Research On Time Series Prediction Based On Empirical Mode Decomposition

Time series prediction has been one of the hot research fields in recent years and has important practical application value.Scientific and accurate forecasting results can provide a key guiding role for social activities.The empirical mode decomposition algorithm is a landmark signal analysis method,which can decompose the signal into a series of oscillating functions adaptively without setting the base function and decomposition scale in advance.Since its introduction,the empirical mode decomposition algorithm has received extensive attention and in-depth research from many scholars,and has now become an important method for time series analysis and prediction research.However,the existing methods for time series prediction based on empirical mode decomposition generally have the problem that the sub-signals have mode mixing,the scale of prediction network is large,as well as the prediction effect of high frequency sub-signal is poor.This paper puts forward two kinds of effective improvement strategies to reduce the scale of prediction network and improve the accuracy of prediction model.To remove mode mixing problem and reduce the scale of prediction network the conventional hybrid methods based on empirical mode decomposition,this paper proposes an improved hybrid forecast model based on ensemble empirical mode decomposition and echo state network.Firstly,ensemble empirical mode decomposition is employed to decompose the original signal adaptively,which can effectively overcome the mode mixing problem existing in empirical mode decomposition.And permutation entropy is used to analyze the complexity of the obtained a series of sub-components.Then combine these neighbouring sub-signals whose complexities are similar for subsequent prediction,to reduce the number of predictors.Finally,these estimates are assembled as an ultimate prediction result.The simulation results based on Lorenz and the annual runoff of Yellow River time series show that the improved model can effectively reduce the scale of prediction network under the premise of guaranteeing the prediction accuracy.For the sake of solving the problem that the prediction model based on single decomposition technique cannot completely deal with the non-stationary and irregularity of original time series,this paper puts forward a hybrid model based on two-layer decomposition technique and optimal neural network.Firstly,complete ensemble empirical mode decomposition with adaptive noise is used for the first layer decomposition of the original signals,then a series of sub-signals from high frequencies to low frequencies are obtained.The large complexity of high-frequency sub-signals results in difficulty for signal tracking and poor effect of modeling and forecasting.In this paper,the variational modal decomposition algorithm is proposed to decompose the high-frequency sub-signals to grasp the characteristics of the original signal more comprehensively,which is beneficial to build more accurate prediction models and improve the overall prediction accuracy.Based on the prediction results of the daily maximum temperatures in Melbourne,it is obviously that the hybrid model based on two-layer decomposition technique presented in this paper has better prediction performance than the hybrid model based on single-layer decomposition technique.