The CEEMD-LSTM-ARIMA Model and Its Application in Time Series Prediction

Abstract

Research shows that there are some data sets with strong timing in reality. After certain processing, they can predict the changes of transactions in the future to some extent and play a positive role in guiding production. However, they often have the characteristics of overall trend variability and seasonal fluctuation, and are also affected by many random factors. Therefore, the data show strong non stationarity. In this paper, a CEEMD-LSTM-ARIMA combined model is proposed. This method reconstructs variables by decomposing time series data, and adopts appropriate methods to deal with variables with different characteristics. In this paper, CEEMD-LSTM-ARIMA combined model is used to predict cigarette data. The experimental results show that the proposed combined model has faster convergence speed and higher accuracy than LSTM and ARIMA, and has more advantages in time series data prediction.