Forecasting Chaotic Time Series Based on Improved Genetic Wnn

The chaotic time series forecast was researched by using wavelet neural networks (WNN) in this paper. An improved training method for WNN was presented. The method combines the genetic algorithm (GA) with gradient descent BP algorithm; the BP method was embedded in the GA operation in order to resolve the GA´s limitation in detail search capability. In the last step of the method the WNN searches the best solution using BP method once again. The experiment on predicting the chaotic time series from Henon map illustrates the performance of the method; the experimental result also shows the method can assure the WNN convergence quickly and have the higher forecasting precision.