Time series forecasting based on the empirical mode decomposition multi-dimensional Taylor network model

A method of the empirical mode decomposition multidimensional Taylor network for establishing the dynamics model and its parameters identification is proposed for time series forecasting. By the empirical mode decomposition algorithm, the time series are decomposed into one residue signal and several intrinsic mode function signals. The multi-dimensional Taylor network models are established for sub-time series with different intrinsic mode functions, respectively. The model parameters are trained by conjugate gradient method, and then the models are used for forecasting. Predictions of all the models are superimposed to obtain the predicted value of the original time series. Experimental results show this method for financial time series forecasting has high prediction accuracy.