A New Time Series Forecasting Model Based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and...

Area of Science:

• Data Science
• Machine Learning
• Time Series Analysis

Background:

• Accurate time series forecasting is crucial for various applications.
• Existing models like LSTM have limitations in capturing complex temporal dependencies.
• Hybrid models offer potential for improved forecasting performance.

Purpose of the Study:

• To propose a novel hybrid time series forecasting model named CEEMDAN-TCN.
• To leverage CEEMDAN for effective time series decomposition and TCN for accurate prediction.
• To evaluate the model's effectiveness in both univariate and multivariate forecasting scenarios.

Main Methods:

• Time series data decomposition using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN).
• Prediction using a Temporal Convolutional Network (TCN).
• Hybrid model construction integrating CEEMDAN and TCN (CEEMDAN-TCN).

Main Results:

• The CEEMDAN-TCN model demonstrated superior performance in univariate time series forecasting.
• The proposed model also showed enhanced accuracy in multivariate time series forecasting tasks.
• Comparative analysis indicated better results than Long Short-Term Memory (LSTM) and other hybrid models.

Conclusions:

• The CEEMDAN-TCN hybrid model is effective for time series forecasting.
• The model offers improved prediction accuracy compared to traditional and existing hybrid approaches.
• This method provides a valuable tool for complex time series analysis and prediction.