Time Series Analysis And Prediction Based On EMD Technology

With the development of real-time observation technology,a large amount of information data has been accumulated in various industries.It is not only the core problem of big data analysis and prediction science,but also the need of practical work,so it has important scientific value and application value to mine useful in-formation that can reveal the general law of the development of things from the mass data measured in the process of development.This paper firstly introduces the HHT method to analyze the time-frequency variation characteristics of time series,reveals the temporal characteristics of event development,and then predicts the fu-ture development trend of time series by combining EMD technology and seasonal ARIMA modelHilbert huang transform(HHT)is a time-frequency analysis method consists of two parts,namely the empirical mode decomposition(EMD)and Hilbert spec-trum analysis(HSA),applied to Husdom Bay company's annual mink sales data,IBM company's common stock trading day's closing price data,the Yellow River in lanzhou station,sediment concentration data,through the study of the HHT time-frequency analysis of these data,get their experience mode decomposition figure,Hilbert spectrum,marginal spectrum,energy spectrum of the Hilbert spectrum and instantaneous frequency and instantaneous energy.The experimental results show that the HHT method accurately describes the nonlinear time-varying characteris-tics of annual mink sales data of Husdom Bay company,the closing price data of IBM common stock trading day and the daily sediment concentration data of lanzhou s-tation of the Yellow River,and is an effective tool for time-frequency analysis of data.A time series prediction model(EMD-ARIMA)based on EMD method is pro-posed to solve the nonlinear,non-stationary and multi-scale complex time series problems.Firstly,by means of EMD method,the time series is decomposed into sev-eral internal modulus functions and trend terms of different time scales.Secondly,the seasonal ARIMA model is used to predict each internal module function,and the trend term is predicted by the trend moving average method.Finally,the predicted results of each sub-time series are compounded to obtain the predicted results of the original time series.The prediction performance of EMD-ARIMA model was compared with arti-ficial neural network(ANN)model and ARIMA model,and three groups of time series of different sizes were selected for short,medium and long term prediction.The average absolute percentage error(MAPE),mean absolute error(MAD)and root mean square error(SDE)of the three models were compared.Numerical re-sults show that the EMD-ARIMA model not only reveals the inherent multi-scale composite characteristics and seasonal variation rules of real time series,but also overcomes the shortcoming that the ARMA model is not suitable for nonlinear and non-stationary time series prediction.Compared with the classical ARIMA model and ANN model,EMD-ARIMA model significantly improves the prediction accura-cy,and is a reliable nonlinear and non-stationary time series prediction method.