A BP multivariate time series neural network-based method for foreign exchange rates modeling and forecasting is presented. Respectively, we used single time series and multivariate time series to train the neural network parameters, connection weight and configuration. By forecasting GBP to USD it shows that the multivariate time series method has reinforced learning properties and mapping capabilities in the circumstances of the deviation of the sample test all below 10. It is useful for modeling and forecasting of uncertain nonlinear systems. |