The application of ANN (Artificial Neural Network) theory combined with other theories to practical problems is focus of recent years' research work.Lack of available theories to instruct settings of BP neural network parameters, this article firstly does some research work on these parameters before performing the simulation of this combined forecasting model, including the optimal compressing range of ANN's input data, the optimal learning rate of different ANN models, and the optimal sample length of periodical input data. These results lay a solid foundation for further research work.Main Component Analysis (MCA) theory filters out the relationship between random vectors of same dimension by calculating covariance, i.e., to get major eigenvalues of multi-dimensional vectors. This article raise a new combined model of MCA and ANN to carry out forecasting. The result of model simulation proves the strong feasibility of this new method. |