A study on combined nonlinear dynamic prediction model

By using a combined nonlinear dynamic system model, a simulation forecast about the sunspot number inflexion for 22nd and 23rd cycle is performed. It is based on a hybrid methodology that combines both Logistic chaotic attraction points and artificial neural network (ANN) models. At first, to calculate the Logistic chaotic attractor according to falling and total solar cycle length for 1610–1986. Aiming at the shortcoming that chaotic time series can not fit the actual fluctuation of small sample discrete data very well, especially for long-term forecast errors, BP neural network is used to predict the fitting errors above, and to correct the final results based on the neural network prediction. The results indicate that prediction accuracy has been improved significantly.