A Time-series Prediction Algorithm Based on a Hybrid Model

Background: In reality, time series is composed of several basic components, which have linear, nonlinear and non-stationary characteristics at the same time. Directly using a single model will show some limitations and the prediction accuracy is difficult to improve. Methods: We propose a mixed forecasting model based on time series decomposition, namely STL-EEMD-LSTM model. First, we use STL filtering algorithm to decompose the time series to obtain the trend component, seasonal component and the remainder component of the time series; then we use EEMD to decompose the seasonal component and the remainder component to obtain multiple sub-sequences. After this, we reconstruct the new seasonal component and the remainder component according to the fluctuation frequency of the sub-sequence. Finally, we use LSTM to build a prediction model for each component obtained by decomposition. Results: We applied the proposed model to simulation data and the time series of satellite calibration parameters and found that the hybrid prediction model proposed in this paper has high prediction accuracy. Conclusion: Therefore, we believe that our proposed model is more suitable for the prediction of time series with complex components.