural networks (ANN) model for runoff forecasting. The traditional methods that not consider the variations of the character of full Lyapunov exponential spectrum in restructure space, are proved of high precision to forecast time series with low-embedded dimension. But, they are not so effective to forecast attractors with high-embedded dimension. The paper proposes a new method, which improves the character of full Lyapunov exponential spectrum in a restructure space with high dimensions. In the mean time, an artificial neural networks model based on determinate mutation evolutionary programming (DMEP) learning algorithm is presented for chaotic runoff series forecasting. DMEP introduces chaos mapping into the mutation operation of EP, which aim to increase its convergence rate and results’ precision. The test result of runoff forecasting series shows that the precision of runoff forecasting is improved by means of the new method when the embedded dimension is high.