Research On Time Series Prediction Algorithm Based On Hybrid Model

Time series prediction is widely used in many fields such as economy,meteorology,industry,etc.Scientific and accurate prediction results can bring important guiding effect to actual production and life.Due to the complexity of time series data in real life,the traditional single prediction model can’t achieve satisfactory prediction effect.Ensemble empirical mode decomposition algorithm can adaptively decompose nonlinear nonstationary signals into multiple simple subsignals,thus reducing the difficulty of modeling and improving the prediction accuracy of the model.At present,the existing methods to predict time series based on ensemble empirical mode decomposition have some problems,such as the end effect,the high frequency component and the residual component,and the prediction effect is poor.In this paper,the following three aspects are studied:Firstly,endpoint effects are not considered in the existing time series prediction studies based on ensemble empirical mode decomposition.In this paper,an endpoint effect suppression method based on grey prediction and polynomial fitting is proposed.In this method,the time points in polynomial fitting function are determined by grey prediction,so that the extreme points are more accurate and the end effects are effectively suppressed.At present,the mainstream end-effect evaluation index needs to compare the decomposed component with each component,but there is no definite expression in the real time series data component,so this evaluation index cannot be used.By analyzing the energy difference of each component,this paper proposes an evaluation index of the end effect of the adaptive energy difference.The effectiveness of the evaluation index is verified by experiments,and it is proved that the method based on grey prediction and polynomial fitting is better than other methods in suppressing the end effect.Secondly,due to the influence of various factors,the real time series data contains a lot of noise.After the ensemble empirical mode decomposition of time series data,noise mainly exists in the connotation mode component with higher frequency.When looking for noise components,due to the disadvantage of poor stability of the original judgment method based on energy criteria,this paper establishes a criterion based on energy and correlation coefficient to judge noise components by analyzing the difference between the energy and autocorrelation coefficient of noise and signal,and proves that the method has good performance through experiments.In this paper,the wavelet threshold denoising method is used to denoise the determined noise components.However,the common threshold functions have their own shortcomings,so this paper proposes a new threshold function,and proves through experiments that the new threshold function has better denoising effect compared with other threshold functions.Finally,after the ensemble empirical mode decomposition of the time series data,a finite number of conformal modal components and a residual component are obtained.The conformal modal component will be predicted using the ARIMA model.The residual component represents the overall trend of the time series and has monotonicity.In this paper,the residual component is based on weighted Markov chain principle,and the improved ward system clustering method is applied to cluster the residual component.The optimized membership degree is introduced to predict the state vector of the reference sample,and the clustering fuzzy weighted Markov chain model is established,and the prediction results are obtained with high precision.